The evolution of a new cell type was associated with competition for a signaling ligandEttensohn, Charles A.;Adomako-Ankomah, Ashrifia
doi: 10.1371/journal.pbio.3000460pmid: 31532765
Introduction Although all adult echinoderms possess calcite-based endoskeletal elements, only euechinoid sea urchins form micromeres and primary mesenchyme cells (PMCs). This newly evolved cell type is rigidly specified early in development through the activity of localized maternal factors and deploys a well-characterized gene regulatory network (GRN) that controls the morphogenetic behaviors and biomineral-forming properties of PMCs [1–3]. Much evidence suggests that the evolutionary appearance of this mesodermal cell lineage was associated with the co-option of an ancestral, adult program of skeletogenesis into the early embryo [3–5]. The regulation of skeletogenesis by vascular endothelial growth factor (VEGF) signaling was likely a component of this ancestral program [4,6–8]. In euechinoid sea urchins, although VEGF signaling is not involved in the maternally entrained, cell-autonomous specification of the micromere-PMC lineage, it plays an important role later in embryogenesis when PMC migration and skeletogenesis come under the regulatory influence of VEGF3 produced by ectoderm cells [9–12]. Morgulis and coworkers [13] recently used whole-embryo RNA sequencing (RNA-seq) to show that VEGF signaling regulates hundreds of genes, including many biomineralization genes expressed selectively by PMCs, and they propose that VEGF regulates an ancient program of tubulogenesis shared by echinoderms and vertebrates. In contrast to PMCs, which are rigidly committed to a single (skeletogenic) fate, blastocoelar cells (BCs) are multipotent. These cells ordinarily give rise to a heterogeneous population of migratory, immunocyte-like cells [14]. They also possess skeletogenic potential, but during normal development, a signal from PMCs suppresses this potential and directs BCs to express an immunocyte fate (Fig 1). If PMCs are ablated at or about the time of ingression, BCs undergo a striking change in phenotype; they adopt PMC-specific morphogenetic behaviors and secrete a correctly patterned skeleton [15–18]. This change in cell phenotype is associated with the molecular reprogramming of BCs, which ectopically deploy the skeletogenic GRN while extinguishing the expression of two regulatory genes, stem cell leukemia (scl) and GATA-binding factor c (gatac), that are associated with the immunocyte fate [14,19]. Although this cellular interaction was first described more than 50 years ago [20], the molecular nature of the PMC-derived signal has not been determined. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. PMC-derived signals suppress the skeletogenic potential of BCs. Upper panel: Each pBC is multipotent and can adopt either a skeletogenic (PMC-like) or BC (immunocyte) fate. Ordinarily, a signal from PMCs (red bar) suppresses the skeletogenic potential of pBCs and causes them to adopt an immunocyte fate. Lower panels: Ablation of PMCs (red cells) at the late (mesenchyme) blastula stage leads to transfating of BCs (green cells). In these PMC(−) embryos, BCs ingress late in gastrulation but migrate to PMC-specific target sites on the blastocoel wall (indicated by white arrows), fuse to form a syncytium, and synthesize a correctly patterned skeleton, beginning with the formation of two triradiate skeletal primordia (shown in yellow). BC, blastocoelar cell; pBC, presumptive BC; PMC, primary mesenchyme cell. https://doi.org/10.1371/journal.pbio.3000460.g001 Results VEGFR signaling is required for the activation and maintenance of the skeletogenic network in BCs There are two VEGF receptor (VEGFR) genes in sea urchins: vegfr-10-Ig is ordinarily expressed at high levels (peak expression > 3,000 transcripts/embryo), whereas vegfr-7-Ig is expressed at much lower levels (<150 transcripts/embryo) [21,22]. During normal development, vegfr-10-Ig is expressed selectively by presumptive PMCs beginning at the early blastula stage [9]. In PMC(−) embryos, however, vegfr-10-Ig is expressed robustly in BCs [19]. The spatial expression pattern of vegfr-7-Ig is not known in detail but, unlike vegfr-10-Ig, this gene is not differentially expressed by PMCs [23,24]. Axitinib is a highly selective inhibitor of VEGF receptors at nanomolar concentrations [25,26]. The effects of axitinib on sea urchin development mimic those of VEGF3 and VEGFR-10-Ig knockdowns, confirming that axitinib is a specific inhibitor of VEGF/VEGFR signaling in this system [9,11]. We found that PMC(−) Lytechinus variegatus embryos treated continuously with axitinib (75 nM) beginning at the time of PMC removal failed to form a skeleton, even after prolonged culture (48 hours postfertilization [hpf]) (Fig 2A and 2A’). Axitinib-treated embryos swam vigorously and gastrulated at the same time as untreated, PMC(−) embryos but never extended arms. Polarization microscopy showed that these embryos completely lacked birefringent skeletal elements (Fig 2B and 2B’). This phenotype was reproducible and highly penetrant; >95% of axitinib-treated, PMC(−) embryos showed this effect across many batches. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Axitinib blocks BC transfating. PMCs were removed from mesenchyme blastula–stage embryos, and the resultant PMC(−) embryos were separated into two cohorts. One cohort was left in plain seawater, whereas the other was transferred to 5 nM axitinib immediately after PMC removal. (A-G) Control PMC(−) embryos. (A’-G’) Axitinib-treated PMC(−) embryos. Axitinib treatment blocked the formation of birefringent skeletal elements (A-B’) and the expression of early skeletogenic regulatory genes by BCs as shown by WMISH analysis of Lv-alx1 (C,C’) and Lv-tbr (D,D’) expression. The expression of skeletogenic effector genes downstream of alx1 was also blocked, as indicated by WMISH analysis of Lv-p58a (E,E’) and Lv-msp130rel2 (F,F’), and by immunostaining with mAb 6a9 (G,G’). Arrowheads indicate expression of skeletogenic genes by transfating BCs. Panel H shows quantification of 6a9-positive cells in control and axitinib-treated PMC(−) embryos at 12 hpd (two independent trials from separate matings). Statistical significance of the data was assessed by two-sided t tests, and p-values < 0.05 are indicated by asterisks. Raw data can be found in S1 Data. BC, blastocoelar cell; DIC, differential interference contrast microscopy; hpd, hours post–PMC depletion; Lv-alx1, L. variegatus aristaless-like 1; Lv-tbr, L. variegatus t-brain; mAb, monoclonal antibody; msp130rel2, L. variegatus mesenchyme specific protein 130-related 2; POL, polarized light microscopy; PMC, primary mesenchyme cell; WMISH, whole-mount in situ hybridization. https://doi.org/10.1371/journal.pbio.3000460.g002 Early molecular events in BC transfating include the activation of aristaless-like homeobox 1 (alx1) and t-brain (tbr), two regulatory (i.e., transcription factor-encoding) genes that function early in the PMC GRN and are ordinarily expressed only in the large micromere-PMC lineage [17,19]. Alx1 is a critically important regulator of the skeletogenic GRN both in the PMC lineage and in transfating BCs, whereas tbr plays a less prominent role [17,24,27–30]. The activation of these two genes in transfating BCs is detectable by whole-mount in situ hybridization (WMISH) as early as 2 hours post–PMC depletion [19]. We confirmed that L. variegatus alx1 (Lv-alx1) and L. variegatus tbr (Lv-tbr) were expressed robustly in the anterior region of the archenteron in control PMC(−) embryos but found that expression of both genes was greatly reduced in embryos that were placed in axitinib at the time of PMC removal (Fig 2C–2D’) (3 trials, n = 15–30 embryos/trial). This effect was readily apparent by 3 hours post–PMC depletion when assayed by conventional (histochemical) WMISH. These findings demonstrated that the expression of alx1 and tbr in BCs was dependent upon VEGF/VEGFR signaling. Consistent with this finding, the expression of two terminal biomineralization genes, p58a and L. variegatus mesenchyme specific protein 130-related 2 (msp130rel2), that receive positive regulatory inputs from alx1 [24,31,32], was also greatly reduced in axitinib-treated, PMC(−) embryos (Fig 2E–2F’, >90% of embryos, n > 40). We also immunostained embryos with monoclonal antibody (mAb) 6a9, a widely used marker of PMC differentiation that recognizes the biomineralization protein MSP130 [15,33,34], and observed a dramatic reduction in the numbers of 6a9(+) cells in axitinib-treated, PMC(−) embryos (Fig 2G and 2H). Culturing such embryos for an additional 24 hours did not increase the numbers of 6a9(+) cells, indicating that axitinib was not simply delaying BC transfating. We next asked whether VEGFR signaling was required only at the onset of BC transfating (i.e., whether it acted like a trigger) or was required continuously during the transfating process. We removed PMCs from mesenchyme blastula–stage embryos and allowed them to develop for varying periods of time before adding axitinib to the medium (Fig 3A). At 12 hours post–PMC depletion, embryos were fixed and immunostained with mAb 6a9. As the time interval between PMC depletion and the addition of axitinib increased, the numbers of 6a9(+) cells increased progressively (Fig 3B). This suggested that VEGFR signaling was required continuously during the first several hours after PMC depletion. Surprisingly, even when axitinib was added 5–6 hours post–PMC depletion, several hours after the activation of skeletogenic regulatory genes in BCs, we observed a significant reduction in the numbers of 6a9(+) cells. This finding raised the question of whether, once activated, early regulatory genes maintained their expression in the presence of axitinib. We removed PMCs at the mesenchyme blastula stage and allowed the resulting PMC(−) embryos to develop for 4–5 hours, sufficient time for alx1 to be activated ectopically in BCs (Fig 3C). Transfer of PMC(−) embryos into 75 nM axitinib at that time led to a decline in alx1 expression to undetectable levels in most embryos (10/13, or 77%) within 5 hours, whereas expression was maintained (or increased) in sibling, untreated, PMC(−) embryos (Fig 3D and 3E). These observations showed that VEGFR signaling was required not only to activate but also to maintain the expression of alx1 in transfating BCs. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Time course of axitinib sensitivity during BC transfating. (A) Experimental design. Solid horizontal bars indicate the duration of axitinib (5 nM) treatment. Start times are shown as a 1-hour interval because microsurgical removal of PMCs required approximately 1 hour for each cohort of embryos. (B) Counts of 6a9-positive cells. Data shown were collected from several independent trials, and untreated controls from all trials were pooled. For all time periods tested, we observed a significant reduction in 6a9(+) cells in axitinib-treated PMC(−) embryos relative to untreated PMC(−) controls. Raw data can be found in S1 Data. (C-E) Signaling through VEGFR is required for the maintenance of Lv-alx1 expression in BCs during reprogramming. PMC(−) embryos were treated with axitinib for 5 hours beginning at 4–5 hpd. At the start of axitinib treatment, Lv-alx1 was expressed robustly by BCs at the tip of the archenteron (arrow in C, 10/11 embryos). Five hours later (9–10 hpd), alx1 continued to be expressed in untreated PMC(−) embryos (arrow in D, 5/5 embryos), but expression was undetectable in most axitinib-treated PMC(−) embryos (E, 10/13 embryos). BC, blastocoelar cell; hpd, hours post–PMC depletion; Lv-alx1, L. variegatus aristaless-like homeobox 1; PMC, primary mesenchyme cell; VEGFR, vascular endothelial growth factor receptor. https://doi.org/10.1371/journal.pbio.3000460.g003 VEGFR signaling acts through VEGFR-10-Ig and VEGF3 Because vegfr-10-Ig is expressed at high levels during embryogenesis and because it has been shown to regulate PMC skeletogenesis, it seemed likely that this receptor was the primary target of axitinib. Using WMISH, we first examined the expression of L. variegatus vegfr-10-Ig (Lv-vegfr-10-Ig) in control embryos. We confirmed that vegfr-10-Ig was expressed robustly in PMCs, as previously reported [9], but also observed faint expression in the wall of the archenteron at the early gastrula stage (S1 Fig). This region represents the presumptive nonskeletogenic mesoderm, including the presumptive BCs. Expression in the nonskeletogenic mesoderm was observed in multiple batches of embryos; however, this expression was apparent in only 30%–40% of embryos in each batch and was not detectable at later developmental stages, suggesting that expression was highly transient. In PMC(−) embryos, we originally documented high levels of expression of vegfr-10-Ig in transfating BCs at a relatively late stage, 10–11 hours post–PMC depletion, several hours after alx1 activation [19]. Cheng and coworkers [35], however, used quantitative PCR (QPCR) and WMISH to show that in micromere(−) embryos, which resemble PMC(−) embryos in many respects, expression of vegfr-10-Ig and alx1 was up-regulated at almost the same time. We therefore reexamined the timing of vegfr-10-Ig activation in PMC(−) embryos. We found that within 2 hours after PMC removal, vegfr-10-Ig mRNA was expressed robustly in the vegetal region of >90% of PMC(−) embryos, in an area that included the BC domain (Fig 4A and 4B) (2 trials, 10–20 embryos/trial). Expression remained strong at 7 hours post–PMC depletion (Fig 4C). The timing of vegfr-10-Ig up-regulation in BCs following PMC removal therefore closely paralleled the timing of alx1 and tbr activation, in agreement with the findings of Cheng and coworkers [35]. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Lv-vegfr-10-Ig expression during BC transfating. (A) Control gastrula. Lv-vegfr-10-Ig is expressed at high levels by PMCs (arrow), as previously reported [9]. (B) PMC(−) embryo, 2 hpd. Expression of Lv-vegfr-10-Ig is evident in the invaginating vegetal plate (arrow), a region that includes presumptive BCs (32/37 embryos). (C) PMC(−) embryo, 7 hpd. Expression of Lv-vegfr-10-Ig is apparent in the wall of the archenteron (arrow). (D) Emetine-treated PMC(−) embryo, 2 hpd. Inhibition of protein synthesis did not prevent the robust expression of Lv-vegfr-10-Ig in the vegetal plate following PMC removal (16/20 embryos). BC, blastocoelar cell; hpd, hours post–PMC depletion; Lv-vegfr-10-Ig, L. variegatus vascular endothelial growth factor receptor-10-Ig; PMC, primary mesenchyme cell. https://doi.org/10.1371/journal.pbio.3000460.g004 The rapidity with which vegfr-10-Ig mRNA accumulated after PMC removal argued against the possibility that activation occurred via a relay mechanism that involved the transcriptional activation of an intermediate regulatory gene. Such a relay mechanism would require new protein synthesis in order to produce the relevant regulatory factor. To explore this possibility, we treated embryos with 100 μM emetine, which rapidly (<25 minutes) blocks >95% of protein synthesis in L. variegatus embryos [36]. Microsurgery was carried out in the presence of the inhibitor, and PMC(−) embryos were allowed to develop for an additional 2 hours in emetine-containing seawater before they were processed for WMISH. Longer incubation times were not possible, as embryos disaggregated when exposed to emetine for longer than approximately 2.5 hours. We observed a strong activation of vegfr-10-Ig in >90% of emetine-treated PMC(−) embryos (3 trials, 10–20 embryos/trial) (Fig 4D), arguing against a transcriptional relay mechanism. To test whether signaling through VEGFR-10-Ig was required for BC transfating, we inhibited the expression of the receptor using a splice-blocking morpholino (MO), following the strategy of Duloquin and coworkers [9] (Fig 5A). Embryos injected with the Lv-vegfr-10-Ig MO showed reduced levels of correctly spliced, wild-type mRNA and expressed high levels of an alternatively spliced RNA species that contained intron 2 (Fig 5B). We cloned and sequenced the prominent 1.1-kb PCR product shown in Fig 5B and confirmed that intron 2 was present, resulting in the introduction of multiple stop codons in all three reading frames. Less abundant PCR products were also detectable (asterisks in Fig 5B), which may reflect utilization of cryptic splice sites within exon 2 and/or intron 2. We concluded that the splice-blocking MO resulted in a marked, but incomplete, knockdown of VEGFR-10-Ig. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Lv-VEGFR-10-Ig is required for BC transfating. (A) MO knockdown strategy. Following the approach of Duloquin and coworkers [9], an MO was directed at the exon 2/intron 2 splice junction of the Lv-vegfr-10-Ig primary transcript. (B) RT-PCR analysis of morphant embryos. The Lv-vegfr-10-Ig MO produced a significant reduction in the level of wild-type mRNA and the appearance of a prominent splicing isoform that included intron 2, as verified by cloning and sequencing of the PCR product. Inclusion of intron 2 resulted in multiple stop codons in all reading frames and the production of a truncated, nonfunctional receptor. Low levels of other mis-spliced forms of Lv-vegfr-10-Ig mRNA (asterisks) were also detected. (C-D) BC transfating and skeletogenesis were suppressed in PMC(−), morphant embryos, as revealed by DIC and polarization microscopy at 40 hpd (C-D’) and by immunostaining with mAb 6a9 at 12 hpd (E-F). BC reprogramming was not affected by an equivalent concentration of an MO directed against Lv-IgTM, a PMC-specific protein that regulates skeletal branching [37]. Statistical significance of the data was assessed by two-sided t tests and p-values < 0.05 are indicated by asterisks. Raw data can be found in S1 Data. BC, blastocoelar cell; DIC, differential interference contrast microscopy; hpd, hours post–PMC depletion; Lv-vegfr-10-Ig, L. variegatus vascular endothelial growth factor receptor-10-Ig; Lv-IgTM, L. variegatus Ig and transmembrane domain protein; mAb, monoclonal antibody; MO, morpholino; PMC, primary mesenchyme cell; POL, polarized light microscopy; RT-PCR, reverse transcription PCR. https://doi.org/10.1371/journal.pbio.3000460.g005 Knockdown of VEGFR-10-Ig in PMC(−) embryos produced a phenotype indistinguishable from axitinib treatment. Morphant PMC(−) embryos gastrulated normally and developed a tripartite gut, ciliary band, pigment cells, and coelomic pouches but lacked birefringent skeletal elements (Fig 5C–5D’). We also observed a striking decrease in the number 6a9(+) cells in PMC(−) morphant embryos (Fig 5E and 5F). This effect was not due to a delay in transfating, as shown by immunostaining morphant embryos at later time points (20 hours post–PMC depletion). In addition, the effect was specific to the VEGFR MO, as injection of an equivalent concentration of an MO complementary to IgTM, a protein that regulates skeletal branching [37], had no effect on BC transfating (Fig 5F). Previous work showed that delta MO also had no effect of BC transfating [19]. These studies showed that signaling through VEGFR-10-Ig was required for BCs to express a skeletogenic fate. They suggested that axitinib blocked transfating by inhibiting the function of this receptor in BCs, where the gene is expressed robustly following PMC removal. Three genes in the sea urchin genome—vegf, vegf2, and vegf3—encode ligands in the VEGF family [21]. During gastrulation, vegf3 is expressed at high levels (peak expression > 1,200 transcripts/embryo), whereas vegf and vegf2 are expressed at very low levels or not at all (<20 transcripts/embryo) [22]. vegf3 is expressed in localized regions of the ectoderm and functions along with vegfr-10-Ig as an essential mediator of PMC migration and skeletogenesis [9–13]. VEGF3 was therefore a likely candidate for the ligand that interacted with VEGFR-10-Ig to activate the skeletogenic GRN in transfating BCs. To test this hypothesis, we knocked down VEGF3 expression using a translation-blocking MO characterized previously [11]. Knockdown of VEGF3 in PMC(−) embryos produced a phenotype indistinguishable from VEGFR-10-Ig knockdown and axitinib treatment. PMC(−) morphant embryos gastrulated on schedule and gave rise to normal, prism-like larvae that lacked arms and birefringent skeletal elements (Fig 6A–6B’). These embryos also contained reduced numbers of 6a9(+) cells (Fig 6C and 6D). Our observations demonstrated that VEGF3 (like VEGFR-10-Ig) was required for BC transfating and strongly supported the hypothesis that VEGF3 is the ligand that interacts with VEGFR-10-Ig. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. Lv-VEGF3 is required for BC transfating. A previously characterized translation-blocking MO was used to interfere with Lv-vegf3 expression [11]. PMC(−) morphant embryos failed to form skeletal elements even at 48 hpd (A-B’), and very few 6a9(+) cells were detectable at 12 hpd (C-D), indicating that BC transfating was largely blocked. Statistical significance of cell count data was assessed by two-sided t tests, and p-values < 0.05 are indicated by asterisks. Raw data can be found in S1 Data. BC, blastocoelar cell; DIC, differential interference contrast microscopy; hpd, hours post–PMC depletion; Lv-vegf3, L. variegatus vascular endothelial growth factor 3; mAb, monoclonal antibody; MO, morpholino; PMC, primary mesenchyme cell; POL, polarized light microscopy. https://doi.org/10.1371/journal.pbio.3000460.g006 PMCs suppress the skeletogenic potential of BCs by sequestering VEGF3 The finding that signaling through the VEGF pathway, mediated by VEGF3 and VEGFR-10-Ig, was essential for the activation of the skeletogenic pathway in BCs suggested a possible mechanism for PMC-to-BC signaling. As noted above, PMCs ordinarily express high levels of vegfr-10-Ig during gastrulation. Our results suggested that PMCs might outcompete BCs for VEGF3, thereby preventing the activation of the skeletogenic program by BCs and directing them into an alternative (immunocyte) pathway. This model generated two testable predictions. First, lowering the level of VEGFR-10-Ig on the PMC surface should reduce the ability of the cells to suppress BC transfating. We tested this hypothesis by generating chimeric embryos in which VEGFR-10-Ig expression was knocked down selectively in PMCs (Fig 7A–7E). The entire complement of PMCs was removed from host embryos, and varying numbers of labeled, donor PMCs were transplanted into the host embryos. PMCs were labeled by incubating donor embryos in a reactive dye, rhodamine isothiocyanate (RITC), which results in the covalent attachment of rhodamine to lysine residues on cell surface proteins [15]. The fluorescently tagged proteins do not readily diffuse from cell to cell after fusion; therefore, this method persistently labels the donor cells. Donor PMCs from control (uninjected) and VEGFR-10-Ig morphant embryos were tested separately. It was shown previously that the number of BCs that express a skeletogenic fate is inversely proportional to the number of PMCs in the blastocoel [15], a finding that we confirmed (Fig 7F). Significantly, we observed that PMCs with reduced expression of VEGFR-10-Ig were much less effective than control PMCs in suppressing the skeletogenic potential of BCs (Fig 7F). The residual signaling activity of morphant PMCs was consistent with the incomplete nature of the VEGFR-10-Ig knockdown (Fig 5B). These experiments showed that signaling by PMCs was dependent upon their expression of VEGFR-10-Ig. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 7. Knockdown of Lv-VEGFR-10-Ig in PMCs reduces their ability to suppress BC transfating. (A) Experimental design. The entire complement of PMCs was removed from a recipient embryo and replaced with 15–40 RITC-labeled PMCs from control or Lv-VEGFR-10-Ig morphant donor embryos. Then, 10.5 hours after transplantation, recipient embryos were fixed and immunostained with mAb 6a9 and a DyLight 488 goat anti-mouse secondary antibody. Donor PMCs were identified by (red + green) fluorescence, whereas transfated BCs exhibited only green fluorescence. (B-E) A representative embryo after immunostaining, viewed with DIC (B) or epifluorescence optics. (C) mAb 6a9 immunostaining showing all skeletogenic cells (donor PMCs and transfated BCs). (D) Rhodamine fluorescence showing donor PMCs. (E) Overlay of the two fluorescent channels. Cells that are green but not red are transfated BCs. (F) Quantification of the numbers of transfated BCs following PMC transplantation. Control PMCs (orange) were more potent at suppressing BC transfating than PMCs that had reduced expression of Lv-VEGFR-10-Ig (teal). In both cases, the number of transfated cells was inversely related to the number of PMCs in the blastocoel. Each point represents a single recipient embryo. Lines of best fit are indicated, and 95% confidence intervals are shown in gray. Raw data can be found in S2 Data. BC, blastocoelar cell; DIC, differential interference contrast microscopy; Lv-VEGFR-10-Ig, L. variegatus vascular endothelial growth factor receptor-10-Ig; mAb, monoclonal antibody; pBC, presumptive BC; PMC, primary mesenchyme cell; RITC, rhodamine isothiocyanate. https://doi.org/10.1371/journal.pbio.3000460.g007 Both control and morphant PMCs migrated actively and accumulated at typical PMC target sites near the embryo equator; however, in about 50% of embryos, morphant PMCs remained localized predominantly at PMC target sites on one side of the blastocoel, in contrast to control PMCs, which formed a bilaterally symmetrical PMC ring pattern. This indicated that knockdown of VEGFR-10-Ig partially compromised PMC migration and/or patterning, as previously reported [9,11]. To test whether VEGFR-10-Ig MO affected signaling indirectly by perturbing PMC migration, we agglutinated PMCs by microinjecting wheat germ agglutinin (WGA) into the blastocoel at the mesenchyme blastula stage [38]. In most embryos, the PMCs quickly (<2 hours) coalesced into 1–2 large masses that persisted for >12 hours (28/31 embryos) (S2 Fig). Nevertheless, PMCs were still capable of fully suppressing BC transfating, as shown by the absence of any 6a9(+) cells at the archenteron tip during gastrulation. Injection of WGA into PMC(−) embryos did not by itself prevent BC transfating, and 6a9(+) cells appeared normally (34/34 embryos). These studies showed that the reduced signaling potency of morphant PMCs could not be attributed to the compromised migratory capacity of these cells. A second prediction of the competition model was that overexpression of VEGF3 should saturate receptors on the PMC surface and provide sufficient unbound ligand to activate the skeletogenic program in BCs. During normal development, expression of VEGF3 is limited to small patches of ectoderm that overlie sites of skeletal growth. We overexpressed VEGF3 by microinjecting mRNA encoding full-length Lv-VEGF3 into fertilized eggs, causing the ligand to be expressed throughout the embryo. Injection of VEGF3 mRNA led to a concentration-dependent increase in the number of skeletogenic (6a9-positive) cells at the late prism stage (24 hpf) (Fig 8A, 8A’ and 8D). The spatial patterning of 6a9(+) cells was markedly perturbed in embryos overexpressing VEGF3, consistent with the role of this molecule in PMC guidance [9]. Strikingly, when we examined Lv-vegf3 mRNA–injected embryos several hours earlier in development, we observed ectopic activation of Lv-alx1 (14/18 embryos, or 78%) and Lv-tbr (24/32 embryos, or 75%) in the wall of the archenteron, in a relatively broad territory that included presumptive mesoderm and possibly also presumptive endoderm cells (Fig 8B–8C’). Activation in presumptive endoderm is consistent with evidence that under appropriate experimental conditions, these cells can also adopt a skeletogenic fate, probably via the re-specification of a BC-like intermediate state [19,39]. Lv-alx1 and Lv-tbr were not activated at the very anterior tip of the archenteron, the location of the small micromere descendants (future germ cells). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 8. Overexpression of Lv-VEGF3 induces BC transfating. Capped mRNA encoding Lv-VEGF3 was injected into fertilized eggs. Many supernumerary 6a9(+) cells were observed in such embryos 24 hpf, when sibling control embryos had reached the late prism stage (A, A’). In control embryos, expression of Lv-alx1 and Lv-tbr was restricted to PMCs, as expected (arrows in B and C), whereas injection of Lv-vegf3 mRNA resulted in the ectopic activation of Lv-alx1 and Lv-tbr in the wall of the archenteron (arrows in B’ and C’). The effect of Lv-vegf3 mRNA was dose-dependent (D). Raw data can be found in S1 Data. BC, blastocoelar cell; hpf, hours postfertilization; Lv-alx1, L. variegatus aristaless-like 1; Lv-tbr, L. variegatus t-brain; Lv-VEGF3, L. variegatus vascular endothelial growth factor 3; mAb, monoclonal antibody; PMC, primary mesenchyme cell. https://doi.org/10.1371/journal.pbio.3000460.g008 VEGFR signaling is required for the activation and maintenance of the skeletogenic network in BCs There are two VEGF receptor (VEGFR) genes in sea urchins: vegfr-10-Ig is ordinarily expressed at high levels (peak expression > 3,000 transcripts/embryo), whereas vegfr-7-Ig is expressed at much lower levels (<150 transcripts/embryo) [21,22]. During normal development, vegfr-10-Ig is expressed selectively by presumptive PMCs beginning at the early blastula stage [9]. In PMC(−) embryos, however, vegfr-10-Ig is expressed robustly in BCs [19]. The spatial expression pattern of vegfr-7-Ig is not known in detail but, unlike vegfr-10-Ig, this gene is not differentially expressed by PMCs [23,24]. Axitinib is a highly selective inhibitor of VEGF receptors at nanomolar concentrations [25,26]. The effects of axitinib on sea urchin development mimic those of VEGF3 and VEGFR-10-Ig knockdowns, confirming that axitinib is a specific inhibitor of VEGF/VEGFR signaling in this system [9,11]. We found that PMC(−) Lytechinus variegatus embryos treated continuously with axitinib (75 nM) beginning at the time of PMC removal failed to form a skeleton, even after prolonged culture (48 hours postfertilization [hpf]) (Fig 2A and 2A’). Axitinib-treated embryos swam vigorously and gastrulated at the same time as untreated, PMC(−) embryos but never extended arms. Polarization microscopy showed that these embryos completely lacked birefringent skeletal elements (Fig 2B and 2B’). This phenotype was reproducible and highly penetrant; >95% of axitinib-treated, PMC(−) embryos showed this effect across many batches. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Axitinib blocks BC transfating. PMCs were removed from mesenchyme blastula–stage embryos, and the resultant PMC(−) embryos were separated into two cohorts. One cohort was left in plain seawater, whereas the other was transferred to 5 nM axitinib immediately after PMC removal. (A-G) Control PMC(−) embryos. (A’-G’) Axitinib-treated PMC(−) embryos. Axitinib treatment blocked the formation of birefringent skeletal elements (A-B’) and the expression of early skeletogenic regulatory genes by BCs as shown by WMISH analysis of Lv-alx1 (C,C’) and Lv-tbr (D,D’) expression. The expression of skeletogenic effector genes downstream of alx1 was also blocked, as indicated by WMISH analysis of Lv-p58a (E,E’) and Lv-msp130rel2 (F,F’), and by immunostaining with mAb 6a9 (G,G’). Arrowheads indicate expression of skeletogenic genes by transfating BCs. Panel H shows quantification of 6a9-positive cells in control and axitinib-treated PMC(−) embryos at 12 hpd (two independent trials from separate matings). Statistical significance of the data was assessed by two-sided t tests, and p-values < 0.05 are indicated by asterisks. Raw data can be found in S1 Data. BC, blastocoelar cell; DIC, differential interference contrast microscopy; hpd, hours post–PMC depletion; Lv-alx1, L. variegatus aristaless-like 1; Lv-tbr, L. variegatus t-brain; mAb, monoclonal antibody; msp130rel2, L. variegatus mesenchyme specific protein 130-related 2; POL, polarized light microscopy; PMC, primary mesenchyme cell; WMISH, whole-mount in situ hybridization. https://doi.org/10.1371/journal.pbio.3000460.g002 Early molecular events in BC transfating include the activation of aristaless-like homeobox 1 (alx1) and t-brain (tbr), two regulatory (i.e., transcription factor-encoding) genes that function early in the PMC GRN and are ordinarily expressed only in the large micromere-PMC lineage [17,19]. Alx1 is a critically important regulator of the skeletogenic GRN both in the PMC lineage and in transfating BCs, whereas tbr plays a less prominent role [17,24,27–30]. The activation of these two genes in transfating BCs is detectable by whole-mount in situ hybridization (WMISH) as early as 2 hours post–PMC depletion [19]. We confirmed that L. variegatus alx1 (Lv-alx1) and L. variegatus tbr (Lv-tbr) were expressed robustly in the anterior region of the archenteron in control PMC(−) embryos but found that expression of both genes was greatly reduced in embryos that were placed in axitinib at the time of PMC removal (Fig 2C–2D’) (3 trials, n = 15–30 embryos/trial). This effect was readily apparent by 3 hours post–PMC depletion when assayed by conventional (histochemical) WMISH. These findings demonstrated that the expression of alx1 and tbr in BCs was dependent upon VEGF/VEGFR signaling. Consistent with this finding, the expression of two terminal biomineralization genes, p58a and L. variegatus mesenchyme specific protein 130-related 2 (msp130rel2), that receive positive regulatory inputs from alx1 [24,31,32], was also greatly reduced in axitinib-treated, PMC(−) embryos (Fig 2E–2F’, >90% of embryos, n > 40). We also immunostained embryos with monoclonal antibody (mAb) 6a9, a widely used marker of PMC differentiation that recognizes the biomineralization protein MSP130 [15,33,34], and observed a dramatic reduction in the numbers of 6a9(+) cells in axitinib-treated, PMC(−) embryos (Fig 2G and 2H). Culturing such embryos for an additional 24 hours did not increase the numbers of 6a9(+) cells, indicating that axitinib was not simply delaying BC transfating. We next asked whether VEGFR signaling was required only at the onset of BC transfating (i.e., whether it acted like a trigger) or was required continuously during the transfating process. We removed PMCs from mesenchyme blastula–stage embryos and allowed them to develop for varying periods of time before adding axitinib to the medium (Fig 3A). At 12 hours post–PMC depletion, embryos were fixed and immunostained with mAb 6a9. As the time interval between PMC depletion and the addition of axitinib increased, the numbers of 6a9(+) cells increased progressively (Fig 3B). This suggested that VEGFR signaling was required continuously during the first several hours after PMC depletion. Surprisingly, even when axitinib was added 5–6 hours post–PMC depletion, several hours after the activation of skeletogenic regulatory genes in BCs, we observed a significant reduction in the numbers of 6a9(+) cells. This finding raised the question of whether, once activated, early regulatory genes maintained their expression in the presence of axitinib. We removed PMCs at the mesenchyme blastula stage and allowed the resulting PMC(−) embryos to develop for 4–5 hours, sufficient time for alx1 to be activated ectopically in BCs (Fig 3C). Transfer of PMC(−) embryos into 75 nM axitinib at that time led to a decline in alx1 expression to undetectable levels in most embryos (10/13, or 77%) within 5 hours, whereas expression was maintained (or increased) in sibling, untreated, PMC(−) embryos (Fig 3D and 3E). These observations showed that VEGFR signaling was required not only to activate but also to maintain the expression of alx1 in transfating BCs. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Time course of axitinib sensitivity during BC transfating. (A) Experimental design. Solid horizontal bars indicate the duration of axitinib (5 nM) treatment. Start times are shown as a 1-hour interval because microsurgical removal of PMCs required approximately 1 hour for each cohort of embryos. (B) Counts of 6a9-positive cells. Data shown were collected from several independent trials, and untreated controls from all trials were pooled. For all time periods tested, we observed a significant reduction in 6a9(+) cells in axitinib-treated PMC(−) embryos relative to untreated PMC(−) controls. Raw data can be found in S1 Data. (C-E) Signaling through VEGFR is required for the maintenance of Lv-alx1 expression in BCs during reprogramming. PMC(−) embryos were treated with axitinib for 5 hours beginning at 4–5 hpd. At the start of axitinib treatment, Lv-alx1 was expressed robustly by BCs at the tip of the archenteron (arrow in C, 10/11 embryos). Five hours later (9–10 hpd), alx1 continued to be expressed in untreated PMC(−) embryos (arrow in D, 5/5 embryos), but expression was undetectable in most axitinib-treated PMC(−) embryos (E, 10/13 embryos). BC, blastocoelar cell; hpd, hours post–PMC depletion; Lv-alx1, L. variegatus aristaless-like homeobox 1; PMC, primary mesenchyme cell; VEGFR, vascular endothelial growth factor receptor. https://doi.org/10.1371/journal.pbio.3000460.g003 VEGFR signaling acts through VEGFR-10-Ig and VEGF3 Because vegfr-10-Ig is expressed at high levels during embryogenesis and because it has been shown to regulate PMC skeletogenesis, it seemed likely that this receptor was the primary target of axitinib. Using WMISH, we first examined the expression of L. variegatus vegfr-10-Ig (Lv-vegfr-10-Ig) in control embryos. We confirmed that vegfr-10-Ig was expressed robustly in PMCs, as previously reported [9], but also observed faint expression in the wall of the archenteron at the early gastrula stage (S1 Fig). This region represents the presumptive nonskeletogenic mesoderm, including the presumptive BCs. Expression in the nonskeletogenic mesoderm was observed in multiple batches of embryos; however, this expression was apparent in only 30%–40% of embryos in each batch and was not detectable at later developmental stages, suggesting that expression was highly transient. In PMC(−) embryos, we originally documented high levels of expression of vegfr-10-Ig in transfating BCs at a relatively late stage, 10–11 hours post–PMC depletion, several hours after alx1 activation [19]. Cheng and coworkers [35], however, used quantitative PCR (QPCR) and WMISH to show that in micromere(−) embryos, which resemble PMC(−) embryos in many respects, expression of vegfr-10-Ig and alx1 was up-regulated at almost the same time. We therefore reexamined the timing of vegfr-10-Ig activation in PMC(−) embryos. We found that within 2 hours after PMC removal, vegfr-10-Ig mRNA was expressed robustly in the vegetal region of >90% of PMC(−) embryos, in an area that included the BC domain (Fig 4A and 4B) (2 trials, 10–20 embryos/trial). Expression remained strong at 7 hours post–PMC depletion (Fig 4C). The timing of vegfr-10-Ig up-regulation in BCs following PMC removal therefore closely paralleled the timing of alx1 and tbr activation, in agreement with the findings of Cheng and coworkers [35]. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Lv-vegfr-10-Ig expression during BC transfating. (A) Control gastrula. Lv-vegfr-10-Ig is expressed at high levels by PMCs (arrow), as previously reported [9]. (B) PMC(−) embryo, 2 hpd. Expression of Lv-vegfr-10-Ig is evident in the invaginating vegetal plate (arrow), a region that includes presumptive BCs (32/37 embryos). (C) PMC(−) embryo, 7 hpd. Expression of Lv-vegfr-10-Ig is apparent in the wall of the archenteron (arrow). (D) Emetine-treated PMC(−) embryo, 2 hpd. Inhibition of protein synthesis did not prevent the robust expression of Lv-vegfr-10-Ig in the vegetal plate following PMC removal (16/20 embryos). BC, blastocoelar cell; hpd, hours post–PMC depletion; Lv-vegfr-10-Ig, L. variegatus vascular endothelial growth factor receptor-10-Ig; PMC, primary mesenchyme cell. https://doi.org/10.1371/journal.pbio.3000460.g004 The rapidity with which vegfr-10-Ig mRNA accumulated after PMC removal argued against the possibility that activation occurred via a relay mechanism that involved the transcriptional activation of an intermediate regulatory gene. Such a relay mechanism would require new protein synthesis in order to produce the relevant regulatory factor. To explore this possibility, we treated embryos with 100 μM emetine, which rapidly (<25 minutes) blocks >95% of protein synthesis in L. variegatus embryos [36]. Microsurgery was carried out in the presence of the inhibitor, and PMC(−) embryos were allowed to develop for an additional 2 hours in emetine-containing seawater before they were processed for WMISH. Longer incubation times were not possible, as embryos disaggregated when exposed to emetine for longer than approximately 2.5 hours. We observed a strong activation of vegfr-10-Ig in >90% of emetine-treated PMC(−) embryos (3 trials, 10–20 embryos/trial) (Fig 4D), arguing against a transcriptional relay mechanism. To test whether signaling through VEGFR-10-Ig was required for BC transfating, we inhibited the expression of the receptor using a splice-blocking morpholino (MO), following the strategy of Duloquin and coworkers [9] (Fig 5A). Embryos injected with the Lv-vegfr-10-Ig MO showed reduced levels of correctly spliced, wild-type mRNA and expressed high levels of an alternatively spliced RNA species that contained intron 2 (Fig 5B). We cloned and sequenced the prominent 1.1-kb PCR product shown in Fig 5B and confirmed that intron 2 was present, resulting in the introduction of multiple stop codons in all three reading frames. Less abundant PCR products were also detectable (asterisks in Fig 5B), which may reflect utilization of cryptic splice sites within exon 2 and/or intron 2. We concluded that the splice-blocking MO resulted in a marked, but incomplete, knockdown of VEGFR-10-Ig. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Lv-VEGFR-10-Ig is required for BC transfating. (A) MO knockdown strategy. Following the approach of Duloquin and coworkers [9], an MO was directed at the exon 2/intron 2 splice junction of the Lv-vegfr-10-Ig primary transcript. (B) RT-PCR analysis of morphant embryos. The Lv-vegfr-10-Ig MO produced a significant reduction in the level of wild-type mRNA and the appearance of a prominent splicing isoform that included intron 2, as verified by cloning and sequencing of the PCR product. Inclusion of intron 2 resulted in multiple stop codons in all reading frames and the production of a truncated, nonfunctional receptor. Low levels of other mis-spliced forms of Lv-vegfr-10-Ig mRNA (asterisks) were also detected. (C-D) BC transfating and skeletogenesis were suppressed in PMC(−), morphant embryos, as revealed by DIC and polarization microscopy at 40 hpd (C-D’) and by immunostaining with mAb 6a9 at 12 hpd (E-F). BC reprogramming was not affected by an equivalent concentration of an MO directed against Lv-IgTM, a PMC-specific protein that regulates skeletal branching [37]. Statistical significance of the data was assessed by two-sided t tests and p-values < 0.05 are indicated by asterisks. Raw data can be found in S1 Data. BC, blastocoelar cell; DIC, differential interference contrast microscopy; hpd, hours post–PMC depletion; Lv-vegfr-10-Ig, L. variegatus vascular endothelial growth factor receptor-10-Ig; Lv-IgTM, L. variegatus Ig and transmembrane domain protein; mAb, monoclonal antibody; MO, morpholino; PMC, primary mesenchyme cell; POL, polarized light microscopy; RT-PCR, reverse transcription PCR. https://doi.org/10.1371/journal.pbio.3000460.g005 Knockdown of VEGFR-10-Ig in PMC(−) embryos produced a phenotype indistinguishable from axitinib treatment. Morphant PMC(−) embryos gastrulated normally and developed a tripartite gut, ciliary band, pigment cells, and coelomic pouches but lacked birefringent skeletal elements (Fig 5C–5D’). We also observed a striking decrease in the number 6a9(+) cells in PMC(−) morphant embryos (Fig 5E and 5F). This effect was not due to a delay in transfating, as shown by immunostaining morphant embryos at later time points (20 hours post–PMC depletion). In addition, the effect was specific to the VEGFR MO, as injection of an equivalent concentration of an MO complementary to IgTM, a protein that regulates skeletal branching [37], had no effect on BC transfating (Fig 5F). Previous work showed that delta MO also had no effect of BC transfating [19]. These studies showed that signaling through VEGFR-10-Ig was required for BCs to express a skeletogenic fate. They suggested that axitinib blocked transfating by inhibiting the function of this receptor in BCs, where the gene is expressed robustly following PMC removal. Three genes in the sea urchin genome—vegf, vegf2, and vegf3—encode ligands in the VEGF family [21]. During gastrulation, vegf3 is expressed at high levels (peak expression > 1,200 transcripts/embryo), whereas vegf and vegf2 are expressed at very low levels or not at all (<20 transcripts/embryo) [22]. vegf3 is expressed in localized regions of the ectoderm and functions along with vegfr-10-Ig as an essential mediator of PMC migration and skeletogenesis [9–13]. VEGF3 was therefore a likely candidate for the ligand that interacted with VEGFR-10-Ig to activate the skeletogenic GRN in transfating BCs. To test this hypothesis, we knocked down VEGF3 expression using a translation-blocking MO characterized previously [11]. Knockdown of VEGF3 in PMC(−) embryos produced a phenotype indistinguishable from VEGFR-10-Ig knockdown and axitinib treatment. PMC(−) morphant embryos gastrulated on schedule and gave rise to normal, prism-like larvae that lacked arms and birefringent skeletal elements (Fig 6A–6B’). These embryos also contained reduced numbers of 6a9(+) cells (Fig 6C and 6D). Our observations demonstrated that VEGF3 (like VEGFR-10-Ig) was required for BC transfating and strongly supported the hypothesis that VEGF3 is the ligand that interacts with VEGFR-10-Ig. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. Lv-VEGF3 is required for BC transfating. A previously characterized translation-blocking MO was used to interfere with Lv-vegf3 expression [11]. PMC(−) morphant embryos failed to form skeletal elements even at 48 hpd (A-B’), and very few 6a9(+) cells were detectable at 12 hpd (C-D), indicating that BC transfating was largely blocked. Statistical significance of cell count data was assessed by two-sided t tests, and p-values < 0.05 are indicated by asterisks. Raw data can be found in S1 Data. BC, blastocoelar cell; DIC, differential interference contrast microscopy; hpd, hours post–PMC depletion; Lv-vegf3, L. variegatus vascular endothelial growth factor 3; mAb, monoclonal antibody; MO, morpholino; PMC, primary mesenchyme cell; POL, polarized light microscopy. https://doi.org/10.1371/journal.pbio.3000460.g006 PMCs suppress the skeletogenic potential of BCs by sequestering VEGF3 The finding that signaling through the VEGF pathway, mediated by VEGF3 and VEGFR-10-Ig, was essential for the activation of the skeletogenic pathway in BCs suggested a possible mechanism for PMC-to-BC signaling. As noted above, PMCs ordinarily express high levels of vegfr-10-Ig during gastrulation. Our results suggested that PMCs might outcompete BCs for VEGF3, thereby preventing the activation of the skeletogenic program by BCs and directing them into an alternative (immunocyte) pathway. This model generated two testable predictions. First, lowering the level of VEGFR-10-Ig on the PMC surface should reduce the ability of the cells to suppress BC transfating. We tested this hypothesis by generating chimeric embryos in which VEGFR-10-Ig expression was knocked down selectively in PMCs (Fig 7A–7E). The entire complement of PMCs was removed from host embryos, and varying numbers of labeled, donor PMCs were transplanted into the host embryos. PMCs were labeled by incubating donor embryos in a reactive dye, rhodamine isothiocyanate (RITC), which results in the covalent attachment of rhodamine to lysine residues on cell surface proteins [15]. The fluorescently tagged proteins do not readily diffuse from cell to cell after fusion; therefore, this method persistently labels the donor cells. Donor PMCs from control (uninjected) and VEGFR-10-Ig morphant embryos were tested separately. It was shown previously that the number of BCs that express a skeletogenic fate is inversely proportional to the number of PMCs in the blastocoel [15], a finding that we confirmed (Fig 7F). Significantly, we observed that PMCs with reduced expression of VEGFR-10-Ig were much less effective than control PMCs in suppressing the skeletogenic potential of BCs (Fig 7F). The residual signaling activity of morphant PMCs was consistent with the incomplete nature of the VEGFR-10-Ig knockdown (Fig 5B). These experiments showed that signaling by PMCs was dependent upon their expression of VEGFR-10-Ig. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 7. Knockdown of Lv-VEGFR-10-Ig in PMCs reduces their ability to suppress BC transfating. (A) Experimental design. The entire complement of PMCs was removed from a recipient embryo and replaced with 15–40 RITC-labeled PMCs from control or Lv-VEGFR-10-Ig morphant donor embryos. Then, 10.5 hours after transplantation, recipient embryos were fixed and immunostained with mAb 6a9 and a DyLight 488 goat anti-mouse secondary antibody. Donor PMCs were identified by (red + green) fluorescence, whereas transfated BCs exhibited only green fluorescence. (B-E) A representative embryo after immunostaining, viewed with DIC (B) or epifluorescence optics. (C) mAb 6a9 immunostaining showing all skeletogenic cells (donor PMCs and transfated BCs). (D) Rhodamine fluorescence showing donor PMCs. (E) Overlay of the two fluorescent channels. Cells that are green but not red are transfated BCs. (F) Quantification of the numbers of transfated BCs following PMC transplantation. Control PMCs (orange) were more potent at suppressing BC transfating than PMCs that had reduced expression of Lv-VEGFR-10-Ig (teal). In both cases, the number of transfated cells was inversely related to the number of PMCs in the blastocoel. Each point represents a single recipient embryo. Lines of best fit are indicated, and 95% confidence intervals are shown in gray. Raw data can be found in S2 Data. BC, blastocoelar cell; DIC, differential interference contrast microscopy; Lv-VEGFR-10-Ig, L. variegatus vascular endothelial growth factor receptor-10-Ig; mAb, monoclonal antibody; pBC, presumptive BC; PMC, primary mesenchyme cell; RITC, rhodamine isothiocyanate. https://doi.org/10.1371/journal.pbio.3000460.g007 Both control and morphant PMCs migrated actively and accumulated at typical PMC target sites near the embryo equator; however, in about 50% of embryos, morphant PMCs remained localized predominantly at PMC target sites on one side of the blastocoel, in contrast to control PMCs, which formed a bilaterally symmetrical PMC ring pattern. This indicated that knockdown of VEGFR-10-Ig partially compromised PMC migration and/or patterning, as previously reported [9,11]. To test whether VEGFR-10-Ig MO affected signaling indirectly by perturbing PMC migration, we agglutinated PMCs by microinjecting wheat germ agglutinin (WGA) into the blastocoel at the mesenchyme blastula stage [38]. In most embryos, the PMCs quickly (<2 hours) coalesced into 1–2 large masses that persisted for >12 hours (28/31 embryos) (S2 Fig). Nevertheless, PMCs were still capable of fully suppressing BC transfating, as shown by the absence of any 6a9(+) cells at the archenteron tip during gastrulation. Injection of WGA into PMC(−) embryos did not by itself prevent BC transfating, and 6a9(+) cells appeared normally (34/34 embryos). These studies showed that the reduced signaling potency of morphant PMCs could not be attributed to the compromised migratory capacity of these cells. A second prediction of the competition model was that overexpression of VEGF3 should saturate receptors on the PMC surface and provide sufficient unbound ligand to activate the skeletogenic program in BCs. During normal development, expression of VEGF3 is limited to small patches of ectoderm that overlie sites of skeletal growth. We overexpressed VEGF3 by microinjecting mRNA encoding full-length Lv-VEGF3 into fertilized eggs, causing the ligand to be expressed throughout the embryo. Injection of VEGF3 mRNA led to a concentration-dependent increase in the number of skeletogenic (6a9-positive) cells at the late prism stage (24 hpf) (Fig 8A, 8A’ and 8D). The spatial patterning of 6a9(+) cells was markedly perturbed in embryos overexpressing VEGF3, consistent with the role of this molecule in PMC guidance [9]. Strikingly, when we examined Lv-vegf3 mRNA–injected embryos several hours earlier in development, we observed ectopic activation of Lv-alx1 (14/18 embryos, or 78%) and Lv-tbr (24/32 embryos, or 75%) in the wall of the archenteron, in a relatively broad territory that included presumptive mesoderm and possibly also presumptive endoderm cells (Fig 8B–8C’). Activation in presumptive endoderm is consistent with evidence that under appropriate experimental conditions, these cells can also adopt a skeletogenic fate, probably via the re-specification of a BC-like intermediate state [19,39]. Lv-alx1 and Lv-tbr were not activated at the very anterior tip of the archenteron, the location of the small micromere descendants (future germ cells). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 8. Overexpression of Lv-VEGF3 induces BC transfating. Capped mRNA encoding Lv-VEGF3 was injected into fertilized eggs. Many supernumerary 6a9(+) cells were observed in such embryos 24 hpf, when sibling control embryos had reached the late prism stage (A, A’). In control embryos, expression of Lv-alx1 and Lv-tbr was restricted to PMCs, as expected (arrows in B and C), whereas injection of Lv-vegf3 mRNA resulted in the ectopic activation of Lv-alx1 and Lv-tbr in the wall of the archenteron (arrows in B’ and C’). The effect of Lv-vegf3 mRNA was dose-dependent (D). Raw data can be found in S1 Data. BC, blastocoelar cell; hpf, hours postfertilization; Lv-alx1, L. variegatus aristaless-like 1; Lv-tbr, L. variegatus t-brain; Lv-VEGF3, L. variegatus vascular endothelial growth factor 3; mAb, monoclonal antibody; PMC, primary mesenchyme cell. https://doi.org/10.1371/journal.pbio.3000460.g008 Discussion It has been proposed that receptor molecules might sequester signaling ligands, thereby influencing their effective range and tissue pattering [40]. In the Drosophila ovary, expression of the Decapentaplegic (Dpp) receptor by escort cells restricts Dpp distribution and its influence on the fates of cells within the stem cell niche [41]. Interactions between Wnt ligands and Frizzled receptors limit the range of Wnt movement in the mammalian intestinal stem cell niche [42]. Other studies have shown that manipulation of Hedgehog and Dpp receptor expression in Drosophila imaginal wing discs influences the range of signaling and subsequent tissue patterning [43,44]. The simplest hypothesis is that receptor molecules sequester ligands, thereby preventing them from reaching more distant cells, although indirect mechanisms may also operate [45]. How widespread such mechanisms are in controlling cell fates, particularly during early embryogenesis, remains an open question. In the present study, we provide evidence that mesodermal cell fates during sea urchin gastrulation are regulated by direct competition for a paracrine factor, VEGF3. Two different groups of cells (PMCs and BCs) are capable of responding to the ligand, but as a result of competition, only one cell population does so, whereas the other is directed into an alternative developmental pathway (Fig 9). The following observations support the view that PMCs regulate BC fate by outcompeting these cells for VEGF3: (1) signaling through VEGFR-10-Ig, which is up-regulated in the presumptive BCs territory immediately following PMC ablation, is essential for the activation of alx1 and the deployment of the skeletogenic GRN in BCs; (2) VEGF3, produced by the ectoderm, is also required for BCs to express a skeletogenic fate, presumably through the interaction of this ligand with VEGFR-10-Ig; (3) PMCs with reduced levels of vegfr-10-Ig have a reduced ability to suppress BC transfating; and (4) overexpression of VEGF3 is sufficient to override the suppressive influence of PMCs and induce BC transfating. It should be noted that injection of VEGF3 mRNA into fertilized eggs probably results in the secretion of the ligand by most or all cells of the embryo. Therefore, it is possible that the effects we see may be caused by VEGF3 produced by the BCs themselves acting in an autocrine fashion. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 9. A model of the PMC–BC interaction. During normal development (top panel), PMCs migrate into the blastocoel at the onset of gastrulation, and VEGFR-Ig-10 on their surfaces sequesters VEGF3, which is expressed by ventrolateral ectoderm cells [9,11]. BCs are hypothesized to express low but functional levels of VEGFR-10-Ig. In PMC(−) embryos (bottom panel), VEGF3 is free to diffuse through the blastocoel and interacts with BCs. As a consequence of this signal, the key selector gene, alx1, is activated along with its many targets. At the same time, alx1 suppresses competing regulatory states [24,29]. alx1, aristaless-like homeobox 1; BC, blastocoelar cell; PMC, primary mesenchyme cell; VEGF, vascular endothelial growth factor; VEGFR, VEGF receptor. https://doi.org/10.1371/journal.pbio.3000460.g009 What is the mechanism by which vegfr-10-Ig is induced in BCs following PMC ablation? One possibility is that there exists a second mechanism of PMC-to-BC communication, entirely distinct from the proposed competition for VEGF3, that underlies the induction of vegfr-10-Ig in BCs. A much simpler hypothesis, however, is that BCs ordinarily express low levels of VEGFR-10-Ig on their surface, consistent with our finding that these cells transiently express low levels of vegfr-10-Ig mRNA (S1 Fig). We hypothesize that signaling by VEGF3 activates a positive-feedback mechanism that rapidly elevates the expression of vegfr-10-Ig. Notably, a similar mechanism normally operates in PMCs, in which vegfr-10-Ig expression is maintained by signaling through VEGF3 and VEGFR-10-Ig [9,11]. Robust activation of vegfr-10-Ig occurs even in the presence of emetine, arguing against any mechanism that requires the expression of new transcriptional regulators and instead suggesting that posttranslational modification of transcription factors that directly regulate vegfr-10-Ig might be responsible. Positive-feedback regulation of VEGFR expression by VEGF signaling has also been described in mammalian cells and has been attributed to the posttranscriptional regulation of several transcription factors that control VEGFR expression [46–48]. An important, unresolved issue is the mechanism by which VEGF signaling activates alx1 in transfating BCs. This may be related to the mechanism by which VEGF3 ordinarily regulates the expression of alx1 (and many other genes) within the PMC syncytium at late developmental stages, although the molecular basis of that control is also unknown. The effects of axitinib on PMC(−) embryos closely resemble those of U0126, an inhibitor of MEK, suggesting that VEGF might act through the MAPK cascade, as it does in other cell types [17,49]. One candidate mediator is E26 transformation-specific 1 (Ets1), an ERK-dependent transcription factor ordinarily expressed by both BCs and PMCs and a positive regulator of alx1 [50]. The processes that have led to the appearance of new embryonic cell types are poorly understood [51–55]. Arendt and colleagues [54] recently formalized an evolution-based definition of cell type, which they consider to be a set of cells in an organism that change together through evolution, partially independent of other cells, and that are evolutionarily more closely related to each other than to other cells. According to this view, the key to the origin of a new cell type is gene regulatory independence—i.e., the capacity for regulating and evolving gene expression independently of other cells. The micromere-PMC lineage is a recent evolutionary invention, and its appearance in euechinoids was associated with the coupling of an ancestral, adult skeletogenic program to the molecular polarity of the oocyte [4]. The intervening steps in the evolutionary appearance of PMCs are difficult to reconstruct, in part because of the great diversity of skeletogenic programs exhibited by modern echinoderms [3]. One hypothesis is that the ancestral echinoid program resembled that of modern cidaroids, the most basal group of living echinoids. These species have variable numbers of micromeres and lack an early-ingressing, skeletogenic mesenchyme; instead, a subpopulation of late-ingressing mesoderm cells produces the embryonic skeleton [56,57]. If this reflects the ancestral echinoid state, then the invention of PMCs may have been associated with mechanisms that suppressed the potential of late-ingressing, skeletogenic cells, thereby shunting them into alternative developmental pathways. It should also be noted that modern euechinoids possess non-micromere-derived, late-ingressing mesoderm cells that express a skeletogenic fate much later in development, after larval feeding begins [58]. This raises the possibility that BCs normally contribute to the late larval or adult skeleton in euechinoids and that PMC removal causes these cells to precociously activate the skeletogenic GRN. In several respects, the gene regulatory programs of PMCs and BCs of modern euechinoids are similar. Many effector genes associated with a mesenchymal phenotype are coexpressed selectively by these two cell populations [59]. Both cell populations also express ets1 and ets-related gene (erg) [60–62], two regulatory genes that provide positive inputs into vegfr-10-Ig in the PMC lineage [29]. The expression of these genes may reflect an ancestral, pan-mesodermal or pan-mesenchymal state, perhaps similar to that seen in the late-ingressing, nonskeletogenic mesenchyme of modern asteroids [63]. One striking exception to this shared pattern of gene expression, however, is the PMC-specific expression of alx1. This gene is activated specifically in the founder cells of the PMC lineage in the first cell cycle after their birth [28]. Alx1 has positive inputs into at least half of the >400 effector genes expressed selectively by PMCs and an even larger proportion of such genes that are expressed at high levels [24]. Recent chromatin immunoprecipitation sequencing (ChIP-seq) studies have shown that many of these inputs are direct [64]. Alx1 also functions in PMCs to suppress alternative mesodermal regulatory states [24,29], and misexpression of alx1 in nonskeletogenic mesoderm cells is sufficient to convert them to a PMC-like fate [17]. Moreover, the developmental expression and skeletogenic function of alx1 are conserved across echinoderms [3]. Together, these findings highlight the pivotal role of Alx1 in specifying skeletogenic cell identity and establish this protein as a “terminal selector” [54,65], i.e., a transcription factor that regulates a cell type–specific suite of effector genes while repressing alternative cell identities. We propose that the evolutionary change that linked alx1 activation to the prefertilization axis of the oocyte bypassed an ancestral mode of regulation that was based on VEGF signaling (Fig 10). The selective coexpression of vegf3, vegfr-10-Ig, and alx1 in embryonic and adult skeletogenic centers in all echinoderms that have been examined strongly suggests that VEGF signaling had an ancient role in skeletogenesis and that expression of both vegfr-Ig-10 and alx1 was associated with the invention of skeletogenic cells very early in echinoderm evolution [4,6–8,66]. Although the evolutionarily derived, initial phase of alx1 expression in euechinoids is independent of VEGF signaling [11], vestiges of an ancestral regulatory mechanism may persist in the apparent, signal-dependent regulation of alx1 in the ventral domain of the PMC syncytium at the late gastrula stage [12]. We hypothesize that the VEGF-dependent expression of alx1 in transfating BCs (see Figs 2C, 2C’, 8B and 8B’) is a manifestation of an ancestral regulatory mode. We further speculate that this mechanism may have involved feedback control whereby alx1, activated by VEGF3, up-regulated the expression of the cognate receptor, vegfr-10-Ig. An attractive feature of a feedback model is that the heterochronic transfer of the skeletogenic program would have only required a shift in the expression of the VEGF3 ligand, rather than separate, coordinated changes in the expression patterns of both the ligand and receptor, which are expressed in different germ layers. In the micromere-PMC lineage, although the ancestral mode of alx1 activation has been superseded by new, maternally entrained mechanisms, vegfr-10-Ig remains a target of alx1 [24,29,59]. One clear and important outcome of the heterochronic shift of the skeletogenic program was the early, robust developmental expression of vegfr-10-Ig by PMCs. This expression begins at the blastula stage, well before PMC ingression [9,22]. Our findings indicate that the early expression of the receptor allowed PMCs to outcompete BCs for VEGF3, effectively isolating BCs from the powerful influence of alx1 and allowing them to adopt an alternative regulatory state. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 10. A provisional model of skeletogenic cell type evolution in euechinoids. (A) Ancestral skeletogenic cell, present in the adult of the last common ancestor of all echinoderms. Ancient roles for alx1, vegf3, and vegfr-10-Ig in echinoderm skeletogenesis are inferred from the conserved expression of these genes in adult and embryonic skeletogenic centers in multiple echinoderm clades and from experimental perturbations of gene function in echinoid and holothuroid embryos (see references in [3]). Red lines are inferred from experimentally determined inputs of VEGF3 signaling into vegfr-10-Ig and biomineralization genes in euechinoid PMCs [9,11,13]. The brokenness of the lines indicates that intervening transcription factors have not been identified. Regulation of alx1 by VEGF signaling is hypothesized based on data presented in this study and from the restricted, ventral expression of alx1 at the late gastrula stage in euechinoids [12]. Inputs from alx1 into vegfr-10-Ig and other biomineralization genes (blue arrows) have been revealed by knockdown/overexpression of alx1 in euechinoid and asteroid embryos [24,29,59,67]. (B) During echinoid evolution, the ancestral skeletogenic gene regulatory system was transferred into mesoderm-derived cells of the larva or late embryo. This may have only required a shift in vegf3 expression in the ectoderm if a widely expressed receptor was linked to a feedback mechanism that up-regulated vegfr-10-Ig. In modern euechinoids, at postgastrula stages of development (“late embryo”), alx1, vegfr-10-Ig, and many biomineralization genes are regulated by VEGF signaling, which we suggest reflects the ancestral mode. Essentially the same regulatory machinery operates in BCs when VEGF3 is available, i.e., in PMC(−) embryos. (C) The evolution of PMCs involved the transfer of alx1 expression into the large micromere lineage (“early embryo”) by linking the activation of this gene to maternal β-catenin, its direct target, pmar1, and unequal cleavage [28,68,69]. The early, cell-autonomous activation of alx1 in the large micromere-PMC lineage resulted in the precocious expression of VEGFR-10-Ig, which in modern euechinoids sequesters VEGF3 and isolates BCs from the exclusionary influence of Alx1, allowing these cells to express an alternative regulatory state. Alx1, Aristaless-like homeobox 1; BC, blastocoelar cell; pmar1, paired-class micromere anti-repressor; PMC, primary mesenchyme cell; VEGF, vascular endothelial growth factor; VEGFR, VEGF receptor. https://doi.org/10.1371/journal.pbio.3000460.g010 Methods General methods L. variegatus embryos were obtained and cultured as previously described [30]. Microsurgical removal of PMCs, PMC transplantation, vital labeling of embryos with RITC, and whole-mount immunostaining with mAb 6a9 were carried out according to Ettensohn and McClay [15]. WMISH using digoxigenin-labeled, antisense RNA probes was carried out as described previously [24]. Treatment with axitinib was carried out according to Adomako-Ankomah and Ettensohn [11]. MO/mRNA injections Microinjection of MOs and mRNA was carried out as described by Cheers and Ettensohn [70]. The sequence of the Lv-VEGFR-10-Ig splice-blocking MO was 5′-TGATTAGGGATTGCTACTTACCTGA-3′. The Lv-VEGF3 and Lv-IgTM MOs were described previously [11,37]. Working solutions (4 mM) of MOs were loaded into microinjection needles, and approximately 2 pl were injected into each egg. VEGF3 mRNA was synthesized as described by Duloquin and coworkers [9] and injected at 2–4 mg/ml (lower concentrations were also used for dose-response studies, as shown in Fig 8). Emetine treatments. For emetine treatments, a stock solution of 100 mM emetine dihydrochloride hydrate (Sigma-Aldrich E2375) was prepared in deionized water and stored at −20 °C. Immediately before use, the stock solution was diluted 1:1,000 in ASW (final [emetine] = 100 μM). After embryos were loaded into a microsurgical chamber, the fluid in the chamber was replaced with 100 μM emetine, and microsurgery was carried out in the presence of the inhibitor. PMC(−) embryos were allowed to develop for an additional 2 hours in emetine-containing seawater before they were processed for WMISH. Injection of WGA into the blastocoel. A 10 mg/ml solution of WGA (Sigma-Aldrich Cat. No. L9640) in seawater was prepared immediately before each use and front-loaded into microinjection pipettes. Control early mesenchyme blastula–stage embryos or PMC(−) embryos were immobilized in microinjection chambers [15], and sufficient WGA solution was injected into the blastocoel of each embryo to cause visible swelling. WGA-injected embryos were marked by co-injecting a droplet of silicon oil into the blastocoel. Embryos were fixed at various time points during gastrulation and immunostained with 6a9 antibody. General methods L. variegatus embryos were obtained and cultured as previously described [30]. Microsurgical removal of PMCs, PMC transplantation, vital labeling of embryos with RITC, and whole-mount immunostaining with mAb 6a9 were carried out according to Ettensohn and McClay [15]. WMISH using digoxigenin-labeled, antisense RNA probes was carried out as described previously [24]. Treatment with axitinib was carried out according to Adomako-Ankomah and Ettensohn [11]. MO/mRNA injections Microinjection of MOs and mRNA was carried out as described by Cheers and Ettensohn [70]. The sequence of the Lv-VEGFR-10-Ig splice-blocking MO was 5′-TGATTAGGGATTGCTACTTACCTGA-3′. The Lv-VEGF3 and Lv-IgTM MOs were described previously [11,37]. Working solutions (4 mM) of MOs were loaded into microinjection needles, and approximately 2 pl were injected into each egg. VEGF3 mRNA was synthesized as described by Duloquin and coworkers [9] and injected at 2–4 mg/ml (lower concentrations were also used for dose-response studies, as shown in Fig 8). Emetine treatments. For emetine treatments, a stock solution of 100 mM emetine dihydrochloride hydrate (Sigma-Aldrich E2375) was prepared in deionized water and stored at −20 °C. Immediately before use, the stock solution was diluted 1:1,000 in ASW (final [emetine] = 100 μM). After embryos were loaded into a microsurgical chamber, the fluid in the chamber was replaced with 100 μM emetine, and microsurgery was carried out in the presence of the inhibitor. PMC(−) embryos were allowed to develop for an additional 2 hours in emetine-containing seawater before they were processed for WMISH. Injection of WGA into the blastocoel. A 10 mg/ml solution of WGA (Sigma-Aldrich Cat. No. L9640) in seawater was prepared immediately before each use and front-loaded into microinjection pipettes. Control early mesenchyme blastula–stage embryos or PMC(−) embryos were immobilized in microinjection chambers [15], and sufficient WGA solution was injected into the blastocoel of each embryo to cause visible swelling. WGA-injected embryos were marked by co-injecting a droplet of silicon oil into the blastocoel. Embryos were fixed at various time points during gastrulation and immunostained with 6a9 antibody. Emetine treatments. For emetine treatments, a stock solution of 100 mM emetine dihydrochloride hydrate (Sigma-Aldrich E2375) was prepared in deionized water and stored at −20 °C. Immediately before use, the stock solution was diluted 1:1,000 in ASW (final [emetine] = 100 μM). After embryos were loaded into a microsurgical chamber, the fluid in the chamber was replaced with 100 μM emetine, and microsurgery was carried out in the presence of the inhibitor. PMC(−) embryos were allowed to develop for an additional 2 hours in emetine-containing seawater before they were processed for WMISH. Injection of WGA into the blastocoel. A 10 mg/ml solution of WGA (Sigma-Aldrich Cat. No. L9640) in seawater was prepared immediately before each use and front-loaded into microinjection pipettes. Control early mesenchyme blastula–stage embryos or PMC(−) embryos were immobilized in microinjection chambers [15], and sufficient WGA solution was injected into the blastocoel of each embryo to cause visible swelling. WGA-injected embryos were marked by co-injecting a droplet of silicon oil into the blastocoel. Embryos were fixed at various time points during gastrulation and immunostained with 6a9 antibody. Supporting information S1 Fig. Expression of Lv-vegfr-Ig-10 at low levels in the presumptive nonskeletogenic mesoderm. Whole-mount in situ hybridization analysis of Lv-vegfr-10-Ig expression in control, early gastrula-stage embryos. (A) Lateral view. (B) Vegetal view. Expression in the wall of the archenteron (presumptive nonskeletogenic mesoderm) is indicated by arrows. Lv-vegfr-10-Ig, L. variegatus vascular endothelial growth factor receptor-10-Ig. https://doi.org/10.1371/journal.pbio.3000460.s001 (TIF) S2 Fig. BC transfating is not triggered by perturbing PMC migration. (A) Control early gastrula-stage embryo. PMCs (arrow) have dispersed from the site of ingression and are migrating along the blastocoel wall. (B) Sibling embryo, 4 hours after microinjection of WGA into the blastocoel. The PMCs (arrow) remain in a single mass at the site of ingression. The embryo is marked with an OD. (C) WGA-injected embryo 12 hours after injection, immunostained with 6a9 antibody. WGA has caused the coalescence of the PMCs into a single large mass from which numerous filopodia extend. No other 6a9(+) cells are present at the tip of the archenteron or in the blastocoel, indicating that BC transfating has not occurred. Analysis of WGA-injected, 6a9-stained embryos at multiple stages during gastrulation confirms that no BC cells transfate when PMC migration and patterning is disrupted. (D) PMC(−) embryo injected with WGA immediately after PMC removal and immunostained with antibody 6a9 after 12 hours. Numerous 6a9(+), transfated BCs are apparent, demonstrating that BCs are capable of transfating in the presence of WGA. BC, blastocoelar cell; OD, oil droplet; PMC, primary mesenchyme cell; WGA, wheat germ agglutinin. https://doi.org/10.1371/journal.pbio.3000460.s002 (TIF) S1 Data. Counts of 6a9-positive cells. Raw cell counts for all experiments presented in this paper. Each value represents the number of 6a9-positive cells in a single embryo. The data are shown graphically in the specific figure panels indicated. https://doi.org/10.1371/journal.pbio.3000460.s003 (DOCX) S2 Data. Raw data underlying Fig 7. https://doi.org/10.1371/journal.pbio.3000460.s004 (XLSX) Acknowledgments The authors thank Dr. Jennifer Guerrero-Santoro for her valuable contributions to this work.
Peptide presentation by bat MHC class I provides new insight into the antiviral immunity of batsLu, Dan;Liu, Kefang;Di Zhang, ;Yue, Can;Lu, Qiong;Cheng, Hao;Wang, Liang;Chai, Yan;Qi, Jianxun;Wang, Lin-Fa;Gao, George F.;Liu, William J.
doi: 10.1371/journal.pbio.3000436pmid: 31498797
Introduction In recent years, emerging and re-emerging viral diseases with high mortality have continuously posed serious threats to human health [1–3]. In etiologic studies with convincing evidence, a series of such fatal diseases in humans have been confirmed or hypothesized to be caused by bat-borne viruses such as Hendra virus (HeV), Ebola virus (EBOV), Marburg virus, and coronaviruses, including severe acute respiratory syndrome coronavirus (SARS-CoV) and Middle East respiratory syndrome coronavirus (MERS-CoV) [4–10]. Similarly, fetal diseases in livestock have also been associated with emerging viruses of bat origin, such as the fatal disease outbreak of pigs in China, which was found to be caused by a novel coronavirus, swine acute diarrhea syndrome coronavirus (SADS-CoV), from bats [11,12]. Furthermore, it was shown that influenza-like viruses, termed H17N10 and H18N11 circulating among bats in Central America, may act as an ancient influenza reservoir [13,14]. It is well accepted now that bats harbor an exceptionally high proportion of zoonotic viruses with interspecies transmission potential [15] that have the potential to become virulent pathogens for humans. However, studies from wild or experimental bats indicate that most of these lethal viruses in humans and other mammals cause only asymptomatic infection in bats, suggesting a potentially special immune system in bats that is different from most other mammals [16,17]. Studies of comparative genomics and transcriptomics confirm that the critical components of the innate and adaptive immune system are conserved and functional in bats [18,19]. However, a large number of bat-unique characteristics related to immunity and antiviral responses have recently been identified. The most notable immune features involve the absence of key natural killer (NK) cell receptors and dampened cell signaling of type I interferons (IFNs) [18–20]. Meanwhile, stimulator of interferon genes (STING), an essential adaptor protein in multiple DNA sensing pathways, has a substitution in bats, compared with other mammals, that leads to decreased IFN activation [21]. Meanwhile, bats also possess special features to maintain an effective immune state. The genome analysis of Rousettus aegyptiacus revealed a dramatic difference from their functional gene counterparts in other mammals [19]. Indeed, bat IFNs and some IFN-stimulated genes are constitutively transcribed or maintain detectable expression levels in the absence of stimulation [22–24]. These characteristics may allow bats to fine-tune innate defense responses against insults by viral, bacterial, or host cytosolic DNA while avoiding excessive inflammation. However, the adaptive immune system in bats is less well studied. Major histocompatibility complex (MHC) class I molecules (MHC I) present antigen peptides to the surface of antigen presenting cells (APCs) to active T cells through interaction with T-cell receptors (TCRs), which play pivotal roles in antiviral defense [25,26]. Epitopes or peptides are accommodated in the six pockets (A–F) of peptides/MHC I complexes, which were initially defined in humans [27]. Pocket A anchors the amine group of the amino terminal residue of the bound peptide, pocket B binds the side chain of peptide residue two, and pocket F accommodates the side chain of the carboxyl terminal residue [28]. In bats, the partial map of the Pteropus alecto MHC I region shows that MHC I genes are highly condensed and present within only one of the three highly conserved class I duplication blocks [29]. Genomic analyses of R. aegyptiacus demonstrate an expanded and diversified set of MHC I genes, with MHC I genes found outside the canonical region [19]. Sequences analyses, based on bat MHC I genes identified thus far, show that many of these bat MHC I molecules have a 3– or 5–amino acid (aa) insertion in the α1 domain compared with other mammals [20,29]. In the R. aegyptiacus genome, it is shown that 11 of the 12 MHC class I loci identified display the 3-aa insertion and only one locus without the insertion. Interestingly, one of the 11 MHC class I loci with 3-aa insertion is located in the canonical MHC alpha loci, indicating that both the MHC class I molecules with and without insertion can present the canonical binding surface [19]. Furthermore, the binding peptide motif of MHC I Ptal-N*01:01 derived from P. alecto has been identified. It displays a preference for peptides with Pro at their C terminus, which has never been seen in MHC I proteins of any other vertebrates [30]. However, the molecular basis for the peptide binding and presentation by bat MHC I remains unclear. In this study, we screened bat MHC I Ptal-N*01:01–binding peptides from different bat-borne viruses (HeV, EBOV, MERS-CoV, and H17N10) and determined the structures of bat MHC class I complexed with these viral peptides. Unusual peptide presentation by bat MHC I with was demonstrated, which may help to understand the greater capacity of bats to coexist with a variety of viruses, from the perspective of adaptive immunity. Results Unusual characteristics of the bat MHC class I peptide binding groove Previous work identified 56 bat MHC I genes from more than seven different species of bats on different continents. Each of these bat MHC I genes has the typical MHC I domains as in other mammals. However, examination of the retrieved set of bat MHC I sequences from GenBank revealed several unusual features. First, bat MHC I genes contain a 3- or 5-aa insertion within their peptide binding groove (PBG) compared with those from a variety of other mammals (i.e., between Trp51 and Ile52 of human HLA-A*0201) (Fig 1A and 1B). Within the bat MHC I genes, 30.36% possess a 3-aa insertion and 57.14% have the 5-aa insertion (Fig 1A, S1 Fig and S1 Table). The 5-aa insertion is unique to bat sequences, while the 3-aa insertion is present in both bats and some marsupials (82.61% of the opossum MHC I, 50% of the koala MHC I, and 100% of both tammar wallaby and Tasmanian devil MHC I contain the 3-aa insertion). All of the higher mammals, such as humans, nonhuman primates (NHPs), mouse, and horse, lack any insertion at this site (Fig 1A and S2A Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Unusual characteristics of bat MHC class I genes. (A) The proportions of MHC class I alleles with 3- or 5-aa insertions in bats, marsupials (opossum, koala, tammar wallaby, and Tasmanian devil), and higher mammals (human, NHP, mouse, and horse). The proportions of MHC class I alleles with the 3-aa insertion, 5-aa insertion, and no insertion are represented with orange, yellow, and cyan columns, respectively. The deletions (mainly including a 3-aa deletion) or insertions (mainly 1-aa insertions) other than the 3- and 5-aa insertions are termed as “others” in gray columns. The numerical data are included in S1 Data. (B) Structure-based sequence alignment of Ptal-N*01:01 and other representative (bats, marsupials, and higher mammals) MHC I molecules covering the residues from positions 49 to 66 (as in Ptal-N*01:01). The full information of the MHC I molecules of these species was listed in S1 Fig and S1 Table. Coils above the sequences indicated α-helices. Residues highlighted in red are completely conserved, and residues in blue boxes are highly (80%) conserved, with consensus amino acids in red. The residues at position 59 and 65 are shown in yellow. Special insertion positions in Ptal-N*01:01 are marked with red arrows above the sequences. The sequence alignment was generated with ClustalX and ESPript. To the right of the sequences, MHC class I alleles with the 3-aa insertion, 5-aa insertion, no insertion, and others are labeled with orange, yellow, cyan, and gray boxes, respectively. (C) The proportions of negatively charged residue Asp59 or Glu59 (“59D/E,” cyan columns), the pairing of the Asp59 and Arg65 (“D59+R65,” yellow columns), and the pairing of the negative-positive charged residues, termed as charge-matching at the two positions (“+/−,” purple columns) at the corresponding locations of MHC I in bats, marsupials, and higher mammals. (D) Statistical analysis of the 3-aa insertion and no 3-aa insertion alleles (no insertion and 5-aa insertion) in bats, respectively. Fisher exact test or the chi-squared test was used for the statistical analyses. **p < 0.01. aa, amino acid; MHC, major histocompatibility complex; NHP, nonhuman primate. https://doi.org/10.1371/journal.pbio.3000436.g001 Another feature of bat MHC I is the higher prevalence of a negatively charged residue at position 59 and a positively charged residue at position 65, as well as their pairing (according to the Ptal-N*01:01 residue code) (Fig 1B and S2A Fig). For the MHC I molecules of the higher mammals, the residue at the position corresponding to 59 (position 56 of human, NHP, mouse, and horse MHC I) is a highly conserved Gly. However, 71.43% of bat MHC Is have an Asp59 or Glu59 (59D/E) at this position (Fig 1C). The pairing of Asp59 and Arg65 (D59+R65) also occurs at a high fraction (35.71%) in bat MHC I compared with the MHC I of humans (0%), NHP (0%), mouse (0%), and horse (0%). Actually, the pairing of the charged residues at these two positions include different types: the negatively charged residues Asp/Glu at 59 paired with positively charged residues Arg/Lys at 65, and even positively charged residue Arg/Lys at 59 paired with Asp/Glu at 65 (as gene EPQ18390.1 in Myotis brandtii). Also, the pairing of the charged residues, termed as charge matching at the two positions (+/−), occurs at an even higher fraction for bat MHC I (39.29%). Interestingly, these features of bat MHC I are also prevalent in marsupials (Fig 1C). To determine whether the insertion has any correlation with the unusual substitutions at positions 59/65, we analyzed the fractions of 59D/E, D59+R65, and the +/− charge matching in the bat MHC I genes with the 3-aa insertion. We found a significantly higher fraction of 59D/E (100%), D59+R65 (64.71%), and the +/− charge matching (70.59%) in the bat MHC I genes with the 3-aa insertion compared with the corresponding fractions of other bat MHC I genes (no insertion and 5-aa insertion) (59% for 59D/E, 23.08% for D59+R65, and 28.21% for the +/− charge matching) (Fig 1D). Collectively, these distinct features suggested an unusual PBG of bat MHC I, which may affect peptide binding and presentation. Ptal-N*01:01–binding peptides from known emerging viruses To verify the peptide binding motif of Ptal-N*01:01 and to screen the potential bat MHC I T-cell epitopes from recently emerging and re-emerging viruses with bats as their potential reservoir, we predicted Ptal-N*01:01–binding peptides from EBOV, MERS-CoV, and H17N10/H18N11 (S2 Table). The previously determined Ptal-N*01:01–binding peptides derived from HeV were also included as a positive control [30]. The binding capacity of these peptides were evaluated by their ability to facilitate the in vitro renaturation of Ptal-N*01:01. Generally, the peptides with higher binding capability to MHCs would have a higher production of the heterotrimer MHC complexes but lower production of β2-microglobulin (β2m). Seven peptides from MERS-CoV, five from EBOV, one from H17N10, and the two HeV-derived peptides helped Ptal-N*01:01 naturally refold (Fig 2). The lengths of the Ptal-N*01:01–binding peptides cover a range from octamers to tridecamer. These peptides possess an Asp at the P1 position (with Gln in one peptide: MERS-CoV-S4), an aromatic aa (Phe or Tyr) at the P2 position, and a Pro or Leu at the PΩ position (C terminus of the peptide) (S2 Table). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. The identification of emerging and re-emerging virus-derived peptides binding to Ptal-N*01:01. The binding of peptides derived from HeV (A), EBOV (B), MERS-CoV (C), and H17N10 influenza-like virus (H17N10) (D) with Ptal-N*01:01 were evaluated by co-refolding. Co-refolding without any peptide was termed as the negative control (No pep), as curves in gray color. After properly refolding, the high-absorbance peaks of the correctly refolded MHC I with the expected molecular mass of 45 kDa were eluted at the estimated volume of 16 mL on a Superdex Increase 200 10/300 GL column. The profile is marked with the approximate positions of the molecular mass standards of 75.0, 44.0, and 13.7 kDa. Inset, reduced SDS–PAGE gel (15%) of Ptal-N*01:01/HeV1 complex for peak 1 (P1), peak 2 (P2), and peak 3 (P3). Lane M contains molecular-mass markers (labeled in kDa). P1, P2, and P3 represent the aggregated heavy chain, the correctly refolded heterotrimer Ptal-N*01:01 complex (45 kDa), and the extra β2m, respectively. β2m, β2-microglobulin; HeV, Hendra virus; EBOV, Ebola virus; MERS-CoV, Middle East respiratory syndrome coronavirus; MHC, major histocompatibility complex; P1, peak 1; P2, peak 2; P3, peak 3. https://doi.org/10.1371/journal.pbio.3000436.g002 The structures of Ptal-N*01:01 verify the motifs of the binding peptides derived from pathogens The binding peptide motif of bat MHC I Ptal-N*01:01, with the negative-charged Asp at the P1 position and a Pro at the PΩ position, is indeed extremely rare in class I binding peptide repertoires of humans and other mammals. To verify the uncommon peptide presentation features of bat MHC I, we determined the crystal structures of Ptal-N*01:01 in complex with the two HeV-derived peptides, HeV1 (DFANTFLP) and HeV2 (DYINTNVLP), two EBOV-derived peptides, EBOV-NP1 (DFQESADSFL) and EBOV-NP2 (DFQESADSFLL), one H17N10-derived peptide, H17N10-NP (DFEKEGYSL), and one MERS-CoV-derived peptide, MERS-CoV-S3 (DFTCSQISP) (Table 1). Download: PPT PowerPoint slide PNG larger image TIFF original image Table 1. X-ray data processing and refinement statistics. https://doi.org/10.1371/journal.pbio.3000436.t001 The overall structures of Ptal-N*01:01 display the common characteristics of classical MHC I molecules in other mammals, with the extracellular region of the heavy chain folding into three different domains. The α1 and α2 domains construct a typical PBG that contains two α1-helices and eight β-sheets, and the α3 domain and β2m display typical immunoglobulin (Ig) domains and underpin the peptide binding domain (Fig 3A). The all-atoms superimposition of Ptal-N*01:01/HeV1 onto the other five structures demonstrates a similar overall conformation, with root mean square deviations (RMSDs) of 0.344–0.517 Å (Fig 3B). The superimposition of Ptal-N*01:01/HeV1 onto human MHC I HLA-A2 and mouse MHC I H-2Kd generated RMSDs of 1.198 and 1.119 Å, respectively (Fig 3C). The most distinct differences between Ptal-N*01:01 and the MHC I from other vertebrates are located in the N terminus of the PBG, with an extension of the α1-helix in Ptal-N*01:01 (Fig 3D). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Overview of Ptal-N*01:01/peptide structures. (A) Overview of the structure of the Ptal-N*01:01/HeV1. The peptide HeV1 is presented as yellow sticks in the peptide-binding cleft. The heavy chain of Ptal-N*01:01 and bat β2m are shown as green and cyan cartoons, respectively. (B) The superimposition of different Ptal-N*01:01 complexes was performed using the determined structures: Ptal-N*01:01/HeV1, Ptal-N*01:01/HeV2, Ptal-N*01:01/EBOV-NP1, Ptal-N*01:01/EBOV-NP2, Ptal-N*01:01/H17N10-NP, and Ptal-N*01:01/MERS-CoV-S3. The peptides were omitted. (C) Structural alignment of MHC I heavy chains and β2m exhibited a similar overall conformation of Ptal-N*01:01 (green, peptide HeV1), HLA-A*0201 (yellow, PDB code: 3I6G), and H-2Kd (purple, PDB code: 5GR7). (D) The superimposition of the α1α2 domains of Ptal-N*01:01 (green) with other vertebrate MHC I molecules: human HLA-A*0201 (3I6G), macaque Mamu-A*02 (3JTT), swine SLA-1*0401 (3QQ4), equine Eqca-N*00602 (4ZUU), bovine N*01801 (3PWV), canine DLA-88*50801 (5F1I), and murine H2-Kd (5GR7) (all in white). The distinct conformations at the N terminus between the α1-helices of Ptal-N*01:01 (green) and the MHC I from other vertebrates (white) are labeled by a pink circle. The peptide HeV1 in Ptal-N*01:01 is presented in yellow loops. (E) The structural alignment of Ptal-N*01:01 (peptide HeV1) when renatured with bat β2m (green) and human β2m (yellow). HeV1 showed similar conformation in the binding groove. (F) The black box in the dashed line indicates the similar conformations of residues in the interface of the bat (green) and human β2m (yellow) binding to α1α2 domains of Ptal-N*01:01 heavy chain. (G) Similar binding of Ptal-N*01:01 heavy chain to bat and human β2m. Trp60 in bat β2m (cyan) and human β2m (orange) binds to Gln99 and Asp125 in Ptal-N*01:01 when forming a complex. (H) The alignment of monomer bat β2m (blue) with HeV-1/Ptal-N*01:01-batβ2m (green) complexes. Resides Ser55 to Tyr63 in monomer bat β2m (light blue) shift minorly after forming a complex (cyan). β2m, β2-microglobulin; EBOV, Ebola virus; HeV, Hendra virus; MERS-CoV, Middle East respiratory syndrome coronavirus; MHC, major histocompatibility complex; PDB, Protein Data Bank. https://doi.org/10.1371/journal.pbio.3000436.g003 To elucidate whether the peptide-presenting features of a bat MHC I molecule can be influenced by binding to human β2m, we solved the structure of the Ptal-N*01:01 heavy chain complexed with human β2m (Ptal-N*01:01-h) at a resolution of 1.6 Å (Table 1). Comparing HeV1/Ptal-N*01:01 renatured with bat β2m and human β2m, the structural conformations of both HeV1 peptides in the two structures are also similar, with an RMSD of 0.228 Å in the two binding grooves (Fig 3E). In addition, the overall structures are quite similar, with the RMSD of 0.564 Å of all atoms (Fig 3F). And the key residues binding to α1α2 domains and α3 domain of the Ptal-N*01:01 heavy chain are highly conserved in human and bat β2m (S2B Fig). Further analysis shows Trp60 in bat β2m binds to Gln99 and Asp125 in Ptal-N*01:01 when forming a complex, which is also conserved in human β2m when binding to Ptal-N*01:01 (Fig 3G). It indicates that the structure of the peptide loaded in the groove of Ptal-N*01:01 was not affected by the substitution of the β2m subunit. We also determined the structure of monomer bat β2m without MHC I (Table 1). The structure alignment of bat β2m monomer with β2m subunit in MHC complexes indicates a minor conformational shift of the loop resides Ser55 to Tyr63 in bat β2m after forming a complex (Fig 3H). Although having different lengths, the 8-mer peptide HeV1, 9-mers HeV2 and H17N10-NP, 10-mer EBOV-NP1, and 11-mer EBOV-NP2 in the PBG of Ptal-N*01:01 all display an M-shaped conformation, with P2 and PΩ residues as the primary anchors and the P5 or P6 residue as the secondary middle anchors (Fig 4A–4F and S3A–S3E Fig). The P1 Asp of all of the peptides adopts a rigid conformation upward to the α1-helix of the heavy chain. In the structure of the Ptal-N*01:01/MERS-CoV-S3 complex, the conformation of the P3–P8 residues of the peptide could not be determined due to their poor electron densities (S3F Fig), revealing a flexible conformation in the middle region of this MERS-CoV-S3 peptide. However, the other three residues (P1-Asp, P2-Phe, and PΩ-Pro) with electron densities available adopt similar conformations as in the other five structurally determined peptides. As a 9-mer peptide, MERS-CoV-S3 possesses a Glu at P6, corresponding to the P6-Asn in 9-mer HeV2, which is the secondary anchor residue. The larger Glu may not be able to locate in the PBG, which leads to a flexible conformation of the MERS-CoV-S3. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. The structural conformation and motif of the binding peptides presented by Ptal-N*01:01. (A-E) The overall conformation of Ptal-N*01:01–binding peptides HeV1 (A) and HeV2 (B) from HeV, H17N10-NP (C) from H17N10 influenza-like virus, and EBOV-NP1 (D) and EBOV-NP2 (E) from EBOV. (F) The superimposition of presented peptides HeV1, HeV2, H17N10-NP, EBOV-NP1, and EBOV-NP2 by Ptal-N*01:01. The peptides are aligned according to the superimposition of the α1 and α2 domains of the five structures of Ptal-N*01:01. The conserved conformations of residues in the P1, P2, and PΩ positions are shown in yellow ellipses. Most conformational distinctions are located in the central region of the peptides, marked by a yellow rectangle. (G-J) The capacity of HeV1 (G), HeV2 (I), and their Ala substitutions for binding to Ptal-N*01:01 was evaluated by in vitro refolding. The thermostabilities of Ptal-N*01:01 with HeV1 (H) and HeV2 (J) peptides and their Ala substitutions were tested by circular dichroism (CD) spectroscopy. The numerical data are included in S1 Data. The curves for the unfolded fractions were determined by monitoring the CD value at 218 nm. The temperature was increased by 1°C/minute. Shown are the fitting data to the denaturation curves using the Origin 8.0 program (OriginLab). The Tms of different peptides are indicated by the dashed gray lines at the 50% fraction unfolded. β2m, β2-microglobulin; CD, circular dichroism; EBOV, Ebola virus; HeV, Hendra virus; Tm, midpoint transition temperature. https://doi.org/10.1371/journal.pbio.3000436.g004 To further validate the role of the unusual binding peptide motif of bat MHC I Ptal-N*01:01, Ala mutations at residues P1, P2, PΩ, and the middle (P5 or P6) positions of peptides HeV1 and HeV2 were evaluated by refolding assays (Fig 4G and 4I). The thermal stability of the resulting heterotrimers was monitored by circular dichroism (CD) spectroscopy (Fig 4H and 4J, S3 Table). Substitution of Ala at P1 (HeV2-D1A), P2 (HeV2-Y2A), and P9 (HeV2-P9A) of peptide HeV2 led to nearly no refolding for Ptal-N*01:01, whereas Ala at P1 (HeV1-D1A), P2 (HeV1-F2A), and P8 (HeV1-P8A) of peptide HeV1 still supported refolding but with significantly lower stability than the original peptide. In contrast, substitution of Ala at P5 of HeV1 (HeV1-T5A) and at P6 of HeV2 (HeV1-N6A) led to similar yield of refolded heterotrimers, indicating a similar stability compared with the original peptides. Thus, through these analyses, the P1 anchor of Ptal-N*01:01–binding peptides seems to be as significant as the P2 and PΩ anchors. The insertion induces uncommon peptide P1 anchoring in bat MHC I Bat Ptal-N*01:01 possesses the Met-Asp-Leu insertion within the N terminus of its α1-helix (between residues 51 and 52 of HLA-A*0201). Compared with the structures available for the MHC I of other mammals, such as humans and mouse, the Ptal-N*01:01 structure displays an extension of the α1-helix of the PBG (Fig 5A). The 3-aa insertion pushes residue Asp59 closer to the N terminus of the binding peptide, which leads to the extension of the negatively charged side chain of Asp59 into the PBG (Fig 5B). The bat MHC I residue Asp59 participates in the formation of the A pocket (S4 Fig). Detailed analysis indicated that Asp59, Arg65, and the P1-Asp of the peptides in all six structures of Ptal-N*01:01 form a triangular network of hydrogen bonds (Fig 5C and S5 Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. The 3-aa (Met52Asp53Leu54) insertion contributes an uncommon P1 anchoring of bat MHC I. (A) The structural superimposition of the N terminus of PBGs from different MHC I molecules. Trp51–Gln61 of Ptal-N*01:01 in green and structurally corresponding residues Trp51–Glu58 of HLA-A*0201 (3I6G), macaque Mamu-A*02 (3JTT), swine SLA-1*0401 (3QQ4), equine Eqca-N*00602 (4ZUU), bovine N*01801 (3PWV), canine DLA-88*50801 (5F1I), and murine H2-Kd (5GR7) in white. The 3-aa Met52Asp53Leu54 insertion in bat MHC I is in purple. (B) The structural superimposition of Ptal-N*01:01 and its mutant Ptal-N*01:01(-3aa) (with Met52Asp53Leu54 deletion). The residues Trp51–Gln61 of Ptal-N*01:01 (green) and Trp51–Gln58 of Ptal-N*01:01(-3aa) (yellow) are presented in surface representation. The heavy chains of these two molecules are shown as white cartoons. The P1 positions of the peptides are shown as sticks. (C) The hydrogen bond network in the A pocket of Ptal-N*01:01/HeV1. Residues Asp59, Arg65, and P1-Asp of HeV1 are shown as blue sticks and form a triangular network of hydrogen bonds in blue dashed lines. (D) Different conformations of the N terminus of the α1-helices of Ptal-N*01:01 (green) and mutant Ptal-N*01:01(-3aa) (yellow). (E) The conformational shift of Asp, Arg, and P1-Asp of HeV1 in the structures Ptal-N*01:01/HeV1 (green) and Ptal-N*01:01(-3aa)/HeV1 (yellow), as indicated by purple arrows. (F) The N-terminal conformation of the α1-helix and hydrogen bonds (red dashed lines) within the structure of mutant Ptal-N*01:01(-3aa)/HeV1. Asp, Arg, and HeV1 P1-Asp are shown as purple sticks. (G) Schematic diagram of the construction of the mutant Ptal-N*01:01(-3aa) with a Met52Asp53Leu54 deletion compared with the wild-type Ptal-N*01:01. (H-I) The capabilities of peptide HeV1 presented by Ptal-N*01:01 and mutant Ptal-N*01:01(-3aa) were evaluated by in vitro refolding (H) and CD spectroscopy (I). The numerical data are included in S1 Data. (J) Schematic diagram of the construction of the mutant HLA-A2M with a Met52Asp53Leu54 insertion and D59+R65 compared with the wild-type HLA-A2. (K-L) The capabilities of peptide DL9 (the first amino acid of GL9 change to D) presented by HLA-A2 and mutant HLA-A2M were evaluated by in vitro refolding (K) and CD spectroscopy (L). aa, amino acid; β2m, β2-microglobulin; CD, circular dichroism; DL9, DILGFVFTL; GL9, GILGFVFTL; HeV, Hendra virus; MHC, major histocompatibility complex; PBG, peptide binding groove; P1, position 1; Tm, midpoint transition temperature; WT, wild type. https://doi.org/10.1371/journal.pbio.3000436.g005 To further investigate the role of the 3-aa insertion in the peptide binding and presentation of Ptal-N*01:01, we constructed the Ptal-N*01:01(-3aa) mutant, which deleted the 3-aa insertion (Fig 5G). The structure of Ptal-N*01:01(-3aa)/HeV1 displayed a shortened α1-helix that is similar to the human HLA-A*0201 (Fig 5D). Meanwhile, although the salt bridge was still observed between Arg65 of Ptal-N*01:01 and the P1-Asp, both the Arg65 and P1-Asp displayed a conformational shift (Fig 5E). The triangular network of the hydrogen bonds between Asp59 (Residue Asp56 in the mutant), Arg65 (Residue Arg62 in the mutant), and the P1-Asp of HeV1 within the structure of the wild-type Ptal-N*01:01 was broken (Fig 5F). One of the hydrogen bonds between the two residues Arg65 (Residue Arg62 in the mutant) and the P1-Asp of HeV1 was lost, and the remaining two hydrogen bonds extended from 2.97 to 3.23 Å and 2.67 to 2.72 Å, respectively. We also investigated whether the 3-aa deletion of Ptal-N*01:01 influenced binding ability to the peptides. Ptal-N*01:01(-3aa) was still renatured in the presence of peptide HeV1 but with a much lower yield of the heterotrimer complex (Fig 5H). CD spectroscopy also indicated a weaker binding of Ptal-N*01:01(-3aa) with the peptide (Fig 5I). To further verify the unusual peptide presentation and preference for peptides with a P1-Asp in bat Ptal-N*01:01, we constructed the HLA-A2M mutant, which has a 3-aa insertion and the charge matching residues at positions 59/65 based on HLA-A*02:01(Fig 5J). Both the refolding assay and CD spectroscopy indicated a stronger binding of DL9 (G1D mutant at P1 of peptide GL9) with the HLA-A2M compared with HLA-A*02:01, and the HLA-A*02:01 has a higher binding capacity to GL9 than DL9 (Fig 5K and 5L). The unusual preference of Pro as the PΩ anchor of Ptal-N*01:01–binding peptides Although Ptal-N*01:01 can bind peptides with Leu as the PΩ anchor, the binding peptides can also possess Pro at this position. This is uncommon in the peptides bound by MHC I molecules from other mammals. The structures of Ptal-N*01:01 with the 9-mer peptides HeV2 (DYINTNVLP) and H17N10-NP (DFEKEGYSL) showed that both of the peptides adopt similar overall conformations as the HLA-A*0201–presented 9-mer peptide GL9 (GILGFVFTL) with a Leu at the PΩ site (Fig 6A). However, to the best of our knowledge, no peptide motif with a Pro at the PΩ position has been reported for HLA-A*0201. Detailed comparative analysis of the F pockets of Ptal-N*01:01 and HLA-A*0201 revealed that the F pocket of Ptal-N*01:01 is shallow but with a wide opening (Fig 6B–6E). Corresponding to Asp77 and Tyr116 of HLA-A*0201, Ptal-N*01:01 possesses smaller residues Gly80 and Leu119, respectively. The Ptal-N*01:01–specific Gly80 in Ptal-N*01:01 is small, which provides more space to accommodate the pyrrolidine ring of PΩ-Pro. In addition, although both HLA-A*0201 and Ptal-N*01:01 have a similar Lys residue in their corresponding positions (146 for HLA-A*0201 and 149 for Ptal-N*01:01), the conformation of the Lys149 of Ptal-N*01:01 is shifted to the C terminus of the PBG, which leaves more space for the open mouth of the F pocket in Ptal-N*01:01. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. The extraordinary preference for Pro as the PΩ anchor of Ptal-N*01:01–binding peptides. (A) The superimposition of 9-mer peptides HeV2 (cyan) and H17N10-NP (green) presented by Ptal-N*01:01 and GL9 (orange) presented by HLA-A*0201 (PDB code: 3I6G). The heavy chains are represented as gray cartoons. The PΩ anchors of peptides are shown as sticks, and the side chains of other residues of the peptides are omitted. (B) Distinct F pockets of Ptal-N*01:01 and HLA-A*0201. The heavy chains are represented as gray loops. Gly80, Leu119, and Lys149 of Ptal-N*01:01 are represented as cyan sticks. The corresponding residues of HLA-A*0201 are shown as orange sticks. The given entrances for F pockets of Ptal-N*01:01 and HLA-A*0201 are labeled by purple and deep blue circles, respectively. (C-D) The side chains of PΩ-Pro from peptide HeV2 (C) and PΩ-Leu from peptide H17N10-NP (D) insert into the F pockets of Ptal-N*01:01. The F pocket of Ptal-N*01:01 is indicated in surface representation with a yellow entrance and purple bottom. (E) The side chain of residue PΩ-Leu from peptide GL9 inserts into the F pocket of HLA-A*0201. The F pocket of HLA-A*0201 is indicated by the representation of a brown entrance with a light blue bottom. (F-G) The stabilities of HLA-A*0201 complexed with peptide GL9 and its mutant GL9-L9P (with a Pro at the P9 position) were evaluated by in vitro refolding (F) and CD spectroscopy (G). The numerical data are included in S1 Data. (H) The relatively lower mutation frequency of amino acid Pro (red bar) as Try and Gly among the 20 component amino acids in the proteins of MERS-CoV and SARS-CoV. The mutation frequency = the number of overall mutations for each amino acid/(the number of occurrences of the amino acid in the reference sequence×total number of sequences). The proteomes of 1,000 MERS-CoV genomes and SARS-CoV genomes were retrieved from GenBank, respectively. The detailed information is included in S2 Data. CD, circular dichroism; GL9, GILGFVFTL; HeV, Hendra virus; MERS-CoV, Middle East respiratory syndrome coronavirus; PDB, Protein Data Bank; SARS-CoV, severe acute respiratory syndrome coronavirus; Tm, midpoint transition temperature. https://doi.org/10.1371/journal.pbio.3000436.g006 To verify the allele-specific preference of Ptal-N*01:01 for peptides with Pro at the PΩ site, we examined the binding ability of HLA-A*0201 to a mutated GL9 peptide, GL9-L9P, with a Pro at P9 (Fig 6F and 6G). We found GL9-L9P has a weaker capacity to help the HLA-A*0201 refold, and the generated heterotrimer complex has lower midpoint transition temperature (Tm) (38.1°C) compared with the wild-type peptide GL9 (55.8°C) with a Leu at PΩ. Thus, although Pro at PΩ may also act as a suboptimal anchor for HLA-A*0201, Ptal-N*01:01 uses Pro as one of its optimal PΩ anchors. The analysis of the 20 different component amino acids for the proteins from the MERS-CoV and SARS-CoV indicated that Pro possesses a relatively low mutation rate compared with the other amino acids (Fig 6H). The mutation rate of Pro is only higher than Trp and Gly, which are the largest and smallest residues, respectively. It indicates that, as special residues Trp and Gly, the residue Pro may also keep the natural conformation and function of viral proteins. In other words, structural constraints favoring conservation of Pro in certain positions of proteins may operate to preserve viral protein conformation and function. Thus, its mutation rate is restricted. The selection of Pro as the anchor residue of bat MHC I Ptal-N*01:01–presented peptides may restrict the viral mutation pushed by T-cell immunity and accelerate virus clearance. The deep B pocket of bat MHC I has a different orientation The P2 positions of Ptal-N*01:01–binding peptides are predominantly aromatic amino acids such as Tyr and Phe as the anchor, which is common in human HLA-A*2402 or mouse H-2Kd. However, when we superimposed the structures of Ptal-N*01:01 onto the previously determined structures of HLA-A*2402 and H-2Kd, we found that the P2 anchors of Ptal-N*01:01 protrude in a different direction (Fig 7A). The HLA-A*2402– and H-2Kd–presented peptides have a Tyr or Phe pointing to the C terminus of the PBG, while the Tyr or Phe of Ptal-N*01:01–binding peptides swing toward the N terminus of the PBG (Fig 7C and 7D). Comparison of the amino acids lining the B pockets of these MHC I molecules from different mammals demonstrated that Ptal-N*01:01 possesses a Tyr9 (compared with the small residues Ser9 in HLA-A*2401 or Val9 in H-2Kd), which takes up space and can push a P2-Tyr or -Phe to the other direction (Fig 7B). Furthermore, compared with the residues Met45 in HLA-A*2402 or Phe45 in H-2Kd, the Ptal-N*01:01–specific Ala45 leaves a large space to accommodate the P2-Tyr or -Phe. Indeed, detailed analyses showed that the P2-Tyr of Ptal-N*01:01–binding peptides form hydrogen bonds directly with the main chain of the β-sheet on the floor of the PBG. In contrast, the P2-Tyr of HLA-A*2402–restricted peptides bind to the side chain of His70 on the α1-helix (Fig 7B). Sequence superimposition of the bat MHC I with other mammal MHC Is indicated that Ala45 is prevalent (41%) among bat MHC I molecules but is never seen in human and mouse MHC I (S2A Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 7. The B pocket of bat MHC I with an uncommon orientation. (A) Structural superimposing of Ptal-N*01:01 with HLA-A*2402 and H-2Kd. The HLA-A*2402– (PBD code: 3I6L and 5WXD) and H-2Kd–presented peptides (PBD code: 5GR7) have a P2-Tyr or -Phe (blue sticks) pointing to the C terminus of the PBG, while the P2-Tyr or -Phe (green sticks) of Ptal-N*01:01–binding peptides (HeV1 and HeV2) swing to the N terminus of the PBGs. (B) Different conformations of P2 anchors (shown in sticks and spheres) of Ptal-N*01:01/HeV2 and HLA-A*2402. The major residues of the B pockets of Ptal-N*01:01/HeV2 and HLA-A*2402 (PBD code: 5WXD) are shown as green and blue sticks, respectively. The hydrogen bonds are shown as red dashed lines. (C) The B pocket of Ptal-N*01:01/HeV2 is shown as a semitransparent surface, with Tyr9 and Ala45 of Ptal-N*01:01 shown as sticks under the surface. Ptal-N*01:01 heavy chain is shown as green cartoons under the semitransparent surface. (D) The B pocket of HLA-A*2402 (PBD code: 5WXD) is shown as a semitransparent surface. The residues Ser9 and Met45 of HLA-A*2402 are shown as sticks under the surfaces. The different orientation of the P2 residue (blue sticks) of HLA-A*2402–presented peptide is compared with the corresponding residue of Ptal-N*01:01–presented peptide HeV2 in panel C. HeV, Hendra virus; MHC, major histocompatibility complex; PBD, peptide binding groove; PBG, peptide binding groove; P2, position 2. https://doi.org/10.1371/journal.pbio.3000436.g007 Ptal-N*01:01 does not bind to long peptides in an N-terminal extended manner Previously, it was indicated that Ptal-N*01:01 has a special preference for the binding of long peptides, together with the common 8–10-mer peptides in other mammal MHC I molecules. To further elucidate whether the 3-aa insertion of Ptal-N*01:01 has an impact on the preference for long peptides through an N-terminal extension manner, we synthesized 20 long peptides (11-mer to 15-mer) that were previously eluted from Ptal-N*01:01–expressing cells (S4 Table) [30]. None of these peptides could facilitate Ptal-N*01:01 refolding in vitro (S6A Fig). Considering that these 20 peptides may not have the typical motif of Ptal-N*01:01–binding peptides, we also synthesized naturally N-terminally extended peptides based on the Ptal-N*01:01–binding peptides from HeV1, MERS-CoV-S7, and EBOV-NP1 (S5 Table). Although these three peptides have a typical motif of Ptal-N*01:01–binding peptides, the N-terminal extension led to a failure in peptide binding (S6B Fig and S5 Table). These data indicate that the 3-aa insertion into Ptal-N*01:01 leads to a more restrictive binding peptide selection for Ptal-N*01:01 but not an extension to longer peptides via the N terminus. Unusual characteristics of the bat MHC class I peptide binding groove Previous work identified 56 bat MHC I genes from more than seven different species of bats on different continents. Each of these bat MHC I genes has the typical MHC I domains as in other mammals. However, examination of the retrieved set of bat MHC I sequences from GenBank revealed several unusual features. First, bat MHC I genes contain a 3- or 5-aa insertion within their peptide binding groove (PBG) compared with those from a variety of other mammals (i.e., between Trp51 and Ile52 of human HLA-A*0201) (Fig 1A and 1B). Within the bat MHC I genes, 30.36% possess a 3-aa insertion and 57.14% have the 5-aa insertion (Fig 1A, S1 Fig and S1 Table). The 5-aa insertion is unique to bat sequences, while the 3-aa insertion is present in both bats and some marsupials (82.61% of the opossum MHC I, 50% of the koala MHC I, and 100% of both tammar wallaby and Tasmanian devil MHC I contain the 3-aa insertion). All of the higher mammals, such as humans, nonhuman primates (NHPs), mouse, and horse, lack any insertion at this site (Fig 1A and S2A Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Unusual characteristics of bat MHC class I genes. (A) The proportions of MHC class I alleles with 3- or 5-aa insertions in bats, marsupials (opossum, koala, tammar wallaby, and Tasmanian devil), and higher mammals (human, NHP, mouse, and horse). The proportions of MHC class I alleles with the 3-aa insertion, 5-aa insertion, and no insertion are represented with orange, yellow, and cyan columns, respectively. The deletions (mainly including a 3-aa deletion) or insertions (mainly 1-aa insertions) other than the 3- and 5-aa insertions are termed as “others” in gray columns. The numerical data are included in S1 Data. (B) Structure-based sequence alignment of Ptal-N*01:01 and other representative (bats, marsupials, and higher mammals) MHC I molecules covering the residues from positions 49 to 66 (as in Ptal-N*01:01). The full information of the MHC I molecules of these species was listed in S1 Fig and S1 Table. Coils above the sequences indicated α-helices. Residues highlighted in red are completely conserved, and residues in blue boxes are highly (80%) conserved, with consensus amino acids in red. The residues at position 59 and 65 are shown in yellow. Special insertion positions in Ptal-N*01:01 are marked with red arrows above the sequences. The sequence alignment was generated with ClustalX and ESPript. To the right of the sequences, MHC class I alleles with the 3-aa insertion, 5-aa insertion, no insertion, and others are labeled with orange, yellow, cyan, and gray boxes, respectively. (C) The proportions of negatively charged residue Asp59 or Glu59 (“59D/E,” cyan columns), the pairing of the Asp59 and Arg65 (“D59+R65,” yellow columns), and the pairing of the negative-positive charged residues, termed as charge-matching at the two positions (“+/−,” purple columns) at the corresponding locations of MHC I in bats, marsupials, and higher mammals. (D) Statistical analysis of the 3-aa insertion and no 3-aa insertion alleles (no insertion and 5-aa insertion) in bats, respectively. Fisher exact test or the chi-squared test was used for the statistical analyses. **p < 0.01. aa, amino acid; MHC, major histocompatibility complex; NHP, nonhuman primate. https://doi.org/10.1371/journal.pbio.3000436.g001 Another feature of bat MHC I is the higher prevalence of a negatively charged residue at position 59 and a positively charged residue at position 65, as well as their pairing (according to the Ptal-N*01:01 residue code) (Fig 1B and S2A Fig). For the MHC I molecules of the higher mammals, the residue at the position corresponding to 59 (position 56 of human, NHP, mouse, and horse MHC I) is a highly conserved Gly. However, 71.43% of bat MHC Is have an Asp59 or Glu59 (59D/E) at this position (Fig 1C). The pairing of Asp59 and Arg65 (D59+R65) also occurs at a high fraction (35.71%) in bat MHC I compared with the MHC I of humans (0%), NHP (0%), mouse (0%), and horse (0%). Actually, the pairing of the charged residues at these two positions include different types: the negatively charged residues Asp/Glu at 59 paired with positively charged residues Arg/Lys at 65, and even positively charged residue Arg/Lys at 59 paired with Asp/Glu at 65 (as gene EPQ18390.1 in Myotis brandtii). Also, the pairing of the charged residues, termed as charge matching at the two positions (+/−), occurs at an even higher fraction for bat MHC I (39.29%). Interestingly, these features of bat MHC I are also prevalent in marsupials (Fig 1C). To determine whether the insertion has any correlation with the unusual substitutions at positions 59/65, we analyzed the fractions of 59D/E, D59+R65, and the +/− charge matching in the bat MHC I genes with the 3-aa insertion. We found a significantly higher fraction of 59D/E (100%), D59+R65 (64.71%), and the +/− charge matching (70.59%) in the bat MHC I genes with the 3-aa insertion compared with the corresponding fractions of other bat MHC I genes (no insertion and 5-aa insertion) (59% for 59D/E, 23.08% for D59+R65, and 28.21% for the +/− charge matching) (Fig 1D). Collectively, these distinct features suggested an unusual PBG of bat MHC I, which may affect peptide binding and presentation. Ptal-N*01:01–binding peptides from known emerging viruses To verify the peptide binding motif of Ptal-N*01:01 and to screen the potential bat MHC I T-cell epitopes from recently emerging and re-emerging viruses with bats as their potential reservoir, we predicted Ptal-N*01:01–binding peptides from EBOV, MERS-CoV, and H17N10/H18N11 (S2 Table). The previously determined Ptal-N*01:01–binding peptides derived from HeV were also included as a positive control [30]. The binding capacity of these peptides were evaluated by their ability to facilitate the in vitro renaturation of Ptal-N*01:01. Generally, the peptides with higher binding capability to MHCs would have a higher production of the heterotrimer MHC complexes but lower production of β2-microglobulin (β2m). Seven peptides from MERS-CoV, five from EBOV, one from H17N10, and the two HeV-derived peptides helped Ptal-N*01:01 naturally refold (Fig 2). The lengths of the Ptal-N*01:01–binding peptides cover a range from octamers to tridecamer. These peptides possess an Asp at the P1 position (with Gln in one peptide: MERS-CoV-S4), an aromatic aa (Phe or Tyr) at the P2 position, and a Pro or Leu at the PΩ position (C terminus of the peptide) (S2 Table). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. The identification of emerging and re-emerging virus-derived peptides binding to Ptal-N*01:01. The binding of peptides derived from HeV (A), EBOV (B), MERS-CoV (C), and H17N10 influenza-like virus (H17N10) (D) with Ptal-N*01:01 were evaluated by co-refolding. Co-refolding without any peptide was termed as the negative control (No pep), as curves in gray color. After properly refolding, the high-absorbance peaks of the correctly refolded MHC I with the expected molecular mass of 45 kDa were eluted at the estimated volume of 16 mL on a Superdex Increase 200 10/300 GL column. The profile is marked with the approximate positions of the molecular mass standards of 75.0, 44.0, and 13.7 kDa. Inset, reduced SDS–PAGE gel (15%) of Ptal-N*01:01/HeV1 complex for peak 1 (P1), peak 2 (P2), and peak 3 (P3). Lane M contains molecular-mass markers (labeled in kDa). P1, P2, and P3 represent the aggregated heavy chain, the correctly refolded heterotrimer Ptal-N*01:01 complex (45 kDa), and the extra β2m, respectively. β2m, β2-microglobulin; HeV, Hendra virus; EBOV, Ebola virus; MERS-CoV, Middle East respiratory syndrome coronavirus; MHC, major histocompatibility complex; P1, peak 1; P2, peak 2; P3, peak 3. https://doi.org/10.1371/journal.pbio.3000436.g002 The structures of Ptal-N*01:01 verify the motifs of the binding peptides derived from pathogens The binding peptide motif of bat MHC I Ptal-N*01:01, with the negative-charged Asp at the P1 position and a Pro at the PΩ position, is indeed extremely rare in class I binding peptide repertoires of humans and other mammals. To verify the uncommon peptide presentation features of bat MHC I, we determined the crystal structures of Ptal-N*01:01 in complex with the two HeV-derived peptides, HeV1 (DFANTFLP) and HeV2 (DYINTNVLP), two EBOV-derived peptides, EBOV-NP1 (DFQESADSFL) and EBOV-NP2 (DFQESADSFLL), one H17N10-derived peptide, H17N10-NP (DFEKEGYSL), and one MERS-CoV-derived peptide, MERS-CoV-S3 (DFTCSQISP) (Table 1). Download: PPT PowerPoint slide PNG larger image TIFF original image Table 1. X-ray data processing and refinement statistics. https://doi.org/10.1371/journal.pbio.3000436.t001 The overall structures of Ptal-N*01:01 display the common characteristics of classical MHC I molecules in other mammals, with the extracellular region of the heavy chain folding into three different domains. The α1 and α2 domains construct a typical PBG that contains two α1-helices and eight β-sheets, and the α3 domain and β2m display typical immunoglobulin (Ig) domains and underpin the peptide binding domain (Fig 3A). The all-atoms superimposition of Ptal-N*01:01/HeV1 onto the other five structures demonstrates a similar overall conformation, with root mean square deviations (RMSDs) of 0.344–0.517 Å (Fig 3B). The superimposition of Ptal-N*01:01/HeV1 onto human MHC I HLA-A2 and mouse MHC I H-2Kd generated RMSDs of 1.198 and 1.119 Å, respectively (Fig 3C). The most distinct differences between Ptal-N*01:01 and the MHC I from other vertebrates are located in the N terminus of the PBG, with an extension of the α1-helix in Ptal-N*01:01 (Fig 3D). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Overview of Ptal-N*01:01/peptide structures. (A) Overview of the structure of the Ptal-N*01:01/HeV1. The peptide HeV1 is presented as yellow sticks in the peptide-binding cleft. The heavy chain of Ptal-N*01:01 and bat β2m are shown as green and cyan cartoons, respectively. (B) The superimposition of different Ptal-N*01:01 complexes was performed using the determined structures: Ptal-N*01:01/HeV1, Ptal-N*01:01/HeV2, Ptal-N*01:01/EBOV-NP1, Ptal-N*01:01/EBOV-NP2, Ptal-N*01:01/H17N10-NP, and Ptal-N*01:01/MERS-CoV-S3. The peptides were omitted. (C) Structural alignment of MHC I heavy chains and β2m exhibited a similar overall conformation of Ptal-N*01:01 (green, peptide HeV1), HLA-A*0201 (yellow, PDB code: 3I6G), and H-2Kd (purple, PDB code: 5GR7). (D) The superimposition of the α1α2 domains of Ptal-N*01:01 (green) with other vertebrate MHC I molecules: human HLA-A*0201 (3I6G), macaque Mamu-A*02 (3JTT), swine SLA-1*0401 (3QQ4), equine Eqca-N*00602 (4ZUU), bovine N*01801 (3PWV), canine DLA-88*50801 (5F1I), and murine H2-Kd (5GR7) (all in white). The distinct conformations at the N terminus between the α1-helices of Ptal-N*01:01 (green) and the MHC I from other vertebrates (white) are labeled by a pink circle. The peptide HeV1 in Ptal-N*01:01 is presented in yellow loops. (E) The structural alignment of Ptal-N*01:01 (peptide HeV1) when renatured with bat β2m (green) and human β2m (yellow). HeV1 showed similar conformation in the binding groove. (F) The black box in the dashed line indicates the similar conformations of residues in the interface of the bat (green) and human β2m (yellow) binding to α1α2 domains of Ptal-N*01:01 heavy chain. (G) Similar binding of Ptal-N*01:01 heavy chain to bat and human β2m. Trp60 in bat β2m (cyan) and human β2m (orange) binds to Gln99 and Asp125 in Ptal-N*01:01 when forming a complex. (H) The alignment of monomer bat β2m (blue) with HeV-1/Ptal-N*01:01-batβ2m (green) complexes. Resides Ser55 to Tyr63 in monomer bat β2m (light blue) shift minorly after forming a complex (cyan). β2m, β2-microglobulin; EBOV, Ebola virus; HeV, Hendra virus; MERS-CoV, Middle East respiratory syndrome coronavirus; MHC, major histocompatibility complex; PDB, Protein Data Bank. https://doi.org/10.1371/journal.pbio.3000436.g003 To elucidate whether the peptide-presenting features of a bat MHC I molecule can be influenced by binding to human β2m, we solved the structure of the Ptal-N*01:01 heavy chain complexed with human β2m (Ptal-N*01:01-h) at a resolution of 1.6 Å (Table 1). Comparing HeV1/Ptal-N*01:01 renatured with bat β2m and human β2m, the structural conformations of both HeV1 peptides in the two structures are also similar, with an RMSD of 0.228 Å in the two binding grooves (Fig 3E). In addition, the overall structures are quite similar, with the RMSD of 0.564 Å of all atoms (Fig 3F). And the key residues binding to α1α2 domains and α3 domain of the Ptal-N*01:01 heavy chain are highly conserved in human and bat β2m (S2B Fig). Further analysis shows Trp60 in bat β2m binds to Gln99 and Asp125 in Ptal-N*01:01 when forming a complex, which is also conserved in human β2m when binding to Ptal-N*01:01 (Fig 3G). It indicates that the structure of the peptide loaded in the groove of Ptal-N*01:01 was not affected by the substitution of the β2m subunit. We also determined the structure of monomer bat β2m without MHC I (Table 1). The structure alignment of bat β2m monomer with β2m subunit in MHC complexes indicates a minor conformational shift of the loop resides Ser55 to Tyr63 in bat β2m after forming a complex (Fig 3H). Although having different lengths, the 8-mer peptide HeV1, 9-mers HeV2 and H17N10-NP, 10-mer EBOV-NP1, and 11-mer EBOV-NP2 in the PBG of Ptal-N*01:01 all display an M-shaped conformation, with P2 and PΩ residues as the primary anchors and the P5 or P6 residue as the secondary middle anchors (Fig 4A–4F and S3A–S3E Fig). The P1 Asp of all of the peptides adopts a rigid conformation upward to the α1-helix of the heavy chain. In the structure of the Ptal-N*01:01/MERS-CoV-S3 complex, the conformation of the P3–P8 residues of the peptide could not be determined due to their poor electron densities (S3F Fig), revealing a flexible conformation in the middle region of this MERS-CoV-S3 peptide. However, the other three residues (P1-Asp, P2-Phe, and PΩ-Pro) with electron densities available adopt similar conformations as in the other five structurally determined peptides. As a 9-mer peptide, MERS-CoV-S3 possesses a Glu at P6, corresponding to the P6-Asn in 9-mer HeV2, which is the secondary anchor residue. The larger Glu may not be able to locate in the PBG, which leads to a flexible conformation of the MERS-CoV-S3. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. The structural conformation and motif of the binding peptides presented by Ptal-N*01:01. (A-E) The overall conformation of Ptal-N*01:01–binding peptides HeV1 (A) and HeV2 (B) from HeV, H17N10-NP (C) from H17N10 influenza-like virus, and EBOV-NP1 (D) and EBOV-NP2 (E) from EBOV. (F) The superimposition of presented peptides HeV1, HeV2, H17N10-NP, EBOV-NP1, and EBOV-NP2 by Ptal-N*01:01. The peptides are aligned according to the superimposition of the α1 and α2 domains of the five structures of Ptal-N*01:01. The conserved conformations of residues in the P1, P2, and PΩ positions are shown in yellow ellipses. Most conformational distinctions are located in the central region of the peptides, marked by a yellow rectangle. (G-J) The capacity of HeV1 (G), HeV2 (I), and their Ala substitutions for binding to Ptal-N*01:01 was evaluated by in vitro refolding. The thermostabilities of Ptal-N*01:01 with HeV1 (H) and HeV2 (J) peptides and their Ala substitutions were tested by circular dichroism (CD) spectroscopy. The numerical data are included in S1 Data. The curves for the unfolded fractions were determined by monitoring the CD value at 218 nm. The temperature was increased by 1°C/minute. Shown are the fitting data to the denaturation curves using the Origin 8.0 program (OriginLab). The Tms of different peptides are indicated by the dashed gray lines at the 50% fraction unfolded. β2m, β2-microglobulin; CD, circular dichroism; EBOV, Ebola virus; HeV, Hendra virus; Tm, midpoint transition temperature. https://doi.org/10.1371/journal.pbio.3000436.g004 To further validate the role of the unusual binding peptide motif of bat MHC I Ptal-N*01:01, Ala mutations at residues P1, P2, PΩ, and the middle (P5 or P6) positions of peptides HeV1 and HeV2 were evaluated by refolding assays (Fig 4G and 4I). The thermal stability of the resulting heterotrimers was monitored by circular dichroism (CD) spectroscopy (Fig 4H and 4J, S3 Table). Substitution of Ala at P1 (HeV2-D1A), P2 (HeV2-Y2A), and P9 (HeV2-P9A) of peptide HeV2 led to nearly no refolding for Ptal-N*01:01, whereas Ala at P1 (HeV1-D1A), P2 (HeV1-F2A), and P8 (HeV1-P8A) of peptide HeV1 still supported refolding but with significantly lower stability than the original peptide. In contrast, substitution of Ala at P5 of HeV1 (HeV1-T5A) and at P6 of HeV2 (HeV1-N6A) led to similar yield of refolded heterotrimers, indicating a similar stability compared with the original peptides. Thus, through these analyses, the P1 anchor of Ptal-N*01:01–binding peptides seems to be as significant as the P2 and PΩ anchors. The insertion induces uncommon peptide P1 anchoring in bat MHC I Bat Ptal-N*01:01 possesses the Met-Asp-Leu insertion within the N terminus of its α1-helix (between residues 51 and 52 of HLA-A*0201). Compared with the structures available for the MHC I of other mammals, such as humans and mouse, the Ptal-N*01:01 structure displays an extension of the α1-helix of the PBG (Fig 5A). The 3-aa insertion pushes residue Asp59 closer to the N terminus of the binding peptide, which leads to the extension of the negatively charged side chain of Asp59 into the PBG (Fig 5B). The bat MHC I residue Asp59 participates in the formation of the A pocket (S4 Fig). Detailed analysis indicated that Asp59, Arg65, and the P1-Asp of the peptides in all six structures of Ptal-N*01:01 form a triangular network of hydrogen bonds (Fig 5C and S5 Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. The 3-aa (Met52Asp53Leu54) insertion contributes an uncommon P1 anchoring of bat MHC I. (A) The structural superimposition of the N terminus of PBGs from different MHC I molecules. Trp51–Gln61 of Ptal-N*01:01 in green and structurally corresponding residues Trp51–Glu58 of HLA-A*0201 (3I6G), macaque Mamu-A*02 (3JTT), swine SLA-1*0401 (3QQ4), equine Eqca-N*00602 (4ZUU), bovine N*01801 (3PWV), canine DLA-88*50801 (5F1I), and murine H2-Kd (5GR7) in white. The 3-aa Met52Asp53Leu54 insertion in bat MHC I is in purple. (B) The structural superimposition of Ptal-N*01:01 and its mutant Ptal-N*01:01(-3aa) (with Met52Asp53Leu54 deletion). The residues Trp51–Gln61 of Ptal-N*01:01 (green) and Trp51–Gln58 of Ptal-N*01:01(-3aa) (yellow) are presented in surface representation. The heavy chains of these two molecules are shown as white cartoons. The P1 positions of the peptides are shown as sticks. (C) The hydrogen bond network in the A pocket of Ptal-N*01:01/HeV1. Residues Asp59, Arg65, and P1-Asp of HeV1 are shown as blue sticks and form a triangular network of hydrogen bonds in blue dashed lines. (D) Different conformations of the N terminus of the α1-helices of Ptal-N*01:01 (green) and mutant Ptal-N*01:01(-3aa) (yellow). (E) The conformational shift of Asp, Arg, and P1-Asp of HeV1 in the structures Ptal-N*01:01/HeV1 (green) and Ptal-N*01:01(-3aa)/HeV1 (yellow), as indicated by purple arrows. (F) The N-terminal conformation of the α1-helix and hydrogen bonds (red dashed lines) within the structure of mutant Ptal-N*01:01(-3aa)/HeV1. Asp, Arg, and HeV1 P1-Asp are shown as purple sticks. (G) Schematic diagram of the construction of the mutant Ptal-N*01:01(-3aa) with a Met52Asp53Leu54 deletion compared with the wild-type Ptal-N*01:01. (H-I) The capabilities of peptide HeV1 presented by Ptal-N*01:01 and mutant Ptal-N*01:01(-3aa) were evaluated by in vitro refolding (H) and CD spectroscopy (I). The numerical data are included in S1 Data. (J) Schematic diagram of the construction of the mutant HLA-A2M with a Met52Asp53Leu54 insertion and D59+R65 compared with the wild-type HLA-A2. (K-L) The capabilities of peptide DL9 (the first amino acid of GL9 change to D) presented by HLA-A2 and mutant HLA-A2M were evaluated by in vitro refolding (K) and CD spectroscopy (L). aa, amino acid; β2m, β2-microglobulin; CD, circular dichroism; DL9, DILGFVFTL; GL9, GILGFVFTL; HeV, Hendra virus; MHC, major histocompatibility complex; PBG, peptide binding groove; P1, position 1; Tm, midpoint transition temperature; WT, wild type. https://doi.org/10.1371/journal.pbio.3000436.g005 To further investigate the role of the 3-aa insertion in the peptide binding and presentation of Ptal-N*01:01, we constructed the Ptal-N*01:01(-3aa) mutant, which deleted the 3-aa insertion (Fig 5G). The structure of Ptal-N*01:01(-3aa)/HeV1 displayed a shortened α1-helix that is similar to the human HLA-A*0201 (Fig 5D). Meanwhile, although the salt bridge was still observed between Arg65 of Ptal-N*01:01 and the P1-Asp, both the Arg65 and P1-Asp displayed a conformational shift (Fig 5E). The triangular network of the hydrogen bonds between Asp59 (Residue Asp56 in the mutant), Arg65 (Residue Arg62 in the mutant), and the P1-Asp of HeV1 within the structure of the wild-type Ptal-N*01:01 was broken (Fig 5F). One of the hydrogen bonds between the two residues Arg65 (Residue Arg62 in the mutant) and the P1-Asp of HeV1 was lost, and the remaining two hydrogen bonds extended from 2.97 to 3.23 Å and 2.67 to 2.72 Å, respectively. We also investigated whether the 3-aa deletion of Ptal-N*01:01 influenced binding ability to the peptides. Ptal-N*01:01(-3aa) was still renatured in the presence of peptide HeV1 but with a much lower yield of the heterotrimer complex (Fig 5H). CD spectroscopy also indicated a weaker binding of Ptal-N*01:01(-3aa) with the peptide (Fig 5I). To further verify the unusual peptide presentation and preference for peptides with a P1-Asp in bat Ptal-N*01:01, we constructed the HLA-A2M mutant, which has a 3-aa insertion and the charge matching residues at positions 59/65 based on HLA-A*02:01(Fig 5J). Both the refolding assay and CD spectroscopy indicated a stronger binding of DL9 (G1D mutant at P1 of peptide GL9) with the HLA-A2M compared with HLA-A*02:01, and the HLA-A*02:01 has a higher binding capacity to GL9 than DL9 (Fig 5K and 5L). The unusual preference of Pro as the PΩ anchor of Ptal-N*01:01–binding peptides Although Ptal-N*01:01 can bind peptides with Leu as the PΩ anchor, the binding peptides can also possess Pro at this position. This is uncommon in the peptides bound by MHC I molecules from other mammals. The structures of Ptal-N*01:01 with the 9-mer peptides HeV2 (DYINTNVLP) and H17N10-NP (DFEKEGYSL) showed that both of the peptides adopt similar overall conformations as the HLA-A*0201–presented 9-mer peptide GL9 (GILGFVFTL) with a Leu at the PΩ site (Fig 6A). However, to the best of our knowledge, no peptide motif with a Pro at the PΩ position has been reported for HLA-A*0201. Detailed comparative analysis of the F pockets of Ptal-N*01:01 and HLA-A*0201 revealed that the F pocket of Ptal-N*01:01 is shallow but with a wide opening (Fig 6B–6E). Corresponding to Asp77 and Tyr116 of HLA-A*0201, Ptal-N*01:01 possesses smaller residues Gly80 and Leu119, respectively. The Ptal-N*01:01–specific Gly80 in Ptal-N*01:01 is small, which provides more space to accommodate the pyrrolidine ring of PΩ-Pro. In addition, although both HLA-A*0201 and Ptal-N*01:01 have a similar Lys residue in their corresponding positions (146 for HLA-A*0201 and 149 for Ptal-N*01:01), the conformation of the Lys149 of Ptal-N*01:01 is shifted to the C terminus of the PBG, which leaves more space for the open mouth of the F pocket in Ptal-N*01:01. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. The extraordinary preference for Pro as the PΩ anchor of Ptal-N*01:01–binding peptides. (A) The superimposition of 9-mer peptides HeV2 (cyan) and H17N10-NP (green) presented by Ptal-N*01:01 and GL9 (orange) presented by HLA-A*0201 (PDB code: 3I6G). The heavy chains are represented as gray cartoons. The PΩ anchors of peptides are shown as sticks, and the side chains of other residues of the peptides are omitted. (B) Distinct F pockets of Ptal-N*01:01 and HLA-A*0201. The heavy chains are represented as gray loops. Gly80, Leu119, and Lys149 of Ptal-N*01:01 are represented as cyan sticks. The corresponding residues of HLA-A*0201 are shown as orange sticks. The given entrances for F pockets of Ptal-N*01:01 and HLA-A*0201 are labeled by purple and deep blue circles, respectively. (C-D) The side chains of PΩ-Pro from peptide HeV2 (C) and PΩ-Leu from peptide H17N10-NP (D) insert into the F pockets of Ptal-N*01:01. The F pocket of Ptal-N*01:01 is indicated in surface representation with a yellow entrance and purple bottom. (E) The side chain of residue PΩ-Leu from peptide GL9 inserts into the F pocket of HLA-A*0201. The F pocket of HLA-A*0201 is indicated by the representation of a brown entrance with a light blue bottom. (F-G) The stabilities of HLA-A*0201 complexed with peptide GL9 and its mutant GL9-L9P (with a Pro at the P9 position) were evaluated by in vitro refolding (F) and CD spectroscopy (G). The numerical data are included in S1 Data. (H) The relatively lower mutation frequency of amino acid Pro (red bar) as Try and Gly among the 20 component amino acids in the proteins of MERS-CoV and SARS-CoV. The mutation frequency = the number of overall mutations for each amino acid/(the number of occurrences of the amino acid in the reference sequence×total number of sequences). The proteomes of 1,000 MERS-CoV genomes and SARS-CoV genomes were retrieved from GenBank, respectively. The detailed information is included in S2 Data. CD, circular dichroism; GL9, GILGFVFTL; HeV, Hendra virus; MERS-CoV, Middle East respiratory syndrome coronavirus; PDB, Protein Data Bank; SARS-CoV, severe acute respiratory syndrome coronavirus; Tm, midpoint transition temperature. https://doi.org/10.1371/journal.pbio.3000436.g006 To verify the allele-specific preference of Ptal-N*01:01 for peptides with Pro at the PΩ site, we examined the binding ability of HLA-A*0201 to a mutated GL9 peptide, GL9-L9P, with a Pro at P9 (Fig 6F and 6G). We found GL9-L9P has a weaker capacity to help the HLA-A*0201 refold, and the generated heterotrimer complex has lower midpoint transition temperature (Tm) (38.1°C) compared with the wild-type peptide GL9 (55.8°C) with a Leu at PΩ. Thus, although Pro at PΩ may also act as a suboptimal anchor for HLA-A*0201, Ptal-N*01:01 uses Pro as one of its optimal PΩ anchors. The analysis of the 20 different component amino acids for the proteins from the MERS-CoV and SARS-CoV indicated that Pro possesses a relatively low mutation rate compared with the other amino acids (Fig 6H). The mutation rate of Pro is only higher than Trp and Gly, which are the largest and smallest residues, respectively. It indicates that, as special residues Trp and Gly, the residue Pro may also keep the natural conformation and function of viral proteins. In other words, structural constraints favoring conservation of Pro in certain positions of proteins may operate to preserve viral protein conformation and function. Thus, its mutation rate is restricted. The selection of Pro as the anchor residue of bat MHC I Ptal-N*01:01–presented peptides may restrict the viral mutation pushed by T-cell immunity and accelerate virus clearance. The deep B pocket of bat MHC I has a different orientation The P2 positions of Ptal-N*01:01–binding peptides are predominantly aromatic amino acids such as Tyr and Phe as the anchor, which is common in human HLA-A*2402 or mouse H-2Kd. However, when we superimposed the structures of Ptal-N*01:01 onto the previously determined structures of HLA-A*2402 and H-2Kd, we found that the P2 anchors of Ptal-N*01:01 protrude in a different direction (Fig 7A). The HLA-A*2402– and H-2Kd–presented peptides have a Tyr or Phe pointing to the C terminus of the PBG, while the Tyr or Phe of Ptal-N*01:01–binding peptides swing toward the N terminus of the PBG (Fig 7C and 7D). Comparison of the amino acids lining the B pockets of these MHC I molecules from different mammals demonstrated that Ptal-N*01:01 possesses a Tyr9 (compared with the small residues Ser9 in HLA-A*2401 or Val9 in H-2Kd), which takes up space and can push a P2-Tyr or -Phe to the other direction (Fig 7B). Furthermore, compared with the residues Met45 in HLA-A*2402 or Phe45 in H-2Kd, the Ptal-N*01:01–specific Ala45 leaves a large space to accommodate the P2-Tyr or -Phe. Indeed, detailed analyses showed that the P2-Tyr of Ptal-N*01:01–binding peptides form hydrogen bonds directly with the main chain of the β-sheet on the floor of the PBG. In contrast, the P2-Tyr of HLA-A*2402–restricted peptides bind to the side chain of His70 on the α1-helix (Fig 7B). Sequence superimposition of the bat MHC I with other mammal MHC Is indicated that Ala45 is prevalent (41%) among bat MHC I molecules but is never seen in human and mouse MHC I (S2A Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 7. The B pocket of bat MHC I with an uncommon orientation. (A) Structural superimposing of Ptal-N*01:01 with HLA-A*2402 and H-2Kd. The HLA-A*2402– (PBD code: 3I6L and 5WXD) and H-2Kd–presented peptides (PBD code: 5GR7) have a P2-Tyr or -Phe (blue sticks) pointing to the C terminus of the PBG, while the P2-Tyr or -Phe (green sticks) of Ptal-N*01:01–binding peptides (HeV1 and HeV2) swing to the N terminus of the PBGs. (B) Different conformations of P2 anchors (shown in sticks and spheres) of Ptal-N*01:01/HeV2 and HLA-A*2402. The major residues of the B pockets of Ptal-N*01:01/HeV2 and HLA-A*2402 (PBD code: 5WXD) are shown as green and blue sticks, respectively. The hydrogen bonds are shown as red dashed lines. (C) The B pocket of Ptal-N*01:01/HeV2 is shown as a semitransparent surface, with Tyr9 and Ala45 of Ptal-N*01:01 shown as sticks under the surface. Ptal-N*01:01 heavy chain is shown as green cartoons under the semitransparent surface. (D) The B pocket of HLA-A*2402 (PBD code: 5WXD) is shown as a semitransparent surface. The residues Ser9 and Met45 of HLA-A*2402 are shown as sticks under the surfaces. The different orientation of the P2 residue (blue sticks) of HLA-A*2402–presented peptide is compared with the corresponding residue of Ptal-N*01:01–presented peptide HeV2 in panel C. HeV, Hendra virus; MHC, major histocompatibility complex; PBD, peptide binding groove; PBG, peptide binding groove; P2, position 2. https://doi.org/10.1371/journal.pbio.3000436.g007 Ptal-N*01:01 does not bind to long peptides in an N-terminal extended manner Previously, it was indicated that Ptal-N*01:01 has a special preference for the binding of long peptides, together with the common 8–10-mer peptides in other mammal MHC I molecules. To further elucidate whether the 3-aa insertion of Ptal-N*01:01 has an impact on the preference for long peptides through an N-terminal extension manner, we synthesized 20 long peptides (11-mer to 15-mer) that were previously eluted from Ptal-N*01:01–expressing cells (S4 Table) [30]. None of these peptides could facilitate Ptal-N*01:01 refolding in vitro (S6A Fig). Considering that these 20 peptides may not have the typical motif of Ptal-N*01:01–binding peptides, we also synthesized naturally N-terminally extended peptides based on the Ptal-N*01:01–binding peptides from HeV1, MERS-CoV-S7, and EBOV-NP1 (S5 Table). Although these three peptides have a typical motif of Ptal-N*01:01–binding peptides, the N-terminal extension led to a failure in peptide binding (S6B Fig and S5 Table). These data indicate that the 3-aa insertion into Ptal-N*01:01 leads to a more restrictive binding peptide selection for Ptal-N*01:01 but not an extension to longer peptides via the N terminus. Discussion The identification of bats as natural reservoirs of several highly pathogenic viruses that impact human and animal health and the fact that these viruses are harmless to bats have resulted in an increasing interest in the investigation of the specificities of the bat immune system. Herein, we screened and identified a series of bat MHC I Ptal-N*01:01–binding peptides derived from four different bat-borne viruses: HeV, EBOV, MERS-CoV, and H17N10. The subsequent determination of the structures of Ptal-N*01:01 complexed with peptides from these viruses revealed unusual peptide presentation features of bat MHC I. Interestingly, this uncommon feature of pocket A of bat MHC I may be shared by the MHC Is from different marsupials. In addition, as the traditional primary anchoring positions for peptides, the B and F pockets of Ptal-N*01:01 also display unconventional conformations that contribute to the distinct peptide presentation and special peptide motif compared with other higher mammals. The sequence combination of the 3-aa insertion at the N terminus of the α1-helix and the charge matching residues at positions 59/65 enable an unusually tight anchoring of the P1-Asp in pocket A of Ptal-N*01:01. But more significantly, this insertion site is located at a position called 310 helix (residues 49–53); newly synthesized MHC I molecules complexed with β2m are poised in the endoplasmic reticulum in a peptide-receptive (PR) form, ready to bind and be stabilized in the mature peptide-loaded (PL) form by peptides destined for display at the cell surface. The movement of a hinged unit containing a conserved 310 helix promotes the PR transition state to a PL mature molecule [31]. Chaperone-mediated loading of high-affinity peptides onto MHC I is a key step in the MHC I antigen presentation pathway. TAP binding protein (TAPBPR; related) remodels the peptide-binding groove of MHC I, resulting in the release of low-affinity peptide [32]. In the absence of TAPBPR, Y84 (Y87 in Ptal-N*01:01) plays a role in closing of the α2-1-helix “latch” by associating with the C terminus of bound peptides in the F pocket, the release of nonoptimal peptides induced by TAPBPR [33]. However, the hydrogen bond network of the A pocket could stabilize peptides with P1-Asp during peptide processing and exchange and high-affinity peptides are guided by initial contacts spanning both the A- and F pockets to form a prolonged interaction within the groove, resulting in a closure of the α2–1 helix latch, which triggers TAPBPR release from the peptide/MHC I complex. Being highly polymorphic, subtle substitutions or insertions in the PBGs of MHC I molecules may dramatically affect the binding peptide pool and also the peptide presentation to T cells [34,35]. Comparative genomic and transcriptomic analysis demonstrated that a series of bat MHC I molecules have an insertion in the N terminus of the PBG compared with other higher mammals [29]. Herein, our structural study visually shows that the 3-aa insertion in the bat MHC I leads to an extension of the α1-helix. This conformational change causes the protrusion of the bat-specific residue Asp59 into the A pocket of the PBG, which forms a hydrogen bond network with the Arg65 and the peptide P1-Asp. Study of the peptide repertoire of key human and mouse MHC I alleles demonstrates that anchor Asp is disfavored at the P1 position of the peptides [36]. In contrast, recently reported data on the peptide elution from bat MHC I molecules [30], together with viral peptide screening results in this study, demonstrate that bat MHC I Ptal-N*01:01 prefers peptides with a P1-Asp. Either depletion of the three inserted residues (Met52Asp53Leu54) or substitution of the P1-Asp with Ala impaired the MHC/peptide binding. Indeed, the electrostatic nature of the contacts between Asp59, Arg65, and P1-Asp and the fact that it is involved in solvent-exposed elements in pocket A of the MHC complex suggest that the P1-Asp acts as a “surface anchor residue.” This surface anchor residue was also previously defined in a phosphopeptide-MHC complex [37] in which the solvent-exposed P4 phosphate moiety can enhance the stability of the peptide-MHC association. The additional surface anchor residue for the bat MHC I binding peptides may have at least two advantages for antiviral T-cell immunity. First, the peptides tightly bind to Ptal-N*01:01 and present a special peptide-MHC landscape at the N terminus of the peptides for the T-cell recognition. Second, negatively charged residues such as Asp/Glu in the peptides may also act as key residues in the original viral proteins for virus replication, which will have lower mutation rates to escape T-cell recognition. In addition, the insertion of Met52Asp53Leu54 also leads to a special exposed landscape of the α1-helix of Ptal-N*01:01. Thus, heterozygous bats with MHC I alleles of both 3-aa insertion and no insertion may possess a T-cell repertoire with broader diversity, which need more work to investigate. Bats are one of the most ancient extant lineages of eutherian mammals, believed to be located in distinct mammalian lineages different from marsupials and other higher eutherians [29]. The evolution of the MHC I gene family is closely tied to the evolution of the vertebrates genome [38]. However, the 3-aa insertion and the charge-matching residues at positions 59/65 of MHC Is are also prevalent among different marsupials: opossum, koala, tammar wallaby, and Tasmanian devil. This may be a phenomenon of convergent evolution under the pressure of related pathogens. To comparatively investigate the peptide presentation of MHC I from marsupials, we have synthesized a marsupial MHC I gene (Trvu-UB*01), which has similar characteristics to 3-aa insertion and charge-matching residues at positions 59/65 to Ptal-N*01:01, and also the small residue G at position 80 (S7A Fig). However, the preference of the Trvu-UB*01 protein for peptides may not be similar to that of Ptal-N*01:01 (S7B Fig). This result indicates that although some alleles of MHC I from the marsupials possess the same key residues in the PBG of Ptal-N*01:01, the preferred peptide motifs are only partially similar between them (Pro as PΩ), which may reflect the contribution of some other adjacent residues. Thus, the detailed peptide motifs and the presentation features of MHC Is from these lower mammals still require further laboratory investigation. Meanwhile, whether the 5-aa insertion in the bat MHC I impacts the peptide presentation in the same manner as 3-aa insertion or with a new molecular mechanism needs more structural studies. In this context, it is also worth mentioning that we also tried two additional online MHC-peptide binding predicting servers, NetMHCpan and Rosetta FlexPepDock, to verify the peptide-binding experiments in the study. However, neither prediction server was able to match the experimental results. This may indicate that current MHC binding peptide predictions were not suitable for non-mouse and nonhuman mammals such as bats, which may have a different manner of peptide binding. Like the human and murine MHC I [2,39], Ptal-N*01:01 has B and F pockets in the PBG to accommodate the primary anchors of the binding peptides, but uncommon conformational features of Ptal-N*01:01–loaded B and F pockets were identified through our structural investigations. The B pocket of Ptal-N*01:01, featuring the bat-specific residue Ala45, has a novel position in the PBG for the P2 anchor. In this position, the P2-Tyr of the peptides form hydrogen bonds directly with the main chain of the β-sheet on the floor of the PBG of Ptal-N*01:01, which leads to a stable anchoring of the P2-Tyr. Meanwhile, Ptal-N*01:01, with a relatively unique Gly at position 80, has an unusually shallow F pocket with a wide entrance, like a large bowl, which can accommodate the uncommon PΩ-Pro anchor. To the best of our knowledge, although the PΩ-Pro anchor is not observed in the previously reported peptides from mammalian MHC I, the precedents for the P2-Pro anchor have been reported in the context of murine H-2Ld and human B*3501 and B*5301 [40–42]. Requirements for Pro in the P3 position as an anchor residue are also observed in the murine H-2Dd- and macaque Mamu-A*01–restricted epitopes [43,44]. A previous study indicates that the antiviral effect of T cells is sufficiently strong to force the virus to adopt a relatively unfavorable mutation, which reduces viral replication [45]. The proline anchor may be a result of special antigen processing in bats. For humans, most products of ornithine decarboxylase degraded in vitro by the 26 S ATP-dependent proteasome, which contained one or two Pro residues, implied that the Pro residue has a role in the escape from random cleavage by proteasomes [46]. In addition, Pro residue(s) within epitopic sequences presumably contribute to efficient production of MHC class I ligands through prevention of their random cleavage by proteasomes [47]. Thus, the peptides with proline as a C terminus are still seldom in human and other common mammals. However, the current research on the proteasome of bats is still blank. Our data also showed that among the 20 different component amino acids for the proteins from the bat-related viruses, the residue Pro possesses a relatively low mutation rate compared with the other amino acids (Fig 6H). Pro, with a unique conformation, may act as a key residue for the structure and function of viral proteins, and thus its mutation rate is low. Therefore, the usage of Pro as the PΩ anchor of the T-cell epitopes in Ptal-N*01:01–carrying bats may also restrict the formation of escape mutations. However, based on the currently limited amount of bat MHC I sequences available, Gly80 in the F pocket does not seem prevalent in bat MHC I. More sequencing of bat genomes and especially MHC I genes are needed to verify whether the accommodation of PΩ-Pro as a peptide anchor is common in bat MHC I or a specific feature of Ptal-N*01:01. In conclusion, through a series of structural and functional investigation, we demonstrated several novel features of bat MHC class I molecules presenting virus-derived peptides. Our results provide new insight into the adaptive immune system of bats, which may contribute to the unique virus–host interactions in these important mammals. Due to the high containment nature of the viruses and the difficulty in conducting live bat infection studies, our current study lacks in vivo functional characterization, which we hope to conduct in the future with international collaborations. Materials and methods Sequence retrieval and analyses The sequences of 56 MHC class I genes (including predicted genes) from bats were retrieved from the NCBI database (S1 Table). Higher mammal MHC I heavy chain sequences were retrieved from the Immuno Polymorphism Database (IPD) (www.ebi.ac.uk/ipd/mhc) and the UniProt database (www.uniprot.org). Previously deposited marsupial (opossum, tammar wallaby, koala, Tasmanian devil) and platypus MHC I transcripts were included in these analyses (S1 Table). Sequence alignments were generated with ClustalX [48] and ESPript [49]. Similarities were calculated using DNAMAN (https://www.lynnon.com/). The proteomes of 1,000 MERS-CoV genomes and 1,000 SARS-CoV genomes were retrieved from GenBank, respectively. After sequence alignment with MAFFT, the dominant amino acid for each site was elected as a reference sequence. The mutation frequency = the number of overall mutations for each amino acid/(the number of occurrences of the amino acid in the reference sequence×total number of sequences). Peptide synthesis and preparation of expression constructs To screen potential peptides for binding to Ptal-N*01:01, the proteomes of the bat-related viruses EBOV (NP: GenBank no. AF054908.1; GP: GenBank no. AKG65250.1), MERS-CoV (GenBank no. AXN92228.1), H17N10 influenza-like virus (A/little yellow-shouldered bat/Guatemala/060/2010(H17N10)), and H18N11 influenza-like virus (A/flat-faced bat/Peru/033/2010(H18N11)) were utilized to predict the candidate peptides. The candidate peptides were predicted and selected according to the recently reported motif, by which the two Ptal-N*01:01–binding peptides, HeV1 and HeV2, derived from HeV were also synthesized (S2 Table) [30]. The potential binding scores of the selected peptides were also predicted through the online NetMHCpan 4.0 server (http://www.cbs.dtu.dk/services/NetMHCpan/) [50] and Rosetta FlexPepDock, which is based on structure modeling [51,52], so that we prefer choose peptides that conform to the motif of Ptal-N*01:01 [30]. The peptide purity was determined to be >95% by analytical HPLC and mass spectrometry. The peptides were stored at −80°C as freeze-dried powders and were dissolved in DMSO before use. The cDNAs for the heavy chain of P. alecto MHC I Ptal-N*01:01 (GenBank no. KT987929) [30] and bat β2m (GenBank no. XP_006920478.1) were synthesized (Genewiz, Beijing, China). Ptal-N*01:01 sequence was deposited to GenBank by Wynne and colleagues, and Ptal-N*01:01–binding peptides HeV1 and HeV2 were identified in their study [30]. Although Ng and colleagues reported the first Ptal-N*01:01 [29], the sequence is not available online. To investigate the function of Met52Asp53Leu54 in Ptal-N*01:01, a mutant termed Ptal-N*01:01(-3aa) with a deletion of these three amino acids was constructed. The amplified products expressing the extracellular domain (residues 1–277) of Ptal-N*01:01 and bat β2m (residues 1–98) were cloned into a pET28a vector (Novagen). The expression plasmid for human β2m (residues 1–99) was previously constructed in our laboratory [53]. Refolding and purification of bat class I complexes Renaturation and purification of Ptal-N*01:01 assembled with peptides were performed as previously described [54,55]. Generally, bat MHC I Ptal-N*01:01 heavy chain and bat β2m were overexpressed as inclusion bodies in the BL21(DE3) strain of Escherichia coli, and the purified inclusion bodies of the proteins were solubilized in 6 M guanidine-HCl buffer with a concentration of 30 mg/mL. Then, injection and dilution of MHC heavy chain, β2m, and peptide occurred at a molar ratio of 1:1:3 in refolding buffer (100 mM Tris-HCl [pH 8], 2 mM EDTA, 400 mM L-Arg, 0.5 mM oxidized glutathione, and 5 mM reduced glutathione) [34]. After 24 hours for protein refolding, the Ptal-N*01:01 complexes were concentrated and exchanged into a buffer of 20 mM Tris-HCl (pH 8) and 50 mM NaCl and then purified using a Superdex 200 16/60 HiLoad (GE Healthcare, Beijing, China) size-exclusion column. Crystallization, data collection, and processing Crystallization was performed using the sitting drop vapor diffusion technique. The Ptal-N*01:01/peptide complexes were screened through Crystal Screen kit I/II, Index Screen kit, PEGIon kit I/II, and the PEGRx kit (Hampton Research). Plates were incubated at 291 K and 277 K and assessed for crystal growth after 1–2 weeks. Ptal-N*01:01/HeV1 crystals were observed in 0.2 M NaCl, 0.1 M Bis-Tris (pH 5.5), and 25% (w/v) polyethylene glycol 3,350 at a concentration of 7.5 mg/mL. Ptal-N*01:01/HeV1(human β2m) crystals were observed in 0.1 M HEPES, pH 7.0, 2% w/v polyethylene glycol 3,350. Monomer bat β2m were grown in 0.1 M BIS-TRIS, pH 6.5, 8% w/v polyethylene glycol monomethyl ether 5,000. Single crystals of Ptal-N*01:01/HeV2 were grown in 0.075 M HEPES (pH 7.5), 15% (w/v) polyethylene glycol 10,000, and 25% (v/v) glycerol at a protein concentration of 10 mg/mL. Single crystals of Ptal-N*01:01/EBOV-NP1 were grown in 0.1 M succinic acid (pH 7.0) and 15% (w/v) polyethylene glycol 3,350. Single crystals of Ptal-N*01:01/EBOV-NP2 were grown in 0.2 M ammonium acetate, 0.1 M Tris (pH 8.0), and 16% (w/v) polyethylene glycol 10,000. Single crystals of Ptal-N*01:01(-3aa)/HeV1 were grown in 0.2 M sodium formate and 20% (w/v) polyethylene glycol 3,350. Single crystals of Ptal-N*01:01/H17N10-NP were grown in 0.075 M HEPES (pH 7.5), 15% (w/v) polyethylene glycol 10,000, and 25% (v/v) glycerol. Single crystals of Ptal-N*01:01/MERS-CoV-S3 were grown in 0.2 M sodium acetate trihydrate and 20% (w/v) polyethylene glycol 3,350. For cryoprotection, crystals were transferred to reservoir solutions containing 20% glycerol and then flash-cooled in a stream of gaseous nitrogen at 100 K. X-ray diffraction data were collected at beamline BL19U of the Shanghai Synchrotron Radiation Facility. The data collection statistics are shown in Table 1. Structure determination and analyses The collected intensities were subsequently processed and scaled using the Denzo program and the HKL2000 software package (HKL Research). The structures were determined using molecular replacement with the program Phaser MR in CCP4 [56]. The model used was the structure coordinates with Protein Data Bank (PDB) code 5F1I [35], and restrained refinement was performed using REFMAC5 from CCP4. Extensive model building was performed by hand using COOT [57]. The stereochemical quality of the final model was assessed with the program REFINE in Phenix or CCP4 (Table 1). Structure-related figures were generated using PyMOL (http://www.pymol.org/) and COOT. Determination of protein thermostability using CD spectroscopy The thermostabilities of Ptal-N*01:01 with two group key peptides were tested by CD spectroscopy. All complexes were refolded, purified, and measured at 0.2 mg/mL in a solution of 20 mM Tris (pH 8) and 50 mM NaCl. CD spectra at 218 nm were measured on a Chirascan spectrometer (Applied Photophysics) using a thermostatically controlled cuvette at temperature intervals of 0.2°C at an ascending rate of 1°C/minute between 20 and 90°C. The unfolded fraction (%) is expressed as (θ−θa)/(θa−θb), where θa and θb are the mean residue ellipticity values in the fully folded and fully unfolded states, respectively. The denaturation curves were generated by nonlinear fitting with OriginPro 8.0 (OriginLab) [58]. The Tm was calculated by fitting data to the denaturation curves and using inflection-determining derivatives. Sequence retrieval and analyses The sequences of 56 MHC class I genes (including predicted genes) from bats were retrieved from the NCBI database (S1 Table). Higher mammal MHC I heavy chain sequences were retrieved from the Immuno Polymorphism Database (IPD) (www.ebi.ac.uk/ipd/mhc) and the UniProt database (www.uniprot.org). Previously deposited marsupial (opossum, tammar wallaby, koala, Tasmanian devil) and platypus MHC I transcripts were included in these analyses (S1 Table). Sequence alignments were generated with ClustalX [48] and ESPript [49]. Similarities were calculated using DNAMAN (https://www.lynnon.com/). The proteomes of 1,000 MERS-CoV genomes and 1,000 SARS-CoV genomes were retrieved from GenBank, respectively. After sequence alignment with MAFFT, the dominant amino acid for each site was elected as a reference sequence. The mutation frequency = the number of overall mutations for each amino acid/(the number of occurrences of the amino acid in the reference sequence×total number of sequences). Peptide synthesis and preparation of expression constructs To screen potential peptides for binding to Ptal-N*01:01, the proteomes of the bat-related viruses EBOV (NP: GenBank no. AF054908.1; GP: GenBank no. AKG65250.1), MERS-CoV (GenBank no. AXN92228.1), H17N10 influenza-like virus (A/little yellow-shouldered bat/Guatemala/060/2010(H17N10)), and H18N11 influenza-like virus (A/flat-faced bat/Peru/033/2010(H18N11)) were utilized to predict the candidate peptides. The candidate peptides were predicted and selected according to the recently reported motif, by which the two Ptal-N*01:01–binding peptides, HeV1 and HeV2, derived from HeV were also synthesized (S2 Table) [30]. The potential binding scores of the selected peptides were also predicted through the online NetMHCpan 4.0 server (http://www.cbs.dtu.dk/services/NetMHCpan/) [50] and Rosetta FlexPepDock, which is based on structure modeling [51,52], so that we prefer choose peptides that conform to the motif of Ptal-N*01:01 [30]. The peptide purity was determined to be >95% by analytical HPLC and mass spectrometry. The peptides were stored at −80°C as freeze-dried powders and were dissolved in DMSO before use. The cDNAs for the heavy chain of P. alecto MHC I Ptal-N*01:01 (GenBank no. KT987929) [30] and bat β2m (GenBank no. XP_006920478.1) were synthesized (Genewiz, Beijing, China). Ptal-N*01:01 sequence was deposited to GenBank by Wynne and colleagues, and Ptal-N*01:01–binding peptides HeV1 and HeV2 were identified in their study [30]. Although Ng and colleagues reported the first Ptal-N*01:01 [29], the sequence is not available online. To investigate the function of Met52Asp53Leu54 in Ptal-N*01:01, a mutant termed Ptal-N*01:01(-3aa) with a deletion of these three amino acids was constructed. The amplified products expressing the extracellular domain (residues 1–277) of Ptal-N*01:01 and bat β2m (residues 1–98) were cloned into a pET28a vector (Novagen). The expression plasmid for human β2m (residues 1–99) was previously constructed in our laboratory [53]. Refolding and purification of bat class I complexes Renaturation and purification of Ptal-N*01:01 assembled with peptides were performed as previously described [54,55]. Generally, bat MHC I Ptal-N*01:01 heavy chain and bat β2m were overexpressed as inclusion bodies in the BL21(DE3) strain of Escherichia coli, and the purified inclusion bodies of the proteins were solubilized in 6 M guanidine-HCl buffer with a concentration of 30 mg/mL. Then, injection and dilution of MHC heavy chain, β2m, and peptide occurred at a molar ratio of 1:1:3 in refolding buffer (100 mM Tris-HCl [pH 8], 2 mM EDTA, 400 mM L-Arg, 0.5 mM oxidized glutathione, and 5 mM reduced glutathione) [34]. After 24 hours for protein refolding, the Ptal-N*01:01 complexes were concentrated and exchanged into a buffer of 20 mM Tris-HCl (pH 8) and 50 mM NaCl and then purified using a Superdex 200 16/60 HiLoad (GE Healthcare, Beijing, China) size-exclusion column. Crystallization, data collection, and processing Crystallization was performed using the sitting drop vapor diffusion technique. The Ptal-N*01:01/peptide complexes were screened through Crystal Screen kit I/II, Index Screen kit, PEGIon kit I/II, and the PEGRx kit (Hampton Research). Plates were incubated at 291 K and 277 K and assessed for crystal growth after 1–2 weeks. Ptal-N*01:01/HeV1 crystals were observed in 0.2 M NaCl, 0.1 M Bis-Tris (pH 5.5), and 25% (w/v) polyethylene glycol 3,350 at a concentration of 7.5 mg/mL. Ptal-N*01:01/HeV1(human β2m) crystals were observed in 0.1 M HEPES, pH 7.0, 2% w/v polyethylene glycol 3,350. Monomer bat β2m were grown in 0.1 M BIS-TRIS, pH 6.5, 8% w/v polyethylene glycol monomethyl ether 5,000. Single crystals of Ptal-N*01:01/HeV2 were grown in 0.075 M HEPES (pH 7.5), 15% (w/v) polyethylene glycol 10,000, and 25% (v/v) glycerol at a protein concentration of 10 mg/mL. Single crystals of Ptal-N*01:01/EBOV-NP1 were grown in 0.1 M succinic acid (pH 7.0) and 15% (w/v) polyethylene glycol 3,350. Single crystals of Ptal-N*01:01/EBOV-NP2 were grown in 0.2 M ammonium acetate, 0.1 M Tris (pH 8.0), and 16% (w/v) polyethylene glycol 10,000. Single crystals of Ptal-N*01:01(-3aa)/HeV1 were grown in 0.2 M sodium formate and 20% (w/v) polyethylene glycol 3,350. Single crystals of Ptal-N*01:01/H17N10-NP were grown in 0.075 M HEPES (pH 7.5), 15% (w/v) polyethylene glycol 10,000, and 25% (v/v) glycerol. Single crystals of Ptal-N*01:01/MERS-CoV-S3 were grown in 0.2 M sodium acetate trihydrate and 20% (w/v) polyethylene glycol 3,350. For cryoprotection, crystals were transferred to reservoir solutions containing 20% glycerol and then flash-cooled in a stream of gaseous nitrogen at 100 K. X-ray diffraction data were collected at beamline BL19U of the Shanghai Synchrotron Radiation Facility. The data collection statistics are shown in Table 1. Structure determination and analyses The collected intensities were subsequently processed and scaled using the Denzo program and the HKL2000 software package (HKL Research). The structures were determined using molecular replacement with the program Phaser MR in CCP4 [56]. The model used was the structure coordinates with Protein Data Bank (PDB) code 5F1I [35], and restrained refinement was performed using REFMAC5 from CCP4. Extensive model building was performed by hand using COOT [57]. The stereochemical quality of the final model was assessed with the program REFINE in Phenix or CCP4 (Table 1). Structure-related figures were generated using PyMOL (http://www.pymol.org/) and COOT. Determination of protein thermostability using CD spectroscopy The thermostabilities of Ptal-N*01:01 with two group key peptides were tested by CD spectroscopy. All complexes were refolded, purified, and measured at 0.2 mg/mL in a solution of 20 mM Tris (pH 8) and 50 mM NaCl. CD spectra at 218 nm were measured on a Chirascan spectrometer (Applied Photophysics) using a thermostatically controlled cuvette at temperature intervals of 0.2°C at an ascending rate of 1°C/minute between 20 and 90°C. The unfolded fraction (%) is expressed as (θ−θa)/(θa−θb), where θa and θb are the mean residue ellipticity values in the fully folded and fully unfolded states, respectively. The denaturation curves were generated by nonlinear fitting with OriginPro 8.0 (OriginLab) [58]. The Tm was calculated by fitting data to the denaturation curves and using inflection-determining derivatives. Supporting information S1 Fig. Structure-based sequence alignment of Ptal-N*01:01 and other bat MHC I molecules. Coils indicate α-helices, and black arrows indicate β-strands. Residues highlighted in red are completely conserved, and residues in blue boxes are highly (80%) conserved, with consensus amino acids in red. Residues forming the B pocket are marked with a yellow background and the F pocket with blue. The key residues in the pocket are marked with red five-pointed stars. Special insertion positions in Ptal-N*01:01 are marked with red arrows. The sequence alignment was generated with MEGA7, ClustalX, and ESPript. MHC, major histocompatibility complex. https://doi.org/10.1371/journal.pbio.3000436.s001 (TIF) S2 Fig. Structure-based sequence alignment of representative MHC I molecules from bats, marsupials, and higher mammals. (A) Coils indicate α-helices, and black arrows indicate β-strands. Residues highlighted in red are completely conserved, and residues in blue boxes are highly (80%) conserved, with consensus amino acids in red. Special insertion positions in Ptal-N*01:01 are marked with red arrows. The sequence alignment was generated with MEGA7, ClustalX, and ESPript. (B) Structure-based sequence alignment of β2m derived from different mammals. Residues binding to α1α2 domains of the Ptal-N*01:01 heavy chain were labeled by black triangles. Residues binding to α3 domains of the Ptal-N*01:01 heavy chain were labeled by red triangles. The conserved residues between bat β2m and human β2m were labeled by filled triangles, and the variable residues between bat β2m and human β2m were labeled by hollow triangles. β2m, β2-microglobulin; MHC, major histocompatibility complex. https://doi.org/10.1371/journal.pbio.3000436.s002 (TIF) S3 Fig. The conformations and electron density maps of Ptal-N*01:01–presented peptides. The authentic conformations of HeV1 (A), HeV2 (B), H17N10-NP (C), EBOV-NP1 (D), EBOV-NP2 (E), and MERS-S3 (F) presented by Ptal-N*01:01 are shown through the 2Fo-Fc electron density maps contoured at a contour of 1.0σ viewed in profile through the α2-helix. The electron density maps were constructed from model phases, omitting the peptides. The peptides are displayed as sticks in different colors. The hypothetical P3–P8 residues of the peptide MERS-S3 with poor electron densities were denoted as dashed orange lines. EBOV, Ebola virus; HeV, Hendra virus; MERS, Middle East respiratory syndrome. https://doi.org/10.1371/journal.pbio.3000436.s003 (TIF) S4 Fig. The surface profile of pocket A of Ptal-N*01:01 compared with the MHC I molecules from other vertebrates. The surface profile of the A pocket of bat MHC I Ptal-N*01:01 was compared with the A pockets of MHC I from other vertebrates, including human HLA-A*0201, HLA-A*2402, murine H-2Kd, chicken BF2*04, and BF2*21. The yellow arrows indicate the vacant edge of the A pockets of the MHC I from vertebrates other than bat. All of the pockets are shown as semitransparent electron density maps, under which the Asp59 and Arg65 of Ptal-N*01:01 and the corresponding residues in other MHC I were shown as sticks. The P1 anchors in different MHC I were represented with gray sticks and spheres. MHC, major histocompatibility complex; P1, position 1. https://doi.org/10.1371/journal.pbio.3000436.s004 (TIF) S5 Fig. The consistent hydrogen bond network in the A pockets of six Ptal-N*01:01 structures. The A pockets of bat MHC I Ptal-N*01:01 structures complexed with peptides HeV1 (A), HeV2 (B), H17N10-NP (C), EBOV-NP1 (D), EBOV-NP2 (E), and MERS-CoV-S3 (F) from different viruses. Residues Asp59 and Arg65 in the A pocket of Ptal-N*01:01 are shown as sticks, and the P1 anchor Asp of these peptides are represented as sticks and spheres. The hydrogen bonds are denoted in dashed lines. The heavy chains of different Ptal-N*01:01 structures are shown in white cartoon. EBOV, Ebola virus; HeV, Hendra virus; MERS-CoV, Middle East respiratory syndrome coronavirus; MHC, major histocompatibility complex; P1, position 1. https://doi.org/10.1371/journal.pbio.3000436.s005 (TIF) S6 Fig. The binding capabilities of long peptides with Ptal-N*01:01. (A) Binding of long peptides (11-mers to 15-mers) to Ptal-N*01:01 elucidated by in vitro refolding. Twenty long peptides (peptide Bat1 to Bat20) that were previously eluted from Ptal-N*01:01–expressing cells were synthesized (S4 Table) [30]. The gray curve is a negative control without any peptide in the refolding reaction. (B) Capability of naturally N-terminally extended peptides HeV1 (DFANTFLP), MERS-CoV-S7 (DFNLTLLEP), and EBOV-NP1 (DFQESADSFL) to renature Ptal-N*01:01. EBOV, Ebola virus; HeV, Hendra virus; MERS-CoV, Middle East respiratory syndrome coronavirus. https://doi.org/10.1371/journal.pbio.3000436.s006 (TIF) S7 Fig. The identification of marsupial related virus-derived peptides binding to Trvu-UB*01. (A) Structure-based sequence alignment of Ptal-N*01:01 and Trvu-UB*01. (B-C) The peptide predictions refer to pocket features. The binding of peptides derived from possum nidovirus with Trvu-UB*01 were evaluated by co-refolding; the peptide sequence is listed. https://doi.org/10.1371/journal.pbio.3000436.s007 (TIF) S1 Table. Sequence information of bats and typical marsupials. Raw data corresponding to Fig 1. https://doi.org/10.1371/journal.pbio.3000436.s008 (DOCX) S2 Table. Characteristics of the virus-peptides used in this study. Raw data corresponding to Fig 2. https://doi.org/10.1371/journal.pbio.3000436.s009 (DOCX) S3 Table. Binding assays of Ptal-N*01:01 with HeV1/HeV2 and its mutants. Peptides’ information corresponding to Fig 4. HeV, Hendra virus. https://doi.org/10.1371/journal.pbio.3000436.s010 (DOCX) S4 Table. Long peptides’ information corresponding to S6 Fig. https://doi.org/10.1371/journal.pbio.3000436.s011 (DOCX) S5 Table. Mutant peptides’ information corresponding to S6 Fig. https://doi.org/10.1371/journal.pbio.3000436.s012 (DOCX) S1 Data. Numerical data underlying Figs 1A, 1C, 1D, 4H, 4J, 5I, 5L and 6G. https://doi.org/10.1371/journal.pbio.3000436.s013 (XLSX) S2 Data. Sequence information and Python script corresponding to Fig 6H. https://doi.org/10.1371/journal.pbio.3000436.s014 (ZIP) Acknowledgments We thank Dr. Jianhui Li (Institute of Biophysics, Chinese Academy of Sciences) for instruction in circular dichroism spectroscopy.
The world needs BRICS countries to build capacity in invasion scienceMeasey, John;Visser, Vernon;Dgebuadze, Yury;Inderjit, ;Li, Bo;Dechoum, Michele;Ziller, Silvia R.;Richardson, David M.
doi: 10.1371/journal.pbio.3000404pmid: 31536486
Introduction All countries suffer from increasing problems with invasive species, but there is a divide between rich and poor nations in terms of progress in tackling this issue [1]. Developing countries are unlikely to meet Aichi Target 9 of the Convention on Biological Diversity (CBD) without help [2]. Emerging economies—such as Brazil, Russia, India, China, and South Africa (BRICS)—sit in between; they are experiencing increasing international and national trade but have limited capacity to translate research needed to inform relevant policy in their context. They are rapidly developing economies, but a large proportion of their populations have subsistence livelihoods although they have a growing footprint in both the developed and developing world. To this end, these countries have formed a forum (BRICS), primarily to discuss economics but also to interact on other issues of common interest. BRICS countries make up 26% of the terrestrial surface of the earth, have 42% of the planet’s human population and 14% of global GDP, and are home to a large proportion of the world’s biodiversity (e.g., the Brazilian Amazon, Cerrado, and Atlantic Forest; Russia’s Caucasus and Far East; Indian Western Ghats, Himalayas; Southwestern China; and South Africa’s Cape Floristic Region, Succulent Karoo, and Maputo-Pondoland-Albany). This biodiversity is under threat from anthropogenic drivers, including habitat conversion, exploitation, climate change, pollution, and species introductions [3]. Rapid economic growth in BRICS countries requires increasing trade, but not at the expense of their natural capital. This presents substantial challenges for legislation and enforcement in the absence of appropriate models in the developed or developing world. One such challenge is invasive species, which were estimated to cost at least 0.1 billion US dollars (USD) per annum to the United States economy 2 decades ago [4], although a more recent annual cost of 1.7 billion USD to the UK economy [5] suggests that the current cost has escalated appreciably. The relationship between increasing economic activity and invasive species is well established [6], but even developed countries have been slow to implement and enact legislation curtailing the increasing effects of invasions. BRICS nations have growing trade both within and outside their borders, which is conducted at both large commercial and small artisanal scales. They are all signatories to the CBD and are thus currently preparing their responses to Aichi Target 9; these include the need to recognize invasive species, as well as determine their pathways of spread by 2020. Once the Aichi targets have been met, the CBD will set new targets relating to invasive species. Meeting new targets will require growing national capacity of invasion scientists with knowledge that relates to specific biomes within each BRICS nation. We posit that coordinating research and building capacity in invasion science deviates from a concentrated institute; instead, we propose the formation of a facilitated network of extant invasion biologists and social scientists with specialties across the biomes of BRICS countries. Furthermore, we present a model for this network and suggest how it could be implemented by BRICS countries to meet the next set of CBD targets in 2030. The invasion paradox in BRICS nations Unlike smaller, less diverse nations, BRICS countries suffer from invasive species that originate both within and outside of their borders. These ‘domestic exotic’ or ‘extra-limital’ invasions are especially relevant in large biodiverse countries such as the BRICS nations because there is often confusion regarding their status within the country [7] and consequently regulatory constraints. BRICS countries share invasive plant and animal species (Table 1A) and are the donors of some of the world’s most highly impacting species (Table 1B). Increasing trade from BRICS countries means that, unless unchecked, these areas are likely to become major donors of invasive species to the broader globe. Among the best predictors of invasive species are propagule pressure, commensurate with increasing trade, and climate suitability [8]. The diverse biomes and climatic conditions within the borders of BRICS countries suggest that they have suitable climatic diversity to cover all but the coldest biomes on Earth (Fig 1; S1–S5 Figs). Our analysis suggests that most of North America and Europe match climates in 3 or more biomes within BRICS countries but that Africa, South America, the Middle East, Asia, and Australia are matched by 4 or more biomes within BRICS countries. Interestingly, BRICS countries themselves have large areas with biomes that match each other. These commonalities among BRICS countries call for an interconnected facilitated network dealing with invasive species. In addition, there is increasing evidence that impacts of invasive species affect the poorest people in emerging economies worst [9], but in a world with a changing climate, effects of invasions have the potential to challenge sustainable economic development. For example, South Africa is estimated to have lost between 1.4 and 2.5 billion m3 of surface water runoff to invasive plant species, impacting drought-stricken cities like Cape Town [10]. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. When biomes from BRICS countries (labelled and with outlines) are projected onto the rest of the world, only the coldest areas are not represented by any biomes (green). That so many of the world’s terrestrial areas are climate matched with BRICS countries suggests that many species from these emerging economies could become invasive elsewhere. Note that the colour scheme adds biomes that are present per BRICS country (and does not indicate different biomes), therefore the maximum number is 5. The volume of trade from BRICS countries (in USD) in 2018 is shown by the thickness of connecting lines (see S1 Text for details). BRICS, Brazil, Russia, India, China, and South Africa; USD, US dollars. https://doi.org/10.1371/journal.pbio.3000404.g001 Download: PPT PowerPoint slide PNG larger image TIFF original image Table 1. Some examples of the world’s worst invasive species that are (a) shared or (b) donated by BRICS countries. ‘N’ is native and ‘X’ is introduced; ‘X [N]’ indicates that the species is both native and introduced in different parts of each BRICS country. https://doi.org/10.1371/journal.pbio.3000404.t001 Facilitated networks in invasion science The importance of global networks for invasion science has already been highlighted [11]. This underlines the urgency for focused translational research in invasion science, as well as the increase in power that results from a collaborative approach that extends the capacity of isolated research groups. Core projects undertaken by the group are complimented by satellite projects, which may result in joint capacity building and skills exchange between institutions and countries; successful examples of such ventures abound [11]. Equal to the requirement of rapid responses to research of emerging alien species is the need to substantially increase the capacity to tackle existing and future invasions in BRICS countries. Achieving this will require more than the transitory international collaborative projects advocated previously but rather in-country institutions that can maintain recruitment and extend beyond the careers of individual researchers, extending scientific knowledge to applied management carried out by government institutions and the nongovernmental organization (NGO) sector. While the standard approach to building institutions is to concentrate resources at a single location, we propose that ecological science demand a facilitated network approach to respond efficiently to invasion threats and to build and maintain capacity for the future. The facilitated network revolves around a hub-and-spoke model drawing on existing excellence in invasion biology research within each country to grow capacity and collaboration over time. The hub (at the institution of the director) contains administrative staff to facilitate the network and disburse finances to Core Team Members (CTMs), already employed through their home institutions, and their associated researchers. While the hub may serve as a physical home representing the network, the network serves to study invasions in multiple contexts within the often-unique cultural and biological situations that exist elsewhere in the countries. Of key importance is the inclusion of social scientists alongside biologists; both economics and psychology are traditionally neglected fields when tackling problems—such as biological invasions—that are by definition linked to human activities and invoke complex human dimensions and the need for cultural changes. Annual research meetings bring all CTMs and students together in a conventional conference. Funding to have international plenaries present contextual, cutting-edge research should include representatives from other BRICS networks. Similarly, student awards should prioritize exchange between national BRICS meetings. Additionally, CTMs come together in a closed meeting at another time of year to discuss growth of the network and strategic directionality of research, including the planning of at least one themed workshop each year that should bring together selected CTMs and international participants to tackle emerging issues in invasion science. We suggest that within 5 years, a facilitated network of researchers in each of the BRICS countries will start making meaningful policy input as well as building capacity within their country and positively influencing their region. These facilitated networks are essentially Centres of Excellence (CoEs) that are distributed throughout the biomes of a country, with an administrative hub. South Africa implemented the CoE model in 2004, to build excellence and capacity in nationally strategic research areas. The Centre of Excellence for Invasion Biology (CIB) was one of the first 5 funded CoEs and started with 14 CTMs in 4 of South Africa’s institutions [12]. The number of CTMs has grown to 27 in 9 institutions, covering the country’s 5 principal biomes as well as freshwater, marine, and terrestrial specialists. Their affiliations span universities, national parks, and government institutes and include staff from partner institutions embedded within the hub. The CIB was initially set up with 0.3 million USD and today receives 0.74 million USD but raises a further 52% (±19%) in co-funding. Of the total, 58.2% of funds are spent on student bursaries and running costs, producing 21 graduates annually in PhD (22%), MSc (35%), Honours (17%), and post doc students (22%). CIB alumni continue in academia (33.1%) or move into governmental and implementing agencies (17.2%), NGOs (5.5%), and other private sectors mostly relating to their fields of study. The CIB has contributed substantially to South African capacity in invasion science but has also had a significant effect on neighbouring countries through capacity development and studies. CTMs and their students and international associates now produce over 200 Web of Science (WoS)-listed publications annually, with more than a quarter in Q1 journals. Research outputs are well cited and respected. For example, almost 10% of references cited in Aichi Target 9 were published by the CIB [12]. Importantly, CTMs are responsible for facilitating production of government policy documents, and South Africa is the first country in the world to produce a national status report on biological invasions and their management [10], including the first framework of indicators for reporting on biological invasions at a country level [13]. While we actively advance this model for BRICS nations, there is no reason why many other countries, both developed and developing, should not adopt a similar model. However, the opportunities provided by the existing forum and agreements that are already in place for BRICS countries make it an attractive starting point. Many developing nations make significant advancements in research and implementation to counter the effects of invasive species [14]. Increased numbers of networks of researchers dedicated to invasion science will ultimately be beneficial to all [11]. Toward a solution The facilitated network approach proposed here offers many advantages to rapidly connect existing academics working on invasions to start building capacity and augment the research foundation on which national policy is formed. Firstly, each is already established and paid within their own institution and can offer biogeographic, cultural, and institutional insights from local invasions within their working context. By connecting these individuals through bi-annual meetings, granting opportunities, and shared bursaries, we expect meaningful collaborations on common problems to arrive intra- and inter-specifically. Capacity built by the networks can be rapidly absorbed into government and NGO sectors, and there will be an assured continuation of invasion biologists in academic positions. Once established, these networks can form cross-network links—building on the response of the global network on biological invasions—to positively influence the global response to invasions among developed and developing countries alike. Supporting information S1 Methods. Supplementary methods for modelling of climate matched areas of BRICS biomes, and the acquisition of trade data. References (Supplementary Only). BRICS, Brazil, Russia, India, China, and South Africa. https://doi.org/10.1371/journal.pbio.3000404.s001 (DOCX) S1 Text. Description of biomes projected onto each of the BRICS nations. BRICS, Brazil, Russia, India, China, and South Africa. https://doi.org/10.1371/journal.pbio.3000404.s002 (DOCX) S1 Fig. Biomes from Brazil projected onto the rest of the world. https://doi.org/10.1371/journal.pbio.3000404.s003 (TIF) S2 Fig. Biomes from Russia projected onto the rest of the world. https://doi.org/10.1371/journal.pbio.3000404.s004 (TIF) S3 Fig. Biomes from India projected onto the rest of the world. https://doi.org/10.1371/journal.pbio.3000404.s005 (TIF) S4 Fig. Biomes from China projected onto the rest of the world. https://doi.org/10.1371/journal.pbio.3000404.s006 (TIF) S5 Fig. Biomes from South Africa projected onto the rest of the world. https://doi.org/10.1371/journal.pbio.3000404.s007 (TIF) Acknowledgments The authors thank the participants at the DST-NRF Centre of Excellence for Invasion Biology BRICS networks workshop in November 2018 for discussion.
TRPC channels regulate Ca2+-signaling and short-term plasticity of fast glutamatergic synapsesSchwarz, Yvonne;Oleinikov, Katharina;Schindeldecker, Barbara;Wyatt, Amanda;Weißgerber, Petra;Flockerzi, Veit;Boehm, Ulrich;Freichel, Marc;Bruns, Dieter
doi: 10.1371/journal.pbio.3000445pmid: 31536487
Introduction Short-term presynaptic plasticity is a widespread and important means of synaptic regulation, but the underlying mechanisms are not fully understood [1,2]. Calcium plays an important role in many use-dependent forms of plasticity, and STE of synaptic signaling can be changed by calcium channel facilitation, saturation of endogenous calcium buffers, or other means downstream of the voltage-gated calcium channel (VGCC)-mediated Ca2+ entry [2]. The canonical transient receptor potential channel (TRPC) family comprises 7 members (TRPC1 to TRPC7) that are able to form Ca2+-permeable nonselective cation channels [3]. TRPC channels can be activated in response to stimulation of phospholipase C-coupled receptors [4], subsequent to store depletion [5–7], or directly by intracellular Ca2+ [8–10]. They have been implicated in diverse neuronal functions, such as excitability [11–13], excitotoxicity [14], neurogenesis [15], and neurite outgrowth [16–18]. Based on phylogenetic analyses and sequence similarities, the TRPC subfamily can be divided into 3 subgroups: TRPC1/TRPC4/TRPC5, TRPC3/TRPC6/TRPC7, and TRPC2 [15,19]. Group-specific interactions between TRPC1, TRPC4, and TRPC5, but not with members of the TRPC3/TRPC6/TRPC7 subgroup, were observed in heterologous expression systems [20]. Furthermore, Trpc1, Trpc4, and Trpc5 genes were found to be co-expressed in subregions of the hippocampus, as documented by immunohistochemistry and in situ hybridization [21–25]. TRPC4 and TRPC5 are strongly potentiated [9,26] or even directly activated by elevations of intracellular calcium concentration ([Ca]i) [8]. The latter study demonstrated that Ca2+ entry through other channels, including VGCCs, directly enhanced TRPC5 activity. In contrast, TRPC1 channels do not respond to intracellular Ca2+ elevations but are inserted into the plasma membrane in response to local Ca2+ entry through Orai1 channels [27]. Although TRPC1, TRPC4, and TRPC5 have been implicated in the generation of Ca2+ signals in different secretory cells [16,18,21,25,28,29], their potential contribution to Ca2+-dependent transmitter release and neuronal communication has remained largely controversial [13,30–36]. In a recent study, we showed that genetic loss of TRPC1, TRPC4, and TRPC5 channels (TRPC triple knockout [tko]) impairs working memory and flexible relearning, most likely by reducing synaptic transmission of hippocampal neurons [37]. Yet, the underlying molecular mechanisms of how these TRPC channels are able to regulate synaptic signaling have remained unclear. Here, we delineate so far unprecedented TRPC actions in the regulation of synaptic vesicle release. Using electrophysiological recordings of hippocampal neurons from TRPC-deficient mice and from a novel Cre knock-in mouse line, which genetically labels primary TRPC5-expressing neurons, along with presynaptic Ca2+ measurements and lentiviral TRPC expression experiments, we provide first evidence that TRPC5 channel activity efficiently regulates presynaptic Ca2+-dynamics and controls synaptic vesicle recruitment. Collectively, our data show that TRPC5 channels play a pivotal role in the regulation of short-term synaptic plasticity in neurons. Results Loss of TRPC1/C4/C5 reduces the replenishment rate and the readily releasable pool size of vesicles To elucidate the mechanisms how genetic loss of TRPC1/C4/C5-channels affects the action potential (AP)-evoked postsynaptic current (EPSC), we used high-frequency train stimulations (20 Hz for 2 s) in autaptic hippocampal wild-type (wt) and TRPC1/C4/C5 tko neurons. Synaptic responses were analyzed with respect to the readily releasable pool (RRP) size, release probability (Pr), paired-pulse ratio (PPR), replenishment rate, and synaptic depression. Loss of TRPC channels significantly reduced the amplitude of the first EPSC during the train (Fig 1A, 1B and 1D) and clearly accelerated the time course of synaptic depression when compared with controls (Fig 1C). Although wt cells showed a clear heterogeneity in synaptic signaling ranging from short-term enhancement (STE, AmpEPSC10/AmpEPSC1 ratio >1) to short-term depression (STD; AmpEPSC10/AmpEPSC1 ratio <1) during HFS, tko neurons predominantly displayed STD (Fig 1E–1G). Furthermore, the synchronous and asynchronous phase of the total synaptic charge transfer were similarly diminished (Fig 1H–1K), indicating that factors upstream of the RRP are affected by genetic loss of the TRPCs. To determine the RRP size, the data plot of the cumulative synchronous EPSC charge was approximated with a linear regression fitting the last 5 stimuli [38] (Fig 1L). Back-extrapolating the linear component of the steady-state phase renders an estimate of the initial RRP size (Fig 1L). Indeed, loss of TRPC channels strongly reduced the RRP size (Fig 1M) but left the Pr (i.e., first EPSC charge divided by the RRP charge) unchanged (Fig 1N). To determine the replenishment rate of vesicles during pool depletion, the slope of the plot of the cumulative total charge was approximated by linear regression fitting over the last 5 stimuli (Fig 1O). The replenishment rate was strongly reduced in tko neurons compared with controls (Fig 1P). Taken together, these results suggest that TRPC channels influence vesicle recruitment and thereby regulate the RRP size and the time course of synaptic depression during HFS. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. TRPC channels modulate the vesicle replenishment rate and the RRP size. (A) Exemplary EPSCs triggered by HFS (20 Hz, 40 AP/2 s) of autaptic wt and TRPC1/C4/C5 tko neurons (inset magnified view on the first 4 EPSCs of the train response. Arrows depict the synchronous and asynchronous phase of secretion). (B and C) TRPC deficiency reduces the EPSC amplitudes and accelerates the time course of synaptic depression (In panel C, data were normalized to the first EPSC amplitude). (D) The first EPSC amplitude is significantly reduced in tko cells (wt, n = 63; tko, n = 39 cells; p = 1.9 × 10−9). (E and F) The AmpEPSC10/AmpEPSC1 ratios of wt neurons range from STD to STE. Loss of TRPCs shifts the AmpEPSC10/AmpEPSC1 ratio towards STD. (G) The AmpEPSC10/AmpEPSC1 ratio is reduced in tko cells (p = 0.00067). (H, I, and J) The total synchronous and asynchronous charges are smaller in tko cells (red) when compared with controls (black). (K) Quantification of the 40th asynchronous charge (p = 0.0000056). (L) Mean cumulative synchronous release components during the 20-Hz train. Continuous line, linear regression of the last 5 data points back-extrapolated to stimulus = 0 to estimate the initial RRP size. Note that the RRP analysis was confined to wt neurons that reached a steady-state response in the late phase of stimulus train (45 out of 63 neurons). (M) The RRP size is significantly decreased in tko cells (wt, n = 45; tko, n = 39; p = 0.0000013). N. The release probability remained unchanged between groups (p = 0.2). (O) Cumulative total charge during 20-Hz train stimulation. The slope of the linear regression from the last 4 stimulation points (continuous line) rendered an estimate of the replenishment rate in panel P. (P) The replenishment rate is significantly reduced in tko neurons (p = 1.6 × 10−10). **p < 0.01; ***p < 0.001; statistical significance was assessed by Mann-Whitney rank sum test. Underlying data can be found in S1 Data. Amp, amplitude; AP, action potential; EPSC, evoked postsynaptic current; HFS, high-frequency stimulation; RRP, readily releasable pool; STD, short-term depression; STE, short-term enhancement; tko, triple knockout; TRPC, transient receptor potential canonical; wt, wild type. https://doi.org/10.1371/journal.pbio.3000445.g001 To study the subcellular distribution of TRPC channels, we co-immunolabeled neuronal cultures with antibodies directed against TRPC5 and the presynaptic marker protein bassoon. Confocal imaging revealed clear TRPC5 staining in somatic regions and axonal branches (S1 Fig, upper panels) as well as in varicosity-like thickenings of fine axonal branches, where it colocalized with bassoon (S1 Fig, lower panels). The specificity of the TRPC labeling was verified in cultures of TRPC1/C4/C5-tko neurons, which did not exhibit a discernable TRPC5 immunosignal (S1B Fig). TRPC5 was detected in about half of the presynaptic terminals (positive for bassoon, S1E Fig), indicating a heterogeneous TRPC5 expression in hippocampal neurons. In any case, these results are in line with our functional analyses showing that TRPC activity alters synaptic efficacy. Primary TRPC5-expressing neurons show strong STE of synaptic signaling To identify primary TRPC5-expressing neurons by genetic labeling, we generated a new TRPC5-Cre knockin (KI) mouse strain (TRPC5-internal ribosomal entry site cre recombinase [IC], S1A and S1B Fig). Here, the trpc5 gene is followed by an internal ribosome entry site (IRES) and by a Cre recombinase cDNA. When crossbred with ROSA26-floxed-stop-τGFP (eR26-τGFP) reporter mice [39], the Cre recombinase excises a floxed termination sequence 5ʹ to the τGFP transgene within the ROSA26 locus resulting in constitutive reporter expression in TRPC5-expressing cells. About 8% of the hippocampal neurons prepared from TRPC5-IC/eR26-τGFP mice were found to be τGFP-positive (7.7% ± 1.5% of 831 total cells, 2 preparations) and showed a clear immunosignal for TRPC5 (S2C and S2D Fig). Yet, among neurons, which were negative for τGFP, we also observed TRPC5-expressing cells (38.5% ± 3.58% of the τGFP-negative neurons; S2D Fig), most likely because the coding sequence 3ʹ of the IRES sequence is known to be expressed at significantly lower levels than the coding sequence 5ʹ of the IRES [40]. A similar frequency of TRPC5 positive cells was found with immunolabeling cultures of wt neurons (52% ± 3%, 195 out of 375 cells, 3 preparations), showing that IRES insertion does not affect the expression of the upstream protein (S2E and S2F Fig). Thus, hippocampal neurons heterogeneously express TRPC5 and τGFP labeling (in cultures of KI mutant mice) enables the identification of neurons that reliably express TRPC5. We comparatively analyzed synaptic signaling from τGFP-positive and τGFP-negative neurons with that of tko neurons. Indeed, τGFP-positive (i.e., TRPC5-expressing) neurons showed strong STE of their synaptic response during HFS and an increased PPR (Fig 2A–2E). In contrast, tko neurons started off with lower first EPSC amplitude and showed classical STD during HFS. The AmpEPSC10/AmpEPSC1 ratio is altered for the entire population of τGFP-positive neurons (Fig 2F and 2G), reaching on average a nearly 3-fold higher ratio than in tko neurons (τGFP-positive neurons: 1.63 ± 0.11; tko neurons 0.6 ± 0.05, p < 0.001, one-way ANOVA on ranks, Fig 2H). Furthermore, the synchronous and asynchronous charge component of the evoked response and the replenishment rate recorded in τGFP-positive cells were strongly enhanced when compared with synaptic responses of tko neurons (Fig 2I–2O). τGFP-negative neurons, instead, displayed an intermediate plasticity phenotype with respect to AmpEPSC10/AmpEPSC1 ratio (0.94 ± 0.09), total synaptic charge transfer, synchronous charge, and asynchronous charge as well as PPR and replenishment rate (Fig 2). These results agree with the observation that about 39% of the τGFP-negative neurons express TRPC5 (S2 Fig). Overall, the systematic changes in STP between τGFP-positive and -negative neurons as well as tko neurons correlate well with the differential expression of TRPC5 in these groups, indicating that endogenous TRPC5 activity is a key factor in regulating synaptic plasticity. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Endogenous TRPC5 channels promote STE of synaptic signaling. (A) Exemplary epifluorescence image of an autaptic TRPC5-IC/eR26-τGFP neuron. (B) Sample recordings from tko, τGFP-negative and τGFP-positive neurons upon 20-Hz HFS. (C–E) τGFP-positive cells show a strong STE of synaptic signaling and an increased PPR (panel E, EPSC2/EPSC1) when compared with tko or τGFP-negative neurons (τGFP-positive versus tko p = 0.00005; τGFP-positive versus τGFP-negative p = 0.003). (F and G) Frequency and cumulative frequency distribution of the AmpEPSC10/AmpEPSC1 ratios for tko (red), τGFP-negative (black), and τGFP-positive neurons (green). (H) The AmpEPSC10/AmpEPSC1 ratio is higher in τGFP-negative cells than in tko neurons and even further increased in τGFP-positive cells (τGFP positive versus tko p = 7.892264 × 10−9; τGFP positive versus τGFP negative p = 0.0009; τGFP negative and tko p = 0.02). (I) Time course of total charge during HFS. (J and K) The replenishment rate determined from the cumulative charge plot shown in panel J gradually increases between tko, τGFP-negative, and τGFP-positive neurons (τGFP positive versus tko p = 1.5 × 10−10; τGFP positive versus τGFP-negative p = 0.0001; τGFP negative and tko p = 0.002). (L and M) Time courses of synchronous and asynchronous charge transfer during the stimulus train. (N and O) The 10th EPSC charge and the 40th asynchronous charge are significantly increased in τGFP-negative and -positive neurons (10th charge: τGFP positive versus tko p = 1.7 × 10−8; τGFP positive versus τGFP negative p = 0.003; τGFP negative and tko p = 0.002; 40th charge: τGFP positive versus tko p = 0.00004; τGFP poisitive versus τGFP negative p = 0.004; τGFP negative and tko p = 0.018). Data were collected from tko (n = 25), τGFP-negative (n = 38), and τGFP-positive (n = 40) neurons. *p < 0.05, **p < 0.01, ***p < 0.001, one-way ANOVA on ranks followed by Dunn’s post hoc test; significance was also tested by Mann-Whitney rank sum test for τGFP-negative cells versus tko. Underlying data can be found in S1 Data. Amp, amplitude; AP, action potential; EPSC, evoked postsynaptic current; eR26, ROSA26-floxed-stop; HFS, high-frequency stimulation; IC, internal ribosomal entry site cre recombinase; PPR, paired-pulse ratio; STD, short-term depression; STE, short-term enhancement; tko, triple knockout; TRPC, transient receptor potential canonical; τGFP, τ-green fluorescent protein. https://doi.org/10.1371/journal.pbio.3000445.g002 TRPC activity promotes presynaptic short-term plasticity in hippocampal neurons To further study the impact of TRPC activity on synaptic signaling, we expressed TRPC1 or TRPC5 in wt neurons using lentiviral transduction. Wt neurons expressing either TRPC1 or TRPC5 developed a robust STE of the synaptic response during the stimulation train (Fig 3A–3C). TRPC channel expression shifts the frequency distribution for AmpEPSC10/AmpEPSC1 ratios to higher values, confirming general changes in STP (Fig 3D–3F). Starting from an unchanged first AP-evoked response (wt: 5.61 ± 0.56 nA; wt + C1: 5.8 ± 0.65 nA; wt + C5: 6.04 ± 0.69 nA; Fig 3B), TRPC expression significantly increased the PPR and elevated the synchronous as well as asynchronous charge transfer when compared with controls (Fig 3G–3N). Overall, a 1.5- to 2-fold higher total synaptic charge transfer (wt: 65.2 ± 5.8 pC; wt + C1: 98.27 ± 7.20 pC; wt + C5: 116.01 ± 8.79 pC) and a corresponding increase in the vesicle supply rate was observed (Fig 3H–3J). Importantly, the synaptic plasticity phenotype observed with lentiviral TRPC5 expression largely mimics that of the τGFP-positive neurons (compare with Fig 2), demonstrating that TRPC5 channels are instrumental in controlling synaptic plasticity. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Expression of either TRPC1 or TRPC5 in wt neurons turns synaptic depression into STE. (A) Sample recordings of EPSCs triggered by HFS (20 Hz, 40 AP/2 s) of autaptic wt neurons and those expressing TRPC5 or TRPC1. (B) Activity-dependent increase in the EPSC amplitudes in wt cells expressing C1 or C5. (C) TRPC expression causes STE of synaptic signaling (data were normalized to the initial peak EPSC amplitude). (D and E) Expression of TRPC1 or TRPC5 shifts the frequency distribution of the AmpEPSC10/AmpEPSC1 ratios to higher values. (F) Mean AmpEPSC10/AmpEPSC1 ratio for the indicated groups (+C1, p = 0.002; +C5, p = 0.0001; versus wt). (G) Mean paired pulse ratio (+C1, p = 0.0036; +C5, p = 0.00585; versus wt). (H) Time course of total charge transfer. (I) Mean cumulative total charge transfer during HFS. The replenishment rate was determined from the slope of the cumulative plot (last 4 data points). (J) The replenishment rate is significantly enhanced in cells expressing either C1 or C5 (+C1, p = 0.003; +C5, p = 0.00001; versus wt). (K and L) Time courses of synchronous (K) and asynchronous charge transfer (L) during the stimulus train. (M and N) The 10th EPSC charge (M) and the 40th asynchronous charge (N) are significantly increased with TRPC expression (10th charge: +C1, p = 0.00001; +C5, p = 0.0000001; versus wt; 40th asynchronous: +C1, p = 0.02; +C5, p = 0.0005; versus wt). Data were collected from the following number of cells: wt, n = 43; wt + TRPC1, n = 30; wt + TRPC5, n = 24; *p < 0.05; **p < 0.01; ***p < 0.001; one-way ANOVA on ranks followed by Dunn’s post hoc test. Underlying data can be found in S1 Data. Amp, amplitude; AP, action potential; EPSC, evoked postsynaptic current; HFS, high-frequency stimulation; PPR, paired-pulse ratio; STD, short-term depression; STE, short-term enhancement; TRPC, transient receptor potential canonical; wt, wild type. https://doi.org/10.1371/journal.pbio.3000445.g003 The short-term plasticity (STP) phenotype of TRPC-expressing cells often hindered the RRP determination because a steady-state response in the late phase of stimulus train was not reached [41]. Therefore, we made use of hypertonic sucrose stimulation to provide an estimate of the RRP size and the Pr with single AP stimulation [42]. Consistent with our results obtained with HFS, loss of TRPC1/C4/C5 reduced the AP-evoked response and the RRP size but left the Pr unchanged (S3 Fig). Furthermore, neither the expression of TRPC5 nor of TRPC1 in wt neurons significantly altered the RRP and the Pr (S3C Fig). Thus, TRPC activity does not affect the basal release probability but rather enhances the vesicle recruitment in an activity-dependent manner during HFS. Given that TRPC channels have been implicated in growth cone guidance and morphology [16,17], one might speculate that TRPC-mediated changes of synaptic signaling may at least in part be due to alterations in the number of synapses. To pursue this issue, we immunolabeled autaptic wt, tko, and wt neurons expressing either TRPC variant with the presynaptic marker protein bassoon (S4 Fig). Yet, synaptogenesis was neither affected by loss of TRPC channels nor by their lentiviral expression. To study whether TRPC channels interfere with the overall synaptic structure, cultured hippocampal neurons were co-immunolabeled for the active zone protein bassoon and the postsynaptic density protein PSD-95. Because bassoon and PSD-95 reside on either side of the synaptic cleft, we determined the degree of juxtaposition (or colocalization) of both signals up to a distance of 770 nm from the bassoon signal (S4C Fig). Quantitation of the percentage of “bassoon area” covered by PSD-95 signal revealed that neither the absence of TRPC1/4/5 nor the expression of TRPC1 or C5 in wt neurons altered the overall colocalization of bassoon and PSD95 puncta (S4D and S4E Fig). Taken together, these results render the possibility unlikely that changes in synapse number or organization contribute to the observed TRPC-mediated alterations in synaptic signaling. TRPC channel activity promotes efficient recovery from synaptic depression To further explore the role of synaptic TRPC channels in vesicle replenishment, we analyzed the recovery kinetics of phasic release from depression after HFS (S5 Fig). The degree of recovery was determined by dividing the EPSC amplitude of the test pulse (given at various time intervals after HFS) by that of the first response during the stimulus train (20 Hz; S5A and S5B Fig). A single exponential function was used to approximate the time course of recovery (see S5 Fig legend for details). The results show that expression of TRPC1 or TRPC5 in wt neurons diminished synaptic depression during the train (S5C Fig), accelerated the recovery kinetics (S5D Fig), and strongly augmented the EPSC amplitude when compared with controls (S5E Fig). In contrast, loss of TRPC activity (tko neurons) slowed down the recovery kinetics from synaptic depression when compared with controls (S5F–S5H Fig). Collectively, these results indicate that activation of synaptic TRPC channels leads to more efficient and faster mobilization of vesicles from the reserve pool. Buffering presynaptic [Ca]i abolishes TRPC-dependent STE Using photolytic uncaging experiments, we have previously shown that TRPC5 channels can directly be activated by an intracellular Ca2+ increase at the millisecond time scale [8]. Thus, it is possible that rapid activation of TRPC-mediated Ca2+ permeabilities further elevate the presynaptic Ca2+ rise during the stimulus train. To minimize the accumulation of intraterminal calcium during HFS, neuronal cultures were preincubated with the membrane-permeable calcium chelator ethylene glycol-bis(β-aminoethyl ether)-N,N,N′,N′-tetraacetic acid acetoxymethyl ester (EGTA-AM; 300 μM for 5 min; Fig 4). Owing to its slow kinetics and high affinity, EGTA buffers global Ca2+ without affecting phasic transmitter release [1]. EGTA did not change the first EPSC amplitude but significantly reduced the PPR of responses for wt neurons and those expressing additional TRPC1 or TRPC5 (Fig 4A, 4B, and 4C). In the same line, EGTA nearly abolished the delayed asynchronous secretion (Fig 4H and 4I) and caused steady-state phasic release (Fig 4D and 4F right panels; see also [38]), indicating efficient buffering of residual [Ca]i in our experiments. Importantly, wt neurons expressing either TRPC variant failed to promote any STE of synaptic signaling in the presence of EGTA (Fig 4D–4G). Both the AmpEPSC10/AmpEPSC1 ratio (Fig 4E) and the synchronous charge (Fig 4G) were significantly reduced when compared with the nontreated controls. These results indicate that enhanced buffering of global [Ca]i by EGTA abolishes the TRPC-mediated STE of synaptic signaling. Collectively, our results support a model wherein TRPC channels are rapidly activated by elevated levels of bulk [Ca]i (in response to Ca2+ entry through presynaptic voltage-gated Ca2+ channels) and in turn may amplify and prolong the presynaptic Ca2+ rise during HFS, thereby increasing the Ca2+ dependent rates of vesicle replenishment [43–45]. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Buffering presynaptic [Ca]i abolishes TRPC-dependent short-term facilitation. (A) Representative EPSC traces for wt neurons and those expressing TRPC1 or TRPC5 after EGTA-AM treatment. EGTA abolishes asynchronous secretion causing a phase of steady-state synchronous secretion and prevents the TRPC-mediated STE of synaptic signaling. (B). Neither TRPC expression (C1 or C5) nor pretreatment with EGTA affects the first EPSC amplitude of the train response (between groups without EGTA: p = 0.7; in-group with versus without EGTA: wt, p = 0.48; +C1, p = 0.49; +C5, p = 0.67). (C) EGTA significantly reduces the PPR for all groups (in-group with versus without EGTA: wt, p = 0.03; +C1, p = 0.001; +C5, p = 2 × 10−6). (D, F, H) Activity-dependent changes of EPSC amplitude (D), synchronous release (F), and asynchronous release (H) for wt neurons and those expressing TRPC1 or TRPC5 without (left panels) and with EGTA treatment (right panels). (E and G) The TRPC-mediated increase in the AmpEPSC10/AmpEPSC1 ratio (E) and the 40th synchronous EPSC charge (panel G, last pulse of the train shown in panel F) are prevented with EGTA treatment (EPSC10/EPSC1 [between groups without EGTA]: +C1, p = 0.07; +C5, p = 0.05, versus wt; in-group with versus without EGTA: wt, p = 0.5; +C1, p = 0.00004; +C5, p = 0.028; 40th synchronous [between groups without EGTA]: +C5, p = 0.002, +C1, p = 0.0014 versus wt; in-group with versus without EGTA: wt, p = 0.36; +C1, p = 0.0004; +C5, p = 5.8 × 10−9). (H and I) EGTA treatment diminishes asynchronous secretion in all groups, indicating effective buffering of presynaptic [Ca]i (+C5, p = 0.0008, +C1, p = 0.05; in-group with versus without EGTA: wt, p = 1.4 × 10−12; +C1, p = 0.0000003; +C5, p = 5.8 × 10−9). Data was collected from EGTA-treated wt (n = 45), wt + TRPC1 (n = 21), and wt + TRPC5 (n = 21) cells and nontreated wt (n = 13), wt + TRPC1 (n = 9), and wt + TRPC5 (n = 13) cells. *p < 0.05, one-way ANOVA on ranks followed by Dunn’s post hoc test for groups without EGTA; **p < 0.01, ***p < 0.001, Mann-Whitney U rank sum test for the in-group comparison with and without EGTA treatment. Underlying data can be found in S1 Data. EGTA-AM, ethylene glycol-bis(β-aminoethyl ether)-N,N,N′,N′-tetraacetic acid acetoxymethyl ester; EPSC, evoked postsynaptic current; ns, not significant; PPR, paired-pulse ratio; STE, short-term enhancement; TRPC, transient receptor potential canonical; wt, wild type. https://doi.org/10.1371/journal.pbio.3000445.g004 Homomeric TRPC5 channels promote STE of synaptic signaling TRPC4 and TRPC5 are strongly potentiated by elevation of [Ca]i [9] and can functionally couple to the Ca2+ entry through voltage-gated Ca2+ channels [8,26]. In contrast, no such regulation has been reported for TRPC1 [27], appearing contradictory to our finding that TRPC1 caused strong STE during repetitive stimulation. To address this issue, we compared the functional consequences of TRPC1 and TRPC5 expression in tko neurons (Fig 5). Expression of TRPC5 caused a strong STE of synaptic signaling (Fig 5A, 5B and 5C) and reproduced the phenotype seen with τGFP-positive wt neurons (Fig 2). TRPC5 expression significantly increased the PPR (tko: 0.9 ± 0.03, tko + C5: 1.44 ± 0.07; p < 0.001, one-way ANOVA on ranks). It shifted the distribution of AmpEPSC10/AmpEPSC1 ratios (mean ± SEM: 1.71 ± 0.14) toward the STE range (Fig 5D, 5E and 5F) and strongly accelerated the rate of vesicle recruitment (Fig 5I), leading to activity-dependent increases in total synchronous as well as asynchronous charge transfer (Fig 5G–5L). Thus, homomeric TRPC5 channels profoundly regulate synaptic plasticity. In contrast, TRPC1 expression in tko neurons failed to cause similar changes in STE-supporting synaptic responses that were indistinguishable from those in controls (Fig 5). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Homomeric TRPC5 channels promote strong STE of synaptic signaling in tko neurons. (A) Sample recordings of EPSCs triggered by HFS (20 Hz, 40 AP/2 s) of tko neurons and those expressing C1 or C5. (B) Starting from similar initial EPSC amplitudes, only TRPC5 expression causes STE of the synaptic response. (C) EPSC amplitude changes during HFS normalized to the amplitude of the first response. (D–F) The AmpEPSC10/AmpEPSC1 ratio of TRPC5-expressing tko neurons is shifted to the STE range and nearly 3-fold higher compared with tko and tko + C1 responses (panel F, +C1, p = 0.14; +C5, p = 1.9 × 10−10, versus tko). (G–I) TRPC5, but not TRPC1, expression elevates the total synaptic charge transfer and increases the replenishment rate (+C1 p = 0.1; +C5, p = 2 × 10−10, versus tko). (J and K) Time courses of synchronous (J) and asynchronous charge transfer (K) during the stimulus train. (L) The 40th asynchronous charge is significantly increased with TRPC5 expression (+C1, p = 0.98; +C5, p = 0.000005; versus tko). Data were collected from the following number of neurons: tko, n = 31; tko + TRPC1, n = 26; tko + TRPC5, n = 20; **p < 0.01, ***p < 0.001; one-way ANOVA on ranks followed by Dunn’s post hoc test. Underlying data can be found in S1 Data. Amp, amplitude; AP, action potential; EPSC, evoked postsynaptic current; HFS, high-frequency stimulation; ns, not significant; STD, short-term depression; STE, short-term enhancement; tko, triple knockout; TRPC, transient receptor potential canonical. https://doi.org/10.1371/journal.pbio.3000445.g005 This observation verifies that the synaptic plasticity phenotype observed with lentiviral expression is not simply a consequence of potential off-target effects. It further suggests that the STE phenotype of TRPC1 expression in wt neurons (Fig 3) most likely relies on the interaction with the other TRPC variants. The latter notion agrees with observations in heterologous expression systems in which no evidence for functional homomeric TRPC1 channels was found [21,46,47]. To extend these findings, we transfected mass cultures of hippocampal wt and tko neurons with either TRPC1 or TRPC5 and recorded spontaneous miniature EPSC (mEPSC) events in the presence of tetrodotoxin (TTX; 10μM; blocking voltage-gated Na+ channels [48]). Expression of TRPC5 or TRPC1 increased the mEPSC frequency in wt cells without changing the kinetics of quantal events, consistent with a presynaptic function of these channels (S6 Fig). In the absence of other TRPC isoforms, only TRPC5, but not TRPC1, expression increased the mEPSC frequency (S6C and S6D Fig). Taken together, homomeric TRPC5 channels profoundly regulate synaptic plasticity and elevate the rate of spontaneous release. In contrast, TRPC1 differentially affects synaptic signaling in wt and tko neurons, most likely because it requires heteromultimerization with other members of its TRPC subgroup to form functional channels. Acute perturbation of TRPC activity interferes with synaptic signaling To study whether acute inhibition of TRPC5 channels impairs synaptic efficacy, neurons were superfused with the TRPC5 inhibitor clemizole (3 μM, S7 Fig) [49]. In tko neurons expressing TRPC5, antagonist application significantly decreased STE of synaptic signaling and strongly reduced the asynchronous release when compared with the previous control response (S7B and S7C Fig). In contrast, clemizole neither affected synaptic depression nor asynchronous release in tko neurons, showing the specificity of the antagonist (S7D and S7E Fig). Collectively, the observed STE phenotype is due to immediate activation of TRPC channels during HFS, rather than being caused by developmental or compensatory mechanisms in response to TRPC channel expression. We next investigated whether direct activation of endogenous TRPC channels by the specific TRPC4/C5-agonist Englerin A [50] influences synaptic vesicle exocytosis. Autaptic wt neurons responded to agonist application in the presence of TTX (10 μM) with a reversible inward current (maximum current amplitude: 558 ± 68 pA, n = 36), whereas no significant current response could be detected in tko cells, verifying the specificity of Englerin A (Fig 6A, 6B and 6C). Interestingly, TRPC-mediated permeability changes were often paralleled by a transient increase in mEPSC frequency (Fig 6D, green line), without changing the amplitude or kinetics of the mEPSCs (Fig 6E). This suggests that activation of endogenous TRPC4/5 channels can directly evoke synaptic vesicle exocytosis. Some wt neurons (12 out of 36 cells) failed to respond to Englerin A (mEPSC frequency increase <1.2-fold), which could be due to heterogeneities in the expression or subcellular localization of TRPC4/C5 in hippocampal neurons. On average, Englerin A evoked a 2-fold increase in the mEPSC frequency of wt neurons (2.0 ± 0.28-fold increase over baseline levels, Fig 6F) but had no effect in tko neurons. To study the ionic basis of the inward current evoked by Englerin A, we repeated these experiments in neurons, which are genetically deficient for vesicular soluble N-ethylmaleimide-sensitive-factor attachment receptor (SNARE) proteins and are devoid of any glutamate release [48,51]. The results show that Englerin A evokes similarly large inward currents in the absence of any neurotransmitter release (S8 Fig). Thus, the observed inward current can be largely attributed to direct activation of TRPC channels rather than to secondary activation of glutamate receptors as a consequence of TRPC activity. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. Englerin A activates presynaptic TRPC channels and increases the mEPSC frequency. (A) Exemplary recordings of autaptic wt and TRPC1/C4/C5-tko neurons superperfused with Ringer (R) solution containing Englerin A (Eng, 1 μM). Englerin A evoked an inward current in wt but not tko cells. Right panels, expanded timescale of the recording at the indicated time points. Note the clear mEPSC frequency increase in wt neurons with Englerin A application (2). (B) Percentage of cells responding to Englerin A with an inward current (p = 0.008). (C) Quantification of the maximum inward current amplitude (wt, n = 36; tko, n = 16; p = 4.3 × 10−9) determined at the end of Englerin A application relative to baseline current. (D) Time course of the averaged inward current (black) and the corresponding mEPSC frequency (green) during Englerin A application (40–70 s) in wt neurons (n = 11). (E) mEPSC amplitudes (determined for the cells shown in panel D) remain unchanged during Englerin A application. Insets depict averaged mEPSCs during Ringer (R, n = 72) and Englerin A (E, n = 65) application. (F) Englerin A evokes a 2-fold increase in mEPSC frequency (relative to the mEPSC frequency before drug application) in wt but not in tko neurons (wt, n = 36; tko, n = 16; p = 0.00004). **p < 0.01; ***p < 0.001, Mann-Whitney rank sum test. Underlying data can be found in S1 Data. mEPSC, miniature excitatory evoked postsynaptic current; tko, triple knockout; TRPC, transient receptor potential canonical; wt, wild type. https://doi.org/10.1371/journal.pbio.3000445.g006 Collectively, these results are in line with an at least partial presynaptic localization of TRPC5 channels (S1 Fig) and compatible with the mEPSC frequency increase observed upon lentiviral TRPC expression (S6 Fig). They show that acute perturbation of TRPC activity is able to regulate STP and synaptic vesicle exocytosis. Enhanced Ca2+-entry through VGCCs does not mimic the TRPC phenotype TRPC channels may either directly mediate Ca2+ influx into synaptic terminals or indirectly modulate synaptic plasticity through facilitated opening of voltage-gated Ca2+ channels. In the latter case, TRPC-channel–mediated depolarization of the membrane potential could ease VGCC opening and enhance Ca2+ entry into the presynaptic terminal. Broadening the presynaptic AP-width with the potassium channel blocker tetraethylammonium (TEA, 300μM) has been shown to enhance the VGCC-mediated Ca2+ influx and synaptic transmission during high-frequency stimulation [45]. Compared to the preceding control response, acute TEA application significantly increased the initial EPSC amplitude and decreased the PPR leading to an overall faster synaptic depression (Fig 7A–7D). These changes are paralleled by an increase in release probability (Fig 7G) and clearly contrast the TRPC phenotype. TEA also enhanced the synchronous release component and the RRP size. It furthermore augmented asynchronous secretion and increased the replenishment rate, consistent with a prolonged and stronger Ca2+ entry into synaptic terminals (Fig 7E–7K). Overall, the enhanced Ca2+ influx through VGCCs leads to stronger STD and clearly differs from the STE phenotype observed with higher TRPC activity. Thus, it is unlikely that synaptic TRPC channels merely increase the Ca2+ influx through depolarization-dependent modulation of VGCCs. The combined set of data suggests that TRPC channels provide an additional Ca2+ entry pathway most likely distal to the active zone, enabling efficient mobilization of dormant vesicles from the reserve pool and leading to STE of synaptic signaling during HFS. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 7. Elevated Ca2+ entry through VGCCs increases the EPSC amplitude and accelerates synaptic depression. (A) Representative EPSC recordings of a wt neuron (20 Hz, 40 AP/2 s) before and after TEA application (300 μM). (B) Time course of the EPSC amplitude for the first (Ringer) and the second train response (+TEA); right panel, data normalized to the first EPSC amplitude. Note that TEA increases the degree of STD. (C and D) TEA increases the initial EPSC amplitude (C) and decreases the PPR (D) (first amp: p = 0.00003; PPR: p = 0.0002). (E) Time course of the EPSC synchronous charge for Ringer and TEA (left) and its cumulative plot (right). Continuous line, linear regression of the last 5 data points to estimate the initial RRP size (shown in panel F). (F and G) The RRP size (F) and the Pr (G) are significantly larger with TEA (RRP: p = 0.0007; Pr: p = 0.24). (H) TEA increases asynchronous release. (I) Mean asynchronous release of the 40th EPSC (p = 0.0001). (J) Time course of the cumulative total synaptic charge transfer; dashed lines, linear regression of the last 4 data points to estimate the replenishment rate shown in panel K. (K) TEA elevates the replenishment rate (p = 0.003). Data was collected from 16 cells, *p < 0.05; **p < 0.01; ***p < 0.001; Student paired t test. Underlying data can be found in S1 Data. AP, action potential; EPSC, evoked postsynaptic current; PPR, paired-pulse ratio; Pr, release probability; RRP, readily releasable pool; STD, short-term depression; TEA, tetraethylammonium; VGCC, voltage-gated calcium channel; wt, wild type. https://doi.org/10.1371/journal.pbio.3000445.g007 TRPC channels augment the presynaptic Ca2+ rise upon HFS To study how TRPC channels influence presynaptic Ca2+ dynamics, we combined electrophysiological recordings of synaptic activity with presynaptic Ca2+-imaging using the vesicle-associated Synaptophysin-GCaMP6s (SyGCaMP6s) fusion protein as a presynaptic Ca2+ reporter (Fig 8). In preparatory work, we verified that SyGCaMP6s is sorted to synaptic sites, as illustrated by its high degree of colocalization with the synaptic vesicle protein synaptobrevin II (SybII; Fig 8A). Autaptic neurons were stimulated with 20 Hz HFS for 2 s, while the synaptic charge transfer and [Ca]i changes at discrete synaptic sites were monitored simultaneously throughout the experiment. Discrete synaptic regions were analyzed when the fluorescence increase (ΔF/F0) exceeded 3 SDs of the background noise (ΔF/F0, mean ± SEM: 0.022 ± 0.00012). Wt neurons responded to electrical stimulation with a robust, activity-dependent increase in SyGCaMP fluorescence (Fig 8B and Fig 8C). Genetic loss of TRPC1/C4/C5 reduced the presynaptic Ca2+ rise, whereas the expression of either TRPC1 or TRPC5 strongly elevated presynaptic Ca2+ dynamics (Fig 8B, 8E and 8D). The slope of the fluorescence increase during HFS was nearly 2-fold higher in TRPC-expressing cells compared with wt neurons (Fig 8E), indicating that TRPC channels directly potentiate the VGCC-mediated increase in presynaptic Ca2+. Even after HFS, when VGCCs have closed, [Ca]i continued to increase more strongly in TRPC-expressing neurons than in wt and tko neurons (Fig 8F and 8G). These observations provide strong evidence that TRPC channels establish an additional Ca2+ entry pathway that functionally couples to the VGCC-mediated Ca2+-influx and is able to prolong the presynaptic Ca2+ signal. Importantly, alterations of the average vesicle replenishment rate determined from simultaneous electrophysiological recordings correlate well with the observed changes in presynaptic Ca2+ levels among the different groups (Fig 8I and 8J). Taken together, these observations demonstrate that TRPC channels augment and prolong the presynaptic Ca2+ increase upon HFS and, by this, set the pace of synaptic vesicle recruitment. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 8. TRPC channels augment the presynaptic Ca2+-rise upon HFS. (A) Exemplary images of neurons expressing the vesicular protein SybII-mRFP (left) and Synaptophysin-GCaMP6s (SyGCaMP6s, middle). Note the high degree of colocalization between synaptobrevinII and SyGCaMP6s (right). (B) Sample difference images (ΔF/F0) of SyGCamp6s signaling recorded (5 Hz) from autaptic neurons (wt, left; tko, middle; wt + TRPC5, right) during the HFS (20 Hz, 2 s); insets are corresponding EPSC recordings from the same cell. Arrows are exemplary ROIs used to monitor ΔF/F0 at single synaptic sites. (C) Loss of TRPC channels decreases the presynaptic Ca2+ rise, whereas expression of either TRPC variant strongly increases the Ca2+ signal. (D) Maximum ΔF/F0 from the data shown in (C), tko, p = 0.006; +C1, p = 0.0076; +C5, p = 0.0023. (E) TRPC deficiency decreases, whereas TRPC expression increases the slope of ΔF/F0 during HFS (slope between the 6th and the 13th data point; tko, p = 0.003; +C1, p = 0.035; +C5, p = 0.002. (F) Expansion of the early phase of the plot shown in panel C illustrating the prolonged Ca2+ rise after HFS. (G) Expression of either TRPC variant significantly enhances the Ca2+ influx right after HFS (slope determined between the 14th and 20th data point; tko, p = 0.48; +C1, p = 0.025; +C5, p = 0.0003). (H) Corresponding mean cumulative total charge transfer of the neurons imaged in panel C. (I and J) The replenishment rate (determined from the slope of the cumulative plot, last 4 data points, shown in panel H, is significantly changed by altering TRPC expression (I) and correlates with changes in SyGCaMP6s (slope of ΔF/F0) during HFS); tko, p = 0.0002; +C1, p = 0.009; +C5, p = 0.0004. Data were collected from wt, n = 15; tko, n = 9; wt + TRPC1, n = 11; wt + TRPC5, n = 15; *p < 0.05; **p < 0.01; ***p < 0.001; one-way ANOVA on ranks followed by Dunn’s post hoc test versus wt. wt versus tko Mann-Whitney rank sum test. Underlying data can be found in S1 Data. AP, action potential; EMCCD, electron multiplying charge coupled device; EPSC, evoked postsynaptic current; HFS, high-frequency stimulation; mRFP, monomeric red fluorescent protein; ROI, region of interest; SybII, synaptobrevin II; SyGCaMP6s, Synaptophysin-GCaMP6s; tko, triple knockout; TRPC, transient receptor potential canonical; VM, membrane voltage; wt, wild type; ΔF, delta fluorescence. https://doi.org/10.1371/journal.pbio.3000445.g008 Loss of TRPC1/C4/C5 reduces the replenishment rate and the readily releasable pool size of vesicles To elucidate the mechanisms how genetic loss of TRPC1/C4/C5-channels affects the action potential (AP)-evoked postsynaptic current (EPSC), we used high-frequency train stimulations (20 Hz for 2 s) in autaptic hippocampal wild-type (wt) and TRPC1/C4/C5 tko neurons. Synaptic responses were analyzed with respect to the readily releasable pool (RRP) size, release probability (Pr), paired-pulse ratio (PPR), replenishment rate, and synaptic depression. Loss of TRPC channels significantly reduced the amplitude of the first EPSC during the train (Fig 1A, 1B and 1D) and clearly accelerated the time course of synaptic depression when compared with controls (Fig 1C). Although wt cells showed a clear heterogeneity in synaptic signaling ranging from short-term enhancement (STE, AmpEPSC10/AmpEPSC1 ratio >1) to short-term depression (STD; AmpEPSC10/AmpEPSC1 ratio <1) during HFS, tko neurons predominantly displayed STD (Fig 1E–1G). Furthermore, the synchronous and asynchronous phase of the total synaptic charge transfer were similarly diminished (Fig 1H–1K), indicating that factors upstream of the RRP are affected by genetic loss of the TRPCs. To determine the RRP size, the data plot of the cumulative synchronous EPSC charge was approximated with a linear regression fitting the last 5 stimuli [38] (Fig 1L). Back-extrapolating the linear component of the steady-state phase renders an estimate of the initial RRP size (Fig 1L). Indeed, loss of TRPC channels strongly reduced the RRP size (Fig 1M) but left the Pr (i.e., first EPSC charge divided by the RRP charge) unchanged (Fig 1N). To determine the replenishment rate of vesicles during pool depletion, the slope of the plot of the cumulative total charge was approximated by linear regression fitting over the last 5 stimuli (Fig 1O). The replenishment rate was strongly reduced in tko neurons compared with controls (Fig 1P). Taken together, these results suggest that TRPC channels influence vesicle recruitment and thereby regulate the RRP size and the time course of synaptic depression during HFS. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. TRPC channels modulate the vesicle replenishment rate and the RRP size. (A) Exemplary EPSCs triggered by HFS (20 Hz, 40 AP/2 s) of autaptic wt and TRPC1/C4/C5 tko neurons (inset magnified view on the first 4 EPSCs of the train response. Arrows depict the synchronous and asynchronous phase of secretion). (B and C) TRPC deficiency reduces the EPSC amplitudes and accelerates the time course of synaptic depression (In panel C, data were normalized to the first EPSC amplitude). (D) The first EPSC amplitude is significantly reduced in tko cells (wt, n = 63; tko, n = 39 cells; p = 1.9 × 10−9). (E and F) The AmpEPSC10/AmpEPSC1 ratios of wt neurons range from STD to STE. Loss of TRPCs shifts the AmpEPSC10/AmpEPSC1 ratio towards STD. (G) The AmpEPSC10/AmpEPSC1 ratio is reduced in tko cells (p = 0.00067). (H, I, and J) The total synchronous and asynchronous charges are smaller in tko cells (red) when compared with controls (black). (K) Quantification of the 40th asynchronous charge (p = 0.0000056). (L) Mean cumulative synchronous release components during the 20-Hz train. Continuous line, linear regression of the last 5 data points back-extrapolated to stimulus = 0 to estimate the initial RRP size. Note that the RRP analysis was confined to wt neurons that reached a steady-state response in the late phase of stimulus train (45 out of 63 neurons). (M) The RRP size is significantly decreased in tko cells (wt, n = 45; tko, n = 39; p = 0.0000013). N. The release probability remained unchanged between groups (p = 0.2). (O) Cumulative total charge during 20-Hz train stimulation. The slope of the linear regression from the last 4 stimulation points (continuous line) rendered an estimate of the replenishment rate in panel P. (P) The replenishment rate is significantly reduced in tko neurons (p = 1.6 × 10−10). **p < 0.01; ***p < 0.001; statistical significance was assessed by Mann-Whitney rank sum test. Underlying data can be found in S1 Data. Amp, amplitude; AP, action potential; EPSC, evoked postsynaptic current; HFS, high-frequency stimulation; RRP, readily releasable pool; STD, short-term depression; STE, short-term enhancement; tko, triple knockout; TRPC, transient receptor potential canonical; wt, wild type. https://doi.org/10.1371/journal.pbio.3000445.g001 To study the subcellular distribution of TRPC channels, we co-immunolabeled neuronal cultures with antibodies directed against TRPC5 and the presynaptic marker protein bassoon. Confocal imaging revealed clear TRPC5 staining in somatic regions and axonal branches (S1 Fig, upper panels) as well as in varicosity-like thickenings of fine axonal branches, where it colocalized with bassoon (S1 Fig, lower panels). The specificity of the TRPC labeling was verified in cultures of TRPC1/C4/C5-tko neurons, which did not exhibit a discernable TRPC5 immunosignal (S1B Fig). TRPC5 was detected in about half of the presynaptic terminals (positive for bassoon, S1E Fig), indicating a heterogeneous TRPC5 expression in hippocampal neurons. In any case, these results are in line with our functional analyses showing that TRPC activity alters synaptic efficacy. Primary TRPC5-expressing neurons show strong STE of synaptic signaling To identify primary TRPC5-expressing neurons by genetic labeling, we generated a new TRPC5-Cre knockin (KI) mouse strain (TRPC5-internal ribosomal entry site cre recombinase [IC], S1A and S1B Fig). Here, the trpc5 gene is followed by an internal ribosome entry site (IRES) and by a Cre recombinase cDNA. When crossbred with ROSA26-floxed-stop-τGFP (eR26-τGFP) reporter mice [39], the Cre recombinase excises a floxed termination sequence 5ʹ to the τGFP transgene within the ROSA26 locus resulting in constitutive reporter expression in TRPC5-expressing cells. About 8% of the hippocampal neurons prepared from TRPC5-IC/eR26-τGFP mice were found to be τGFP-positive (7.7% ± 1.5% of 831 total cells, 2 preparations) and showed a clear immunosignal for TRPC5 (S2C and S2D Fig). Yet, among neurons, which were negative for τGFP, we also observed TRPC5-expressing cells (38.5% ± 3.58% of the τGFP-negative neurons; S2D Fig), most likely because the coding sequence 3ʹ of the IRES sequence is known to be expressed at significantly lower levels than the coding sequence 5ʹ of the IRES [40]. A similar frequency of TRPC5 positive cells was found with immunolabeling cultures of wt neurons (52% ± 3%, 195 out of 375 cells, 3 preparations), showing that IRES insertion does not affect the expression of the upstream protein (S2E and S2F Fig). Thus, hippocampal neurons heterogeneously express TRPC5 and τGFP labeling (in cultures of KI mutant mice) enables the identification of neurons that reliably express TRPC5. We comparatively analyzed synaptic signaling from τGFP-positive and τGFP-negative neurons with that of tko neurons. Indeed, τGFP-positive (i.e., TRPC5-expressing) neurons showed strong STE of their synaptic response during HFS and an increased PPR (Fig 2A–2E). In contrast, tko neurons started off with lower first EPSC amplitude and showed classical STD during HFS. The AmpEPSC10/AmpEPSC1 ratio is altered for the entire population of τGFP-positive neurons (Fig 2F and 2G), reaching on average a nearly 3-fold higher ratio than in tko neurons (τGFP-positive neurons: 1.63 ± 0.11; tko neurons 0.6 ± 0.05, p < 0.001, one-way ANOVA on ranks, Fig 2H). Furthermore, the synchronous and asynchronous charge component of the evoked response and the replenishment rate recorded in τGFP-positive cells were strongly enhanced when compared with synaptic responses of tko neurons (Fig 2I–2O). τGFP-negative neurons, instead, displayed an intermediate plasticity phenotype with respect to AmpEPSC10/AmpEPSC1 ratio (0.94 ± 0.09), total synaptic charge transfer, synchronous charge, and asynchronous charge as well as PPR and replenishment rate (Fig 2). These results agree with the observation that about 39% of the τGFP-negative neurons express TRPC5 (S2 Fig). Overall, the systematic changes in STP between τGFP-positive and -negative neurons as well as tko neurons correlate well with the differential expression of TRPC5 in these groups, indicating that endogenous TRPC5 activity is a key factor in regulating synaptic plasticity. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Endogenous TRPC5 channels promote STE of synaptic signaling. (A) Exemplary epifluorescence image of an autaptic TRPC5-IC/eR26-τGFP neuron. (B) Sample recordings from tko, τGFP-negative and τGFP-positive neurons upon 20-Hz HFS. (C–E) τGFP-positive cells show a strong STE of synaptic signaling and an increased PPR (panel E, EPSC2/EPSC1) when compared with tko or τGFP-negative neurons (τGFP-positive versus tko p = 0.00005; τGFP-positive versus τGFP-negative p = 0.003). (F and G) Frequency and cumulative frequency distribution of the AmpEPSC10/AmpEPSC1 ratios for tko (red), τGFP-negative (black), and τGFP-positive neurons (green). (H) The AmpEPSC10/AmpEPSC1 ratio is higher in τGFP-negative cells than in tko neurons and even further increased in τGFP-positive cells (τGFP positive versus tko p = 7.892264 × 10−9; τGFP positive versus τGFP negative p = 0.0009; τGFP negative and tko p = 0.02). (I) Time course of total charge during HFS. (J and K) The replenishment rate determined from the cumulative charge plot shown in panel J gradually increases between tko, τGFP-negative, and τGFP-positive neurons (τGFP positive versus tko p = 1.5 × 10−10; τGFP positive versus τGFP-negative p = 0.0001; τGFP negative and tko p = 0.002). (L and M) Time courses of synchronous and asynchronous charge transfer during the stimulus train. (N and O) The 10th EPSC charge and the 40th asynchronous charge are significantly increased in τGFP-negative and -positive neurons (10th charge: τGFP positive versus tko p = 1.7 × 10−8; τGFP positive versus τGFP negative p = 0.003; τGFP negative and tko p = 0.002; 40th charge: τGFP positive versus tko p = 0.00004; τGFP poisitive versus τGFP negative p = 0.004; τGFP negative and tko p = 0.018). Data were collected from tko (n = 25), τGFP-negative (n = 38), and τGFP-positive (n = 40) neurons. *p < 0.05, **p < 0.01, ***p < 0.001, one-way ANOVA on ranks followed by Dunn’s post hoc test; significance was also tested by Mann-Whitney rank sum test for τGFP-negative cells versus tko. Underlying data can be found in S1 Data. Amp, amplitude; AP, action potential; EPSC, evoked postsynaptic current; eR26, ROSA26-floxed-stop; HFS, high-frequency stimulation; IC, internal ribosomal entry site cre recombinase; PPR, paired-pulse ratio; STD, short-term depression; STE, short-term enhancement; tko, triple knockout; TRPC, transient receptor potential canonical; τGFP, τ-green fluorescent protein. https://doi.org/10.1371/journal.pbio.3000445.g002 TRPC activity promotes presynaptic short-term plasticity in hippocampal neurons To further study the impact of TRPC activity on synaptic signaling, we expressed TRPC1 or TRPC5 in wt neurons using lentiviral transduction. Wt neurons expressing either TRPC1 or TRPC5 developed a robust STE of the synaptic response during the stimulation train (Fig 3A–3C). TRPC channel expression shifts the frequency distribution for AmpEPSC10/AmpEPSC1 ratios to higher values, confirming general changes in STP (Fig 3D–3F). Starting from an unchanged first AP-evoked response (wt: 5.61 ± 0.56 nA; wt + C1: 5.8 ± 0.65 nA; wt + C5: 6.04 ± 0.69 nA; Fig 3B), TRPC expression significantly increased the PPR and elevated the synchronous as well as asynchronous charge transfer when compared with controls (Fig 3G–3N). Overall, a 1.5- to 2-fold higher total synaptic charge transfer (wt: 65.2 ± 5.8 pC; wt + C1: 98.27 ± 7.20 pC; wt + C5: 116.01 ± 8.79 pC) and a corresponding increase in the vesicle supply rate was observed (Fig 3H–3J). Importantly, the synaptic plasticity phenotype observed with lentiviral TRPC5 expression largely mimics that of the τGFP-positive neurons (compare with Fig 2), demonstrating that TRPC5 channels are instrumental in controlling synaptic plasticity. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Expression of either TRPC1 or TRPC5 in wt neurons turns synaptic depression into STE. (A) Sample recordings of EPSCs triggered by HFS (20 Hz, 40 AP/2 s) of autaptic wt neurons and those expressing TRPC5 or TRPC1. (B) Activity-dependent increase in the EPSC amplitudes in wt cells expressing C1 or C5. (C) TRPC expression causes STE of synaptic signaling (data were normalized to the initial peak EPSC amplitude). (D and E) Expression of TRPC1 or TRPC5 shifts the frequency distribution of the AmpEPSC10/AmpEPSC1 ratios to higher values. (F) Mean AmpEPSC10/AmpEPSC1 ratio for the indicated groups (+C1, p = 0.002; +C5, p = 0.0001; versus wt). (G) Mean paired pulse ratio (+C1, p = 0.0036; +C5, p = 0.00585; versus wt). (H) Time course of total charge transfer. (I) Mean cumulative total charge transfer during HFS. The replenishment rate was determined from the slope of the cumulative plot (last 4 data points). (J) The replenishment rate is significantly enhanced in cells expressing either C1 or C5 (+C1, p = 0.003; +C5, p = 0.00001; versus wt). (K and L) Time courses of synchronous (K) and asynchronous charge transfer (L) during the stimulus train. (M and N) The 10th EPSC charge (M) and the 40th asynchronous charge (N) are significantly increased with TRPC expression (10th charge: +C1, p = 0.00001; +C5, p = 0.0000001; versus wt; 40th asynchronous: +C1, p = 0.02; +C5, p = 0.0005; versus wt). Data were collected from the following number of cells: wt, n = 43; wt + TRPC1, n = 30; wt + TRPC5, n = 24; *p < 0.05; **p < 0.01; ***p < 0.001; one-way ANOVA on ranks followed by Dunn’s post hoc test. Underlying data can be found in S1 Data. Amp, amplitude; AP, action potential; EPSC, evoked postsynaptic current; HFS, high-frequency stimulation; PPR, paired-pulse ratio; STD, short-term depression; STE, short-term enhancement; TRPC, transient receptor potential canonical; wt, wild type. https://doi.org/10.1371/journal.pbio.3000445.g003 The short-term plasticity (STP) phenotype of TRPC-expressing cells often hindered the RRP determination because a steady-state response in the late phase of stimulus train was not reached [41]. Therefore, we made use of hypertonic sucrose stimulation to provide an estimate of the RRP size and the Pr with single AP stimulation [42]. Consistent with our results obtained with HFS, loss of TRPC1/C4/C5 reduced the AP-evoked response and the RRP size but left the Pr unchanged (S3 Fig). Furthermore, neither the expression of TRPC5 nor of TRPC1 in wt neurons significantly altered the RRP and the Pr (S3C Fig). Thus, TRPC activity does not affect the basal release probability but rather enhances the vesicle recruitment in an activity-dependent manner during HFS. Given that TRPC channels have been implicated in growth cone guidance and morphology [16,17], one might speculate that TRPC-mediated changes of synaptic signaling may at least in part be due to alterations in the number of synapses. To pursue this issue, we immunolabeled autaptic wt, tko, and wt neurons expressing either TRPC variant with the presynaptic marker protein bassoon (S4 Fig). Yet, synaptogenesis was neither affected by loss of TRPC channels nor by their lentiviral expression. To study whether TRPC channels interfere with the overall synaptic structure, cultured hippocampal neurons were co-immunolabeled for the active zone protein bassoon and the postsynaptic density protein PSD-95. Because bassoon and PSD-95 reside on either side of the synaptic cleft, we determined the degree of juxtaposition (or colocalization) of both signals up to a distance of 770 nm from the bassoon signal (S4C Fig). Quantitation of the percentage of “bassoon area” covered by PSD-95 signal revealed that neither the absence of TRPC1/4/5 nor the expression of TRPC1 or C5 in wt neurons altered the overall colocalization of bassoon and PSD95 puncta (S4D and S4E Fig). Taken together, these results render the possibility unlikely that changes in synapse number or organization contribute to the observed TRPC-mediated alterations in synaptic signaling. TRPC channel activity promotes efficient recovery from synaptic depression To further explore the role of synaptic TRPC channels in vesicle replenishment, we analyzed the recovery kinetics of phasic release from depression after HFS (S5 Fig). The degree of recovery was determined by dividing the EPSC amplitude of the test pulse (given at various time intervals after HFS) by that of the first response during the stimulus train (20 Hz; S5A and S5B Fig). A single exponential function was used to approximate the time course of recovery (see S5 Fig legend for details). The results show that expression of TRPC1 or TRPC5 in wt neurons diminished synaptic depression during the train (S5C Fig), accelerated the recovery kinetics (S5D Fig), and strongly augmented the EPSC amplitude when compared with controls (S5E Fig). In contrast, loss of TRPC activity (tko neurons) slowed down the recovery kinetics from synaptic depression when compared with controls (S5F–S5H Fig). Collectively, these results indicate that activation of synaptic TRPC channels leads to more efficient and faster mobilization of vesicles from the reserve pool. Buffering presynaptic [Ca]i abolishes TRPC-dependent STE Using photolytic uncaging experiments, we have previously shown that TRPC5 channels can directly be activated by an intracellular Ca2+ increase at the millisecond time scale [8]. Thus, it is possible that rapid activation of TRPC-mediated Ca2+ permeabilities further elevate the presynaptic Ca2+ rise during the stimulus train. To minimize the accumulation of intraterminal calcium during HFS, neuronal cultures were preincubated with the membrane-permeable calcium chelator ethylene glycol-bis(β-aminoethyl ether)-N,N,N′,N′-tetraacetic acid acetoxymethyl ester (EGTA-AM; 300 μM for 5 min; Fig 4). Owing to its slow kinetics and high affinity, EGTA buffers global Ca2+ without affecting phasic transmitter release [1]. EGTA did not change the first EPSC amplitude but significantly reduced the PPR of responses for wt neurons and those expressing additional TRPC1 or TRPC5 (Fig 4A, 4B, and 4C). In the same line, EGTA nearly abolished the delayed asynchronous secretion (Fig 4H and 4I) and caused steady-state phasic release (Fig 4D and 4F right panels; see also [38]), indicating efficient buffering of residual [Ca]i in our experiments. Importantly, wt neurons expressing either TRPC variant failed to promote any STE of synaptic signaling in the presence of EGTA (Fig 4D–4G). Both the AmpEPSC10/AmpEPSC1 ratio (Fig 4E) and the synchronous charge (Fig 4G) were significantly reduced when compared with the nontreated controls. These results indicate that enhanced buffering of global [Ca]i by EGTA abolishes the TRPC-mediated STE of synaptic signaling. Collectively, our results support a model wherein TRPC channels are rapidly activated by elevated levels of bulk [Ca]i (in response to Ca2+ entry through presynaptic voltage-gated Ca2+ channels) and in turn may amplify and prolong the presynaptic Ca2+ rise during HFS, thereby increasing the Ca2+ dependent rates of vesicle replenishment [43–45]. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Buffering presynaptic [Ca]i abolishes TRPC-dependent short-term facilitation. (A) Representative EPSC traces for wt neurons and those expressing TRPC1 or TRPC5 after EGTA-AM treatment. EGTA abolishes asynchronous secretion causing a phase of steady-state synchronous secretion and prevents the TRPC-mediated STE of synaptic signaling. (B). Neither TRPC expression (C1 or C5) nor pretreatment with EGTA affects the first EPSC amplitude of the train response (between groups without EGTA: p = 0.7; in-group with versus without EGTA: wt, p = 0.48; +C1, p = 0.49; +C5, p = 0.67). (C) EGTA significantly reduces the PPR for all groups (in-group with versus without EGTA: wt, p = 0.03; +C1, p = 0.001; +C5, p = 2 × 10−6). (D, F, H) Activity-dependent changes of EPSC amplitude (D), synchronous release (F), and asynchronous release (H) for wt neurons and those expressing TRPC1 or TRPC5 without (left panels) and with EGTA treatment (right panels). (E and G) The TRPC-mediated increase in the AmpEPSC10/AmpEPSC1 ratio (E) and the 40th synchronous EPSC charge (panel G, last pulse of the train shown in panel F) are prevented with EGTA treatment (EPSC10/EPSC1 [between groups without EGTA]: +C1, p = 0.07; +C5, p = 0.05, versus wt; in-group with versus without EGTA: wt, p = 0.5; +C1, p = 0.00004; +C5, p = 0.028; 40th synchronous [between groups without EGTA]: +C5, p = 0.002, +C1, p = 0.0014 versus wt; in-group with versus without EGTA: wt, p = 0.36; +C1, p = 0.0004; +C5, p = 5.8 × 10−9). (H and I) EGTA treatment diminishes asynchronous secretion in all groups, indicating effective buffering of presynaptic [Ca]i (+C5, p = 0.0008, +C1, p = 0.05; in-group with versus without EGTA: wt, p = 1.4 × 10−12; +C1, p = 0.0000003; +C5, p = 5.8 × 10−9). Data was collected from EGTA-treated wt (n = 45), wt + TRPC1 (n = 21), and wt + TRPC5 (n = 21) cells and nontreated wt (n = 13), wt + TRPC1 (n = 9), and wt + TRPC5 (n = 13) cells. *p < 0.05, one-way ANOVA on ranks followed by Dunn’s post hoc test for groups without EGTA; **p < 0.01, ***p < 0.001, Mann-Whitney U rank sum test for the in-group comparison with and without EGTA treatment. Underlying data can be found in S1 Data. EGTA-AM, ethylene glycol-bis(β-aminoethyl ether)-N,N,N′,N′-tetraacetic acid acetoxymethyl ester; EPSC, evoked postsynaptic current; ns, not significant; PPR, paired-pulse ratio; STE, short-term enhancement; TRPC, transient receptor potential canonical; wt, wild type. https://doi.org/10.1371/journal.pbio.3000445.g004 Homomeric TRPC5 channels promote STE of synaptic signaling TRPC4 and TRPC5 are strongly potentiated by elevation of [Ca]i [9] and can functionally couple to the Ca2+ entry through voltage-gated Ca2+ channels [8,26]. In contrast, no such regulation has been reported for TRPC1 [27], appearing contradictory to our finding that TRPC1 caused strong STE during repetitive stimulation. To address this issue, we compared the functional consequences of TRPC1 and TRPC5 expression in tko neurons (Fig 5). Expression of TRPC5 caused a strong STE of synaptic signaling (Fig 5A, 5B and 5C) and reproduced the phenotype seen with τGFP-positive wt neurons (Fig 2). TRPC5 expression significantly increased the PPR (tko: 0.9 ± 0.03, tko + C5: 1.44 ± 0.07; p < 0.001, one-way ANOVA on ranks). It shifted the distribution of AmpEPSC10/AmpEPSC1 ratios (mean ± SEM: 1.71 ± 0.14) toward the STE range (Fig 5D, 5E and 5F) and strongly accelerated the rate of vesicle recruitment (Fig 5I), leading to activity-dependent increases in total synchronous as well as asynchronous charge transfer (Fig 5G–5L). Thus, homomeric TRPC5 channels profoundly regulate synaptic plasticity. In contrast, TRPC1 expression in tko neurons failed to cause similar changes in STE-supporting synaptic responses that were indistinguishable from those in controls (Fig 5). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Homomeric TRPC5 channels promote strong STE of synaptic signaling in tko neurons. (A) Sample recordings of EPSCs triggered by HFS (20 Hz, 40 AP/2 s) of tko neurons and those expressing C1 or C5. (B) Starting from similar initial EPSC amplitudes, only TRPC5 expression causes STE of the synaptic response. (C) EPSC amplitude changes during HFS normalized to the amplitude of the first response. (D–F) The AmpEPSC10/AmpEPSC1 ratio of TRPC5-expressing tko neurons is shifted to the STE range and nearly 3-fold higher compared with tko and tko + C1 responses (panel F, +C1, p = 0.14; +C5, p = 1.9 × 10−10, versus tko). (G–I) TRPC5, but not TRPC1, expression elevates the total synaptic charge transfer and increases the replenishment rate (+C1 p = 0.1; +C5, p = 2 × 10−10, versus tko). (J and K) Time courses of synchronous (J) and asynchronous charge transfer (K) during the stimulus train. (L) The 40th asynchronous charge is significantly increased with TRPC5 expression (+C1, p = 0.98; +C5, p = 0.000005; versus tko). Data were collected from the following number of neurons: tko, n = 31; tko + TRPC1, n = 26; tko + TRPC5, n = 20; **p < 0.01, ***p < 0.001; one-way ANOVA on ranks followed by Dunn’s post hoc test. Underlying data can be found in S1 Data. Amp, amplitude; AP, action potential; EPSC, evoked postsynaptic current; HFS, high-frequency stimulation; ns, not significant; STD, short-term depression; STE, short-term enhancement; tko, triple knockout; TRPC, transient receptor potential canonical. https://doi.org/10.1371/journal.pbio.3000445.g005 This observation verifies that the synaptic plasticity phenotype observed with lentiviral expression is not simply a consequence of potential off-target effects. It further suggests that the STE phenotype of TRPC1 expression in wt neurons (Fig 3) most likely relies on the interaction with the other TRPC variants. The latter notion agrees with observations in heterologous expression systems in which no evidence for functional homomeric TRPC1 channels was found [21,46,47]. To extend these findings, we transfected mass cultures of hippocampal wt and tko neurons with either TRPC1 or TRPC5 and recorded spontaneous miniature EPSC (mEPSC) events in the presence of tetrodotoxin (TTX; 10μM; blocking voltage-gated Na+ channels [48]). Expression of TRPC5 or TRPC1 increased the mEPSC frequency in wt cells without changing the kinetics of quantal events, consistent with a presynaptic function of these channels (S6 Fig). In the absence of other TRPC isoforms, only TRPC5, but not TRPC1, expression increased the mEPSC frequency (S6C and S6D Fig). Taken together, homomeric TRPC5 channels profoundly regulate synaptic plasticity and elevate the rate of spontaneous release. In contrast, TRPC1 differentially affects synaptic signaling in wt and tko neurons, most likely because it requires heteromultimerization with other members of its TRPC subgroup to form functional channels. Acute perturbation of TRPC activity interferes with synaptic signaling To study whether acute inhibition of TRPC5 channels impairs synaptic efficacy, neurons were superfused with the TRPC5 inhibitor clemizole (3 μM, S7 Fig) [49]. In tko neurons expressing TRPC5, antagonist application significantly decreased STE of synaptic signaling and strongly reduced the asynchronous release when compared with the previous control response (S7B and S7C Fig). In contrast, clemizole neither affected synaptic depression nor asynchronous release in tko neurons, showing the specificity of the antagonist (S7D and S7E Fig). Collectively, the observed STE phenotype is due to immediate activation of TRPC channels during HFS, rather than being caused by developmental or compensatory mechanisms in response to TRPC channel expression. We next investigated whether direct activation of endogenous TRPC channels by the specific TRPC4/C5-agonist Englerin A [50] influences synaptic vesicle exocytosis. Autaptic wt neurons responded to agonist application in the presence of TTX (10 μM) with a reversible inward current (maximum current amplitude: 558 ± 68 pA, n = 36), whereas no significant current response could be detected in tko cells, verifying the specificity of Englerin A (Fig 6A, 6B and 6C). Interestingly, TRPC-mediated permeability changes were often paralleled by a transient increase in mEPSC frequency (Fig 6D, green line), without changing the amplitude or kinetics of the mEPSCs (Fig 6E). This suggests that activation of endogenous TRPC4/5 channels can directly evoke synaptic vesicle exocytosis. Some wt neurons (12 out of 36 cells) failed to respond to Englerin A (mEPSC frequency increase <1.2-fold), which could be due to heterogeneities in the expression or subcellular localization of TRPC4/C5 in hippocampal neurons. On average, Englerin A evoked a 2-fold increase in the mEPSC frequency of wt neurons (2.0 ± 0.28-fold increase over baseline levels, Fig 6F) but had no effect in tko neurons. To study the ionic basis of the inward current evoked by Englerin A, we repeated these experiments in neurons, which are genetically deficient for vesicular soluble N-ethylmaleimide-sensitive-factor attachment receptor (SNARE) proteins and are devoid of any glutamate release [48,51]. The results show that Englerin A evokes similarly large inward currents in the absence of any neurotransmitter release (S8 Fig). Thus, the observed inward current can be largely attributed to direct activation of TRPC channels rather than to secondary activation of glutamate receptors as a consequence of TRPC activity. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. Englerin A activates presynaptic TRPC channels and increases the mEPSC frequency. (A) Exemplary recordings of autaptic wt and TRPC1/C4/C5-tko neurons superperfused with Ringer (R) solution containing Englerin A (Eng, 1 μM). Englerin A evoked an inward current in wt but not tko cells. Right panels, expanded timescale of the recording at the indicated time points. Note the clear mEPSC frequency increase in wt neurons with Englerin A application (2). (B) Percentage of cells responding to Englerin A with an inward current (p = 0.008). (C) Quantification of the maximum inward current amplitude (wt, n = 36; tko, n = 16; p = 4.3 × 10−9) determined at the end of Englerin A application relative to baseline current. (D) Time course of the averaged inward current (black) and the corresponding mEPSC frequency (green) during Englerin A application (40–70 s) in wt neurons (n = 11). (E) mEPSC amplitudes (determined for the cells shown in panel D) remain unchanged during Englerin A application. Insets depict averaged mEPSCs during Ringer (R, n = 72) and Englerin A (E, n = 65) application. (F) Englerin A evokes a 2-fold increase in mEPSC frequency (relative to the mEPSC frequency before drug application) in wt but not in tko neurons (wt, n = 36; tko, n = 16; p = 0.00004). **p < 0.01; ***p < 0.001, Mann-Whitney rank sum test. Underlying data can be found in S1 Data. mEPSC, miniature excitatory evoked postsynaptic current; tko, triple knockout; TRPC, transient receptor potential canonical; wt, wild type. https://doi.org/10.1371/journal.pbio.3000445.g006 Collectively, these results are in line with an at least partial presynaptic localization of TRPC5 channels (S1 Fig) and compatible with the mEPSC frequency increase observed upon lentiviral TRPC expression (S6 Fig). They show that acute perturbation of TRPC activity is able to regulate STP and synaptic vesicle exocytosis. Enhanced Ca2+-entry through VGCCs does not mimic the TRPC phenotype TRPC channels may either directly mediate Ca2+ influx into synaptic terminals or indirectly modulate synaptic plasticity through facilitated opening of voltage-gated Ca2+ channels. In the latter case, TRPC-channel–mediated depolarization of the membrane potential could ease VGCC opening and enhance Ca2+ entry into the presynaptic terminal. Broadening the presynaptic AP-width with the potassium channel blocker tetraethylammonium (TEA, 300μM) has been shown to enhance the VGCC-mediated Ca2+ influx and synaptic transmission during high-frequency stimulation [45]. Compared to the preceding control response, acute TEA application significantly increased the initial EPSC amplitude and decreased the PPR leading to an overall faster synaptic depression (Fig 7A–7D). These changes are paralleled by an increase in release probability (Fig 7G) and clearly contrast the TRPC phenotype. TEA also enhanced the synchronous release component and the RRP size. It furthermore augmented asynchronous secretion and increased the replenishment rate, consistent with a prolonged and stronger Ca2+ entry into synaptic terminals (Fig 7E–7K). Overall, the enhanced Ca2+ influx through VGCCs leads to stronger STD and clearly differs from the STE phenotype observed with higher TRPC activity. Thus, it is unlikely that synaptic TRPC channels merely increase the Ca2+ influx through depolarization-dependent modulation of VGCCs. The combined set of data suggests that TRPC channels provide an additional Ca2+ entry pathway most likely distal to the active zone, enabling efficient mobilization of dormant vesicles from the reserve pool and leading to STE of synaptic signaling during HFS. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 7. Elevated Ca2+ entry through VGCCs increases the EPSC amplitude and accelerates synaptic depression. (A) Representative EPSC recordings of a wt neuron (20 Hz, 40 AP/2 s) before and after TEA application (300 μM). (B) Time course of the EPSC amplitude for the first (Ringer) and the second train response (+TEA); right panel, data normalized to the first EPSC amplitude. Note that TEA increases the degree of STD. (C and D) TEA increases the initial EPSC amplitude (C) and decreases the PPR (D) (first amp: p = 0.00003; PPR: p = 0.0002). (E) Time course of the EPSC synchronous charge for Ringer and TEA (left) and its cumulative plot (right). Continuous line, linear regression of the last 5 data points to estimate the initial RRP size (shown in panel F). (F and G) The RRP size (F) and the Pr (G) are significantly larger with TEA (RRP: p = 0.0007; Pr: p = 0.24). (H) TEA increases asynchronous release. (I) Mean asynchronous release of the 40th EPSC (p = 0.0001). (J) Time course of the cumulative total synaptic charge transfer; dashed lines, linear regression of the last 4 data points to estimate the replenishment rate shown in panel K. (K) TEA elevates the replenishment rate (p = 0.003). Data was collected from 16 cells, *p < 0.05; **p < 0.01; ***p < 0.001; Student paired t test. Underlying data can be found in S1 Data. AP, action potential; EPSC, evoked postsynaptic current; PPR, paired-pulse ratio; Pr, release probability; RRP, readily releasable pool; STD, short-term depression; TEA, tetraethylammonium; VGCC, voltage-gated calcium channel; wt, wild type. https://doi.org/10.1371/journal.pbio.3000445.g007 TRPC channels augment the presynaptic Ca2+ rise upon HFS To study how TRPC channels influence presynaptic Ca2+ dynamics, we combined electrophysiological recordings of synaptic activity with presynaptic Ca2+-imaging using the vesicle-associated Synaptophysin-GCaMP6s (SyGCaMP6s) fusion protein as a presynaptic Ca2+ reporter (Fig 8). In preparatory work, we verified that SyGCaMP6s is sorted to synaptic sites, as illustrated by its high degree of colocalization with the synaptic vesicle protein synaptobrevin II (SybII; Fig 8A). Autaptic neurons were stimulated with 20 Hz HFS for 2 s, while the synaptic charge transfer and [Ca]i changes at discrete synaptic sites were monitored simultaneously throughout the experiment. Discrete synaptic regions were analyzed when the fluorescence increase (ΔF/F0) exceeded 3 SDs of the background noise (ΔF/F0, mean ± SEM: 0.022 ± 0.00012). Wt neurons responded to electrical stimulation with a robust, activity-dependent increase in SyGCaMP fluorescence (Fig 8B and Fig 8C). Genetic loss of TRPC1/C4/C5 reduced the presynaptic Ca2+ rise, whereas the expression of either TRPC1 or TRPC5 strongly elevated presynaptic Ca2+ dynamics (Fig 8B, 8E and 8D). The slope of the fluorescence increase during HFS was nearly 2-fold higher in TRPC-expressing cells compared with wt neurons (Fig 8E), indicating that TRPC channels directly potentiate the VGCC-mediated increase in presynaptic Ca2+. Even after HFS, when VGCCs have closed, [Ca]i continued to increase more strongly in TRPC-expressing neurons than in wt and tko neurons (Fig 8F and 8G). These observations provide strong evidence that TRPC channels establish an additional Ca2+ entry pathway that functionally couples to the VGCC-mediated Ca2+-influx and is able to prolong the presynaptic Ca2+ signal. Importantly, alterations of the average vesicle replenishment rate determined from simultaneous electrophysiological recordings correlate well with the observed changes in presynaptic Ca2+ levels among the different groups (Fig 8I and 8J). Taken together, these observations demonstrate that TRPC channels augment and prolong the presynaptic Ca2+ increase upon HFS and, by this, set the pace of synaptic vesicle recruitment. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 8. TRPC channels augment the presynaptic Ca2+-rise upon HFS. (A) Exemplary images of neurons expressing the vesicular protein SybII-mRFP (left) and Synaptophysin-GCaMP6s (SyGCaMP6s, middle). Note the high degree of colocalization between synaptobrevinII and SyGCaMP6s (right). (B) Sample difference images (ΔF/F0) of SyGCamp6s signaling recorded (5 Hz) from autaptic neurons (wt, left; tko, middle; wt + TRPC5, right) during the HFS (20 Hz, 2 s); insets are corresponding EPSC recordings from the same cell. Arrows are exemplary ROIs used to monitor ΔF/F0 at single synaptic sites. (C) Loss of TRPC channels decreases the presynaptic Ca2+ rise, whereas expression of either TRPC variant strongly increases the Ca2+ signal. (D) Maximum ΔF/F0 from the data shown in (C), tko, p = 0.006; +C1, p = 0.0076; +C5, p = 0.0023. (E) TRPC deficiency decreases, whereas TRPC expression increases the slope of ΔF/F0 during HFS (slope between the 6th and the 13th data point; tko, p = 0.003; +C1, p = 0.035; +C5, p = 0.002. (F) Expansion of the early phase of the plot shown in panel C illustrating the prolonged Ca2+ rise after HFS. (G) Expression of either TRPC variant significantly enhances the Ca2+ influx right after HFS (slope determined between the 14th and 20th data point; tko, p = 0.48; +C1, p = 0.025; +C5, p = 0.0003). (H) Corresponding mean cumulative total charge transfer of the neurons imaged in panel C. (I and J) The replenishment rate (determined from the slope of the cumulative plot, last 4 data points, shown in panel H, is significantly changed by altering TRPC expression (I) and correlates with changes in SyGCaMP6s (slope of ΔF/F0) during HFS); tko, p = 0.0002; +C1, p = 0.009; +C5, p = 0.0004. Data were collected from wt, n = 15; tko, n = 9; wt + TRPC1, n = 11; wt + TRPC5, n = 15; *p < 0.05; **p < 0.01; ***p < 0.001; one-way ANOVA on ranks followed by Dunn’s post hoc test versus wt. wt versus tko Mann-Whitney rank sum test. Underlying data can be found in S1 Data. AP, action potential; EMCCD, electron multiplying charge coupled device; EPSC, evoked postsynaptic current; HFS, high-frequency stimulation; mRFP, monomeric red fluorescent protein; ROI, region of interest; SybII, synaptobrevin II; SyGCaMP6s, Synaptophysin-GCaMP6s; tko, triple knockout; TRPC, transient receptor potential canonical; VM, membrane voltage; wt, wild type; ΔF, delta fluorescence. https://doi.org/10.1371/journal.pbio.3000445.g008 Discussion Changing the strength of synaptic connections between neurons is subject to different short-lived or long-lasting regulation processes, with the underlying cellular mechanisms often remaining enigmatic. In the present study, we employed a variety of genetic and lentiviral expression strategies to unravel the functional impact of Ca2+-permeable TRPC channels on synaptic efficacy of hippocampal neurons. Our experiments provide first evidence that systematic alterations in the expression of endogenous TRPC5 channels lead to graded changes in the synaptic plasticity behavior of glutamatergic synapses. Lentiviral expression of TRPC proteins in combination with presynaptic Ca2+-imaging unveiled a TRPC-dependent augmentation and prolongation of the intraterminal Ca2+ signal that accelerates the rate of vesicle replenishment and boosts STE of synaptic signaling. Overall, the results identify hitherto unknown TRPC-mediated mechanisms which profoundly alter synaptic plasticity at synapses in the central nervous system (CNS). Presynaptic TRPCs govern synaptic plasticity Short-term synaptic plasticity regulates the strength of neurotransmission through facilitation and depression at the millisecond time scale and plays a key role in encoding information in the nervous system. VGCCs channels are the major source of Ca2+ entry for neurotransmission in the central nervous system. Yet, little is known about other Ca2+-entry channels, which may activate in response to Ca2+ influx through VGCCs and in turn further elevate the presynaptic Ca2+ rise. Such a scenario has been proposed for the activity-dependent activation of Ca2+-permeable transient receptor potential channels of the vanilloid subtype (TRPV1) receptors that promote asynchronous release from solitary tract afferents [52]. We have previously shown that [Ca]i elevations are essential and sufficient for the activation of TRPC5 [8]. TRPC5 channels are activated in a dose-dependent manner at submicromolar [Ca]i (effective concentration 50 [EC50] 635nM [8]) and are therefore well suited to sense changes in bulk [Ca]i. Furthermore, they respond to stepwise changes in [Ca]i at the millisecond time scale and thereby meet the speed requirements for regulating synaptic STP. Using independent experimental strategies, we show that TRPC channels reside at presynaptic sites and are able to regulate synaptic transmission. First, endogenous TRPC5 channels could be found in varicosity-like thickenings in which they colocalized with the presynaptic marker protein bassoon (S1 Fig). Second, genetic loss of TRPC1/C4/C5 reduced basal synaptic transmission, diminished the RRP size, and sped up the rate of synaptic depression during the HFS because of impeded vesicle recruitment (Fig 1). Third, acute activation of TRPC4/C5 channels by the agonist Englerin A stimulated an increase in mEPSC frequency, indicating that Ca2+ entry through TRPC channels is able to directly evoke small synaptic vesicle (SSV) exocytosis (Fig 6). Fourth, TRPC deficiency neither affected properties of quantal signaling, indicative of unchanged postsynaptic signal generation (S6 Fig), nor did it cause any changes in synapse number or organization (S4 Fig). Furthermore, by taking advantage of a TRPC5 channel reporter mouse strain (TRPC5-IC/eR26-τGFP), we avoided heterogeneities in channel expression and confined our measurements to neurons that express endogenous TRPC5 channels (Fig 2). These neurons exhibit strong STE of their synaptic response, enhanced asynchronous release, and an elevated replenishment rate. This behavior contrasts the classical synaptic depression during HFS usually observed in wt neurons and is characteristic for enhanced vesicle supply, leading to transient overfilling of the RRP during the stimulus train [53]. Moreover, loss of TRPC channels reduced presynaptic Ca2+ entry as documented by SyGCaMP6s measurements (Fig 8). Overall, these results show that endogenous TRPC channels crucially modulate presynaptic Ca2+ levels which in turn govern the rate of synaptic vesicle recruitment to release sites. These new observations also provide an attractive explanation for the functional deficits that underlie the impaired transient potentiation after 100 Hz stimulation in acute hippocampal slices of Trpc1/4/5−/− tko mice [37]. Expression of either TRPC variant (C1 or C5) profoundly influenced STP and switched short-term depression into STE of synaptic signaling (Fig 3). The activity-dependent increase in synchronous and asynchronous release during HFS together with a higher PPR is remarkably similar to the phenotype of τGFP-positive neurons expressing endogenous TRPC5. This indicates that TRPC5 activity is crucial for changing the mode of synaptic plasticity and excludes the possibility that STP changes with lentiviral expression are simply due to an excess of these ion channels. Furthermore, it stands to reason that TRPC5 expression in tko neurons did not merely rescue the plasticity phenotype of wt neurons but instead induced substantial STE (compare Fig 5 and Fig 1). This result is most likely due to the absence of TRPC5 in a subpopulation of wt neurons (S2 Fig) and may also be the consequence of higher TRPC5 expression levels upon lentiviral transduction. In tko neurons, expression of TRPC5 suffices to produce strong changes in STP, whereas TRPC1 neither changed evoked nor spontaneous release (Fig 5; S6 Fig). Thus, TRPC5 channels can function as homomeric channels to control presynaptic Ca2+ dynamics, but TRPC1 requires other members of its subfamily, as has been also observed in heterologous expression systems [20]. Given that isolated TRPC1 channels neither function as Ca2+-entry channels [54] nor are they directly activated by Ca2+ [27], the effects of TRPC1 expression in wt neurons may be explained by alterations in the stoichiometry of presynaptic TRPC channels or by changes in their subcellular distribution. Furthermore, other functionally relevant interactions of TRPC1 either with members of the store operated calcium entry (SOCE) family (Orai1, and stromal interaction molecule 1 [STIM1], [27]) or with those of a different TRPC subgroup (e.g., TRPC3/C6/C7 [54]) cannot be rigorously excluded. Nevertheless, our results demonstrate that loss and surplus of TRPC activity affect STP of hippocampal neurons in an opposite manner, either increasing or decreasing synaptic depression. A new model for the TRPC function in presynaptic Ca2+ homeostasis Given that TRPC5 channels functionally couple to VGCCs in heterologous expression systems [8], we hypothesize that Ca2+ entry through VGCCs may trigger TRPC channel openings, which lead to additional Ca2+ influx and thereby modulate presynaptic Ca2+ homeostasis and synaptic plasticity. In good agreement, expression of either TRPC1 or C5 elevated the presynaptic Ca2+ rise right after the HFS, providing direct evidence that these ion channels are able to prolong the Ca2+ entry into presynaptic terminals (Fig 8). These observations are consistent with acute TRPC-mediated changes in presynaptic Ca2+ dynamics rather than overall alterations in presynaptic Ca2+ buffering. Moreover, the TRPC5 inhibitor clemizole acutely affects STE of synaptic signaling and asynchronous release, reinforcing the view that TRPC5 activity directly shapes presynaptic Ca2+ dynamics (S7 Fig). Notably, the changes in presynaptic [Ca]i during HFS correlated well with alterations in the replenishment rates among the different groups (as observed with simultaneous electrophysiological recordings). In the same line, TRPC5 and C1 expression accelerated and strongly augmented the degree of recovery from synaptic depression, confirming that the TRPC-mediated Ca2+ permeability facilitates the refilling of the RRP through mobilization of synaptic vesicles from the reserve pool (S5 Fig). In contrast to genetic TRPC deficiency, lentiviral transduction of TRPC channels in wt or tko neurons neither affected the first EPSC amplitude of the train response nor the RRP size (see Figs 1, 3, 5 and S3 Fig). It is possible that such changes are counterbalanced by high levels of spontaneous vesicle fusion as documented by the elevated mEPSC frequency of TRPC-expressing neurons (S6 Fig). The comparative analyses between the TRPC1/C5 expression phenotype and the enhanced Ca2+ influx through VGCCs by TEA treatment revealed important functional differences (Fig 7). TEA-application enhanced asynchronous release and the vesicle replenishment rate but also increased the first EPSC amplitude, reduced the PPR, and showed a stronger synaptic depression during HFS. The latter results are characteristic of a higher release probability in the presence of TEA (see also [45,55]) and are most likely due to increased presynaptic [Ca]i in close proximity to the fusion site. Yet, they clearly differ from the TRPC expression phenotype, indicating that activation of distinct synaptic Ca2+ permeabilities differentially affects STP characteristics varying between STD and STE. An attractive explanation could be that VGCCs cluster at active zones [56], whereas TRPCs may reside at perisynaptic sites. In support of such a hypothesis, we found that preincubation with EGTA fully abolishes the TRPC effects on short-term plasticity (Fig 4). Assuming low millimolar concentrations of EGTA (approximately 1 mM) at the synaptic site, the Ca2+ buffer will chelate Ca2+ ions with a time constant of about 100 μs (τchelation [57]). Under these conditions and for a Ca2+ diffusion coefficient (DCa) of 2.2 × 10−6 cm2sec−1 [58] Ca2+ can diffuse more than 350 nm (r2 = 6 × DCa × τchelation) before efficient chelation by EGTA. These estimates are consistent with the idea that TRPC channels may reside at farther distances from the active zone, making them well positioned to enhance preferentially the Ca2+-dependent mobilization of reluctant vesicles from the reserve pool. Collectively, these experiments show that different types of synaptic plasticity result from enhanced Ca2+ influx through either VGCCs or TRPCs. They suggest that presynaptic Ca2+ homeostasis is a highly orchestrated process based on different compartmentalized Ca2+ entry pathways to ensure that efficient vesicle recruitment matches the high rates of vesicle fusion upon ongoing synaptic activity. Alterations in TRPC channel expression may contribute to the known variability of synaptic strength and plasticity in hippocampal neurons [59]. Furthermore, the STP characteristics of enhanced TRPC activity are remarkably similar to those observed in mammalian uncoordinated 13–2 (Munc13-2) dependent synapses [60], including the frequency-dependent facilitation during HFS and the post-train augmentation (i.e., increased EPSC amplitudes after high-frequency train). Munc13-2 has been shown to facilitate basal synaptic vesicle (SV) priming, STE of synaptic signaling [61], and is recruited to synapses by the active zone protein glutamic acid/leucine/lysine/serine‐rich protein (ELKS1) [62]. Intriguingly, synaptic vesicle priming factor proteins like Munc13-2 or Double C2-protein (Doc2) [63], which sense Ca2+ either with their C2-domains or in a complex with calmodulin, are translocated to the plasma membrane in a Ca2+-dependent manner [64] and may serve as transducers of the TRPC-mediated effects on vesicle replenishment and short-term plasticity. Similarly, other EF-hand proteins such as the neuronal calcium sensor 1 (NCS1) protein may contribute to the TRPC phenotype. NCS1 is enriched in presynaptic terminals [65], interacts with TRPC channels [18], and has a strong impact on short-term plasticity of hippocampal neurons. Interestingly, overexpression of NCS1 in hippocampal neurons mimics the phenotype of STE [65] observed with TRPC5 expression, making it possible that both proteins are engaged in the same signaling pathway. Taken together, our results identify TRPC channels as important modulators of presynaptic Ca2+ homeostasis and plasticity of fast glutamatergic synapses. They indicate that TRPC channels amplify the presynaptic Ca2+ rise and elevate the rate of synaptic vesicle recruitment to cope with vesicle consumption during high neuronal activity. Thus, TRPC-mediated changes in synaptic plasticity are well suited to play a central role for information processing in the central nervous system. Presynaptic TRPCs govern synaptic plasticity Short-term synaptic plasticity regulates the strength of neurotransmission through facilitation and depression at the millisecond time scale and plays a key role in encoding information in the nervous system. VGCCs channels are the major source of Ca2+ entry for neurotransmission in the central nervous system. Yet, little is known about other Ca2+-entry channels, which may activate in response to Ca2+ influx through VGCCs and in turn further elevate the presynaptic Ca2+ rise. Such a scenario has been proposed for the activity-dependent activation of Ca2+-permeable transient receptor potential channels of the vanilloid subtype (TRPV1) receptors that promote asynchronous release from solitary tract afferents [52]. We have previously shown that [Ca]i elevations are essential and sufficient for the activation of TRPC5 [8]. TRPC5 channels are activated in a dose-dependent manner at submicromolar [Ca]i (effective concentration 50 [EC50] 635nM [8]) and are therefore well suited to sense changes in bulk [Ca]i. Furthermore, they respond to stepwise changes in [Ca]i at the millisecond time scale and thereby meet the speed requirements for regulating synaptic STP. Using independent experimental strategies, we show that TRPC channels reside at presynaptic sites and are able to regulate synaptic transmission. First, endogenous TRPC5 channels could be found in varicosity-like thickenings in which they colocalized with the presynaptic marker protein bassoon (S1 Fig). Second, genetic loss of TRPC1/C4/C5 reduced basal synaptic transmission, diminished the RRP size, and sped up the rate of synaptic depression during the HFS because of impeded vesicle recruitment (Fig 1). Third, acute activation of TRPC4/C5 channels by the agonist Englerin A stimulated an increase in mEPSC frequency, indicating that Ca2+ entry through TRPC channels is able to directly evoke small synaptic vesicle (SSV) exocytosis (Fig 6). Fourth, TRPC deficiency neither affected properties of quantal signaling, indicative of unchanged postsynaptic signal generation (S6 Fig), nor did it cause any changes in synapse number or organization (S4 Fig). Furthermore, by taking advantage of a TRPC5 channel reporter mouse strain (TRPC5-IC/eR26-τGFP), we avoided heterogeneities in channel expression and confined our measurements to neurons that express endogenous TRPC5 channels (Fig 2). These neurons exhibit strong STE of their synaptic response, enhanced asynchronous release, and an elevated replenishment rate. This behavior contrasts the classical synaptic depression during HFS usually observed in wt neurons and is characteristic for enhanced vesicle supply, leading to transient overfilling of the RRP during the stimulus train [53]. Moreover, loss of TRPC channels reduced presynaptic Ca2+ entry as documented by SyGCaMP6s measurements (Fig 8). Overall, these results show that endogenous TRPC channels crucially modulate presynaptic Ca2+ levels which in turn govern the rate of synaptic vesicle recruitment to release sites. These new observations also provide an attractive explanation for the functional deficits that underlie the impaired transient potentiation after 100 Hz stimulation in acute hippocampal slices of Trpc1/4/5−/− tko mice [37]. Expression of either TRPC variant (C1 or C5) profoundly influenced STP and switched short-term depression into STE of synaptic signaling (Fig 3). The activity-dependent increase in synchronous and asynchronous release during HFS together with a higher PPR is remarkably similar to the phenotype of τGFP-positive neurons expressing endogenous TRPC5. This indicates that TRPC5 activity is crucial for changing the mode of synaptic plasticity and excludes the possibility that STP changes with lentiviral expression are simply due to an excess of these ion channels. Furthermore, it stands to reason that TRPC5 expression in tko neurons did not merely rescue the plasticity phenotype of wt neurons but instead induced substantial STE (compare Fig 5 and Fig 1). This result is most likely due to the absence of TRPC5 in a subpopulation of wt neurons (S2 Fig) and may also be the consequence of higher TRPC5 expression levels upon lentiviral transduction. In tko neurons, expression of TRPC5 suffices to produce strong changes in STP, whereas TRPC1 neither changed evoked nor spontaneous release (Fig 5; S6 Fig). Thus, TRPC5 channels can function as homomeric channels to control presynaptic Ca2+ dynamics, but TRPC1 requires other members of its subfamily, as has been also observed in heterologous expression systems [20]. Given that isolated TRPC1 channels neither function as Ca2+-entry channels [54] nor are they directly activated by Ca2+ [27], the effects of TRPC1 expression in wt neurons may be explained by alterations in the stoichiometry of presynaptic TRPC channels or by changes in their subcellular distribution. Furthermore, other functionally relevant interactions of TRPC1 either with members of the store operated calcium entry (SOCE) family (Orai1, and stromal interaction molecule 1 [STIM1], [27]) or with those of a different TRPC subgroup (e.g., TRPC3/C6/C7 [54]) cannot be rigorously excluded. Nevertheless, our results demonstrate that loss and surplus of TRPC activity affect STP of hippocampal neurons in an opposite manner, either increasing or decreasing synaptic depression. A new model for the TRPC function in presynaptic Ca2+ homeostasis Given that TRPC5 channels functionally couple to VGCCs in heterologous expression systems [8], we hypothesize that Ca2+ entry through VGCCs may trigger TRPC channel openings, which lead to additional Ca2+ influx and thereby modulate presynaptic Ca2+ homeostasis and synaptic plasticity. In good agreement, expression of either TRPC1 or C5 elevated the presynaptic Ca2+ rise right after the HFS, providing direct evidence that these ion channels are able to prolong the Ca2+ entry into presynaptic terminals (Fig 8). These observations are consistent with acute TRPC-mediated changes in presynaptic Ca2+ dynamics rather than overall alterations in presynaptic Ca2+ buffering. Moreover, the TRPC5 inhibitor clemizole acutely affects STE of synaptic signaling and asynchronous release, reinforcing the view that TRPC5 activity directly shapes presynaptic Ca2+ dynamics (S7 Fig). Notably, the changes in presynaptic [Ca]i during HFS correlated well with alterations in the replenishment rates among the different groups (as observed with simultaneous electrophysiological recordings). In the same line, TRPC5 and C1 expression accelerated and strongly augmented the degree of recovery from synaptic depression, confirming that the TRPC-mediated Ca2+ permeability facilitates the refilling of the RRP through mobilization of synaptic vesicles from the reserve pool (S5 Fig). In contrast to genetic TRPC deficiency, lentiviral transduction of TRPC channels in wt or tko neurons neither affected the first EPSC amplitude of the train response nor the RRP size (see Figs 1, 3, 5 and S3 Fig). It is possible that such changes are counterbalanced by high levels of spontaneous vesicle fusion as documented by the elevated mEPSC frequency of TRPC-expressing neurons (S6 Fig). The comparative analyses between the TRPC1/C5 expression phenotype and the enhanced Ca2+ influx through VGCCs by TEA treatment revealed important functional differences (Fig 7). TEA-application enhanced asynchronous release and the vesicle replenishment rate but also increased the first EPSC amplitude, reduced the PPR, and showed a stronger synaptic depression during HFS. The latter results are characteristic of a higher release probability in the presence of TEA (see also [45,55]) and are most likely due to increased presynaptic [Ca]i in close proximity to the fusion site. Yet, they clearly differ from the TRPC expression phenotype, indicating that activation of distinct synaptic Ca2+ permeabilities differentially affects STP characteristics varying between STD and STE. An attractive explanation could be that VGCCs cluster at active zones [56], whereas TRPCs may reside at perisynaptic sites. In support of such a hypothesis, we found that preincubation with EGTA fully abolishes the TRPC effects on short-term plasticity (Fig 4). Assuming low millimolar concentrations of EGTA (approximately 1 mM) at the synaptic site, the Ca2+ buffer will chelate Ca2+ ions with a time constant of about 100 μs (τchelation [57]). Under these conditions and for a Ca2+ diffusion coefficient (DCa) of 2.2 × 10−6 cm2sec−1 [58] Ca2+ can diffuse more than 350 nm (r2 = 6 × DCa × τchelation) before efficient chelation by EGTA. These estimates are consistent with the idea that TRPC channels may reside at farther distances from the active zone, making them well positioned to enhance preferentially the Ca2+-dependent mobilization of reluctant vesicles from the reserve pool. Collectively, these experiments show that different types of synaptic plasticity result from enhanced Ca2+ influx through either VGCCs or TRPCs. They suggest that presynaptic Ca2+ homeostasis is a highly orchestrated process based on different compartmentalized Ca2+ entry pathways to ensure that efficient vesicle recruitment matches the high rates of vesicle fusion upon ongoing synaptic activity. Alterations in TRPC channel expression may contribute to the known variability of synaptic strength and plasticity in hippocampal neurons [59]. Furthermore, the STP characteristics of enhanced TRPC activity are remarkably similar to those observed in mammalian uncoordinated 13–2 (Munc13-2) dependent synapses [60], including the frequency-dependent facilitation during HFS and the post-train augmentation (i.e., increased EPSC amplitudes after high-frequency train). Munc13-2 has been shown to facilitate basal synaptic vesicle (SV) priming, STE of synaptic signaling [61], and is recruited to synapses by the active zone protein glutamic acid/leucine/lysine/serine‐rich protein (ELKS1) [62]. Intriguingly, synaptic vesicle priming factor proteins like Munc13-2 or Double C2-protein (Doc2) [63], which sense Ca2+ either with their C2-domains or in a complex with calmodulin, are translocated to the plasma membrane in a Ca2+-dependent manner [64] and may serve as transducers of the TRPC-mediated effects on vesicle replenishment and short-term plasticity. Similarly, other EF-hand proteins such as the neuronal calcium sensor 1 (NCS1) protein may contribute to the TRPC phenotype. NCS1 is enriched in presynaptic terminals [65], interacts with TRPC channels [18], and has a strong impact on short-term plasticity of hippocampal neurons. Interestingly, overexpression of NCS1 in hippocampal neurons mimics the phenotype of STE [65] observed with TRPC5 expression, making it possible that both proteins are engaged in the same signaling pathway. Taken together, our results identify TRPC channels as important modulators of presynaptic Ca2+ homeostasis and plasticity of fast glutamatergic synapses. They indicate that TRPC channels amplify the presynaptic Ca2+ rise and elevate the rate of synaptic vesicle recruitment to cope with vesicle consumption during high neuronal activity. Thus, TRPC-mediated changes in synaptic plasticity are well suited to play a central role for information processing in the central nervous system. Material and methods Cell culture and animals TRPC1/C4/C5-tko mice were generated as described previously by Broker-Lai and colleagues [37]. Autaptic cultures of hippocampal neurons from age-matched TRPC tko, wt (C56BI/6N strain), and TRPC5-IC/eR26-τGFP mice were prepared from P0-1 animals as described previously by Schwarz and colleagues [66]. v-SNARE knockout animals (syb2−/− or syb2−/−/ceb−/−-dko) and their littermate control were prepared at E18.5. Briefly, hippocampi were dissected from the brain and digested for 20 mins at 37°C with 10 units of papain (Worthington, NJ), followed by gentle mechanical trituration. Neurons were seeded at low density (1,000 cells/ml) onto a layer of glial microislands, resulting in co‐cultures of glia and neurons. For electrophysiological recordings, only islands with single neurons were used. For mass cultures, neurons were seeded with a density of 300 cells/mm2 on 25-mm cover slips coated with 0.5 mg/ml of poly‐D‐lysine (Sigma, Germany). Cultures were maintained at 37°C in an incubator, humidified with 95% air and 5% CO2 in NBA (Invitrogen), supplemented with 2% B‐27 (Sigma, Germany), 1% Glutamax (Invitrogen, Germany), and 1% penicillin/streptomycin (Invitrogen, Germany). Recordings were performed at room temperature on 9 to 13 days of culture. Construction of the TRPC5-IRES-Cre (TRPC5-IC) targeting vector The final targeting construct includes a 5ʹ TRPC5 homology arm, an IRES-Cre-pgk promoter-driven Flp recombination target (FRT)–neomycin (neo)–FRT cassette and a 3ʹ TRPC5 homology arm. Using genomic R1 mouse embryonic stem (ES) cell DNA, the 2587-bp 3ʹ homology arm containing sequence downstream of the final exon of TRPC5 (exon 11) was amplified by polymerase chain reaction (PCR). By incorporation of restriction enzyme sequences within the primers, an AscI site was added 5ʹ to the homology arm and a BamHI site added at the 3ʹ end, and the fragment was subcloned into pKO-DTA. Using a similar strategy, the 2417-bp 5ʹ homology arm containing the stop codon of TRPC5 was generated and also cloned into the vector using XhoI and AscI (New England Biolabs, Germany) sites. PCR amplification of both homology arms was undertaken using the high-fidelity pfu DNA polymerase to minimize PCR-induced mutations, and any nucleotides that differed from the database sequence upon sequence analysis were verified by independent PCR amplification and sequencing. In a final step, the IRES-Cre-pgk promoter-driven FRT-neo-FRT cassette was cloned into the AscI (New England Biolabs, Germany) site found at the junction of the 2 homology arms. The completed targeting construct was then further verified by a complete sequence analysis and restriction mapping. Gene targeting Following verification of the integrity of the targeting construct, plasmid DNA was linearized using the NotI enzyme and then electroporated into R1 ES cells at the GIGA institute, University of Liege. Following electroporation, Southern blot analysis was utilized to identify correctly targeted clones, and these were then used to generate mice following standard protocols (injection of ES cells [129/Sv] into blastocysts [C57BL/6], implantation of injected blastocysts into foster mothers, backcross of male chimeras with C57BL6 females). F1 animals resulting from backcrosses were then crossed with FLP-deleter mice, which contain a ubiquitously expressed FLP recombinase gene, to facilitate removal of the neomycin selection cassette. Viral contructs/transfections The cDNAs encoding for TRPC1, TRPC5 (NCBI accession numbers NM_011643 and NM_009428.2), and SyGcamp6s (Addgene #26124) were subcloned into the plenti-hsynapsin lentiviral transfer vector. mRFP was fused to the C-terminal end of TRPC5 with a flexible 12 amino acid linker sequence. All constructs were verified by DNA sequence analysis (MWG Germany). Lentiviral particles were produced as previously described by Schwarz and colleagues [66]. Primary neurons were transfected with 100 to 300 μL of viral suspension 1-3DIC. Drug treatment All chemicals were purchased from Sigma-Aldrich unless stated otherwise. For Englerin A (Roth, Germany 0.2–1 μM), TEA, clemizole (Tocris, United Kingdom), and sucrose experiments, neurons were rapidly superfused using a gravity-fed fast-flow system. For EGTA-AM pretreatment, neurons were incubated for 5 min in nominally Ca2+ free Ringer’s solution in the presence of 300 μ M EGTA-AM (Merck, Germany) and subsequently washed with Ca2+-containing Ringer’s solution before recording. Electrophysiological measurements of synaptic currents Synaptic currents were recorded in the whole-cell voltage clamp mode from autaptic neurons. Patch pipettes (Rtip 4–5.5 MOhm) were filled with the following intracellular solution (in mM): 137.5 K-gluconate, 11 NaCl, 2 MgATP, 0.2 Na2GTP, 1.1 EGTA, 11 HEPES, and 11 D-glucose (pH 7.3) with KOH. The standard extracellular solution containing (in mM) 130 NaCl, 10 NaHCO3, 2.4 KCl, 1 to 2 CaCl2, 2 MgCl2, 10 HEPES, and 10 D-glucose (pH 7.3) with NaOH, osmolarity 299 mOsm was used. D-2-amino-5-phosphonopentanoate (APV; 50 μM) was added to inhibit NMDA-receptors and prevent synaptic plasticity changes. The reversal potential of chloride-mediated currents was adjusted to the holding potential to avoid the potential contribution of GABAergic currents. Exemplary neurons were treated with 25 μ M DNQX (Sigma, Germany) to confirm the recordings of AMPA receptor (AMPAR) currents. Neurons were voltage-clamped at -70 mV with an EPC10 amplifier (HEKA Electronic, Germany) under control of Pulse 8.5 program (HEKA Electronic, Germany) and stimulated by membrane depolarizations to +10 mV for 0.7 ms every 5 s (0.2 Hz). Cells with an average access resistance of 6 to 15 mOhm, with 75% to 80% resistance compensation and <100 pA leak-current were analyzed. Current signals were low-pass filtered at 2.9 kHz (4 pole Bessel filter EPSC10) and digitized at a rate of 10 or 50 kHz. Spontaneous mEPSCs were recorded prior to stimulation. mEPSC recordings in mass neuronal cultures were performed in the presence of TTX (10μM). To determine mEPSC properties with reasonable fidelity and to prevent the detection of “false events” (due to random noise fluctuations), spontaneous mEPSCs with a peak amplitude exceeding >5 times the standard deviation of the baseline noise and a charge criterion >25 fC were analyzed using a commercial software (Mini Analysis, Synaptosoft, Version 6.0.3). The AP-evoked EPSC amplitude and charge were determined from the average of 10 EPSCs recorded at 0.2 Hz. In hypertonic sucrose experiments, neurons were rapidly superfused with sucrose solution (500 mM sucrose, 5 s) using a gravity-fed fast-flow system. The RRP size was quantified from the charge integral of the current signal in response to sucrose application after subtracting the steady-state current component (determined at the end of hypertonic sucrose application), which probably reflects steady-state vesicle replenishment and exocytosis during hypertonic challenges [42]. The RRP size during HFS was quantified from the cumulative synchronous EPSC charge integral during the 20 Hz train. For this, the total charge of each AP evoked response within the train was corrected by subtracting the integral of the steady-state current component (= asynchronous charge) determined at the end of the stimulus interval [66]. The cumulative plot of the resulting synchronous release component reports the decrement in RRP size followed by a sustained phase of charge increase reflecting the steady-state phase of ongoing RRP vesicle replenishment. The sustained phase of secretion (last 5 data points) was approximated by linear regression. When back-extrapolated to time 0, its y-intercept provides an estimate of the RRP size with minimal contribution of refilling (see Fig 1L). Imaging SyGCaMP6s responses SyGCaMP6s fluorescence was acquired with an Evolve EMCCD camera (Visitron, Germany) using a Zeiss Plan Apochromat 40× oil immersion objective (NA 1.3) on a Axiovert200 microscope (Zeiss). Autaptic neurons were stimulated with 20 Hz for 2 s. Fluorescent images were captured at 5 Hz with custom written macros in VisiView (Visitron, Germany), processed offline using ImageJ 1.43 software and SigmaPlot 13. The background subtraction was done by subtracting the F0 image (average of 3 prestimulus images) from all subsequent images (ΔFn = Fn − F0). Regions of interest of identical size (4 × 4 pixels) were placed over single synapses reacting to electrical stimulation, and fluorescence changes were tracked throughout the stack. Immuncytochemistry Neurons were fixed for 10 min (RT) in PBS containing 4% paraformaldehyde. Cells were quenched for 10 min with 50mM NH4Cl in PBS, blocked for 30 min in PBS containing 3% BSA and 0.1% TritonX100. Primary (anti-TRPC5, 1:100, affinity-purified rabbit polyclonal home-made; anti-bassoon, 1:500, mouse monoclonal, Synaptic Systems; anti-GFP, 1:500, guinea pig; Synaptic Systems, anti-PSD95, 1:500, rabbit polyclonal, Synaptic Systems, Germany) and secondary antibodies (1:1,000, Alexa-Fluor 555, Alexa Fluor 488 and Alexa Fluor 643-conjugated goat anti-mouse or goat anti-rabbit; Invitrogen, Germany) were diluted in blocking buffer. The anti-TRPC5 antibody was generated by immunization of rabbits with a C-terminal protein fragment (amino acids 766–969) of mouse TRPC5 (NCBI accession number NM_009428.2). Cells were incubated with primary and secondary antibodies overnight and for 1.5 h at RT, respectively. In some experiments, cells were treated with DAPI (200 nM; Invitrogen, Germany) for 5 min at RT before mounting in glycerol cells were imaged either on a confocal microscope (LSM 710; Zeiss, Germany) using AxioVision 2008 software (Zeiss, Germany) or an Axiovert200 (Zeiss, Germany), fluorescence was elicited with a Polychrome V monochromator (Till Photonics, Germany) and captured with a EMCCD camera (Evolve, Visitron, Germany). The following objectives and filter sets were used: 100×, 1.3 NA; 63×, 1.4 NA; 40×, 1.3 NA; 25×, 0.8 NA oil objectives, Zeiss 10 (BP 450–490; FT 510, BP 515–565), Zeiss 09 (BP 450–490, FT 510, LP 515), Zeiss 15 (BP 546/12, FT 580, LP 590), Zeiss 38 (BP 470/40, FT 495, LP BP 525/50), respectively. Image analysis Images were analyzed with the software package ImageJ (version 1.45), AxioVision 2008 software (Zeiss, Germany), and SigmaPlot 13.0 (Systat Software, Inc.). For confocal imaging, optical sectioning was achieved with a 1 airy unit pinhole setting. In case of z-stack acquisition, horizontal image planes were separated by 0.480 μm in the stack. The acquisition settings were optimized to avoid underexposure and oversaturation effects and kept equal throughout image acquisition of control versus knockout samples. Thresholding and background corrections were performed with identical settings for a given set of images acquired from control versus knockout samples. Presynaptic boutons were identified in single planes of the anti-bassoon image stacks in fine axonal structures to avoid contaminations with somatic TRPC5 fluorescence. Identified regions of interest (ROIs) were stored and used to quantify the Mander´s colocalization coefficient between bassoon and TRPC5. For confocal imaging, the percentage of TRPC5 and τGFP immunopositive cells was quantified relative to the number of nuclei profiles visible in the corresponding 40× bright field image. The number of TRPC5 positive cells was quantified by counting the number of DAPI positive nuclei that showed somatic TRPC5 staining. The number of synapses was determined after the images were subjected to uniform background subtraction (23 ± 2.1 au). Identical ROIs (3 × 3 pixels) were placed around bassoon-positive puncta on the entire autaptic neuron cell surface area throughout the stack (15–28 sections). Presynaptic bassoon-positive puncta were counted manually. To examine the spatial relationship between pre- and postsynaptic marker proteins, images were thresholded with 3× SD of the background fluorescence for PSD-95 and with a uniform threshold of 75 au for bassoon. Bassoon-positive puncta were detected using the Analyze Particle function in ImageJ. Enlarged ROIs (up to a 770 nm distance from the bassoon signal = “bassoon area”) were superimposed onto the PSD95 channel to determine the degree of colocalization and/or juxtaposition of both immunosignals within the ‘bassoon area’. Statistical analysis Values are given as means ± SEM. To determine statistically significant differences, one-way ANOVA and a Student t test for comparing groups were used, if not indicated otherwise. The similarity of variances between groups was tested when performing statistics in SigmaPlot, and statistical tests were chosen accordingly. Normality was tested (Shapiro-Wilk). Multiple comparisons were performed using the Tukey-Kramer post hoc test. No statistical methods were used to predetermine sample sizes; however, sample sizes were similar to those employed in the field. Data collection and analyses were not performed blind to the experimental condition; no method of randomization was done. Online supplemental material S1 Fig shows the subcellular distribution of TRPC5 channels in hippocampal neurons. S2 Fig shows the genetic targeting strategy used to express Cre recombinase under control of the TRPC5 promoter and the heterogeneous expression of TRPC5 in hippocampal neurons. S3 Fig shows that TRPC channels do not affect the basal release probability. S4 Fig shows that neither loss nor expression of TRPC channels affects the synapse number. S5 Fig shows that increased TRPC activity increases the rate of recovery from short-term depression. S6 Fig shows that increased TRPC5 activity elevates the mEPSC frequency. S7 Fig shows that the TRPC antagonist clemizole reduces STE of synaptic signaling. S8 Fig shows the Englerin A evoked inward current reports direct activation of TRPC5 channels. Cell culture and animals TRPC1/C4/C5-tko mice were generated as described previously by Broker-Lai and colleagues [37]. Autaptic cultures of hippocampal neurons from age-matched TRPC tko, wt (C56BI/6N strain), and TRPC5-IC/eR26-τGFP mice were prepared from P0-1 animals as described previously by Schwarz and colleagues [66]. v-SNARE knockout animals (syb2−/− or syb2−/−/ceb−/−-dko) and their littermate control were prepared at E18.5. Briefly, hippocampi were dissected from the brain and digested for 20 mins at 37°C with 10 units of papain (Worthington, NJ), followed by gentle mechanical trituration. Neurons were seeded at low density (1,000 cells/ml) onto a layer of glial microislands, resulting in co‐cultures of glia and neurons. For electrophysiological recordings, only islands with single neurons were used. For mass cultures, neurons were seeded with a density of 300 cells/mm2 on 25-mm cover slips coated with 0.5 mg/ml of poly‐D‐lysine (Sigma, Germany). Cultures were maintained at 37°C in an incubator, humidified with 95% air and 5% CO2 in NBA (Invitrogen), supplemented with 2% B‐27 (Sigma, Germany), 1% Glutamax (Invitrogen, Germany), and 1% penicillin/streptomycin (Invitrogen, Germany). Recordings were performed at room temperature on 9 to 13 days of culture. Construction of the TRPC5-IRES-Cre (TRPC5-IC) targeting vector The final targeting construct includes a 5ʹ TRPC5 homology arm, an IRES-Cre-pgk promoter-driven Flp recombination target (FRT)–neomycin (neo)–FRT cassette and a 3ʹ TRPC5 homology arm. Using genomic R1 mouse embryonic stem (ES) cell DNA, the 2587-bp 3ʹ homology arm containing sequence downstream of the final exon of TRPC5 (exon 11) was amplified by polymerase chain reaction (PCR). By incorporation of restriction enzyme sequences within the primers, an AscI site was added 5ʹ to the homology arm and a BamHI site added at the 3ʹ end, and the fragment was subcloned into pKO-DTA. Using a similar strategy, the 2417-bp 5ʹ homology arm containing the stop codon of TRPC5 was generated and also cloned into the vector using XhoI and AscI (New England Biolabs, Germany) sites. PCR amplification of both homology arms was undertaken using the high-fidelity pfu DNA polymerase to minimize PCR-induced mutations, and any nucleotides that differed from the database sequence upon sequence analysis were verified by independent PCR amplification and sequencing. In a final step, the IRES-Cre-pgk promoter-driven FRT-neo-FRT cassette was cloned into the AscI (New England Biolabs, Germany) site found at the junction of the 2 homology arms. The completed targeting construct was then further verified by a complete sequence analysis and restriction mapping. Gene targeting Following verification of the integrity of the targeting construct, plasmid DNA was linearized using the NotI enzyme and then electroporated into R1 ES cells at the GIGA institute, University of Liege. Following electroporation, Southern blot analysis was utilized to identify correctly targeted clones, and these were then used to generate mice following standard protocols (injection of ES cells [129/Sv] into blastocysts [C57BL/6], implantation of injected blastocysts into foster mothers, backcross of male chimeras with C57BL6 females). F1 animals resulting from backcrosses were then crossed with FLP-deleter mice, which contain a ubiquitously expressed FLP recombinase gene, to facilitate removal of the neomycin selection cassette. Viral contructs/transfections The cDNAs encoding for TRPC1, TRPC5 (NCBI accession numbers NM_011643 and NM_009428.2), and SyGcamp6s (Addgene #26124) were subcloned into the plenti-hsynapsin lentiviral transfer vector. mRFP was fused to the C-terminal end of TRPC5 with a flexible 12 amino acid linker sequence. All constructs were verified by DNA sequence analysis (MWG Germany). Lentiviral particles were produced as previously described by Schwarz and colleagues [66]. Primary neurons were transfected with 100 to 300 μL of viral suspension 1-3DIC. Drug treatment All chemicals were purchased from Sigma-Aldrich unless stated otherwise. For Englerin A (Roth, Germany 0.2–1 μM), TEA, clemizole (Tocris, United Kingdom), and sucrose experiments, neurons were rapidly superfused using a gravity-fed fast-flow system. For EGTA-AM pretreatment, neurons were incubated for 5 min in nominally Ca2+ free Ringer’s solution in the presence of 300 μ M EGTA-AM (Merck, Germany) and subsequently washed with Ca2+-containing Ringer’s solution before recording. Electrophysiological measurements of synaptic currents Synaptic currents were recorded in the whole-cell voltage clamp mode from autaptic neurons. Patch pipettes (Rtip 4–5.5 MOhm) were filled with the following intracellular solution (in mM): 137.5 K-gluconate, 11 NaCl, 2 MgATP, 0.2 Na2GTP, 1.1 EGTA, 11 HEPES, and 11 D-glucose (pH 7.3) with KOH. The standard extracellular solution containing (in mM) 130 NaCl, 10 NaHCO3, 2.4 KCl, 1 to 2 CaCl2, 2 MgCl2, 10 HEPES, and 10 D-glucose (pH 7.3) with NaOH, osmolarity 299 mOsm was used. D-2-amino-5-phosphonopentanoate (APV; 50 μM) was added to inhibit NMDA-receptors and prevent synaptic plasticity changes. The reversal potential of chloride-mediated currents was adjusted to the holding potential to avoid the potential contribution of GABAergic currents. Exemplary neurons were treated with 25 μ M DNQX (Sigma, Germany) to confirm the recordings of AMPA receptor (AMPAR) currents. Neurons were voltage-clamped at -70 mV with an EPC10 amplifier (HEKA Electronic, Germany) under control of Pulse 8.5 program (HEKA Electronic, Germany) and stimulated by membrane depolarizations to +10 mV for 0.7 ms every 5 s (0.2 Hz). Cells with an average access resistance of 6 to 15 mOhm, with 75% to 80% resistance compensation and <100 pA leak-current were analyzed. Current signals were low-pass filtered at 2.9 kHz (4 pole Bessel filter EPSC10) and digitized at a rate of 10 or 50 kHz. Spontaneous mEPSCs were recorded prior to stimulation. mEPSC recordings in mass neuronal cultures were performed in the presence of TTX (10μM). To determine mEPSC properties with reasonable fidelity and to prevent the detection of “false events” (due to random noise fluctuations), spontaneous mEPSCs with a peak amplitude exceeding >5 times the standard deviation of the baseline noise and a charge criterion >25 fC were analyzed using a commercial software (Mini Analysis, Synaptosoft, Version 6.0.3). The AP-evoked EPSC amplitude and charge were determined from the average of 10 EPSCs recorded at 0.2 Hz. In hypertonic sucrose experiments, neurons were rapidly superfused with sucrose solution (500 mM sucrose, 5 s) using a gravity-fed fast-flow system. The RRP size was quantified from the charge integral of the current signal in response to sucrose application after subtracting the steady-state current component (determined at the end of hypertonic sucrose application), which probably reflects steady-state vesicle replenishment and exocytosis during hypertonic challenges [42]. The RRP size during HFS was quantified from the cumulative synchronous EPSC charge integral during the 20 Hz train. For this, the total charge of each AP evoked response within the train was corrected by subtracting the integral of the steady-state current component (= asynchronous charge) determined at the end of the stimulus interval [66]. The cumulative plot of the resulting synchronous release component reports the decrement in RRP size followed by a sustained phase of charge increase reflecting the steady-state phase of ongoing RRP vesicle replenishment. The sustained phase of secretion (last 5 data points) was approximated by linear regression. When back-extrapolated to time 0, its y-intercept provides an estimate of the RRP size with minimal contribution of refilling (see Fig 1L). Imaging SyGCaMP6s responses SyGCaMP6s fluorescence was acquired with an Evolve EMCCD camera (Visitron, Germany) using a Zeiss Plan Apochromat 40× oil immersion objective (NA 1.3) on a Axiovert200 microscope (Zeiss). Autaptic neurons were stimulated with 20 Hz for 2 s. Fluorescent images were captured at 5 Hz with custom written macros in VisiView (Visitron, Germany), processed offline using ImageJ 1.43 software and SigmaPlot 13. The background subtraction was done by subtracting the F0 image (average of 3 prestimulus images) from all subsequent images (ΔFn = Fn − F0). Regions of interest of identical size (4 × 4 pixels) were placed over single synapses reacting to electrical stimulation, and fluorescence changes were tracked throughout the stack. Immuncytochemistry Neurons were fixed for 10 min (RT) in PBS containing 4% paraformaldehyde. Cells were quenched for 10 min with 50mM NH4Cl in PBS, blocked for 30 min in PBS containing 3% BSA and 0.1% TritonX100. Primary (anti-TRPC5, 1:100, affinity-purified rabbit polyclonal home-made; anti-bassoon, 1:500, mouse monoclonal, Synaptic Systems; anti-GFP, 1:500, guinea pig; Synaptic Systems, anti-PSD95, 1:500, rabbit polyclonal, Synaptic Systems, Germany) and secondary antibodies (1:1,000, Alexa-Fluor 555, Alexa Fluor 488 and Alexa Fluor 643-conjugated goat anti-mouse or goat anti-rabbit; Invitrogen, Germany) were diluted in blocking buffer. The anti-TRPC5 antibody was generated by immunization of rabbits with a C-terminal protein fragment (amino acids 766–969) of mouse TRPC5 (NCBI accession number NM_009428.2). Cells were incubated with primary and secondary antibodies overnight and for 1.5 h at RT, respectively. In some experiments, cells were treated with DAPI (200 nM; Invitrogen, Germany) for 5 min at RT before mounting in glycerol cells were imaged either on a confocal microscope (LSM 710; Zeiss, Germany) using AxioVision 2008 software (Zeiss, Germany) or an Axiovert200 (Zeiss, Germany), fluorescence was elicited with a Polychrome V monochromator (Till Photonics, Germany) and captured with a EMCCD camera (Evolve, Visitron, Germany). The following objectives and filter sets were used: 100×, 1.3 NA; 63×, 1.4 NA; 40×, 1.3 NA; 25×, 0.8 NA oil objectives, Zeiss 10 (BP 450–490; FT 510, BP 515–565), Zeiss 09 (BP 450–490, FT 510, LP 515), Zeiss 15 (BP 546/12, FT 580, LP 590), Zeiss 38 (BP 470/40, FT 495, LP BP 525/50), respectively. Image analysis Images were analyzed with the software package ImageJ (version 1.45), AxioVision 2008 software (Zeiss, Germany), and SigmaPlot 13.0 (Systat Software, Inc.). For confocal imaging, optical sectioning was achieved with a 1 airy unit pinhole setting. In case of z-stack acquisition, horizontal image planes were separated by 0.480 μm in the stack. The acquisition settings were optimized to avoid underexposure and oversaturation effects and kept equal throughout image acquisition of control versus knockout samples. Thresholding and background corrections were performed with identical settings for a given set of images acquired from control versus knockout samples. Presynaptic boutons were identified in single planes of the anti-bassoon image stacks in fine axonal structures to avoid contaminations with somatic TRPC5 fluorescence. Identified regions of interest (ROIs) were stored and used to quantify the Mander´s colocalization coefficient between bassoon and TRPC5. For confocal imaging, the percentage of TRPC5 and τGFP immunopositive cells was quantified relative to the number of nuclei profiles visible in the corresponding 40× bright field image. The number of TRPC5 positive cells was quantified by counting the number of DAPI positive nuclei that showed somatic TRPC5 staining. The number of synapses was determined after the images were subjected to uniform background subtraction (23 ± 2.1 au). Identical ROIs (3 × 3 pixels) were placed around bassoon-positive puncta on the entire autaptic neuron cell surface area throughout the stack (15–28 sections). Presynaptic bassoon-positive puncta were counted manually. To examine the spatial relationship between pre- and postsynaptic marker proteins, images were thresholded with 3× SD of the background fluorescence for PSD-95 and with a uniform threshold of 75 au for bassoon. Bassoon-positive puncta were detected using the Analyze Particle function in ImageJ. Enlarged ROIs (up to a 770 nm distance from the bassoon signal = “bassoon area”) were superimposed onto the PSD95 channel to determine the degree of colocalization and/or juxtaposition of both immunosignals within the ‘bassoon area’. Statistical analysis Values are given as means ± SEM. To determine statistically significant differences, one-way ANOVA and a Student t test for comparing groups were used, if not indicated otherwise. The similarity of variances between groups was tested when performing statistics in SigmaPlot, and statistical tests were chosen accordingly. Normality was tested (Shapiro-Wilk). Multiple comparisons were performed using the Tukey-Kramer post hoc test. No statistical methods were used to predetermine sample sizes; however, sample sizes were similar to those employed in the field. Data collection and analyses were not performed blind to the experimental condition; no method of randomization was done. Online supplemental material S1 Fig shows the subcellular distribution of TRPC5 channels in hippocampal neurons. S2 Fig shows the genetic targeting strategy used to express Cre recombinase under control of the TRPC5 promoter and the heterogeneous expression of TRPC5 in hippocampal neurons. S3 Fig shows that TRPC channels do not affect the basal release probability. S4 Fig shows that neither loss nor expression of TRPC channels affects the synapse number. S5 Fig shows that increased TRPC activity increases the rate of recovery from short-term depression. S6 Fig shows that increased TRPC5 activity elevates the mEPSC frequency. S7 Fig shows that the TRPC antagonist clemizole reduces STE of synaptic signaling. S8 Fig shows the Englerin A evoked inward current reports direct activation of TRPC5 channels. Supporting information S1 Fig. TRPC5 is localized to presynaptic terminals. (A) Co-immunolabeling of wt and TRPC1/C4/C5-tko neurons with the presynaptic marker protein bassoon (left panels) and TRPC5 (middle). Endogenous TRPC5 is found in somatic, dendritic, and axonal regions. It colocalizes in varicosity-like thickenings of soma-proximal and soma-distal fine axonal branches with the presynaptic marker protein bassoon (middle panels, magnified view of the boxed area shown in panel A. (B) No discernable immunosignal for TRPC5 could be detected in TRPC1/C4/C5-tko neurons. (C and D) Line-scan analyses (dashed lines shown in panels A and B) confirm the high degree of colocalization between TRPC5 (red) and bassoon (green). (E) TRPC5 colocalizes with bassoon in presynaptic terminals in wt cells. No TRPC5 signal could be detected in tko neurons. The colocalization was determined in thin axonal structures to avoid the contribution of somatic or dendritic TRPC5 staining. Data were collected from wt neurons (19 images, 179 ROIs) and tko neurons (21 images, 168 ROIs) prepared from 3 independent cultures. ***p = 1.5 × 10−11, Mann-Whitney rank sum test. Underlying data can be found in S1 Data. ROI, region of interest; tko, triple knockout; TRPC, transient receptor potential canonical; wt, wild type. https://doi.org/10.1371/journal.pbio.3000445.s001 (TIF) S2 Fig. Heterogeneous expression of TRPC5 in hippocampal neurons. (A) Generation of the TRPC5-IRES-Cre mouse strain. Schematic representation of the targeting strategy used to express Cre recombinase under control of the TRPC5 promoter. From top to bottom, the targeting vector, the TRPC5 wt allele and the targeted TRPC5 allele before (neo+) and after (neo−) removal of the neomycin cassette are shown. The restriction sites for HpaI and the location of the probe are indicated. The inserted cassette is composed of an IRES followed by the coding sequence for Cre recombinase (Cre) and a pgk promoter-driven neomycin selection cassette flanked by FRT sites. (B) Genotyping of TRPC5-IRES-Cre mice. Representative genotyping result from mice already carrying the neoallele, expected product sizes are 364 bp for wt and 500 bp for neoallele. Lanes 4, 5, and 6 represent TRPC5-IRES-Cre neohomozygous, heterozygous, and wt mice, respectively. Sites of primer annealing are shown in panel A; primers 1 and 2 in combination produce the wt band, and primers 3 and 2 produce the TRPC5-IRES-Cre neoband. (C) Confocal images of τGFP immunoreactivity (left panel) and TRPC5 staining (right panel) in hippocampal neurons (12 div) prepared from TRPC5-IC/eR26-τGFP mice. All τGFP-positive neurons (arrows) express TRPC5. Some cells positive for TRPC5 by immunostaining do not show τGFP staining (asterisk), indicating that τGFP-positive neurons represent a subset of TRPC5 expressing cells. (D) τGFP labeling faithfully reports TRPC5 expression, yet 38.5% ± 3.58% of the τGFP-negative cells were found to TRPC5 positive (data were collected from 831 cells, 2 preparations). No discernable immunosignal for TRPC5 was detected in tko neurons (n = 101 cells, 2 preparations). (E) Hippocampal wt neurons stained with DAPI (left panel, overlaid epifluorescence + brightfield channel) and immunolabeled with TRPC5 antibody (right panel). Only a subpopulation of hippocampal neurons (52% ± 3%, 195 out of 375 cells, 3 preparations) is immunopositive for TRPC5. No discernable staining could be detected within tko neurons, supporting the specificity of antibody reaction (lower panels). (F) Quantification of TRPC5 expressing cells as percentage of DAPI-stained cells. Underlying data can be found in S1 Data. Cre, cre-recombinase; DAPI, 4′,6-Diamidine-2′-phenylindole dihydrochloride; eR26, ROSA26-floxed-stop; FRT, Flp recombination target; IC, internal ribosomal entry site cre recombinase; IRES, internal ribosome entry site; neo, neomycin; tko, triple knockout; TRPC, transient receptor potential canonical; wt, wild type; τGFP, τ-green fluorescent protein. https://doi.org/10.1371/journal.pbio.3000445.s002 (TIF) S3 Fig. TRPC channels do not affect the basal release probability. (A) Representative traces of the averaged evoked response (left panel) and the secretory response to 5 s application of 500 mM hypertonic sucrose solution (left panel) for wt, tko, wt + TRPC1, and wt + TRPC5 cells. (B) The evoked EPSC amplitude (left) and EPSC charge (right) are significantly reduced in tko neurons (amp: wt versus tko, p = 0.026; wt versus +C1, p = 0.31; wt versus +C5, p = 1.0; charge: wt versus tko, p = 0.01; wt versus +C1, p = 0.11; wt versus +C5, p = 0.9). (C) The RRP charge (left panel), determined by the time integral over hypertonic response, is reduced in tko cells (wt versus tko, p = 0.0004; wt versus +C1, p = 0.9; wt versus +C5, p = 0.9). The release probability determined by the ratio of EPSCcharge/RRPcharge is unchanged (right panel; wt versus tko, p = 0.9; wt versus +C1, p = 0.2; wt versus +C5, p = 0.9). Data were collected from wt, n = 53; tko, n = 42; wt + TRPC1, n = 27; wt + TRPC5, n = 26; *p < 0.05; one-way ANOVA on ranks followed by Dunn’s post hoc test. Underlying data can be found in S1 Data. EPSC, evoked postsynaptic current; RRP, readily releasable pool; tko, triple knockout; TRPC, transient receptor potential canonical; wt, wild type. https://doi.org/10.1371/journal.pbio.3000445.s003 (TIF) S4 Fig. Neither loss nor expression of TRPC channels changes the synapse number or organization. (A) Exemplary confocal images (displayed as maximum intensity projections over 15 sections) of autaptic wt, tko, and wt neurons expressing TRPC1 or TRPC5. Immunolabeling with the presynaptic marker protein bassoon revealed no differences in the number of synapses. (Right panels) Higher magnification of presynaptic terminals from a single confocal section from the corresponding image on the left(dashed box). (B) The number of synapses (identified by counting single bassoon-positive puncta throughout the stack) was unchanged among the different groups (one-way ANOVA on ranks followed by Dunn’s post hoc test, p = 0.742). Data were collected from wt, n = 19; tko, n = 20; wt + TRPC1, n = 18; wt + TRPC5, n = 18. (C) Exemplary confocal image of cultured hippocampal wt neurons immunostained for the active zone protein bassoon and the postsynaptic density protein PSD-95. Bassoon and PSD-95 partially colocalize but more often occupy adjacent domains, consistent with their localization in pre- and postsynaptic compartments. Solid lines define the area wherein the colocalization or apposition of bassoon and PSD95 was quantified. Images are displayed as maximum intensity projection of 3 z-planes. (D) Neither loss of TRPC1/C4/C5 (tko) nor expression of TRPC1 or C5 in wt neurons affects the apposition of bassoon and PSD-95 when compared with controls. Values are given as mean of median determined from the parameter’s frequency distribution for each cell. Data were collected from wt, n = 16; tko, n = 14; wt + TRPC1, n = 12; wt + TRPC5, n = 12; p = 0.985; one-way ANOVA on ranks followed by Dunn’s post hoc test. (E) Mean frequency distributions of the relative “bassoon area” covered by PSD-95 staining for wt, wt neurons expressing TRPC1 or TRPC5, and tko neurons. Statistical power (pr) was determined with a post hoc power analysis. Underlying data can be found in S1 Data. PSD-95, post synaptic density protein 95; tko, triple knockout; TRPC, transient receptor potential canonical; wt, wild type. https://doi.org/10.1371/journal.pbio.3000445.s004 (TIF) S5 Fig. Elevated TRPC activity increases the rate of recovery from short-term depression and leads to post-train augmentation of the synaptic response. (A) Representative EPSC traces (20 Hz, 2 s) of wt (black) and TRPC5-expressing wt neurons (green). Following HFS, the time course of recovery was determined by test pulses given after 20, 50, 150, 350, 650, 1,150, 2,150, and 4,150 ms intervals. (B) Time course for the recovery of the EPSC amplitude. Fraction of recovery was determined by dividing EPSC amplitudes of the test pulses by the first amplitude of the HF train. Data were fitted with a single exponential function (EPSC(t) = koff + EPSC(∞) × (1-exp-(t/τ)), revealing reduced depression during HFS (koff, C), faster recovery kinetics (τ, D), and transient post-train augmentation of the synaptic response (EPSC(4s), E) with expression of TRPC1 or TRPC5. (Inset) expansion of the early phase of the plot after normalization to the wt response. (C) TRPC1 or C5 expression increases koff because of diminished synaptic depression during HFS (+C5, *p = 0.025; +C1, *p = 0.018; versus wt). (D and E) TRPC1 or TRPC5 expression significantly speeds up the recovery from depression (D; wt versus +C5, *p = 0.039, wt versus +C1, *p = 0.049; one-way ANOVA on ranks followed by Dunn’s post hoc test) and leads to post-train augmentation of the synaptic response (E; +C5, p = 0.0421, wt versus +C1, p = 0.0009, versus wt). Data were collected from wt, n = 29, wt + TRPC1, n = 21, wt + TRPC5, n = 16. (F) tko neurons show a slower rate of recovery from depression. Analysis was restricted to cells with significant depression during HFS to determine the time course of recovery with reasonable fidelity (wt, n = 44; tko, n = 25 cells; ***p = 0.000003; Mann-Whitney rank sum test). (G) The post-train augmentation of the EPSC was not significantly altered for tko neurons. (H) Time course of recovery for wt and tko cells (p = 0.73, Mann-Whitney rank sum test). Loss of TRPC channels slows down the rate of recovery from HFS; data were collected from wt, n = 57; tko, n = 25 cells. Underlying data can be found in S1 Data. EPSC, evoked postsynaptic current; HF, high frequency; HFS, high-frequency stimulation; tko, triple knockout; TRPC, transient receptor potential canonical; wt, wild type https://doi.org/10.1371/journal.pbio.3000445.s005 (TIF) S6 Fig. TRPC5 expression suffices to elevate spontaneous glutamatergic signaling in tko neurons. (A) Exemplary recordings of spontaneous mEPSC signaling in neuronal mass cultures recorded in the presence of 10 μ M TTX. (B) Expression of either TRPC1 or TRPC5 significantly increases the mEPSC frequency (left panel). Corresponding cumulative frequency distributions of the mEPSC interevent interval (right panel). Data were collected from wt, n = 25; wt + C1, n = 22; p = 0.006; wt + C5, n = 20; p = 0.0001 cells. (C) The kinetics of the quantal events as illustrated for the indicated parameters remained unaffected by TRPC expression. (D) Sample recordings of mEPSC signaling from tko neurons expressing either TRPC1 or TRPC5. (E) Expression of TRPC5 but not of TRPC1 promotes an increase in mEPSC frequency (left panel). Corresponding cumulative frequency distributions of the mEPSC interevent interval (right panel). (F) mEPSC signaling with respect to amplitude and kinetics remained unchanged. Data were collected from tko, n = 20; tko + C1, n = 18; p = 0.91; tko + C5, n = 20; p = 0.002. **p < 0.01, ***p < 0.001; one-way ANOVA on ranks followed by Dunn’s post hoc test. Underlying data can be found in S1 Data. mEPSC, miniature evoked excitatory postsynaptic current; tko, triple knockout; TRPC, transient receptor potential canonical; TTX, tetrodotoxin; wt, wild type. https://doi.org/10.1371/journal.pbio.3000445.s006 (TIF) S7 Fig. TRPC5 antagonist clemizole reduces STE of synaptic signaling. (A) Sample EPSC recordings of TRPC5-expressing tko neurons during 20 Hz HFS before (left) and after application of clemizole (3 μM, 90 s; right). (B) Time course of the EPSC amplitudes during HFS for Ringer and subsequent clemizole application. (Right panel) Clemizole significantly diminishes the EPSC amplitude (10th pulse; p = 0.016). (C) The asynchronous charge (determined for the 40th AP during the train) is significantly decreased with clemizole superfusion (p = 0.031). (D and E) In tko neurons, clemizole application neither affects the EPSC amplitude (10th pulse) nor the asynchronous charge (E). Data were collected from 3 independent preparations, tko, n = 16; tko + TRPC5, n = 17; *p < 0.05; paired t test. Underlying data can be found in S1 Data. AP, action potential; EPSC, evoked postsynaptic current; STE, short-term enhancement; tko, triple knockout; TRPC, transient receptor potential canonical. https://doi.org/10.1371/journal.pbio.3000445.s007 (TIF) S8 Fig. The EA evoked inward current reports direct activation of TRPC5 channels. (A) Exemplary recordings of autaptic wt and synaptobrevin2-deficient neurons (syb2-/-) superfused with Ringer (R) solution containing EA (1 μM). No mEPSCs could be recorded in syb2−/− neurons indicative of the exocytotic block upon genetic loss of syb2. (B and C) Neither the time course (B) nor the peak amplitude (C) of the EA-evoked current response was significantly altered in neurons lacking syb2 or in the common absence of syb2 and its homolog cellubrevin (syb2−/−/ceb−/−, dko). Data were pooled from syb2 ko and dko animals because loss of syb2 is crucial for abolishing transmitter release, and no significant differences between the phenotype of the corresponding littermate controls (wt neurons and ceb ko neurons) were detected. Data were collected from littermates (n = 13) and syb−/− or syb−/−/ceb−/− (n = 16); p = 0.983; Mann-Whitney rank sum test. Underlying data can be found in S1 Data. ceb, cellubrevin; dko, synaptobrevin2 and cellubrevin double knock out; EA, Englerin A; ko, knock out; mEPSC, miniature evoked excitatory postsynaptic current; syb2, synaptobrevin2; TRPC, transient receptor potential canonical; wt, wild type. https://doi.org/10.1371/journal.pbio.3000445.s008 (TIF) S1 Data. Data underlying Figs 1–8 and S1–S8 Figs. https://doi.org/10.1371/journal.pbio.3000445.s009 (XLSX) Acknowledgments We thank M. Wirth, V. Schmitt, and W. Frisch for excellent technical assistance and Drs. M. Dhara, R. Mohrmann, D. Stevens, and J. Rettig for helpful discussions.
Single-molecule correlated chemical probing reveals large-scale structural communication in the ribosome and the mechanism of the antibiotic spectinomycin in living cellsSengupta, Arnab;Rice, Greggory M.;Weeks, Kevin M.
doi: 10.1371/journal.pbio.3000393pmid: 31487286
Introduction The ribosome is a megadalton complex that undergoes two large-scale motions during translation [1,2]. First, the two ribosome subunits rotate relative to each other in a ratchet-like manner. The bacterial ribosomal small subunit rotates by 6°–9° relative to the large subunit during translocation, pivoting about an intersubunit bridge, termed B3 [3–6]. B3 is unique among the intersubunit bridges (termed B1–B8, with several subtypes) in that it remains intact during intersubunit rotation whereas other bridges undergo conformational changes during dynamic cycles of disruption and reformation [5,7]. A second large-scale movement, small-subunit head-domain swiveling, accompanies translocation and involves pivoting at two “hinge” sites [6,8–10]. The antibiotic spectinomycin (Spc; molecular weight = 332 g/mol) is thought to freeze the head domain in a partially swiveled state that blocks translation [8,11,12], emphasizing the importance of this movement during translocation for ribosome function. Past investigations have inferred features of ribosome dynamics during translocation by visualizing stable and semistable intermediate states using high-resolution approaches [5,6,13–15]. In addition, the dynamics of specific pair-wise elements have been examined by single-molecule approaches [15,16]. These studies revealed that the ratchet-like intersubunit rotation of the ribosomal subunits occurs contemporaneously with the swiveling motion of the head domain of the small ribosomal subunit [8,9]. However, structural communication between head swiveling and intersubunit rotation within the ribosome is not well understood, and internal ribosome motions have not been analyzed in living cells. Moreover, although ribosome activity can be strongly affected by low–molecular-mass antibiotics, mechanistic understanding of how small-molecule binding influences global ribosomal RNA (rRNA) dynamics is limited. We recently developed the RNA interaction groups analyzed by mutational profiling (RING-MaP) chemical probing strategy that makes it possible to detect multiple chemical modification events on the same strand of RNA using massively parallel sequencing and to analyze these events for correlations (Fig 1A) [17]. The RING-MaP experiment is thus a single-molecule recording of co-occurring modifications in the same RNA molecule. By analyzing correlations between chemical modification events, it is possible to measure the through-space structural communication between nucleotides and to group these interactions in network communities (Fig 1B) [17–20]. A variety of chemical probes, including dimethyl sulfate (DMS), penetrate cell membranes and can be used to interrogate RNA structure in living cells. Here, we used RING-MaP with DMS to characterize hundreds of occurrences of through-space internucleotide communication in the 16S rRNA in Escherichia coli cells to define structural communities within the ribosome small subunit and to examine the effects of binding of the antibiotic Spc on ribosome structural dynamics. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. RING-MaP single-molecule correlated chemical probing and network analysis of a large RNA. (A) Overview of the RING-MaP experiment and network analysis. DMS chemical adducts on an RNA are detected by MaP, which records adduct sites as mutations and indels in the cDNA sequence generated by reverse transcription. Mutations co-occurring on the same read are analyzed for correlations that report RINGs, which are then visualized using a network graph. (B) Workflow for network analysis and detection of structural communities. Network analysis was performed using Gephi. DMS, dimethyl sulfate; MaP, mutational profiling; RING, RNA interaction group; RING-MaP, RNA interaction groups analyzed by mutational profiling. https://doi.org/10.1371/journal.pbio.3000393.g001 Results Multisite chemical probing of the 16S rRNA in living E. coli cells E. coli cells in the mid-log phase of growth were treated with DMS under three conditions: unperturbed, treated with rifampicin (Rif), or first treated with Rif and then with Spc. Rif treatment inhibits transcription by RNA polymerase and allows the 30S ribosomal subunit to assemble into stable, fully formed complexes [21–23]. Rif treatment reduces intermediate states of ribosome assembly to a small fraction of the total population of ribosome complexes; subsequent Spc treatment thus has minimal impact on ribosome assembly [22]. Rif also promotes degradation of polysomes [24], and pre-treatment with Rif reduces subsequent Spc-induced accumulation of polysomes [25]. After DMS treatment, the 16S rRNA was purified and subjected to mutational profiling (MaP). MaP is a high-throughput sequencing technology that enables detection of sites of chemical modifications in an RNA strand as internal sequence changes in a cDNA synthesized during reverse transcription [17,26] (Fig 1A). In this work, we developed new experimental conditions to allow efficient generation of long sequencing reads despite a high level of chemical modification and to detect through-space correlated nucleotide reactivities over distances spanning roughly 500 nucleotides. We also developed a new algorithm for detecting correlated nucleotide reactivities for randomly primed data over long sequence distances (Methods and S1 Fig). DMS reactivity is not strongly correlated with solvent accessibility (R2 = 0.007), and DMS reacts broadly with the 16S rRNA, affording good coverage of structural communities across the RNA. Profound differences in chemical probing data as assessed by per-nucleotide versus correlated reactivities In the simplest interpretation, MaP data can be used to generate straightforward reactivity versus position profiles, analogous to data obtained in conventional chemical probing experiments. We examined the effect of Spc binding to the small subunit by comparing the in-cell chemical probing signal in samples treated with Rif (+Rif) with that of samples treated with Rif and Spc (+Spc). Under the conditions employed in this study, DMS reacts primarily with A and C nucleotides that are (at least transiently) accessible at their base pairing face. As assessed by in-cell probing, Spc binding protected a single site in the 16S rRNA, C1192 (Fig 2A). Protection at C1192 is consistent with prior footprinting experiments with DMS [27–29] and with high-resolution structures of the E. coli ribosome complexed with Spc, which show that Spc binds in the minor groove of helix h34 (Fig 2B and 2C) [11,12]. We did not observe DMS-mediated protection at C1063, as observed previously using purified ribosomes [28], because DMS does not react with this nucleotide in the cellular environment in the absence of Spc. In addition, the overall near-identical per-nucleotide DMS reactivities of the +Rif and +Spc samples revealed that the structure of the in-cell 30S subunit and average association with the 50S subunit and with translation factors was similar in the presence and absence of Spc (Fig 3A). Overall, the per-nucleotide DMS chemical probing data suggest that Spc binding induces little change to ribosome structure in cells and that alterations are limited to its localized binding site. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Spc binding site in the 16S rRNA in E. coli cells revealed by DMS footprinting. (A) Per-nucleotide DMS-induced mutation rate profiles for 16S rRNA in cells treated with Rif (+Rif) and with Rif and Spc (+Spc). The plot shows a 550-nucleotide region spanning the 3' domain of the 16S rRNA; a single nucleotide at C1192 had a notable difference between the two experiments. The underlying data for this figure are available at https://doi.org/10.6084/m9.figshare.9252995.v1. (B) Model of Spc bound near C1192 (in red) of the 16S rRNA. (C) Structure of the small ribosomal subunit. 16S rRNA is light purple, ribosomal proteins are brown, mRNA is green, tRNA is black, and Spc is yellow. 30S subunits domains are labeled by morphology. Ribosome structure in all figures are from PDB 4v56 and 5afi [5,12]. DMS, dimethyl sulfate; PDB, Protein Data Bank; Rif, rifampicin; rRNA, ribosomal RNA; Spc, spectinomycin. https://doi.org/10.1371/journal.pbio.3000393.g002 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Network analysis applied to RING-MaP correlation data reveals that Spc induces extensive changes in through-space structural communication in the ribosome. (A) Correlation of RING-MaP data for independent experiments based on Spc-untreated versus Spc-treated samples (top panel; both samples pretreated with Rif); and between biological replicates (bottom panel) (B–C) Network analysis of structural communication in the 16S rRNA in (B) the absence and (C) the presence of Spc. Network analysis reveals four structural communities (blue, red, yellow, and green). Nodes are colored by community and are sized based on the number of correlations with other nodes in the network. Edges indicate internucleotide correlations, and edge weights indicate correlation strength. For both–Spc and +Spc conditions, cells were pretreated with Rif, which allows the rRNAs to fully assemble into complete subunits [22]. (D) Superposition of network nodes on the three-dimensional structure of the 30S subunit. Nodes were categorized by strength into strong, medium, and weak indicated by large, medium, and small spheres, respectively; strong interactions are shown as colored lines. (E) Edges that were strengthened (left) or weakened (right) upon addition of Spc. The underlying data for this figure are available at https://doi.org/10.6084/m9.figshare.9252995.v1. Rif, rifampicin; RING-MaP, RNA interaction groups analyzed by mutational profiling; rRNA, ribosomal RNA; Spc, spectinomycin. https://doi.org/10.1371/journal.pbio.3000393.g003 The MaP strategy also allows chemical probing data to be analyzed to detect correlated chemical events between any two nucleotides on a single strand (Fig 1A). We refer to correlated chemical modification reactions as RNA interaction groups (RINGs). RINGs report through-space structural communication in RNA [17]. We obtained two full biological replicates for 16S rRNA probed under two states: in unperturbed cells and for E. coli grown in +Rif and the +Spc conditions using the RING-MaP strategy; correlations between biological replicates were high (Fig 3A). RING correlation networks were identified in each biological replicate and then merged into a single dataset containing correlations that occurred in both replicates (S2 Fig). RING correlations were not appreciably different between the untreated and +Rif cells (S2 Fig), suggesting that under both conditions most ribosomes are fully assembled. In the +Rif condition, many RING correlations were relatively weak (Fig 3B), consistent with correlations that originate from an averaged ensemble of cellular ribosomes in different conformations. We next examined correlations for in-cell ribosomes probed after addition of Spc to the fully assembled ribosomes. In-cell treatment with Spc caused extensive and profound changes in through-space correlations across the length of the 16S rRNA, with especially strong enrichment in correlation density at the 5' and 3' ends of the RNA (Fig 3B and 3C, S3 Fig). Based on prior work, the averaged ensemble of cellular ribosomes in the +Spc condition would include Spc-arrested ribosomes that were trapped in an intermediate state of tRNA translocation, also bound by elongation factor-G [15]. Many correlations are shared between the +Rif and the +Spc states. However, in the presence of Spc, many correlations already present in the +Rif state were strengthened, and new correlations were observed (Fig 3C, S3 Fig). The observed changes in the number and strength of RING correlations are consistent with Spc-mediated inhibition of swiveling of the head domain. To summarize, in strong contrast to a conventional interpretation of chemical reactivity on a per-nucleotide basis, as a function of position, which suggested little change in ribosome structure (Fig 2), the correlated chemical probing experiment revealed that Spc induces extensive and profound global changes in the 30S ribosome subunit (Fig 3, S3 Fig). The RING experiment detected extensive through-space interactions throughout the 16S rRNA and revealed dramatic changes in through-space interactions upon addition of Spc. The extensive through-space structural communication revealed in our study reflects both changes in higher-order intramolecular RNA–RNA tertiary structure interactions and also structural communication modulated by contacts with neighboring rRNA, translation factors, and other proteins. 30S domain architecture as revealed by network communities The RING-MaP experiment identified many cases in which one nucleotide interacts with several other nucleotides in the 16S rRNA as revealed by mutual correlations in chemical reactivity patterns. We analyzed the correlated chemical reactivity RING data as a network graph with nucleotides represented as nodes and correlation strengths between nucleotides as edges (Fig 1B) [30]. The network graph representation of RING data allowed us to identify strong nodes and correlations and to group nucleotides into communities in an unbiased way. Network analysis divided the internucleotide communities in the 16S rRNA into 4 groups (Fig 3). This representation also allowed us to evaluate global changes in through-space RNA structural communication upon ligand binding. The 16S rRNA contains four domains based on the organization of the secondary structure of the RNA, conventionally termed the 5', central, 3'-major, and 3'-minor domains. These secondary structure domains also largely overlap with the physical structure of the 30S subunit such that the 5', central, and 3' RNA secondary structure domains correspond approximately to the body, platform, and head structures, respectively, and the 3'-minor domain forms an extended helix (h44) that extends from the head across the body (Fig 4A) [12]. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Domain architecture of the 16S rRNA as defined by network communities. Spheres indicate strong network nodes, colored by community. (A) 16S rRNA domain architecture based on secondary structure. Note that part of the central (platform) domain, as defined by the 16S rRNA secondary structure, is a structural component of the body. (B) Network communities correctly define the body domain and the more compact platform domain (yellow). These domains emerged naturally from the network analysis, even though no three-dimensional structural information was included in the network analysis. (C) The conventional head domain contains two RING network communities occupying distinct regions, defined here as the outer-head domain (red) and the inner-head and spine domain (blue). (D) Small-subunit domain model based on RING network communities in the 16S rRNA. The domains shown correspond to RING-MaP network community nodes of high correlation observed for the +Spc (+Rif) treatment condition; however, the same communities are observed in the absence of Spc. Rif, rifampicin; RING, RNA interaction group; RING-MaP, RNA interaction groups analyzed by mutational profiling; rRNA, ribosomal RNA; Spc, spectinomycin. https://doi.org/10.1371/journal.pbio.3000393.g004 When we superimposed strong nodes identified in our network analysis of the 16S rRNA on the 30S subunit structure, we observed that the four RNA communities are roughly centered on the body (green), platform (yellow), and head (red and blue) domains in 30S subunit (Fig 4A). This is a satisfying and important result because we did not impose the conventional domain organization on the network analysis. We can detect RING interactions over approximately 500 nucleotides, which would have allowed us to detect an alternative long-range organization of the rRNA. The RING data correctly detected an important nuance of the domain organization of the small subunit. Helices 19, 20, and 21 are part of the central domain as defined by RNA secondary structure (Fig 4B). The central domain largely overlaps with the platform of the 30S subunit; however, helices 19–21 contain multiple nodes assigned to the green community (Fig 4B). RING analysis thus correctly assigned helices 19, 20, and 21 to the body domain (Fig 4B, green). The platform domain in the RING-directed model contains only helices 22 to 27 (Fig 4B and 4D, yellow). This difference in domain organization between the established 16S rRNA secondary structure and domain organization in three-dimensional space was identified in the first crystal structure of the 30S subunit [31]. Strikingly, this same nonintuitive organization emerged naturally from our network analysis. Thus, network analysis based on single-molecule correlated chemical probing can correctly identify structurally distinct and cohesive structural domains. RING analysis reveals a new domain organization for the 30S subunit RING-based networks also revealed two major discrepancies with the conventional organization of ribosome domains. First, the conventional head domain (helices 28 to 43) contains two distinct network communities (Fig 4C). Second, the long helix h44, which is considered to be a separate 3' minor domain, is structurally connected via RINGs to nodes in the head domain (Fig 4C, blue community). The remainder of the conventional head domain is occupied exclusively by nodes that belong to the red community, and this region includes some of the highest strength nodes observed in the network analysis. The RING-based network analysis thus supports a physical model in which helices 28–40 at the 3' end of the 16S rRNA form what we call the outer-head domain, and helices 41–45 form the inner-head and spine domain (Fig 4C and 4D; red and blue communities, respectively). Based on a previous (and elegant) analysis of hinge motions in the small subunit, a region similar to our outer-head domain was previously identified as a distinct structural element in the head domain [10]. Nucleotides of the newly identified inner-head and spine domain have network connectivity suggestive of a cohesive structural entity that is distinct from the outer-head domain. The cohesive inner-head and spine domain spans nearly the entire length of the small ribosomal subunit (approximately 200 Å). The inner-head and spine domain is located at the intersubunit interface and makes extensive contacts within the active translational complex with the 50S subunit, tRNAs, mRNA, and translation factors. These external contacts involve coordinated rearrangements along the length of the inner-head and spine and, during translation, likely contribute to the experimentally observed cohesive nature of this long, structurally integrated domain. Additional notable features are revealed by superimposing the network-based domain architecture on the three-dimensional structure [12] of the 16S rRNA (Fig 4D). In the RING-based model, the core constituents of each domain are compact, and notably more compact than the full RNA structure. Nodes are most dense in the center of the ribosome and in the head domain. Roughly one-third of the 16S rRNA does not contain RING-based nodes, and these node-less regions fall in the outer edges of the 30S subunit. RING analysis thus correctly identifies the functional core of the 30S ribosome subunit. Finally, the inner-head and spine domain (Fig 4C, blue) extends approximately 200 Å between the most distant nodes, revealing structural integration of inner regions of the conventional head domain and the h44 spine (Fig 4D). Correlated chemical probing thus supports a 4-domain model of the 30S subunit. Spc remodels intra-network communities and long-range interactions In the absence of Spc, nucleotide A1111 is the largest node in the network; A1111 is engaged in extensive structural communication with other nodes within the red community (Fig 3B). Most structural communication in this no-Spc state is confined within the individual blue, red, yellow, and green communities. A1111 has the highest cross-community interactions of all nodes, and these are primarily with the blue community. A1111 and three other strong nodes of the red community (C1109, A1067, A1188) are part of a 3-helix junction region that is adjacent to the site of Spc binding (Fig 3D). Binding by Spc induced a large increase in the number of connections and enhanced the connections of A1111 both within the red community and across community boundaries to the blue community relative to the absence of Spc (Fig 3C, S3 Fig). Binding by Spc also strengthened correlations throughout the inner-head and spine regions. In the presence of Spc, nodes in the inner-head community network became denser, and a modest remodeling of the long distance structural communication with the spine component of this community was observed (Fig 3D and 3E). The body (green) and platform (yellow) domains had relatively few correlations with other domains in the absence of Spc (Fig 3B). Upon binding by Spc, new incidences of strong structural communication were observed that radiate from the body domain. These correlations are centered at A563, which is the predominant node for structural communication between the green and yellow communities (Fig 3C and 3D). Cross-community interactions connect head swiveling and intersubunit rotation Next, we created maps of the 16S rRNA secondary structure showing all interaction edges between red and blue network communities in the absence and presence of Spc (Fig 5A). C1192, at the Spc binding site, is adjacent to the 3-helix junction formed by helices h34, h35, and h38 [12]. In the presence of Spc, there are long-distance connections between this 3-helix junction, located in the outer-head domain, and nucleotides A1418 and A1483 in h44 in the inner-head and spine domain. In the absence of Spc, these connections are relatively few in number and occur almost exclusively with A1418 in h44 (Fig 5A, gray lines). Upon binding by Spc, the strength and number of interactions involving the h44 nucleotides, especially A1483, increase. Numerous other interactions linking the outer-head domain with the inner-head and spine domain increase in number and become stronger upon binding by Spc (Figs 3E and 5B). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Cross-community interactions strengthened upon Spc binding. (A, B) Cross-community correlations connecting the outer-head (red) and inner-head and spine (blue) communities are shown as grey lines on secondary structure diagrams for experiments performed (A) without and (B) with Spc treatment. The strongest cross-community interactions link nodes in the outer-head domain at the h34-h35-h38 3-helix junction to nodes in helix 44 of the inner-head and spine domain. Orange spheres indicate the two hinges that allow head domain swiveling [10]. (C) Cross-community interactions visualized with respect to the three-dimensional structure of bacterial ribosome. The two dashed lines mark the axes for head swiveling (passing through hinges 1 and 2) and for intersubunit rotation (at bridge B3). Bridge B3 links helix 44 of 30S rRNA with helix 71 of the 23S rRNA of the large subunit. rRNA, ribosomal RNA; Spc, spectinomycin. https://doi.org/10.1371/journal.pbio.3000393.g005 Strikingly, the cross-community interactions stabilized by Spc connect the two “hinges” that mediate large-scale motions in the 30S subunit. The 3-helix junction region is populated with strong nodes from the outer-head (red) community that overlap with one of the two hinge points (hinge 2, at C1066) [10] that mediate head swiveling (Fig 5A). The second hinge site for head rotation (hinge 1) is located in helix 28 at C1390, and the coordinated movement about these two hinges results in swiveling of the head domain (Fig 5C) [10]. At the other end of the cross-community interaction, A1418 and A1483 form the intersubunit bridge, termed B3, with helix 71 of the large subunit 23S rRNA. The axis of intersubunit rotation passes through bridge B3 (Fig 5C), and this structural feature is conserved in both prokaryotic and eukaryotic ribosomes [5,32,33]. The nucleotides that form this intersubunit bridge remain in contact during ratcheting of the two subunits during translation [5]. The major through-space cross-community interaction stabilized by Spc binding (Fig 5B) specifically connects the axis of head swiveling with the axis of intersubunit rotation (Fig 5C). This major consequence of Spc binding is invisible to standard chemical probing (Fig 2) and has not been detected in high-resolution structural studies. This cross-community connection, readily detected by single-molecule in-cell correlated chemical probing, reveals direct coordination between the two crucial pivot movements fundamental to ribosome-mediated translation. Multisite chemical probing of the 16S rRNA in living E. coli cells E. coli cells in the mid-log phase of growth were treated with DMS under three conditions: unperturbed, treated with rifampicin (Rif), or first treated with Rif and then with Spc. Rif treatment inhibits transcription by RNA polymerase and allows the 30S ribosomal subunit to assemble into stable, fully formed complexes [21–23]. Rif treatment reduces intermediate states of ribosome assembly to a small fraction of the total population of ribosome complexes; subsequent Spc treatment thus has minimal impact on ribosome assembly [22]. Rif also promotes degradation of polysomes [24], and pre-treatment with Rif reduces subsequent Spc-induced accumulation of polysomes [25]. After DMS treatment, the 16S rRNA was purified and subjected to mutational profiling (MaP). MaP is a high-throughput sequencing technology that enables detection of sites of chemical modifications in an RNA strand as internal sequence changes in a cDNA synthesized during reverse transcription [17,26] (Fig 1A). In this work, we developed new experimental conditions to allow efficient generation of long sequencing reads despite a high level of chemical modification and to detect through-space correlated nucleotide reactivities over distances spanning roughly 500 nucleotides. We also developed a new algorithm for detecting correlated nucleotide reactivities for randomly primed data over long sequence distances (Methods and S1 Fig). DMS reactivity is not strongly correlated with solvent accessibility (R2 = 0.007), and DMS reacts broadly with the 16S rRNA, affording good coverage of structural communities across the RNA. Profound differences in chemical probing data as assessed by per-nucleotide versus correlated reactivities In the simplest interpretation, MaP data can be used to generate straightforward reactivity versus position profiles, analogous to data obtained in conventional chemical probing experiments. We examined the effect of Spc binding to the small subunit by comparing the in-cell chemical probing signal in samples treated with Rif (+Rif) with that of samples treated with Rif and Spc (+Spc). Under the conditions employed in this study, DMS reacts primarily with A and C nucleotides that are (at least transiently) accessible at their base pairing face. As assessed by in-cell probing, Spc binding protected a single site in the 16S rRNA, C1192 (Fig 2A). Protection at C1192 is consistent with prior footprinting experiments with DMS [27–29] and with high-resolution structures of the E. coli ribosome complexed with Spc, which show that Spc binds in the minor groove of helix h34 (Fig 2B and 2C) [11,12]. We did not observe DMS-mediated protection at C1063, as observed previously using purified ribosomes [28], because DMS does not react with this nucleotide in the cellular environment in the absence of Spc. In addition, the overall near-identical per-nucleotide DMS reactivities of the +Rif and +Spc samples revealed that the structure of the in-cell 30S subunit and average association with the 50S subunit and with translation factors was similar in the presence and absence of Spc (Fig 3A). Overall, the per-nucleotide DMS chemical probing data suggest that Spc binding induces little change to ribosome structure in cells and that alterations are limited to its localized binding site. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Spc binding site in the 16S rRNA in E. coli cells revealed by DMS footprinting. (A) Per-nucleotide DMS-induced mutation rate profiles for 16S rRNA in cells treated with Rif (+Rif) and with Rif and Spc (+Spc). The plot shows a 550-nucleotide region spanning the 3' domain of the 16S rRNA; a single nucleotide at C1192 had a notable difference between the two experiments. The underlying data for this figure are available at https://doi.org/10.6084/m9.figshare.9252995.v1. (B) Model of Spc bound near C1192 (in red) of the 16S rRNA. (C) Structure of the small ribosomal subunit. 16S rRNA is light purple, ribosomal proteins are brown, mRNA is green, tRNA is black, and Spc is yellow. 30S subunits domains are labeled by morphology. Ribosome structure in all figures are from PDB 4v56 and 5afi [5,12]. DMS, dimethyl sulfate; PDB, Protein Data Bank; Rif, rifampicin; rRNA, ribosomal RNA; Spc, spectinomycin. https://doi.org/10.1371/journal.pbio.3000393.g002 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Network analysis applied to RING-MaP correlation data reveals that Spc induces extensive changes in through-space structural communication in the ribosome. (A) Correlation of RING-MaP data for independent experiments based on Spc-untreated versus Spc-treated samples (top panel; both samples pretreated with Rif); and between biological replicates (bottom panel) (B–C) Network analysis of structural communication in the 16S rRNA in (B) the absence and (C) the presence of Spc. Network analysis reveals four structural communities (blue, red, yellow, and green). Nodes are colored by community and are sized based on the number of correlations with other nodes in the network. Edges indicate internucleotide correlations, and edge weights indicate correlation strength. For both–Spc and +Spc conditions, cells were pretreated with Rif, which allows the rRNAs to fully assemble into complete subunits [22]. (D) Superposition of network nodes on the three-dimensional structure of the 30S subunit. Nodes were categorized by strength into strong, medium, and weak indicated by large, medium, and small spheres, respectively; strong interactions are shown as colored lines. (E) Edges that were strengthened (left) or weakened (right) upon addition of Spc. The underlying data for this figure are available at https://doi.org/10.6084/m9.figshare.9252995.v1. Rif, rifampicin; RING-MaP, RNA interaction groups analyzed by mutational profiling; rRNA, ribosomal RNA; Spc, spectinomycin. https://doi.org/10.1371/journal.pbio.3000393.g003 The MaP strategy also allows chemical probing data to be analyzed to detect correlated chemical events between any two nucleotides on a single strand (Fig 1A). We refer to correlated chemical modification reactions as RNA interaction groups (RINGs). RINGs report through-space structural communication in RNA [17]. We obtained two full biological replicates for 16S rRNA probed under two states: in unperturbed cells and for E. coli grown in +Rif and the +Spc conditions using the RING-MaP strategy; correlations between biological replicates were high (Fig 3A). RING correlation networks were identified in each biological replicate and then merged into a single dataset containing correlations that occurred in both replicates (S2 Fig). RING correlations were not appreciably different between the untreated and +Rif cells (S2 Fig), suggesting that under both conditions most ribosomes are fully assembled. In the +Rif condition, many RING correlations were relatively weak (Fig 3B), consistent with correlations that originate from an averaged ensemble of cellular ribosomes in different conformations. We next examined correlations for in-cell ribosomes probed after addition of Spc to the fully assembled ribosomes. In-cell treatment with Spc caused extensive and profound changes in through-space correlations across the length of the 16S rRNA, with especially strong enrichment in correlation density at the 5' and 3' ends of the RNA (Fig 3B and 3C, S3 Fig). Based on prior work, the averaged ensemble of cellular ribosomes in the +Spc condition would include Spc-arrested ribosomes that were trapped in an intermediate state of tRNA translocation, also bound by elongation factor-G [15]. Many correlations are shared between the +Rif and the +Spc states. However, in the presence of Spc, many correlations already present in the +Rif state were strengthened, and new correlations were observed (Fig 3C, S3 Fig). The observed changes in the number and strength of RING correlations are consistent with Spc-mediated inhibition of swiveling of the head domain. To summarize, in strong contrast to a conventional interpretation of chemical reactivity on a per-nucleotide basis, as a function of position, which suggested little change in ribosome structure (Fig 2), the correlated chemical probing experiment revealed that Spc induces extensive and profound global changes in the 30S ribosome subunit (Fig 3, S3 Fig). The RING experiment detected extensive through-space interactions throughout the 16S rRNA and revealed dramatic changes in through-space interactions upon addition of Spc. The extensive through-space structural communication revealed in our study reflects both changes in higher-order intramolecular RNA–RNA tertiary structure interactions and also structural communication modulated by contacts with neighboring rRNA, translation factors, and other proteins. 30S domain architecture as revealed by network communities The RING-MaP experiment identified many cases in which one nucleotide interacts with several other nucleotides in the 16S rRNA as revealed by mutual correlations in chemical reactivity patterns. We analyzed the correlated chemical reactivity RING data as a network graph with nucleotides represented as nodes and correlation strengths between nucleotides as edges (Fig 1B) [30]. The network graph representation of RING data allowed us to identify strong nodes and correlations and to group nucleotides into communities in an unbiased way. Network analysis divided the internucleotide communities in the 16S rRNA into 4 groups (Fig 3). This representation also allowed us to evaluate global changes in through-space RNA structural communication upon ligand binding. The 16S rRNA contains four domains based on the organization of the secondary structure of the RNA, conventionally termed the 5', central, 3'-major, and 3'-minor domains. These secondary structure domains also largely overlap with the physical structure of the 30S subunit such that the 5', central, and 3' RNA secondary structure domains correspond approximately to the body, platform, and head structures, respectively, and the 3'-minor domain forms an extended helix (h44) that extends from the head across the body (Fig 4A) [12]. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Domain architecture of the 16S rRNA as defined by network communities. Spheres indicate strong network nodes, colored by community. (A) 16S rRNA domain architecture based on secondary structure. Note that part of the central (platform) domain, as defined by the 16S rRNA secondary structure, is a structural component of the body. (B) Network communities correctly define the body domain and the more compact platform domain (yellow). These domains emerged naturally from the network analysis, even though no three-dimensional structural information was included in the network analysis. (C) The conventional head domain contains two RING network communities occupying distinct regions, defined here as the outer-head domain (red) and the inner-head and spine domain (blue). (D) Small-subunit domain model based on RING network communities in the 16S rRNA. The domains shown correspond to RING-MaP network community nodes of high correlation observed for the +Spc (+Rif) treatment condition; however, the same communities are observed in the absence of Spc. Rif, rifampicin; RING, RNA interaction group; RING-MaP, RNA interaction groups analyzed by mutational profiling; rRNA, ribosomal RNA; Spc, spectinomycin. https://doi.org/10.1371/journal.pbio.3000393.g004 When we superimposed strong nodes identified in our network analysis of the 16S rRNA on the 30S subunit structure, we observed that the four RNA communities are roughly centered on the body (green), platform (yellow), and head (red and blue) domains in 30S subunit (Fig 4A). This is a satisfying and important result because we did not impose the conventional domain organization on the network analysis. We can detect RING interactions over approximately 500 nucleotides, which would have allowed us to detect an alternative long-range organization of the rRNA. The RING data correctly detected an important nuance of the domain organization of the small subunit. Helices 19, 20, and 21 are part of the central domain as defined by RNA secondary structure (Fig 4B). The central domain largely overlaps with the platform of the 30S subunit; however, helices 19–21 contain multiple nodes assigned to the green community (Fig 4B). RING analysis thus correctly assigned helices 19, 20, and 21 to the body domain (Fig 4B, green). The platform domain in the RING-directed model contains only helices 22 to 27 (Fig 4B and 4D, yellow). This difference in domain organization between the established 16S rRNA secondary structure and domain organization in three-dimensional space was identified in the first crystal structure of the 30S subunit [31]. Strikingly, this same nonintuitive organization emerged naturally from our network analysis. Thus, network analysis based on single-molecule correlated chemical probing can correctly identify structurally distinct and cohesive structural domains. RING analysis reveals a new domain organization for the 30S subunit RING-based networks also revealed two major discrepancies with the conventional organization of ribosome domains. First, the conventional head domain (helices 28 to 43) contains two distinct network communities (Fig 4C). Second, the long helix h44, which is considered to be a separate 3' minor domain, is structurally connected via RINGs to nodes in the head domain (Fig 4C, blue community). The remainder of the conventional head domain is occupied exclusively by nodes that belong to the red community, and this region includes some of the highest strength nodes observed in the network analysis. The RING-based network analysis thus supports a physical model in which helices 28–40 at the 3' end of the 16S rRNA form what we call the outer-head domain, and helices 41–45 form the inner-head and spine domain (Fig 4C and 4D; red and blue communities, respectively). Based on a previous (and elegant) analysis of hinge motions in the small subunit, a region similar to our outer-head domain was previously identified as a distinct structural element in the head domain [10]. Nucleotides of the newly identified inner-head and spine domain have network connectivity suggestive of a cohesive structural entity that is distinct from the outer-head domain. The cohesive inner-head and spine domain spans nearly the entire length of the small ribosomal subunit (approximately 200 Å). The inner-head and spine domain is located at the intersubunit interface and makes extensive contacts within the active translational complex with the 50S subunit, tRNAs, mRNA, and translation factors. These external contacts involve coordinated rearrangements along the length of the inner-head and spine and, during translation, likely contribute to the experimentally observed cohesive nature of this long, structurally integrated domain. Additional notable features are revealed by superimposing the network-based domain architecture on the three-dimensional structure [12] of the 16S rRNA (Fig 4D). In the RING-based model, the core constituents of each domain are compact, and notably more compact than the full RNA structure. Nodes are most dense in the center of the ribosome and in the head domain. Roughly one-third of the 16S rRNA does not contain RING-based nodes, and these node-less regions fall in the outer edges of the 30S subunit. RING analysis thus correctly identifies the functional core of the 30S ribosome subunit. Finally, the inner-head and spine domain (Fig 4C, blue) extends approximately 200 Å between the most distant nodes, revealing structural integration of inner regions of the conventional head domain and the h44 spine (Fig 4D). Correlated chemical probing thus supports a 4-domain model of the 30S subunit. Spc remodels intra-network communities and long-range interactions In the absence of Spc, nucleotide A1111 is the largest node in the network; A1111 is engaged in extensive structural communication with other nodes within the red community (Fig 3B). Most structural communication in this no-Spc state is confined within the individual blue, red, yellow, and green communities. A1111 has the highest cross-community interactions of all nodes, and these are primarily with the blue community. A1111 and three other strong nodes of the red community (C1109, A1067, A1188) are part of a 3-helix junction region that is adjacent to the site of Spc binding (Fig 3D). Binding by Spc induced a large increase in the number of connections and enhanced the connections of A1111 both within the red community and across community boundaries to the blue community relative to the absence of Spc (Fig 3C, S3 Fig). Binding by Spc also strengthened correlations throughout the inner-head and spine regions. In the presence of Spc, nodes in the inner-head community network became denser, and a modest remodeling of the long distance structural communication with the spine component of this community was observed (Fig 3D and 3E). The body (green) and platform (yellow) domains had relatively few correlations with other domains in the absence of Spc (Fig 3B). Upon binding by Spc, new incidences of strong structural communication were observed that radiate from the body domain. These correlations are centered at A563, which is the predominant node for structural communication between the green and yellow communities (Fig 3C and 3D). Cross-community interactions connect head swiveling and intersubunit rotation Next, we created maps of the 16S rRNA secondary structure showing all interaction edges between red and blue network communities in the absence and presence of Spc (Fig 5A). C1192, at the Spc binding site, is adjacent to the 3-helix junction formed by helices h34, h35, and h38 [12]. In the presence of Spc, there are long-distance connections between this 3-helix junction, located in the outer-head domain, and nucleotides A1418 and A1483 in h44 in the inner-head and spine domain. In the absence of Spc, these connections are relatively few in number and occur almost exclusively with A1418 in h44 (Fig 5A, gray lines). Upon binding by Spc, the strength and number of interactions involving the h44 nucleotides, especially A1483, increase. Numerous other interactions linking the outer-head domain with the inner-head and spine domain increase in number and become stronger upon binding by Spc (Figs 3E and 5B). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Cross-community interactions strengthened upon Spc binding. (A, B) Cross-community correlations connecting the outer-head (red) and inner-head and spine (blue) communities are shown as grey lines on secondary structure diagrams for experiments performed (A) without and (B) with Spc treatment. The strongest cross-community interactions link nodes in the outer-head domain at the h34-h35-h38 3-helix junction to nodes in helix 44 of the inner-head and spine domain. Orange spheres indicate the two hinges that allow head domain swiveling [10]. (C) Cross-community interactions visualized with respect to the three-dimensional structure of bacterial ribosome. The two dashed lines mark the axes for head swiveling (passing through hinges 1 and 2) and for intersubunit rotation (at bridge B3). Bridge B3 links helix 44 of 30S rRNA with helix 71 of the 23S rRNA of the large subunit. rRNA, ribosomal RNA; Spc, spectinomycin. https://doi.org/10.1371/journal.pbio.3000393.g005 Strikingly, the cross-community interactions stabilized by Spc connect the two “hinges” that mediate large-scale motions in the 30S subunit. The 3-helix junction region is populated with strong nodes from the outer-head (red) community that overlap with one of the two hinge points (hinge 2, at C1066) [10] that mediate head swiveling (Fig 5A). The second hinge site for head rotation (hinge 1) is located in helix 28 at C1390, and the coordinated movement about these two hinges results in swiveling of the head domain (Fig 5C) [10]. At the other end of the cross-community interaction, A1418 and A1483 form the intersubunit bridge, termed B3, with helix 71 of the large subunit 23S rRNA. The axis of intersubunit rotation passes through bridge B3 (Fig 5C), and this structural feature is conserved in both prokaryotic and eukaryotic ribosomes [5,32,33]. The nucleotides that form this intersubunit bridge remain in contact during ratcheting of the two subunits during translation [5]. The major through-space cross-community interaction stabilized by Spc binding (Fig 5B) specifically connects the axis of head swiveling with the axis of intersubunit rotation (Fig 5C). This major consequence of Spc binding is invisible to standard chemical probing (Fig 2) and has not been detected in high-resolution structural studies. This cross-community connection, readily detected by single-molecule in-cell correlated chemical probing, reveals direct coordination between the two crucial pivot movements fundamental to ribosome-mediated translation. Discussion We analyzed hundreds of occurrences of pair-wise through-space structural communication within the 30S subunit RNA structure in an unbiased way using RING-MaP single-molecule correlated chemical probing [17] in E. coli cells. We used network partitioning analysis to visualize nucleotides, domains, and communities that show interelement correlations (Fig 1). This network approach recapitulated the overall domain architecture of the 16S rRNA, correctly assigned elements of the central domain as defined by the secondary structure to the body domain (Fig 4B) and directly detected the structural and functional core of the 30S subunit (Fig 4D). Network analysis revealed that the conventionally assigned head domain contains two separate structural regions: an outer-head domain, which has been previously described as the structural core of the head domain [10], and a second community that connects the head domain with helix h44 to form a distinct structural unit that we call the inner-head and spine domain (Fig 4D). Treating E. coli cells with the antibiotic Spc, a translocation inhibitor that disrupts proper movement of the small subunit head domain [8,10,12], induced a highly localized change in DMS reactivity of the small subunit RNA in ribosomes probed in cells (Fig 2A). In contrast, analyses of RING connectivities revealed that binding by Spc induced a dramatic strengthening of through-space interactions across the entire 16S rRNA (Figs 3 and 5). The antibiotic induced widespread enrichment of interactions in the outer-head, inner-head, and spine domains indicating that Spc binding stabilizes long-range interactions throughout the entire small subunit (Fig 3). Thus, the effect of Spc binding extends far beyond a localized effect at its direct binding site. We identified strong through-space RNA interactions linking head swiveling with intersubunit rotation over distances spanning 95 Å (Fig 5). The principle pivot for head swiveling (hinge 2) lies in the h34-h35-h38 3-helix junction located in the outer-head (red) domain [1,2,10] (Fig 5A). Spc binds adjacent to this pivot point (Fig 5B and 5C). Although most Spc-resistance mutations occur proximal to this binding site, two (A1351C [34] and G1386A [35]) are distant from the binding site. Additionally Spc inhibits ultraviolet light–induced crosslinking between C934 and U1345 [36]. A1351, G1386, and U1345 all lie within the inner-head (blue) domain defined by our network analysis, corroborating our model that Spc (which binds in the outer-head domain, in red) has long-range effects on the 16S rRNA that span the outer-head and inner-head and spine domains. In the presence of Spc, the 3-helix junction in the outer-head domain had strong through-space RING interactions with the inner-head region, as well as with 2 nucleotides in the h44 spine region, A1418 and A1483 (Fig 5, gray lines). A1418 and A1483 form the intersubunit bridge B3 by interacting with helix 71 of the 23S rRNA, the anchoring pivot for intersubunit rotation [5,7,32,33,37]. Thus, our work reveals that head swiveling is directly correlated with intersubunit rotation, two motions essential for tRNA translocation, even though the sites of these two large-scale conformational changes are far apart in three-dimensional space. Together, these data support the following model for how Spc functions to inhibit translation. In the absence of Spc, through-space interactions between the pivot regions for head swiveling and intersubunit rotation are sufficiently weak to allow relatively fluid motions, including free head domain swiveling, intersubunit rotation, L1 stalk movement, and tRNA translocation [15] (Fig 6). In this state, several intersubunit bridges are dynamically disrupted and reformed allowing free intersubunit rotation. Head domain swiveling in particular involves disruption of intersubunit bridge B1a/b, an important RNA–protein and protein–protein bridge (connecting the universal small subunit protein uS13 to the central protuberance of the large subunit) [7,38]. Notably, the anchoring intersubunit bridge B3 remains intact throughout these movements [7]. Spc binding specifically strengthens interactions adjacent to the two hinge elements that allow head swiveling and creates strong interactions between hinge 2 and the intersubunit bridge B3 (Fig 5). Thus, Spc binding modulates intersubunit rotation at bridge B3 by strengthening through-space RNA–RNA interactions, which restricts the relative movement of the subunits during translocation (Fig 6). L1 stalk movement and tRNA translocation are also inhibited [39] (Fig 6), emphasizing that the observed through-space internucleotide communication we observe is likely mediated by contacts with other components of the ribosomal translation complex. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. Effect of Spc on ribosome dynamics. Spc binding blocks free swiveling of the head domain and halts intersubunit rotation due to strengthened through-space interactions between pivot regions, thereby trapping the ribosome in a partially swiveled state. Black dashed lines indicate movements involving head swiveling and L1 stalk movement during translocation. Red dashed lines outline the portion of the outer-head domain concealed behind the inner-head domain. Spc, spectinomycin. https://doi.org/10.1371/journal.pbio.3000393.g006 In the Spc-arrested state, head swiveling is no longer able to guide translocating tRNAs from the hybrid to the post-translocation positions [8,10,12]. Specifically, Spc traps the ribosomal complex when the tRNAs assume an intermediate conformation, with the 2 tRNA anticodon loops on the 30S A and P sites, and their respective acceptor arms in the 50S P and E sites [2,12] (Fig 6). Elongation factor-G normally enters the 30S A site and facilitates displacement of the tRNA anticodon to the P site [6,13,16] but cannot do so without the 30S P-site tRNA moving to the E site, a movement guided by head swiveling [8]. Critically, these external contacts of the small subunit RNA—involving the large subunit, tRNAs, domain IV of elongation factor-G, and other intersubunit bridges [7]—are mediated by the inner-head and spine domain, which our study reveals is a structurally integrated entity (Fig 6). Structural and biophysical studies also indicate that Spc traps the head domain in a partially swiveled state [12] that contains tRNAs in a compacted conformation stabilized in an intermediate state of translocation [5,15]. This compaction occurs because the elbow regions of tRNAs are blocked by the 50S subunit L1 stalk, which normally moves to allow tRNA translocation on the 50S at a coordinated rate with respect to head swiveling (Fig 6) [39]. This restricted tRNA movement in the Spc-bound state prevents L1 stalk movement, resulting in no intersubunit rotation (Fig 6) [1,2]. Our model specifically depicts the distance and strength of this interaction (Fig 6). Thus, Spc binding locks the gears (intersubunit rotation and swiveling) of ribosome dynamics by effectively tying two distinct pivots to each other. RING-MaP, a single-molecule experiment, revealed extensive long-range structural communication throughout the small subunit of the ribosome in living cells. RING analysis enabled discovery of previously unobserved rRNA domain structure and through-space interactions. Analysis of Spc binding revealed how a small molecule can bind to and block the function of a large megadalton complex and suggests why it has proven so difficult to develop new ribosome-targeting antibiotics that function by novel mechanisms. This work reveals that Spc does not function by simple occlusion, local stabilization, or trapping but instead alters structural communication networks throughout the ribosome. Targeting RNA–RNA interactions, such as those that create the cross-domain connection we observe between h34-h35-h38 3-helix junction and the B3 bridge (Figs 4 and 5) might be a compelling drug-development strategy. We anticipate that single-molecule correlated probing will reveal new principles of large RNA domain and RNA-protein organization and structural communication when applied to functional motifs in viral RNAs, mRNAs, and long noncoding RNAs. Methods Bacterial cell growth and antibiotic treatment Overnight cultures (2 mL) were added to 48 mL of LB media. Cells were incubated with shaking until the culture reached an OD600 of approximately 0.5 (~30 min), at which time cells were collected by centrifugation. For antibiotic treated cells, 5.55 mL of a 187.5 μg/mL Rif solution were added, and cells were incubated with shaking for 10 min. Following incubation, 27 mL of each culture were transferred to a new culture flask. To each culture, either 3 mL of water or 3 mL of a 494 μg/mL of Spc were added. Cultures were incubated with shaking for 10 min. Cells were pelleted in 25-mL aliquots by centrifugation at 4,000g for 20 min. Supernatants were discarded, and the cell pellet was resuspended in 200 μL of folding buffer containing 300 mM cacodylate (pH 7.0), 200 mM potassium acetate (pH 7.0), and 10 mM MgCl2 and incubated at 37 °C for 5 min. Note that all +Spc samples were also pretreated with Rif. DMS treatment and purification of rRNA Aliquots (90 μL) of cells were added to 10 μL DMS (1:5 dilution in neat ethanol) ([+] reaction) or 10 μL neat ethanol ([–] reaction) and incubated at 37 °C for 6 min. Following incubation, an equal volume (100 μL) of neat 2-mercaptoethanol was added to quench the DMS. To each sample, 1 mL of TRIzol (Invitrogen, Carlsbad, CA) was added, and the reaction tubes were incubated at room temperature. After 5 min, 200 μL of cold chloroform was added, and tubes were shaken vigorously by hand for 15 s. Samples were incubated at room temperature for 2 to 3 min. Tubes were centrifuged at 12,000g for 15 min at 4 °C. The aqueous (upper) layer was transferred to a new tube, and 1.1 mL isopropanol was added. Reactions were incubated at −20 °C for 30 min and then centrifuged at 15,000g for 30 min at 4 °C. The supernatant was discarded, and pellets were carefully washed twice with 500 μL 80% ethanol, centrifuging 5 min at 15,000g between washes. Following the washes, the supernatant was discarded, and pellets were dried in air for 5 min. After resuspension in RNase-free water, samples were then treated with DNase I (Ambion, Carlsbad, CA) to remove any contaminating genomic DNA and subjected to affinity purification (RNeasy Mini Kit, Qiagen, Germantown, MD). Reverse transcription under MaP conditions To 700 ng of RNA was added 200 ng of random 9-mer primer, 2 μL of 10 mM dNTPs (Fermentas, Waltham, MA), and water to a final volume of 10 μL. Primers were annealed at 65 °C for 5 min followed by incubation at 4 °C for 2 min. Next, 9 μL of buffer master mix (2 μL of 500 mM Tris [pH 8.0], 750 mM KCl, and 100 mM DTT; 2.76 μL water and 4 μL 5 M betaine (Sigma, St Louis, MO); and 0.24 μL 500 mM MnCl2) was added to the annealed reaction mix. These conditions incorporate high concentrations of betaine and a primer annealing protocol that yield improved and efficient reverse transcription of DMS-modified products [40]. After incubation at 25 °C for 2 min, 1 μL of SuperScript II (Invitrogen, Carlsbad, CA) was added, and samples were incubated according to a stepped primer extension protocol: 25 °C for 10 min followed by 42 °C for 90 min, and then 10 cycles of 2 min at 50 °C and 2 min at 42 °C. The reverse transcriptase enzyme was inactivated by incubating the samples at 70 °C for 10 min. Following primer extension, cDNA products were purified (RNAclean beads, 1.8 bead to sample ratio; Beckman Coulter, Indianapolis, IN). Purified RNA was eluted from the beads in 68 μL nuclease-free water and converted to double-stranded DNA (dsDNA) using a second-strand synthesis enzyme mix (NEB, Ipswitch, MA). Following second-strand synthesis, dsDNA was purified (AmpureXP beads, 0.7:1 bead to sample ratio; Beckman Coulter, Indianapolis, IN). Product sizes following second-strand synthesis were analyzed (Agilent Bioanalyzer 2100). Library preparation and sequencing For library preparation, 1 ng of each second-strand synthesis product was used to create libraries for sequencing (NexteraXT, Illumina, San Diego, CA); final libraries were size selected (AmpureXP beads, using a 0.5:1 bead to sample ratio; Beckman Coulter, Indianapolis, IN) and quantified (Agilent Bioanalyzer 2100 and QuBit high-sensitivity dsDNA assays). Sequencing was performed on an Illumina NextSeq 500 system with a loading concentration of 1.4 pM, yielding approximately 400 million reads. Data processing and alignment Adapter sequences were removed from raw FASTQ files using the program scythe (version 0.991; available at https://github.com/vsbuffalo/scythe) with default parameters [41]. Reads were then trimmed for quality using sickle (version 1.33; available at https://github.com/najoshi/sickle) in paired-end mode with a Phred quality cutoff of 20 and a minimum length of 20 [42]. Only pairs for which both mates passed filtering were used in downstream stages. Following adapter removal and quality trimming, ShapeMapper (version 1.2) was used to map the processed FASTQ files to the 16S and 23S sequences [26,43]. No further quality trimming was performed during the quality trimming stage in ShapeMapper. During the read alignment stage, 2 additional flags in Bowtie2 were used to force concordant alignments: “—no-discordant” and “—no-mixed.” The following options were changed from the defaults to optimize for long insert sizes: “maxInsertSize = 1200,” “minMapQual = 30,” and “minPhredToCount = 30.” Analysis framework for randomly primed RING-MaP reads The previously reported RING analysis approach [17] required that all sequencing reads be stored in computer memory in order to perform association analyses. This approach is appropriate for small RNAs but impractical for RNAs as long as the 16S rRNA. Rather than retaining all sequencing reads in memory, it is possible to create a simplified representation of alignment and mutation location information (S1 Fig). This representation is a two-dimensional array with each element containing a contingency table of the counts and kinds of interactions (S1 Fig). Information about pairwise observations of mutations within each read can thus be counted and stored independently. Using this strategy, the total amount of memory needed for analysis depends only on the RNA length and not on the number of sequencing reads. During analysis of sequencing data, only reads that meet Phred quality cutoffs were included. Since interactions are stored as i-j interaction pairs, long stretches of incomplete information (such as gaps between sequencing reads relative to the reference sequence) were allowed, with each i-j point in the matrix representing the contingency table for all reads that contained both nucleotides i and j. Using the contingency table, a Yates chi-squared statistic and Pearson correlation (phi) was calculated (S1 Fig). Correlations with Yates chi-squared values above 20 were considered significant. Based on this threshold for chi-squared statistics, the probability that correlated nucleotides were independent was less than 10−5. In paired-end sequencing, both ends of the DNA library are sequenced even though they may be separated by several hundred nucleotides. Modern sequencing platforms keep paired reads “together,” effectively allowing detection of interactions at a distance up to the size of the inserts of the sequencing library. In this work, large DNA fragments were selected, and sequencing libraries were constructed with insert sizes between 500 and 700 nucleotides. Approximately 50,000 reads between 2 locations of the RNA were needed to reliably detect correlated interactions. At this sequencing depth, for each of our samples, we can reliably detect interactions between nucleotides that are separated by 450 to 650 nucleotides in sequence space with the specific number depending on biases resulting from random priming (S1 Fig). Correlation analysis of randomly primed reads The “mutation strings” files from the ShapeMapper pipeline were used as input for randomly primed correlation analysis; these files contain a simplified representation of the read alignment location, mutation locations, and sequencing instrument quality calls. A square matrix was constructed with a size equal to the length of the aligned RNA (S1 Fig). Each element (P) contains a 2 × 2 contingency matrix representing the possible outcomes in comparing 2 nucleotides (S1 Fig). In each read, all i-j combinations of nucleotides were used to index the storage matrix. Mutations (scored as 1) and matching nucleotides (scored as 0) were used to index the contingency table. Only nucleotides with a phred score above 30 were counted. We also excluded 26 residues that showed a high background mutation rate (in the non–DMS-treated sample), most of which correspond to known single nucleotide polymorphisms in the E. coli 16S rRNA gene. After all reads were processed, the storage matrix contained an easily indexed representation of the entire sequencing dataset, and each i-j element in the matrix contained a snapshot of all the reads that span nucleotides i and j. The total number of times nucleotides i and j were read together (Ni,j) is the sum of all the elements in the contingency table. Next, each i < j pair in the read storage matrix was tested for significance using the Yates chi-squared test with a significance criterion of 20 (S1 Fig); the strength of the correlation was measured using the Pearson r metric. Correlations from the two 2 biological replicates, performed for each condition, were pooled by requiring that a correlation pair occur in both replicates to be included (S2 Fig). Network analysis of correlations in the 16S rRNA Correlation values from the +Rif and +Spc samples were analyzed using the network visualization software Gephi (version 0.9.2; available at https://gephi.org) (S1 Fig) [30]. A network diagram was drawn as an “undirected graph,” which produces a network of nodes with connecting edges such that all edges are bidirectional. Filtering was used to restrict the network population requiring that (i) the strength of the correlation must be greater than 0.015 and that (ii) each node must have at least 3 connections (k-core = 3). Nucleotides were treated as network nodes, and edges depict correlation strength between connected 2 nodes. Network diagrams were arranged such that nodes linked by stronger connecting weights attract each other, while nodes with weaker connecting weights are pushed apart (Force Atlas option). Node color and size and edge color and thickness were set (Ranking Module, using Degree as the ranking parameter). In an undirected graph, the degree of a node is simply the sum of all its edges. Communities were detected by Modularity, using the Louvian method [44]; graph modularity was calculated with sensitivity setting of 1.0. This community detection algorithm creates a Modularity Class for each node, which was used to partition the network into communities, represented by different colors. To visualize the network graph on the structure of the ribosome, the Average weighted degree parameter was used to generate a list of nodes ranked by the sum of the total weight (correlation strength) of all its edges. Nodes were categorized as strong (greater than 0.25), medium (between 0.1 and 0.25), or weak (lower than 0.1) based on average weighted degree values. Nucleotides corresponding to these nodes are represented as spheres (large, medium, or small) on the structure of the ribosomal small subunit (PDB 4v56) [6]. Network edges were ranked by correlation strength, and categorized as strong (75th percentile), medium (50th–75th percentile), or weak (below 50th percentile). Strong edges are represented on the three-dimensional structure of the ribosomal small subunit as colored lines. Small ribosomal subunit domains by network community data were defined by including regions occupied by high strength nodes in both the absence and presence of Spc. Nodes from the more extensive +Spc network graph were then superimposed on the network defined domains (Fig 3). Bacterial cell growth and antibiotic treatment Overnight cultures (2 mL) were added to 48 mL of LB media. Cells were incubated with shaking until the culture reached an OD600 of approximately 0.5 (~30 min), at which time cells were collected by centrifugation. For antibiotic treated cells, 5.55 mL of a 187.5 μg/mL Rif solution were added, and cells were incubated with shaking for 10 min. Following incubation, 27 mL of each culture were transferred to a new culture flask. To each culture, either 3 mL of water or 3 mL of a 494 μg/mL of Spc were added. Cultures were incubated with shaking for 10 min. Cells were pelleted in 25-mL aliquots by centrifugation at 4,000g for 20 min. Supernatants were discarded, and the cell pellet was resuspended in 200 μL of folding buffer containing 300 mM cacodylate (pH 7.0), 200 mM potassium acetate (pH 7.0), and 10 mM MgCl2 and incubated at 37 °C for 5 min. Note that all +Spc samples were also pretreated with Rif. DMS treatment and purification of rRNA Aliquots (90 μL) of cells were added to 10 μL DMS (1:5 dilution in neat ethanol) ([+] reaction) or 10 μL neat ethanol ([–] reaction) and incubated at 37 °C for 6 min. Following incubation, an equal volume (100 μL) of neat 2-mercaptoethanol was added to quench the DMS. To each sample, 1 mL of TRIzol (Invitrogen, Carlsbad, CA) was added, and the reaction tubes were incubated at room temperature. After 5 min, 200 μL of cold chloroform was added, and tubes were shaken vigorously by hand for 15 s. Samples were incubated at room temperature for 2 to 3 min. Tubes were centrifuged at 12,000g for 15 min at 4 °C. The aqueous (upper) layer was transferred to a new tube, and 1.1 mL isopropanol was added. Reactions were incubated at −20 °C for 30 min and then centrifuged at 15,000g for 30 min at 4 °C. The supernatant was discarded, and pellets were carefully washed twice with 500 μL 80% ethanol, centrifuging 5 min at 15,000g between washes. Following the washes, the supernatant was discarded, and pellets were dried in air for 5 min. After resuspension in RNase-free water, samples were then treated with DNase I (Ambion, Carlsbad, CA) to remove any contaminating genomic DNA and subjected to affinity purification (RNeasy Mini Kit, Qiagen, Germantown, MD). Reverse transcription under MaP conditions To 700 ng of RNA was added 200 ng of random 9-mer primer, 2 μL of 10 mM dNTPs (Fermentas, Waltham, MA), and water to a final volume of 10 μL. Primers were annealed at 65 °C for 5 min followed by incubation at 4 °C for 2 min. Next, 9 μL of buffer master mix (2 μL of 500 mM Tris [pH 8.0], 750 mM KCl, and 100 mM DTT; 2.76 μL water and 4 μL 5 M betaine (Sigma, St Louis, MO); and 0.24 μL 500 mM MnCl2) was added to the annealed reaction mix. These conditions incorporate high concentrations of betaine and a primer annealing protocol that yield improved and efficient reverse transcription of DMS-modified products [40]. After incubation at 25 °C for 2 min, 1 μL of SuperScript II (Invitrogen, Carlsbad, CA) was added, and samples were incubated according to a stepped primer extension protocol: 25 °C for 10 min followed by 42 °C for 90 min, and then 10 cycles of 2 min at 50 °C and 2 min at 42 °C. The reverse transcriptase enzyme was inactivated by incubating the samples at 70 °C for 10 min. Following primer extension, cDNA products were purified (RNAclean beads, 1.8 bead to sample ratio; Beckman Coulter, Indianapolis, IN). Purified RNA was eluted from the beads in 68 μL nuclease-free water and converted to double-stranded DNA (dsDNA) using a second-strand synthesis enzyme mix (NEB, Ipswitch, MA). Following second-strand synthesis, dsDNA was purified (AmpureXP beads, 0.7:1 bead to sample ratio; Beckman Coulter, Indianapolis, IN). Product sizes following second-strand synthesis were analyzed (Agilent Bioanalyzer 2100). Library preparation and sequencing For library preparation, 1 ng of each second-strand synthesis product was used to create libraries for sequencing (NexteraXT, Illumina, San Diego, CA); final libraries were size selected (AmpureXP beads, using a 0.5:1 bead to sample ratio; Beckman Coulter, Indianapolis, IN) and quantified (Agilent Bioanalyzer 2100 and QuBit high-sensitivity dsDNA assays). Sequencing was performed on an Illumina NextSeq 500 system with a loading concentration of 1.4 pM, yielding approximately 400 million reads. Data processing and alignment Adapter sequences were removed from raw FASTQ files using the program scythe (version 0.991; available at https://github.com/vsbuffalo/scythe) with default parameters [41]. Reads were then trimmed for quality using sickle (version 1.33; available at https://github.com/najoshi/sickle) in paired-end mode with a Phred quality cutoff of 20 and a minimum length of 20 [42]. Only pairs for which both mates passed filtering were used in downstream stages. Following adapter removal and quality trimming, ShapeMapper (version 1.2) was used to map the processed FASTQ files to the 16S and 23S sequences [26,43]. No further quality trimming was performed during the quality trimming stage in ShapeMapper. During the read alignment stage, 2 additional flags in Bowtie2 were used to force concordant alignments: “—no-discordant” and “—no-mixed.” The following options were changed from the defaults to optimize for long insert sizes: “maxInsertSize = 1200,” “minMapQual = 30,” and “minPhredToCount = 30.” Analysis framework for randomly primed RING-MaP reads The previously reported RING analysis approach [17] required that all sequencing reads be stored in computer memory in order to perform association analyses. This approach is appropriate for small RNAs but impractical for RNAs as long as the 16S rRNA. Rather than retaining all sequencing reads in memory, it is possible to create a simplified representation of alignment and mutation location information (S1 Fig). This representation is a two-dimensional array with each element containing a contingency table of the counts and kinds of interactions (S1 Fig). Information about pairwise observations of mutations within each read can thus be counted and stored independently. Using this strategy, the total amount of memory needed for analysis depends only on the RNA length and not on the number of sequencing reads. During analysis of sequencing data, only reads that meet Phred quality cutoffs were included. Since interactions are stored as i-j interaction pairs, long stretches of incomplete information (such as gaps between sequencing reads relative to the reference sequence) were allowed, with each i-j point in the matrix representing the contingency table for all reads that contained both nucleotides i and j. Using the contingency table, a Yates chi-squared statistic and Pearson correlation (phi) was calculated (S1 Fig). Correlations with Yates chi-squared values above 20 were considered significant. Based on this threshold for chi-squared statistics, the probability that correlated nucleotides were independent was less than 10−5. In paired-end sequencing, both ends of the DNA library are sequenced even though they may be separated by several hundred nucleotides. Modern sequencing platforms keep paired reads “together,” effectively allowing detection of interactions at a distance up to the size of the inserts of the sequencing library. In this work, large DNA fragments were selected, and sequencing libraries were constructed with insert sizes between 500 and 700 nucleotides. Approximately 50,000 reads between 2 locations of the RNA were needed to reliably detect correlated interactions. At this sequencing depth, for each of our samples, we can reliably detect interactions between nucleotides that are separated by 450 to 650 nucleotides in sequence space with the specific number depending on biases resulting from random priming (S1 Fig). Correlation analysis of randomly primed reads The “mutation strings” files from the ShapeMapper pipeline were used as input for randomly primed correlation analysis; these files contain a simplified representation of the read alignment location, mutation locations, and sequencing instrument quality calls. A square matrix was constructed with a size equal to the length of the aligned RNA (S1 Fig). Each element (P) contains a 2 × 2 contingency matrix representing the possible outcomes in comparing 2 nucleotides (S1 Fig). In each read, all i-j combinations of nucleotides were used to index the storage matrix. Mutations (scored as 1) and matching nucleotides (scored as 0) were used to index the contingency table. Only nucleotides with a phred score above 30 were counted. We also excluded 26 residues that showed a high background mutation rate (in the non–DMS-treated sample), most of which correspond to known single nucleotide polymorphisms in the E. coli 16S rRNA gene. After all reads were processed, the storage matrix contained an easily indexed representation of the entire sequencing dataset, and each i-j element in the matrix contained a snapshot of all the reads that span nucleotides i and j. The total number of times nucleotides i and j were read together (Ni,j) is the sum of all the elements in the contingency table. Next, each i < j pair in the read storage matrix was tested for significance using the Yates chi-squared test with a significance criterion of 20 (S1 Fig); the strength of the correlation was measured using the Pearson r metric. Correlations from the two 2 biological replicates, performed for each condition, were pooled by requiring that a correlation pair occur in both replicates to be included (S2 Fig). Network analysis of correlations in the 16S rRNA Correlation values from the +Rif and +Spc samples were analyzed using the network visualization software Gephi (version 0.9.2; available at https://gephi.org) (S1 Fig) [30]. A network diagram was drawn as an “undirected graph,” which produces a network of nodes with connecting edges such that all edges are bidirectional. Filtering was used to restrict the network population requiring that (i) the strength of the correlation must be greater than 0.015 and that (ii) each node must have at least 3 connections (k-core = 3). Nucleotides were treated as network nodes, and edges depict correlation strength between connected 2 nodes. Network diagrams were arranged such that nodes linked by stronger connecting weights attract each other, while nodes with weaker connecting weights are pushed apart (Force Atlas option). Node color and size and edge color and thickness were set (Ranking Module, using Degree as the ranking parameter). In an undirected graph, the degree of a node is simply the sum of all its edges. Communities were detected by Modularity, using the Louvian method [44]; graph modularity was calculated with sensitivity setting of 1.0. This community detection algorithm creates a Modularity Class for each node, which was used to partition the network into communities, represented by different colors. To visualize the network graph on the structure of the ribosome, the Average weighted degree parameter was used to generate a list of nodes ranked by the sum of the total weight (correlation strength) of all its edges. Nodes were categorized as strong (greater than 0.25), medium (between 0.1 and 0.25), or weak (lower than 0.1) based on average weighted degree values. Nucleotides corresponding to these nodes are represented as spheres (large, medium, or small) on the structure of the ribosomal small subunit (PDB 4v56) [6]. Network edges were ranked by correlation strength, and categorized as strong (75th percentile), medium (50th–75th percentile), or weak (below 50th percentile). Strong edges are represented on the three-dimensional structure of the ribosomal small subunit as colored lines. Small ribosomal subunit domains by network community data were defined by including regions occupied by high strength nodes in both the absence and presence of Spc. Nodes from the more extensive +Spc network graph were then superimposed on the network defined domains (Fig 3). Supporting information S1 Fig. Algorithmic innovations to enable RING-MaP analysis of large-scale interactions using randomly primed reverse transcription. (A) Data storage matrix used to show counts of within-read interactions. (B) Contingency table for each data storage element (P). (C) Equations for calculating significance of each interaction performed after read counting using Yates chi-squared test. Correlation strength was calculated using the Pearson r metric. (D) Detection interval for DMS-modified 16S rRNA using the optimized cDNA synthesis protocol. The effective maximum detection interval for correlation analysis requires approximately 50,000 reads. This interval is enclosed with a dashed line. https://doi.org/10.1371/journal.pbio.3000393.s001 (TIF) S2 Fig. RING-MaP correlations measured for replicate in-cell states of the small ribosomal subunit. (A) Correlations from 2 biological replicates were merged; only those that occurred in both replicates were retained. (B) RING-MaP correlations for merged replicates as a function of cellular state. Conventional domain boundaries for the 16S rRNA are indicated. Base pairs (top) present in the structure established by covariation analysis are shown as gray arcs. Correlations for each of the merged datasets for the 3 in-cell conditions are shown as red and blue arcs for positive and negative correlations, respectively. The underlying data for this figure are available at: https://doi.org/10.6084/m9.figshare.9252995.v1. https://doi.org/10.1371/journal.pbio.3000393.s002 (TIF) S3 Fig. In-cell binding by Spc increases numbers of and strengths of correlations. (A) Edges present in +Spc network diagram colored by correlation with the same edges in the +Rif network. (B) Correlations strengthened in the presence of Spc relative to the +Rif network. Correlation strength is illustrated by edge thickness. The underlying data for this figure are available at: https://doi.org/10.6084/m9.figshare.9252995.v1. https://doi.org/10.1371/journal.pbio.3000393.s003 (TIF)
EXP1 is critical for nutrient uptake across the parasitophorous vacuole membrane of malaria parasitesMesén-Ramírez, Paolo;Bergmann, Bärbel;Tran, Thuy Tuyen;Garten, Matthias;Stäcker, Jan;Naranjo-Prado, Isabel;Höhn, Katharina;Zimmerberg, Joshua;Spielmann, Tobias
doi: 10.1371/journal.pbio.3000473pmid: 31568532
Introduction Malaria pathology results from the replication of Plasmodium parasites in red blood cells (RBCs). Malaria parasites proliferate within a parasitophorous vacuole (PV) surrounded by the membrane of the PV (PVM) [1,2], the interface between parasite and host cell. The PVM harbors protein complexes that serve key functions for the intracellular survival of the parasite [1]. The Plasmodium translocon of exported proteins (PTEX) mediates the transport of parasite proteins across the PVM into the host cell [3,4,5]. A nonselective pore permits passage of nutrients such as monosaccharides, folates, and amino acids through the PVM [6,7]. While the composition and function of PTEX has been studied in some detail, the molecular basis for the nutrient-permeable channel activity is much less defined. PTEX comprises oligomers of 3 core components, including exported protein 2 (EXP2), which forms a heptameric PVM-spanning channel [8]. Conditional knockdown and patch-clamp measurements indicated that EXP2 has a dual role in both protein export as part of PTEX and in the nutrient-permeable channel activity of the PVM [7], a conclusion initially indicated by the homology and complementation capacity of EXP2 to 2 proteins required for the solute pore activity in Toxoplasma gondii [9]. Other PVM proteins include highly expressed single-pass transmembrane (TM) proteins such as early transcribed membrane proteins (ETRAMPs) [10] and exported protein 1 (EXP1) [11]. EXP1 was the first protein localized to the PVM of blood and liver stages where it forms homo-oligomers with the N-terminus facing the PV lumen and the C-terminus exposed to the RBC cytosol [1,11,12,13]. Unsuccessful attempts to genetically disrupt exp1 suggested an essential role for EXP1 in parasite development [14]. Bioinformatic analyses classified EXP1 as a member of the superfamily of membrane-associated proteins in eicosanoid and glutathione metabolism (MAPEG) [15,16]. Enzymatic in vitro assays indicated that EXP1 possesses glutathione S-transferase (GST) activity, and it was proposed to protect the parasite from oxidative damage via detoxification of hemoglobin byproducts by conjugation with reduced glutathione (GSH) [16]. It was further proposed that EXP1 is associated with artemisinin (ART) resistance, as its transcription was up-regulated in ART-resistant parasite strains [16]. A study in P. berghei indicated that EXP1 may also play a role in the uptake of lipids during intrahepatic development [17]. Here, using a conditional knockout (KO) of exp1, we show that EXP1 is essential for the growth of P. falciparum blood stages independently of the previously proposed GST activity. We find that EXP1 is required for the nutrient-permeable channel activity of the PVM. Consistently, parasites with reduced levels of EXP1 became hypersensitive to nutrient-limiting conditions, associating the permeability measured at the PVM with nutrient acquisition. EXP1 interacts with EXP2 at the PVM and is required for EXP2’s proper distribution and function as a nutrient-permeable channel but not for the function of EXP2 in protein export. Hence, EXP1 defines the function of EXP2 as a nutrient-permeable channel and is critical for nutrient uptake across the PVM, which now can be specifically studied. Results EXP1 is essential for parasite development in RBCs To first test whether EXP1 is required for the survival of P. falciparum erythrocytic stages, we generated a conditional exp1 KO based on the Dimerizable Cre (DiCre) system [18,19, 20] using selection-linked integration (SLI) [20]. The endogenous exp1 was disrupted before the region encoding the TM domain, and at the same time, a second functional copy of exp1 flanked by loxP sites was introduced in the exp1 locus. The new floxed copy of EXP1 is hemagglutinin (HA) tagged and can be conditionally excised by DiCre upon addition of rapalog (Fig 1A and S1A and S1B Fig). Immunofluorescence assays (IFAs) showed that the corresponding cell line (condΔEXP1) correctly expressed the functional EXP1-HA in the PVM (Fig 1B). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. EXP1 is essential for blood stage development. (A) Simplified schematic of DiCre-based conditional exp1 KO using SLI. Arrows indicate primers P1 and P2 (see S1 Fig for details). (B) IFA of compound 2-arrested condΔEXP1 late-stage schizonts show localization of EXP1*-HA at the PVM. α-MSP1 (MSP1) labels the PPM. Nuclei were stained with DAPI; scale bars, 5 μm. (C) Strategy for depletion of EXP1 from the PVM of synchronized condΔEXP1 ring stages divided into a culture with rapalog (“rap”) and one without rapalog (“control”). Top: schematic: green boxes and green line signify PVM with EXP1. Mid: PCR with primers P1 and P2 from gDNA 24 and 48 hours after addition of rapalog. Original: band for intact exp1 locus (1,919 bp); excised: band after excision of exp1* (1,326 bp). Bottom: western blot probed with α-HA to detect EXP1*-HA and α-BIP as loading control. Asterisk: unskipped (first T2A) product that is present before excision and hence has no impact on parasites. (D) Quantification of levels of wtEXP1-HA in the first ("24 h") and second ("48 h") cycle after addition of rapalog based on densitometric analysis of anti-HA immunoblots (mean of n = 2 independent experiments) of which one is shown in (C). HA signal was normalized to BIP. Error bars indicate SD. (E) FC growth curves of synchronous ring stage condΔEXP1 parasites starting in cycle 2 (ΔEXP1 parasites) as shown in (C). One representative of n = 5 independent experiments. (F) Stage distribution of condΔEXP1 parasites in Giemsa smears of synchronous parasites grown with (+) and one without (−) rapalog (“rap”) at different time points (average time post invasion) after adding rapalog. Light blue arrowheads show blebs. One of n = 4 independent experiments is shown. Blue arrow in (C–F) indicates start of a new cycle without EXP1. 2A, T2A skip peptide; BIP, binding immunoglobulin protein; DIC, differential interference contrast; DiCre, Dimerizable Cre; EXP1, exported protein 1; EXP1*, recodonized exp1; FC, flow cytometry; gDNA, genomic DNA; HA, triple hemagglutinin tag; KO, knockout; MSP1, merozoite surface protein 1; PPM, parasite plasma membrane; PVM, parasitophorous vacuolar membrane; SLI, selection-linked integration; SP, signal peptide; TM, transmembrane domain; wt, wild-type. https://doi.org/10.1371/journal.pbio.3000473.g001 To investigate the effect of the loss of EXP1 on parasite survival, synchronous condΔEXP1 ring-stage parasites were grown in the presence of rapalog, alongside a control culture (Fig 1C). PCR confirmed efficient excision of the functional copy of exp1 within one growth cycle (48 hours) upon addition of rapalog (Fig 1C). No growth defect was observed in this first cycle during which the exp1 gene was excised (S1C Fig). In this cycle, EXP1 protein levels remained at approximately 70% due to protein expressed in the PVM before excision was complete (Fig 1C and 1D and S1D Fig). Western blot confirmed loss of EXP1 (approximately 5% residual protein detected at 48 h; Fig 1D) in parasites (henceforth termed ΔEXP1 parasites) after invasion and start of a new cycle (Fig 1C, blue arrow). The ΔEXP1 parasites failed to replicate (Fig 1E and S1C Fig), demonstrating that EXP1 is essential for propagation in RBC. The N-terminal fragment remaining after excision was nonfunctional as it does not rescue growth and was only detectable by IFA (S1D Fig), likely because of its small size and possible low stability. Giemsa smears taken in regular intervals from synchronous parasites showed that ΔEXP1 ring stages were much slower to reach the trophozoite stage than controls (Fig 1F). ΔEXP1 trophozoites often displayed protrusions reaching into the host cell cytoplasm (“blebs,” light blue arrowheads, Fig 1F) and frequently had an aberrant condensed morphology (Fig 1F). ΔEXP1 trophozoites did not complete schizogony as evident by a significantly reduced number of nuclei per ΔEXP1 parasite compared to controls (S1E Fig) and by an almost complete absence of new rings in the next cycle (Fig 1F). To test whether very slowly growing ΔEXP1 parasites persisted, we carried out an extended growth assay with the ΔEXP1 parasites. These experiments revealed a resurfacing of parasites in the ΔEXP1 culture grown in the presence of rapalog 9 days after loss of EXP1, but PCR identified this as a population with a nonexcised exp1 locus. These parasites were therefore breakthroughs, further indicating that loss of EXP1 abolishes parasite propagation in RBCs (S1F Fig). To observe the phenotype of EXP1 loss in more detail, we compared the development of ΔEXP1 parasites to controls, using long-term time-lapse imaging [21] (Fig 2A). This confirmed a severe delay of the ring stage (Fig 2B) and a very slow development of trophozoites without completion of the cycle (Fig 2A). Time-lapse imaging also revealed phenotypes in ring stages: ΔEXP1 parasites changed position in the host cell less frequently and rarely showed amoeboid shapes (Fig 2A and 2C), 2 typical features of ring stages [21] regularly observed in controls (Fig 2A and 2C). In addition, ΔEXP1 parasites were often found closely adjoined to the RBC membrane, a phenomenon that we here termed “hugging” (Fig 2A and 2C, red arrowheads), also evident by electron microscopy (S1G Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Morphological phenotypes of ΔEXP1 parasites. (A) Time-lapse imaging of condΔEXP1 parasites starting a cycle with (control) and without EXP1 (rapalog) imaged side by side. Single DIC z-sections of selected time points typical phenotypes are shown: top row: slow development; mid row: failure to reach full trophozoite stage; bottom row: cell showing extensive hugging. Arrowheads: white, ring stages after reinvasion; blue, amoeboid ring; light blue, blebs; red, hugging. (B) Number of hours (after start of the time lapse experiment) to reach young trophozoite stage in n = 24 control and N = 27 ΔEXP1 parasites. (C) Frequency of the indicated events in individual ring stages observed by time lapse microscopy (presence or absence of event was scored every hour); n = 25 parasites each for ΔEXP1 and control. (D, E) Left, live cell images of control and ΔEXP1 (rapalog) parasites expressing Lyn-mCherry (PPM marker) (D) or labelled with Bodipy TR ceramide (BODIPY) (E). Light blue arrowheads, blebs; yellow arrowheads, TVN. Graphs: quantification of number of blebs per cell (controls N = 25, 73 for (D) and (E), respectively; ΔEXP1 (rapalog) n = 28 and 156 for (D) and (E), respectively. (F) Left, IFA of control and ΔEXP1 (rapalog) late-stage gametocytes using α-HA to detect EXP1*-HA and α-Pfg377 (late-stage gametocyte marker). Right, PCR using primers P1 and P2 (Fig 1A) confirms excision of exp1 in late gametocytes. (G) Left, gametocytemia of control and ΔEXP1 parasites early after induction (Pfs16 positive cells) and 8 days later (Pfg377 positive cells) based on IFAs. Right, fold reduction in the number of early (Pfs16) and late (Pfg377) ΔEXP1 gametocytes versus control; n = 3. (D, E, and F); scale bars, 5 μm. In (F), nuclei were stained with DAPI. In (B–E and G), green lines indicate mean and error bars (SD); two-tailed unpaired t test, P values indicated. DIC, differential interference contrast; EXP1, exported protein 1; HA, triple hemagglutinin tag; IFA, immunofluorescence assay; PPM, parasite plasma membrane. https://doi.org/10.1371/journal.pbio.3000473.g002 The protrusions observed in Giemsa smears were also visible in differential interference contrast (DIC) and also in ΔEXP1 rings (Fig 2A, light blue arrowheads). These “blebs” were bounded by PPM and present almost exclusively in ΔEXP1 parasites as confirmed with a parasite plasma membrane (PPM) marker (Lyn-mCherry) [22] (Fig 2D). The total number of membrane-bounded protrusions detected using in Bodipy-TR-ceramide or Lyso-PC was also mildly enriched in ΔEXP1 parasites (Fig 2E and S1H Fig). Despite these morphological alterations, the PVM integrity in ΔEXP1 parasites was not compromised, because a soluble PV marker (SP-mScarlet) was retained in the PV (S1I Fig). Next, we evaluated the impact of EXP1 loss on gametocytogenesis. Interestingly, we detected stage III–V gametocytes lacking EXP1 (Fig 2F), although the number of late-stage gametocytes was reduced by more than 50% 8 days after induction (Fig 2G). Nevertheless, this indicated a lower effect of loss of EXP1 on the development of gametocytes than on asexual blood stages. All regions of EXP1 are important for its function To confirm that the observed growth phenotype is specific for EXP1 loss, we complemented the condΔEXP1 parasites with a Ty-tagged full-length copy of EXP1 (EXP1wt-Ty) expressed under the constitutive nmd3 promoter. EXP1wt-Ty was correctly located at the PVM (Fig 3A) and, after removal of the endogenous EXP1, restored parasite growth to 80% of the unexcised control (Fig 3B and S2A–S2C Fig). The level of complementation correlated with the level of expression of the EXP1wt-Ty construct, as demonstrated using promoters driving different levels of expression (Fig 3B–3D and S2A–S2C Fig). The complemented parasites also showed a similar duration of the ring stage compared to the wild type, indicating that the delay to reach the trophozoite was reverted by the complementation (S2D Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Complementation pinpoints important regions of EXP1. (A) IFA images of condΔEXP1 schizont stages expressing EXP1 wt-Ty probed with α-HA and α-Ty show localization of EXP1*-HA and EXP1 wt-Ty in the PVM. α-MSP1 (MSP1) labels the PPM. DAPI, nuclei; scale bars, 5 μm. Right: immunoblot of extracts of cell line on the left probed with α-Ty to detect EXP1wt-Ty and α-SBP1 as control for a TM protein. Saponin was used to separate the parasite pellet (“P”) from the supernatant (“SN”) containing PV and host cell content. See S3 Fig for IFAs and immunoblots of all complementation constructs. (B) Relative activity of the EXP1 complementation constructs indicated. Except where otherwise indicated, constructs were expressed under the nmd3 (mid) promoter. The complementation capacity of every tested construct was calculated as relative activity to the EXP1wt-Ty construct under the nmd3 promoter, which was set as 100% (right dotted green line). ΔEXP1 was set to 0% (left dotted green line). Each data point (red dot) shows growth of rapalog-treated versus unexcised parasites at the end of a 5-day growth assay relative to the growth of the wt construct; n ≥ 4 independent experiments per cell line. Error bars: SD. See S2 Fig for activity of all complementation constructs. (C) Immunoblot of lysates of condΔEXP1 parasites expressing EXP1wtlow, EXP1mid, and EXP1high probed with α-Ty (EXP1wt-Ty), α-HA (EXP1*-HA), and α-BIP (loading control). (D) Densitometric analysis of EXP1wt expression levels (C) under low, mid, and high promoters relative to the mid promoter (green). Mean of 3 independent experiments. Error bars: SD. (E) Immunoblot of extracts of +/−formaldehyde (PFA) treated cell lines expressing the indicated constructs probed with α-Ty. Single asterisk: monomer; double asterisk: dimer. BIP, binding immunoglobulin protein; DIC, differential interference contrast; EXP1, exported protein 1; HA, triple hemagglutinin tag; IFA, immunofluorescence assay; MSP1, merozoite surface protein 1; PPM, parasite plasma membrane; PV, parasitophorous vacuole; PVM, parasitophorous vacuolar membrane; SBP1, skeleton binding protein 1; TM, transmembrane; wt, wild type. https://doi.org/10.1371/journal.pbio.3000473.g003 To pinpoint the functional regions in EXP1, we tested a series of modified Ty-tagged EXP1 constructs for their capacity to complement loss of endogenous EXP1. Except for EXP1wt-mScarlet, all constructs were correctly inserted into the PVM, and deletions or replacements in the N- or C-terminus mostly led to severe loss of function (S2A-S2C and S3A and S3B Fig). Interestingly, EXP1 of the rodent malaria parasite P. berghei (PbEXP1) only partially (46.1% ± 9.4% activity) rescued loss of EXP1 in P. falciparum (Fig 3B), indicating limited functional conservation between species. Deletion of the entire C-terminus of EXP1 reduced its activity to 59.37% ± 13.82% (Fig 3B). EXP1 lacking an 11 amino acid stretch (sequence SGVSSKKKNKK) in the N-terminus named E-domain (EXP1ΔED), a region proposed to be necessary for the dimerization based on similarity with MAPEGs [16,23], complemented only poorly (Fig 3B). Previous work indicated that EXP1 homo-oligomerizes [12]. However, loss of function of EXP1ΔED was not due to profound alterations in its capacity to oligomerize, as dimers were still detectable after formaldehyde crosslinking (Fig 3E). We noticed that the TM domains of integral PVM proteins such as EXP1 and ETRAMPs in different malaria species are particularly rich in G, S, and T residues (S2E Fig). G, S, and T are known to be important as TM interaction interfaces [24, 25]. Strikingly, point mutations of the first glycine residues (from G to L) of the two GXXG motifs of the P. falciparum EXP1 (PfEXP1) TM abolished EXP1 function (Fig 3B) although the protein was still correctly trafficked (S3A and S3B Fig) and capable of dimerizing (Fig 3E). Collectively, we conclude that all parts of EXP1 are required for its function. EXP1 GST activity is dispensable for parasite growth Previous work showed that recombinant EXP1 conjugates hematin to GSH in vitro and thereby may protect the parasite from heme-induced oxidative damage [16]. Based on homology with other MAPEGs, it was postulated that the catalytic center of the GST activity of EXP1 resides in 3 N-terminal residues (Fig 4A). Mutation of one of these residues (R70) led to a reduced enzymatic activity in vitro [16]. We used our complementation approach to assess the importance of these residues (and hence of the proposed GST activity) for parasite development. Constructs with mutations of 1 (EXP1R70A) or all 3 residues (EXP1-3xmut) of the proposed catalytic site showed 69.0% ± 9.2% and 62.5% ± 16.2% complementation activity, respectively (Fig 4B), indicating that these mutations had only a moderate effect on EXP1 function compared to most other complementation constructs (Fig 3B and S2A Fig). IFAs and solubility assays showed correct targeting of the complementation constructs to the PVM (S3A and S3B Fig). EXP1 R70A expressed under the stronger promoter resulted in nearly identical complementation capacity to the wild-type construct (Fig 4B). These data indicate that if EXP1 is a MAPEG, its GST activity plays only a minor role for growth of blood stage parasites. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Dispensability of EXP1 GST activity and lack of oxidative stress in ΔEXP1 parasites. (A) Schematic of the region of EXP1 with the proposed catalytic site of the GST activity and of the mutations introduced. (B) Relative activity of the EXP1 complementation constructs indicated. Green lines: activity of EXP1wt (nmd3, mid) (set as 100%) and absence of activity (ΔEXP1) set as 0%; n ≥ 4 independent experiments per cell line. (C) Live cell images of condΔEXP1 parasites incubated with CM-H2DCFDA. Scale bars: 5 μm. (D) Fluorescence intensity of matching stages of control and ΔEXP1 parasites (rapalog) incubated with CM-H2DCFDA (C). Results from 3 independent experiments with a total of n = 55 control and n = 53 ΔEXP1 cells. Green line, mean; error bars, SD. (E) FC growth curves of synchronous condΔEXP1 parasites and the complementation cell line EXP1 wtlow after one cycle with (red) and without (black) rapalog (see Fig 1D) grown in RPMI alone or supplemented with the compounds indicated. One representative of n = 3 independent biological replicas. (F) Left, growth curves of EXP1wtlow parasites ± E64 starting after one growth cycle ± rapalog (see Fig 1D). Right, Giemsa smears of parasites on day 2 (after incubation with E64) and day 3 (after removal). Arrowheads: swollen food vacuole. One representative of n = 3 biological replicas. (G) Effect of E64 treatment on survival of DHA and rapalog-treated EXP1wtlow parasites versus untreated. Mean of n = 3 independent experiments. (H) Left, dose-response curves of the parasite lines indicated treated with DHA (0–50 nM). Right, DHA IC50 values for these cell lines ± rapalog. Mean of n ≥ 3 experiments. (I) RSAs of the indicated complementation cell lines and a Kelch13 C580Y mutant line. Data points are percent survival of DHA treated versus untreated parasites ± rapalog. Mean of n ≥ 3 per cell line. (D, G, and H), two-tailed unpaired t test; P values indicated; (B, D, and G), error bars, SD. CM-H2DCFDA,; DHA, dihydroartemisinin; DIC, differential interference contrast; EXP1, exported protein 1; FC, flow cytometry; GST, glutathione S-transferase; IC50, half maximal inhibitory concentration; RPMI, Roswell Park Memorial Institute; RSA, ring-stage survival assay; wt, wild type. https://doi.org/10.1371/journal.pbio.3000473.g004 Loss of EXP1 is not associated with elevated oxidative stress If EXP1 is a GST that protects from heme-induced oxidative damage, loss of its activity should lead to increase oxidative stress in the parasite [16]. To test whether EXP1, irrespective of our complementation data, may act as a GST in the parasite, we measured the oxidative stress status in ΔEXP1 and control age-matched trophozoites using the fluorescent reporter chloromethyl dihydrochloro fluorescein (CM2-DCFDA) [26] (Fig 4C) to quantify intracellular reactive oxygen species (ROSs). Interestingly, the levels of fluorescence in ΔEXP1 parasites were not significantly different from those of control parasites (Fig 4D and S4A and S4B Fig), suggesting that ΔEXP1 parasites are not under elevated oxidative stress. Consistently, ΔEXP1 parasites were not rescued in the presence of antioxidants (Trolox, ascorbic acid) and glutathione precursors (N-acetylcysteine and cysteine) (Fig 4E). Growth of parasites complemented with limiting amounts of EXP1 (EXP1wtlow) was also not ameliorated in the presence of any of the supplements (Fig 4E). Overall, these data indicate that the growth defect of ΔEXP1 parasites is not caused by elevated oxidative stress and reduction of oxidative damage does not rescue loss of EXP1. The oxidative insult generated by hemoglobin byproducts can be diminished by inhibiting hemoglobin digestion with the cysteine protease inhibitor E64 [26, 27]. If EXP1 is involved in protecting the parasites from heme-mediated oxidative damage as proposed [16], growth of ΔEXP1 parasites should improve after inhibiting hemoglobin degradation. To test this, we treated ring stages expressing limiting amounts of EXP1 (EXP1wtlow) with E64 (before start of hemoglobin ingestion) and removed the inhibitor 12 hours later. Efficient inhibition of hemoglobin degradation was evident by swollen food vacuoles [28] (Fig 4F, arrowheads). However, E64 treatment did not restore growth of these parasites (Fig 4F). To confirm that E64 can protect against oxidative stress, we treated these parasites with dihydroartemisinin (DHA). While E64 ameliorated the effect of DHA as previously reported [29], E64 did not improve the growth defect of parasites expressing limiting levels of EXP1 (Fig 4G). Hence, the growth of parasites with reduced levels of EXP1 was not ameliorated by lower levels of hemoglobin-byproduct–induced oxidative stress, indicating that the proposed detoxification of hemoglobin metabolites is not a major function of EXP1. EXP1 does not influence ART resistance EXP1 was proposed to be involved in ART resistance by conjugating ART to GSH and thereby lowering oxidative damage. ART resistance was also associated with up-regulation of EXP1 [16]. We exploited our conditional ΔEXP1 parasites to determine whether the expression levels of EXP1 influence the susceptibility of the parasites to DHA. The levels of EXP1 in the parasites used for these experiments ranged from very low and growth limiting (EXP1wtlow in ΔEXP1 parasites) to overexpression (EXP1wthigh on top of the endogenous EXP1-HA) (Fig 3B–3D). Determination of the half maximal inhibitory concentration (IC50) for DHA showed no significant difference between these cell lines after removal of the endogenous EXP1 (Fig 4H). Naturally occurring ART resistance can only be measured using ring-stage survival assays (RSAs) [30]. To detect lower levels of resistance, we used a lower than usual concentration of DHA (350 nM). While a previously established DHA-resistant line [20] displayed reduced susceptibility to DHA in the RSA, the parasites expressing higher levels of EXP1 than wild type displayed no better survival than parasites expressing limiting levels of EXP1 or the GST catalytic site mutant EXP1 (Fig 4I), indicating that EXP1 levels did not affect ART responsiveness. The nutrient-permeable channel activity but not PTEX function is defective in ΔEXP1 parasites As the previously postulated function as a heme-detoxifying GST did not appear to be responsible for the phenotype in ΔEXP1 parasites, we looked for other possible functions of EXP1. In previous work, we identified EXP1 in immunoprecipitations (IPs) of EXP2 [31]. Therefore, EXP1 might be involved in functions attributed to EXP2, either protein export or the nutrient-permeable channel activity at the PVM. First, we tested whether ablation of EXP1 affects protein export. ΔEXP1 parasites showed no defect in the export of skeleton binding protein 1 (SBP1), ring exported protein 1 (REX1), REX2 (early-stage exported), and knob-associated histidine-rich protein (KAHRP) and MSP7-related protein 6 (MSRP6) (late-stage exported) (Fig 5A and 5B and S4C Fig). Thus, PTEX is still functional in ΔEXP1 parasites. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. PVM nutrient-permeable channel but not protein export or PSAC is impaired in ΔEXP1 parasites. (A) IFA images of control and ΔEXP1 parasites (rapalog) probed with α-HA (EXP1*-HA) and α-KAHRP. Graph to the right shows quantification of export (n = 20 cells). See S5C Fig for other exported proteins. (B) Live cell imaging of control and ΔEXP1 parasites (rapalog) expressing SBP1-mScarlet. Graph to the right shows quantification of export (n = 20 cells). (C) Live cell images of a ΔEXP1 trophozoite expressing SP-mScarlet liberated from its host RBC. (D) Left, current recorded from liberated control and ΔEXP1 parasites (rapalog) at 30 mV applied potential to the pipette electrode. Scale bar shows time in seconds and current in pA. The dotted line indicates a current reference level. At 30 mV, the PVM channels have an open probably of about one-half, therefore the flicker is offset in the control example as multiple channels are in the recording. The shown recordings are representative of the experiments done in each condition and show 1-second details from longer recordings to resolve the typical channel flicker in print. Right, probability of detecting at least one PVM channel per sealed patch (fchan) in ΔEXP1 parasites (rapalog, n = 14) and controls (n = 12). Fischer's exact test was used to estimate P value. Error bars indicate SD. In (A) and (B), nuclei were stained with DAPI; scale bars: 5 μm. DIC, differential interference contrast; EXP1, exported protein 1; HA, triple hemagglutinin tag; KAHRP, knob-associated histidine-rich protein; pA, Picoampere; PSAC, parasite surface anion channel; PVM, parasitophorous vacuolar membrane; RBC, red blood cell; SBP1, skeleton binding protein 1. https://doi.org/10.1371/journal.pbio.3000473.g005 To test whether EXP1 affects the nutrient-permeable channel activity at the PVM, patch-clamp measurements were performed on the PVM of ΔEXP1 parasites liberated from their host cell. After liberation, the PVM remained intact, as evidenced by the retention of a co-expressed soluble PV marker (Fig 5C and S1I Fig). The PVM of the liberated parasites was now accessible to a patch-clamp pipette. After giga-seal formation, each individual sample was inspected for channel activity, defined as a current flicker from closing channels at 30 mV applied voltage to the patch pipette [7]. While channel activity was often immediately apparent in the control sample, channel activity was mostly absent in the ΔEXP1 parasites (Fig 5D left). The frequency to detect at least one channel at the PVM (fchan) of ΔEXP1 parasites was significantly reduced compared to controls (Fig 5D right). Together, these results demonstrate that EXP1 is important for the nutrient-permeable channel activity at the PVM but not for protein export. Loss of EXP1 alters the distribution of EXP2, and both proteins interact at the PVM To investigate the effect of EXP1 loss on EXP2 localization as a possible reason for the reduction of nutrient-permeable channel activity, we expressed a green fluorescent protein (GFP) fusion of EXP2 (EXP2GFP) in condΔEXP1 parasites. No differences were obvious between controls and ΔEXP1 ring stages (S5A and S5B Fig). However, at the beginning of the trophozoite stage, EXP2-GFP displayed a profoundly altered distribution in 75% of the ΔEXP1 cells (Fig 6A–6C) as evident by the concentration of EXP2 in small regions of the PVM, frequently in loop-like Bodipy-TR-ceramide positive protrusions (Fig 6A and 6C). In control cells, EXP2-GFP was predominantly found in a circular pattern around the parasite (Fig 6A–6C). IFAs with specific antibodies confirmed this phenotype and showed an altered distribution of the endogenous EXP2 in ΔEXP1 parasites (Fig 6D and 6E and S5B Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. Localization of EXP2 and PVM proteins in ΔEXP1 parasites and interaction analysis. (A) Live cell images of ΔEXP1 (rapalog) and control trophozoites expressing EXP2-GFP. Light blue arrowhead, loop-like protrusions. (B) Quantification of the phenotype of the cells from (A). Green, signal around the parasite; black, aberrant distribution of signal. Mean of a total of n = 109 control and n = 99 ΔEXP1 cells derived from 4 biological replicas. (C) Live cell images of Bodipy-TR-ceramide labelled ΔEXP1 (rapalog) and control parasites expressing EXP2-GFP. (D) IFA images of ΔEXP1 (rapalog) and control parasites probed with α-HA to detect EXP1*-HA and α-EXP2 for endogenous EXP2. (E) Quantification of the phenotype of the cells from (D). Red, signal around the parasite; black, aberrant distribution of signal. Mean of a total of N = 108 control and N = 135 ΔEXP1 cells derived from 3 biological replicas. (F–H) IFA images of ΔEXP1 (rapalog) and control ring stages (F) or trophozoites (G, H) probed with α-HA (EXP1*-HA) and α-ETRAMP10.1 (F), α-GFP (EXP2-GFP) with α-ETRAMP4 (G), or with α-ETRAMP5 (H). (I) IFA images of ΔEXP2 (rapalog) and control trophozoites expressing EXP1-Ty probed with α-Ty and α-HA to detect EXP1-Ty and EXP2-HA, respectively. (J) Quantification of phenotypes of the parasites shown in (I). Red, signal around the parasite; black, aberrant distribution of signal. Mean of a total of n = 86 control and n = 110 ΔEXP1 cells derived from 3 biological replicas. (K) IFA images of ΔEXP2 (rapalog) and control parasites probed with α-HA and α-ETRAMP5. In (I) and (K), α-HA detects full (control) or truncated (rapalog) EXP2-HA. In (A, C, D, F–I, K), scale bar: 5 μm. DAPI, nuclei. (l) Western blot of a co-IP experiment in the cell line condΔEXP1 expressing EXP2-GFP (IP of EXP1*-HA with α-HA). α-HA detects EXP1*-HA (monomer: asterisk; dimer: double asterisk); α-GFP, EXP2-GFP (arrowhead); α-SERP, soluble PV protein; α-ETRAMP4, integral PVM protein; α-aldolase, cytosolic parasite protein. Input (I): total lysate before IP; post IP lysate (PI); Eluate (E). One representative of n = 3 independent biological replicas. In (B, E, and J), P values were calculated with a Fischer’s exact test. P < 0.05, significant. ETRAMP, early transcribed membrane protein; EXP1, exported protein 1; GFP, green fluorescent protein; HA, triple hemagglutinin tag; IFA, immunofluorescence assay; IP, immunoprecipitation; PV, parsitophorous vacuole; PVM, parasitophorous vacuolar membrane; SERP, serine-rich antigen also known as serine repeat antigen 5 (SERA5). https://doi.org/10.1371/journal.pbio.3000473.g006 The distribution of other PVM proteins such as early (ETRAMP10.1) or later (ETRAMP4) integral PVM markers [10] showed no apparent differences in ΔEXP1 parasites compared to controls (Fig 6F and 6G). In contrast, ETRAMP5 accumulated in small regions of the PVM where it colocalized with the aberrantly distributed EXP2-GFP (Fig 6H). Interestingly, ETRAMP5 was previously found in co-IPs as a potential interaction partner of EXP2 [31]. However, in contrast to EXP1, ETRAMP5 was dispensable for parasite growth (S6 Fig) and may not be needed for an essential process such as the nutrient-permeable channel activity at the PVM. As EXP1 influenced the distribution of EXP2, we tested whether this effect was reciprocal by knocking out EXP2. For this, we used the same strategy as for EXP1 to generate a conditional EXP2 KO (S7 Fig). As previously published, KO of EXP2 led to loss of protein export and arrested development at the trophozoite stage [7,32]. However, neither the distribution of EXP1-Ty expressed in these parasites nor that of endogenous ETRAMP5 was markedly altered in ΔEXP2 parasites (Fig 6I–6K). The minor effect observed on the distribution of EXP1 likely is due to the impact on parasite morphology in the EXP2 KO. Hence, the correct localization of EXP1 does not appear to depend on EXP2. Prompted by the altered distribution of EXP2 in the ΔEXP1 parasites, we performed co-IPs in crosslinked parasites to test whether the two proteins interact. IP of the endogenously HA-tagged EXP1 in condΔEXP1 parasites expressing EXP2-GFP resulted in copurification of EXP2-GFP (Fig 6L), while a soluble PV protein (serine-rich antigen 5 [SERA5]) and an integral PVM protein (ETRAMP4) were not co-immunoprecipitated. The reciprocal experiment by immunoprecipitating EXP2-GFP corroborated these findings (S8A Fig). Thus, EXP1 interacts with EXP2 at the PVM, suggesting that their activity may be linked. We also found EXP1 and EXP2 in structures within merozoites (S8B Fig), in agreement with previous results showing that both proteins are stored in dense granules [33, 34]. Finally, we tested whether loss of EXP1 affected membrane association of EXP2. In the EXP1 KO parasites, EXP2 was still membrane associated as indicated by retention of the protein after lysing parasites with saponin (S8C Fig). Next, we carried out more detailed tests using carbonate and urea to solubilize peripheral membrane proteins. Previous work showed that a small fraction of EXP2 can be extracted with carbonate [34]. This property was not changed after removal of EXP1 (S8D Fig). While the extractability of EXP2 by urea appeared to be increased in the ΔEXP1 parasites compared to the control, this was not significant (S8D Fig). This indicated that loss of EXP1 does not profoundly alter the membrane association of EXP2. Levels of EXP1 influence the capacity of parasites to respond to amino acid starvation Due to its additional role in protein export, knocking out EXP2 precludes specific analysis of the PVM nutrient-permeable channel function. In contrast, EXP1 loss only affected this activity (Fig 5D). In agreement with a role in nutrient uptake, the phenotype observed in the ΔEXP1 parasites (condensed trophozoites, growth retardation, and blebbing) resembled starvation phenotypes caused by amino acid depletion [35] and by the loss of the parasite surface anion channel (PSAC) [36,37], the activity at the RBC membrane for uptake of nutrients from the serum [38]. To first confirm that the phenotype of ΔEXP1 parasites was solely due to loss of the channel activity at the PVM, not at the RBC membrane, we assessed the uptake of 5-aminolevulinic acid (5-ALA) [39] into ΔEXP1 infected RBCs, an indicator for PSAC activity. Uptake of 5-ALA into RBC infected with ΔEXP1 trophozoites was not significantly different from controls (Fig 7A and 7B). The small trend for reduced uptake was likely due to the growth retardation of ΔEXP1 parasites (S8E Fig). Thus, PSAC activity is not impaired in ΔEXP1 parasites, and the starvation-like phenotype is caused by the loss of the nutrient permeability at the PVM. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 7. Reduced levels of EXP1 mimic starvation and lead to low nutrient hypersensitivity. (A) Live cell images of control and ΔEXP1 parasites (rapalog) incubated with 5-ALA. (B) Quantification of PPIX fluorescence in control and ΔEXP1 parasites (rapalog) using FC. Mean (green line) of n = 11 independent experiments. Error bars indicate SD. (C) Stage distribution of tightly synchronous parasite cell lines 24 h.p.i. (after an initial cycle ± rapalog) grown in amino acid–limited and complete medium. Mean of n = 3 independent experiments. (D) Percentage of rings at different time points post invasion (after 1 cycle ± rapalog) of condΔEXP1 and EXP1wtlow parasites grown in amino acid–limited and complete medium. Mean of n = 3 independent experiments. (E) Growth on day 4 (after 2 cycles) of the indicated parasite lines and condition (±rapalog) in the presence of azide in proportion to growth of the same parasites in medium without azide (left, NaN3 versus no NaN3) or growth in amino acid–limited medium in proportion to the same parasites grown in complete medium (right, limiting versus complete medium). Growth of the control culture (medium without NaN3 or complete medium) was set as 100%. Rapalog was added 1 cycle prior to the growth test to start with the corresponding KO parasites. Green line indicates mean of at least n = 5 independent experiments. In (B, C, and E), two-tailed unpaired t test, P values are indicated. 5-ALA, 5-aminolevulinic acid; DIC, differential interference contrast; EXP1, exported protein 1; FC, flow cytometry; h.p.i., hours post invasion; KO, knockout; PPIX, protoporphyrin IX; rapa, rapalog; wt, wild type. https://doi.org/10.1371/journal.pbio.3000473.g007 To more specifically test the association of EXP1 with nutrient acquisition, we compared parasite growth in medium containing limiting concentrations of amino acids with parasites grown in complete medium in cell lines expressing different levels of EXP1. Interestingly, growth in limited medium mimicked the prolonged ring-phase phenotype observed in ΔEXP1 parasites (Fig 7C and 7D). Furthermore, while all cell lines expressing physiological (or higher) levels of EXP1 tolerated the limiting medium over 2 growth cycles to a similar extent, the cell line expressing limiting levels (EXP1low on rapalog) was hypersensitive to low levels of amino acids (Fig 7E). This hypersensitivity was specifically related to the lack of amino acids because its response to an unrelated growth inhibition (using sodium azide) was similar to the other cell lines (Fig 7E). We conclude that EXP1 is critical for nutrient acquisition across the PVM and that the presence of the nutrient-permeable channel detected by patch clamping at the PVM correlates with nutrient uptake. Parasites relying only on EXP1 with the mutated catalytic site showed an intermediate sensitivity to the limited medium, indicating that the growth reduction in these parasites (Fig 4B) was also due to reduced nutrient acquisition with this version of EXP1, although there was no significant difference to the line expressing both, the mutated and the wild-type form of EXP1 (Fig 7E). EXP1 is essential for parasite development in RBCs To first test whether EXP1 is required for the survival of P. falciparum erythrocytic stages, we generated a conditional exp1 KO based on the Dimerizable Cre (DiCre) system [18,19, 20] using selection-linked integration (SLI) [20]. The endogenous exp1 was disrupted before the region encoding the TM domain, and at the same time, a second functional copy of exp1 flanked by loxP sites was introduced in the exp1 locus. The new floxed copy of EXP1 is hemagglutinin (HA) tagged and can be conditionally excised by DiCre upon addition of rapalog (Fig 1A and S1A and S1B Fig). Immunofluorescence assays (IFAs) showed that the corresponding cell line (condΔEXP1) correctly expressed the functional EXP1-HA in the PVM (Fig 1B). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. EXP1 is essential for blood stage development. (A) Simplified schematic of DiCre-based conditional exp1 KO using SLI. Arrows indicate primers P1 and P2 (see S1 Fig for details). (B) IFA of compound 2-arrested condΔEXP1 late-stage schizonts show localization of EXP1*-HA at the PVM. α-MSP1 (MSP1) labels the PPM. Nuclei were stained with DAPI; scale bars, 5 μm. (C) Strategy for depletion of EXP1 from the PVM of synchronized condΔEXP1 ring stages divided into a culture with rapalog (“rap”) and one without rapalog (“control”). Top: schematic: green boxes and green line signify PVM with EXP1. Mid: PCR with primers P1 and P2 from gDNA 24 and 48 hours after addition of rapalog. Original: band for intact exp1 locus (1,919 bp); excised: band after excision of exp1* (1,326 bp). Bottom: western blot probed with α-HA to detect EXP1*-HA and α-BIP as loading control. Asterisk: unskipped (first T2A) product that is present before excision and hence has no impact on parasites. (D) Quantification of levels of wtEXP1-HA in the first ("24 h") and second ("48 h") cycle after addition of rapalog based on densitometric analysis of anti-HA immunoblots (mean of n = 2 independent experiments) of which one is shown in (C). HA signal was normalized to BIP. Error bars indicate SD. (E) FC growth curves of synchronous ring stage condΔEXP1 parasites starting in cycle 2 (ΔEXP1 parasites) as shown in (C). One representative of n = 5 independent experiments. (F) Stage distribution of condΔEXP1 parasites in Giemsa smears of synchronous parasites grown with (+) and one without (−) rapalog (“rap”) at different time points (average time post invasion) after adding rapalog. Light blue arrowheads show blebs. One of n = 4 independent experiments is shown. Blue arrow in (C–F) indicates start of a new cycle without EXP1. 2A, T2A skip peptide; BIP, binding immunoglobulin protein; DIC, differential interference contrast; DiCre, Dimerizable Cre; EXP1, exported protein 1; EXP1*, recodonized exp1; FC, flow cytometry; gDNA, genomic DNA; HA, triple hemagglutinin tag; KO, knockout; MSP1, merozoite surface protein 1; PPM, parasite plasma membrane; PVM, parasitophorous vacuolar membrane; SLI, selection-linked integration; SP, signal peptide; TM, transmembrane domain; wt, wild-type. https://doi.org/10.1371/journal.pbio.3000473.g001 To investigate the effect of the loss of EXP1 on parasite survival, synchronous condΔEXP1 ring-stage parasites were grown in the presence of rapalog, alongside a control culture (Fig 1C). PCR confirmed efficient excision of the functional copy of exp1 within one growth cycle (48 hours) upon addition of rapalog (Fig 1C). No growth defect was observed in this first cycle during which the exp1 gene was excised (S1C Fig). In this cycle, EXP1 protein levels remained at approximately 70% due to protein expressed in the PVM before excision was complete (Fig 1C and 1D and S1D Fig). Western blot confirmed loss of EXP1 (approximately 5% residual protein detected at 48 h; Fig 1D) in parasites (henceforth termed ΔEXP1 parasites) after invasion and start of a new cycle (Fig 1C, blue arrow). The ΔEXP1 parasites failed to replicate (Fig 1E and S1C Fig), demonstrating that EXP1 is essential for propagation in RBC. The N-terminal fragment remaining after excision was nonfunctional as it does not rescue growth and was only detectable by IFA (S1D Fig), likely because of its small size and possible low stability. Giemsa smears taken in regular intervals from synchronous parasites showed that ΔEXP1 ring stages were much slower to reach the trophozoite stage than controls (Fig 1F). ΔEXP1 trophozoites often displayed protrusions reaching into the host cell cytoplasm (“blebs,” light blue arrowheads, Fig 1F) and frequently had an aberrant condensed morphology (Fig 1F). ΔEXP1 trophozoites did not complete schizogony as evident by a significantly reduced number of nuclei per ΔEXP1 parasite compared to controls (S1E Fig) and by an almost complete absence of new rings in the next cycle (Fig 1F). To test whether very slowly growing ΔEXP1 parasites persisted, we carried out an extended growth assay with the ΔEXP1 parasites. These experiments revealed a resurfacing of parasites in the ΔEXP1 culture grown in the presence of rapalog 9 days after loss of EXP1, but PCR identified this as a population with a nonexcised exp1 locus. These parasites were therefore breakthroughs, further indicating that loss of EXP1 abolishes parasite propagation in RBCs (S1F Fig). To observe the phenotype of EXP1 loss in more detail, we compared the development of ΔEXP1 parasites to controls, using long-term time-lapse imaging [21] (Fig 2A). This confirmed a severe delay of the ring stage (Fig 2B) and a very slow development of trophozoites without completion of the cycle (Fig 2A). Time-lapse imaging also revealed phenotypes in ring stages: ΔEXP1 parasites changed position in the host cell less frequently and rarely showed amoeboid shapes (Fig 2A and 2C), 2 typical features of ring stages [21] regularly observed in controls (Fig 2A and 2C). In addition, ΔEXP1 parasites were often found closely adjoined to the RBC membrane, a phenomenon that we here termed “hugging” (Fig 2A and 2C, red arrowheads), also evident by electron microscopy (S1G Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Morphological phenotypes of ΔEXP1 parasites. (A) Time-lapse imaging of condΔEXP1 parasites starting a cycle with (control) and without EXP1 (rapalog) imaged side by side. Single DIC z-sections of selected time points typical phenotypes are shown: top row: slow development; mid row: failure to reach full trophozoite stage; bottom row: cell showing extensive hugging. Arrowheads: white, ring stages after reinvasion; blue, amoeboid ring; light blue, blebs; red, hugging. (B) Number of hours (after start of the time lapse experiment) to reach young trophozoite stage in n = 24 control and N = 27 ΔEXP1 parasites. (C) Frequency of the indicated events in individual ring stages observed by time lapse microscopy (presence or absence of event was scored every hour); n = 25 parasites each for ΔEXP1 and control. (D, E) Left, live cell images of control and ΔEXP1 (rapalog) parasites expressing Lyn-mCherry (PPM marker) (D) or labelled with Bodipy TR ceramide (BODIPY) (E). Light blue arrowheads, blebs; yellow arrowheads, TVN. Graphs: quantification of number of blebs per cell (controls N = 25, 73 for (D) and (E), respectively; ΔEXP1 (rapalog) n = 28 and 156 for (D) and (E), respectively. (F) Left, IFA of control and ΔEXP1 (rapalog) late-stage gametocytes using α-HA to detect EXP1*-HA and α-Pfg377 (late-stage gametocyte marker). Right, PCR using primers P1 and P2 (Fig 1A) confirms excision of exp1 in late gametocytes. (G) Left, gametocytemia of control and ΔEXP1 parasites early after induction (Pfs16 positive cells) and 8 days later (Pfg377 positive cells) based on IFAs. Right, fold reduction in the number of early (Pfs16) and late (Pfg377) ΔEXP1 gametocytes versus control; n = 3. (D, E, and F); scale bars, 5 μm. In (F), nuclei were stained with DAPI. In (B–E and G), green lines indicate mean and error bars (SD); two-tailed unpaired t test, P values indicated. DIC, differential interference contrast; EXP1, exported protein 1; HA, triple hemagglutinin tag; IFA, immunofluorescence assay; PPM, parasite plasma membrane. https://doi.org/10.1371/journal.pbio.3000473.g002 The protrusions observed in Giemsa smears were also visible in differential interference contrast (DIC) and also in ΔEXP1 rings (Fig 2A, light blue arrowheads). These “blebs” were bounded by PPM and present almost exclusively in ΔEXP1 parasites as confirmed with a parasite plasma membrane (PPM) marker (Lyn-mCherry) [22] (Fig 2D). The total number of membrane-bounded protrusions detected using in Bodipy-TR-ceramide or Lyso-PC was also mildly enriched in ΔEXP1 parasites (Fig 2E and S1H Fig). Despite these morphological alterations, the PVM integrity in ΔEXP1 parasites was not compromised, because a soluble PV marker (SP-mScarlet) was retained in the PV (S1I Fig). Next, we evaluated the impact of EXP1 loss on gametocytogenesis. Interestingly, we detected stage III–V gametocytes lacking EXP1 (Fig 2F), although the number of late-stage gametocytes was reduced by more than 50% 8 days after induction (Fig 2G). Nevertheless, this indicated a lower effect of loss of EXP1 on the development of gametocytes than on asexual blood stages. All regions of EXP1 are important for its function To confirm that the observed growth phenotype is specific for EXP1 loss, we complemented the condΔEXP1 parasites with a Ty-tagged full-length copy of EXP1 (EXP1wt-Ty) expressed under the constitutive nmd3 promoter. EXP1wt-Ty was correctly located at the PVM (Fig 3A) and, after removal of the endogenous EXP1, restored parasite growth to 80% of the unexcised control (Fig 3B and S2A–S2C Fig). The level of complementation correlated with the level of expression of the EXP1wt-Ty construct, as demonstrated using promoters driving different levels of expression (Fig 3B–3D and S2A–S2C Fig). The complemented parasites also showed a similar duration of the ring stage compared to the wild type, indicating that the delay to reach the trophozoite was reverted by the complementation (S2D Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Complementation pinpoints important regions of EXP1. (A) IFA images of condΔEXP1 schizont stages expressing EXP1 wt-Ty probed with α-HA and α-Ty show localization of EXP1*-HA and EXP1 wt-Ty in the PVM. α-MSP1 (MSP1) labels the PPM. DAPI, nuclei; scale bars, 5 μm. Right: immunoblot of extracts of cell line on the left probed with α-Ty to detect EXP1wt-Ty and α-SBP1 as control for a TM protein. Saponin was used to separate the parasite pellet (“P”) from the supernatant (“SN”) containing PV and host cell content. See S3 Fig for IFAs and immunoblots of all complementation constructs. (B) Relative activity of the EXP1 complementation constructs indicated. Except where otherwise indicated, constructs were expressed under the nmd3 (mid) promoter. The complementation capacity of every tested construct was calculated as relative activity to the EXP1wt-Ty construct under the nmd3 promoter, which was set as 100% (right dotted green line). ΔEXP1 was set to 0% (left dotted green line). Each data point (red dot) shows growth of rapalog-treated versus unexcised parasites at the end of a 5-day growth assay relative to the growth of the wt construct; n ≥ 4 independent experiments per cell line. Error bars: SD. See S2 Fig for activity of all complementation constructs. (C) Immunoblot of lysates of condΔEXP1 parasites expressing EXP1wtlow, EXP1mid, and EXP1high probed with α-Ty (EXP1wt-Ty), α-HA (EXP1*-HA), and α-BIP (loading control). (D) Densitometric analysis of EXP1wt expression levels (C) under low, mid, and high promoters relative to the mid promoter (green). Mean of 3 independent experiments. Error bars: SD. (E) Immunoblot of extracts of +/−formaldehyde (PFA) treated cell lines expressing the indicated constructs probed with α-Ty. Single asterisk: monomer; double asterisk: dimer. BIP, binding immunoglobulin protein; DIC, differential interference contrast; EXP1, exported protein 1; HA, triple hemagglutinin tag; IFA, immunofluorescence assay; MSP1, merozoite surface protein 1; PPM, parasite plasma membrane; PV, parasitophorous vacuole; PVM, parasitophorous vacuolar membrane; SBP1, skeleton binding protein 1; TM, transmembrane; wt, wild type. https://doi.org/10.1371/journal.pbio.3000473.g003 To pinpoint the functional regions in EXP1, we tested a series of modified Ty-tagged EXP1 constructs for their capacity to complement loss of endogenous EXP1. Except for EXP1wt-mScarlet, all constructs were correctly inserted into the PVM, and deletions or replacements in the N- or C-terminus mostly led to severe loss of function (S2A-S2C and S3A and S3B Fig). Interestingly, EXP1 of the rodent malaria parasite P. berghei (PbEXP1) only partially (46.1% ± 9.4% activity) rescued loss of EXP1 in P. falciparum (Fig 3B), indicating limited functional conservation between species. Deletion of the entire C-terminus of EXP1 reduced its activity to 59.37% ± 13.82% (Fig 3B). EXP1 lacking an 11 amino acid stretch (sequence SGVSSKKKNKK) in the N-terminus named E-domain (EXP1ΔED), a region proposed to be necessary for the dimerization based on similarity with MAPEGs [16,23], complemented only poorly (Fig 3B). Previous work indicated that EXP1 homo-oligomerizes [12]. However, loss of function of EXP1ΔED was not due to profound alterations in its capacity to oligomerize, as dimers were still detectable after formaldehyde crosslinking (Fig 3E). We noticed that the TM domains of integral PVM proteins such as EXP1 and ETRAMPs in different malaria species are particularly rich in G, S, and T residues (S2E Fig). G, S, and T are known to be important as TM interaction interfaces [24, 25]. Strikingly, point mutations of the first glycine residues (from G to L) of the two GXXG motifs of the P. falciparum EXP1 (PfEXP1) TM abolished EXP1 function (Fig 3B) although the protein was still correctly trafficked (S3A and S3B Fig) and capable of dimerizing (Fig 3E). Collectively, we conclude that all parts of EXP1 are required for its function. EXP1 GST activity is dispensable for parasite growth Previous work showed that recombinant EXP1 conjugates hematin to GSH in vitro and thereby may protect the parasite from heme-induced oxidative damage [16]. Based on homology with other MAPEGs, it was postulated that the catalytic center of the GST activity of EXP1 resides in 3 N-terminal residues (Fig 4A). Mutation of one of these residues (R70) led to a reduced enzymatic activity in vitro [16]. We used our complementation approach to assess the importance of these residues (and hence of the proposed GST activity) for parasite development. Constructs with mutations of 1 (EXP1R70A) or all 3 residues (EXP1-3xmut) of the proposed catalytic site showed 69.0% ± 9.2% and 62.5% ± 16.2% complementation activity, respectively (Fig 4B), indicating that these mutations had only a moderate effect on EXP1 function compared to most other complementation constructs (Fig 3B and S2A Fig). IFAs and solubility assays showed correct targeting of the complementation constructs to the PVM (S3A and S3B Fig). EXP1 R70A expressed under the stronger promoter resulted in nearly identical complementation capacity to the wild-type construct (Fig 4B). These data indicate that if EXP1 is a MAPEG, its GST activity plays only a minor role for growth of blood stage parasites. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Dispensability of EXP1 GST activity and lack of oxidative stress in ΔEXP1 parasites. (A) Schematic of the region of EXP1 with the proposed catalytic site of the GST activity and of the mutations introduced. (B) Relative activity of the EXP1 complementation constructs indicated. Green lines: activity of EXP1wt (nmd3, mid) (set as 100%) and absence of activity (ΔEXP1) set as 0%; n ≥ 4 independent experiments per cell line. (C) Live cell images of condΔEXP1 parasites incubated with CM-H2DCFDA. Scale bars: 5 μm. (D) Fluorescence intensity of matching stages of control and ΔEXP1 parasites (rapalog) incubated with CM-H2DCFDA (C). Results from 3 independent experiments with a total of n = 55 control and n = 53 ΔEXP1 cells. Green line, mean; error bars, SD. (E) FC growth curves of synchronous condΔEXP1 parasites and the complementation cell line EXP1 wtlow after one cycle with (red) and without (black) rapalog (see Fig 1D) grown in RPMI alone or supplemented with the compounds indicated. One representative of n = 3 independent biological replicas. (F) Left, growth curves of EXP1wtlow parasites ± E64 starting after one growth cycle ± rapalog (see Fig 1D). Right, Giemsa smears of parasites on day 2 (after incubation with E64) and day 3 (after removal). Arrowheads: swollen food vacuole. One representative of n = 3 biological replicas. (G) Effect of E64 treatment on survival of DHA and rapalog-treated EXP1wtlow parasites versus untreated. Mean of n = 3 independent experiments. (H) Left, dose-response curves of the parasite lines indicated treated with DHA (0–50 nM). Right, DHA IC50 values for these cell lines ± rapalog. Mean of n ≥ 3 experiments. (I) RSAs of the indicated complementation cell lines and a Kelch13 C580Y mutant line. Data points are percent survival of DHA treated versus untreated parasites ± rapalog. Mean of n ≥ 3 per cell line. (D, G, and H), two-tailed unpaired t test; P values indicated; (B, D, and G), error bars, SD. CM-H2DCFDA,; DHA, dihydroartemisinin; DIC, differential interference contrast; EXP1, exported protein 1; FC, flow cytometry; GST, glutathione S-transferase; IC50, half maximal inhibitory concentration; RPMI, Roswell Park Memorial Institute; RSA, ring-stage survival assay; wt, wild type. https://doi.org/10.1371/journal.pbio.3000473.g004 Loss of EXP1 is not associated with elevated oxidative stress If EXP1 is a GST that protects from heme-induced oxidative damage, loss of its activity should lead to increase oxidative stress in the parasite [16]. To test whether EXP1, irrespective of our complementation data, may act as a GST in the parasite, we measured the oxidative stress status in ΔEXP1 and control age-matched trophozoites using the fluorescent reporter chloromethyl dihydrochloro fluorescein (CM2-DCFDA) [26] (Fig 4C) to quantify intracellular reactive oxygen species (ROSs). Interestingly, the levels of fluorescence in ΔEXP1 parasites were not significantly different from those of control parasites (Fig 4D and S4A and S4B Fig), suggesting that ΔEXP1 parasites are not under elevated oxidative stress. Consistently, ΔEXP1 parasites were not rescued in the presence of antioxidants (Trolox, ascorbic acid) and glutathione precursors (N-acetylcysteine and cysteine) (Fig 4E). Growth of parasites complemented with limiting amounts of EXP1 (EXP1wtlow) was also not ameliorated in the presence of any of the supplements (Fig 4E). Overall, these data indicate that the growth defect of ΔEXP1 parasites is not caused by elevated oxidative stress and reduction of oxidative damage does not rescue loss of EXP1. The oxidative insult generated by hemoglobin byproducts can be diminished by inhibiting hemoglobin digestion with the cysteine protease inhibitor E64 [26, 27]. If EXP1 is involved in protecting the parasites from heme-mediated oxidative damage as proposed [16], growth of ΔEXP1 parasites should improve after inhibiting hemoglobin degradation. To test this, we treated ring stages expressing limiting amounts of EXP1 (EXP1wtlow) with E64 (before start of hemoglobin ingestion) and removed the inhibitor 12 hours later. Efficient inhibition of hemoglobin degradation was evident by swollen food vacuoles [28] (Fig 4F, arrowheads). However, E64 treatment did not restore growth of these parasites (Fig 4F). To confirm that E64 can protect against oxidative stress, we treated these parasites with dihydroartemisinin (DHA). While E64 ameliorated the effect of DHA as previously reported [29], E64 did not improve the growth defect of parasites expressing limiting levels of EXP1 (Fig 4G). Hence, the growth of parasites with reduced levels of EXP1 was not ameliorated by lower levels of hemoglobin-byproduct–induced oxidative stress, indicating that the proposed detoxification of hemoglobin metabolites is not a major function of EXP1. EXP1 does not influence ART resistance EXP1 was proposed to be involved in ART resistance by conjugating ART to GSH and thereby lowering oxidative damage. ART resistance was also associated with up-regulation of EXP1 [16]. We exploited our conditional ΔEXP1 parasites to determine whether the expression levels of EXP1 influence the susceptibility of the parasites to DHA. The levels of EXP1 in the parasites used for these experiments ranged from very low and growth limiting (EXP1wtlow in ΔEXP1 parasites) to overexpression (EXP1wthigh on top of the endogenous EXP1-HA) (Fig 3B–3D). Determination of the half maximal inhibitory concentration (IC50) for DHA showed no significant difference between these cell lines after removal of the endogenous EXP1 (Fig 4H). Naturally occurring ART resistance can only be measured using ring-stage survival assays (RSAs) [30]. To detect lower levels of resistance, we used a lower than usual concentration of DHA (350 nM). While a previously established DHA-resistant line [20] displayed reduced susceptibility to DHA in the RSA, the parasites expressing higher levels of EXP1 than wild type displayed no better survival than parasites expressing limiting levels of EXP1 or the GST catalytic site mutant EXP1 (Fig 4I), indicating that EXP1 levels did not affect ART responsiveness. The nutrient-permeable channel activity but not PTEX function is defective in ΔEXP1 parasites As the previously postulated function as a heme-detoxifying GST did not appear to be responsible for the phenotype in ΔEXP1 parasites, we looked for other possible functions of EXP1. In previous work, we identified EXP1 in immunoprecipitations (IPs) of EXP2 [31]. Therefore, EXP1 might be involved in functions attributed to EXP2, either protein export or the nutrient-permeable channel activity at the PVM. First, we tested whether ablation of EXP1 affects protein export. ΔEXP1 parasites showed no defect in the export of skeleton binding protein 1 (SBP1), ring exported protein 1 (REX1), REX2 (early-stage exported), and knob-associated histidine-rich protein (KAHRP) and MSP7-related protein 6 (MSRP6) (late-stage exported) (Fig 5A and 5B and S4C Fig). Thus, PTEX is still functional in ΔEXP1 parasites. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. PVM nutrient-permeable channel but not protein export or PSAC is impaired in ΔEXP1 parasites. (A) IFA images of control and ΔEXP1 parasites (rapalog) probed with α-HA (EXP1*-HA) and α-KAHRP. Graph to the right shows quantification of export (n = 20 cells). See S5C Fig for other exported proteins. (B) Live cell imaging of control and ΔEXP1 parasites (rapalog) expressing SBP1-mScarlet. Graph to the right shows quantification of export (n = 20 cells). (C) Live cell images of a ΔEXP1 trophozoite expressing SP-mScarlet liberated from its host RBC. (D) Left, current recorded from liberated control and ΔEXP1 parasites (rapalog) at 30 mV applied potential to the pipette electrode. Scale bar shows time in seconds and current in pA. The dotted line indicates a current reference level. At 30 mV, the PVM channels have an open probably of about one-half, therefore the flicker is offset in the control example as multiple channels are in the recording. The shown recordings are representative of the experiments done in each condition and show 1-second details from longer recordings to resolve the typical channel flicker in print. Right, probability of detecting at least one PVM channel per sealed patch (fchan) in ΔEXP1 parasites (rapalog, n = 14) and controls (n = 12). Fischer's exact test was used to estimate P value. Error bars indicate SD. In (A) and (B), nuclei were stained with DAPI; scale bars: 5 μm. DIC, differential interference contrast; EXP1, exported protein 1; HA, triple hemagglutinin tag; KAHRP, knob-associated histidine-rich protein; pA, Picoampere; PSAC, parasite surface anion channel; PVM, parasitophorous vacuolar membrane; RBC, red blood cell; SBP1, skeleton binding protein 1. https://doi.org/10.1371/journal.pbio.3000473.g005 To test whether EXP1 affects the nutrient-permeable channel activity at the PVM, patch-clamp measurements were performed on the PVM of ΔEXP1 parasites liberated from their host cell. After liberation, the PVM remained intact, as evidenced by the retention of a co-expressed soluble PV marker (Fig 5C and S1I Fig). The PVM of the liberated parasites was now accessible to a patch-clamp pipette. After giga-seal formation, each individual sample was inspected for channel activity, defined as a current flicker from closing channels at 30 mV applied voltage to the patch pipette [7]. While channel activity was often immediately apparent in the control sample, channel activity was mostly absent in the ΔEXP1 parasites (Fig 5D left). The frequency to detect at least one channel at the PVM (fchan) of ΔEXP1 parasites was significantly reduced compared to controls (Fig 5D right). Together, these results demonstrate that EXP1 is important for the nutrient-permeable channel activity at the PVM but not for protein export. Loss of EXP1 alters the distribution of EXP2, and both proteins interact at the PVM To investigate the effect of EXP1 loss on EXP2 localization as a possible reason for the reduction of nutrient-permeable channel activity, we expressed a green fluorescent protein (GFP) fusion of EXP2 (EXP2GFP) in condΔEXP1 parasites. No differences were obvious between controls and ΔEXP1 ring stages (S5A and S5B Fig). However, at the beginning of the trophozoite stage, EXP2-GFP displayed a profoundly altered distribution in 75% of the ΔEXP1 cells (Fig 6A–6C) as evident by the concentration of EXP2 in small regions of the PVM, frequently in loop-like Bodipy-TR-ceramide positive protrusions (Fig 6A and 6C). In control cells, EXP2-GFP was predominantly found in a circular pattern around the parasite (Fig 6A–6C). IFAs with specific antibodies confirmed this phenotype and showed an altered distribution of the endogenous EXP2 in ΔEXP1 parasites (Fig 6D and 6E and S5B Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. Localization of EXP2 and PVM proteins in ΔEXP1 parasites and interaction analysis. (A) Live cell images of ΔEXP1 (rapalog) and control trophozoites expressing EXP2-GFP. Light blue arrowhead, loop-like protrusions. (B) Quantification of the phenotype of the cells from (A). Green, signal around the parasite; black, aberrant distribution of signal. Mean of a total of n = 109 control and n = 99 ΔEXP1 cells derived from 4 biological replicas. (C) Live cell images of Bodipy-TR-ceramide labelled ΔEXP1 (rapalog) and control parasites expressing EXP2-GFP. (D) IFA images of ΔEXP1 (rapalog) and control parasites probed with α-HA to detect EXP1*-HA and α-EXP2 for endogenous EXP2. (E) Quantification of the phenotype of the cells from (D). Red, signal around the parasite; black, aberrant distribution of signal. Mean of a total of N = 108 control and N = 135 ΔEXP1 cells derived from 3 biological replicas. (F–H) IFA images of ΔEXP1 (rapalog) and control ring stages (F) or trophozoites (G, H) probed with α-HA (EXP1*-HA) and α-ETRAMP10.1 (F), α-GFP (EXP2-GFP) with α-ETRAMP4 (G), or with α-ETRAMP5 (H). (I) IFA images of ΔEXP2 (rapalog) and control trophozoites expressing EXP1-Ty probed with α-Ty and α-HA to detect EXP1-Ty and EXP2-HA, respectively. (J) Quantification of phenotypes of the parasites shown in (I). Red, signal around the parasite; black, aberrant distribution of signal. Mean of a total of n = 86 control and n = 110 ΔEXP1 cells derived from 3 biological replicas. (K) IFA images of ΔEXP2 (rapalog) and control parasites probed with α-HA and α-ETRAMP5. In (I) and (K), α-HA detects full (control) or truncated (rapalog) EXP2-HA. In (A, C, D, F–I, K), scale bar: 5 μm. DAPI, nuclei. (l) Western blot of a co-IP experiment in the cell line condΔEXP1 expressing EXP2-GFP (IP of EXP1*-HA with α-HA). α-HA detects EXP1*-HA (monomer: asterisk; dimer: double asterisk); α-GFP, EXP2-GFP (arrowhead); α-SERP, soluble PV protein; α-ETRAMP4, integral PVM protein; α-aldolase, cytosolic parasite protein. Input (I): total lysate before IP; post IP lysate (PI); Eluate (E). One representative of n = 3 independent biological replicas. In (B, E, and J), P values were calculated with a Fischer’s exact test. P < 0.05, significant. ETRAMP, early transcribed membrane protein; EXP1, exported protein 1; GFP, green fluorescent protein; HA, triple hemagglutinin tag; IFA, immunofluorescence assay; IP, immunoprecipitation; PV, parsitophorous vacuole; PVM, parasitophorous vacuolar membrane; SERP, serine-rich antigen also known as serine repeat antigen 5 (SERA5). https://doi.org/10.1371/journal.pbio.3000473.g006 The distribution of other PVM proteins such as early (ETRAMP10.1) or later (ETRAMP4) integral PVM markers [10] showed no apparent differences in ΔEXP1 parasites compared to controls (Fig 6F and 6G). In contrast, ETRAMP5 accumulated in small regions of the PVM where it colocalized with the aberrantly distributed EXP2-GFP (Fig 6H). Interestingly, ETRAMP5 was previously found in co-IPs as a potential interaction partner of EXP2 [31]. However, in contrast to EXP1, ETRAMP5 was dispensable for parasite growth (S6 Fig) and may not be needed for an essential process such as the nutrient-permeable channel activity at the PVM. As EXP1 influenced the distribution of EXP2, we tested whether this effect was reciprocal by knocking out EXP2. For this, we used the same strategy as for EXP1 to generate a conditional EXP2 KO (S7 Fig). As previously published, KO of EXP2 led to loss of protein export and arrested development at the trophozoite stage [7,32]. However, neither the distribution of EXP1-Ty expressed in these parasites nor that of endogenous ETRAMP5 was markedly altered in ΔEXP2 parasites (Fig 6I–6K). The minor effect observed on the distribution of EXP1 likely is due to the impact on parasite morphology in the EXP2 KO. Hence, the correct localization of EXP1 does not appear to depend on EXP2. Prompted by the altered distribution of EXP2 in the ΔEXP1 parasites, we performed co-IPs in crosslinked parasites to test whether the two proteins interact. IP of the endogenously HA-tagged EXP1 in condΔEXP1 parasites expressing EXP2-GFP resulted in copurification of EXP2-GFP (Fig 6L), while a soluble PV protein (serine-rich antigen 5 [SERA5]) and an integral PVM protein (ETRAMP4) were not co-immunoprecipitated. The reciprocal experiment by immunoprecipitating EXP2-GFP corroborated these findings (S8A Fig). Thus, EXP1 interacts with EXP2 at the PVM, suggesting that their activity may be linked. We also found EXP1 and EXP2 in structures within merozoites (S8B Fig), in agreement with previous results showing that both proteins are stored in dense granules [33, 34]. Finally, we tested whether loss of EXP1 affected membrane association of EXP2. In the EXP1 KO parasites, EXP2 was still membrane associated as indicated by retention of the protein after lysing parasites with saponin (S8C Fig). Next, we carried out more detailed tests using carbonate and urea to solubilize peripheral membrane proteins. Previous work showed that a small fraction of EXP2 can be extracted with carbonate [34]. This property was not changed after removal of EXP1 (S8D Fig). While the extractability of EXP2 by urea appeared to be increased in the ΔEXP1 parasites compared to the control, this was not significant (S8D Fig). This indicated that loss of EXP1 does not profoundly alter the membrane association of EXP2. Levels of EXP1 influence the capacity of parasites to respond to amino acid starvation Due to its additional role in protein export, knocking out EXP2 precludes specific analysis of the PVM nutrient-permeable channel function. In contrast, EXP1 loss only affected this activity (Fig 5D). In agreement with a role in nutrient uptake, the phenotype observed in the ΔEXP1 parasites (condensed trophozoites, growth retardation, and blebbing) resembled starvation phenotypes caused by amino acid depletion [35] and by the loss of the parasite surface anion channel (PSAC) [36,37], the activity at the RBC membrane for uptake of nutrients from the serum [38]. To first confirm that the phenotype of ΔEXP1 parasites was solely due to loss of the channel activity at the PVM, not at the RBC membrane, we assessed the uptake of 5-aminolevulinic acid (5-ALA) [39] into ΔEXP1 infected RBCs, an indicator for PSAC activity. Uptake of 5-ALA into RBC infected with ΔEXP1 trophozoites was not significantly different from controls (Fig 7A and 7B). The small trend for reduced uptake was likely due to the growth retardation of ΔEXP1 parasites (S8E Fig). Thus, PSAC activity is not impaired in ΔEXP1 parasites, and the starvation-like phenotype is caused by the loss of the nutrient permeability at the PVM. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 7. Reduced levels of EXP1 mimic starvation and lead to low nutrient hypersensitivity. (A) Live cell images of control and ΔEXP1 parasites (rapalog) incubated with 5-ALA. (B) Quantification of PPIX fluorescence in control and ΔEXP1 parasites (rapalog) using FC. Mean (green line) of n = 11 independent experiments. Error bars indicate SD. (C) Stage distribution of tightly synchronous parasite cell lines 24 h.p.i. (after an initial cycle ± rapalog) grown in amino acid–limited and complete medium. Mean of n = 3 independent experiments. (D) Percentage of rings at different time points post invasion (after 1 cycle ± rapalog) of condΔEXP1 and EXP1wtlow parasites grown in amino acid–limited and complete medium. Mean of n = 3 independent experiments. (E) Growth on day 4 (after 2 cycles) of the indicated parasite lines and condition (±rapalog) in the presence of azide in proportion to growth of the same parasites in medium without azide (left, NaN3 versus no NaN3) or growth in amino acid–limited medium in proportion to the same parasites grown in complete medium (right, limiting versus complete medium). Growth of the control culture (medium without NaN3 or complete medium) was set as 100%. Rapalog was added 1 cycle prior to the growth test to start with the corresponding KO parasites. Green line indicates mean of at least n = 5 independent experiments. In (B, C, and E), two-tailed unpaired t test, P values are indicated. 5-ALA, 5-aminolevulinic acid; DIC, differential interference contrast; EXP1, exported protein 1; FC, flow cytometry; h.p.i., hours post invasion; KO, knockout; PPIX, protoporphyrin IX; rapa, rapalog; wt, wild type. https://doi.org/10.1371/journal.pbio.3000473.g007 To more specifically test the association of EXP1 with nutrient acquisition, we compared parasite growth in medium containing limiting concentrations of amino acids with parasites grown in complete medium in cell lines expressing different levels of EXP1. Interestingly, growth in limited medium mimicked the prolonged ring-phase phenotype observed in ΔEXP1 parasites (Fig 7C and 7D). Furthermore, while all cell lines expressing physiological (or higher) levels of EXP1 tolerated the limiting medium over 2 growth cycles to a similar extent, the cell line expressing limiting levels (EXP1low on rapalog) was hypersensitive to low levels of amino acids (Fig 7E). This hypersensitivity was specifically related to the lack of amino acids because its response to an unrelated growth inhibition (using sodium azide) was similar to the other cell lines (Fig 7E). We conclude that EXP1 is critical for nutrient acquisition across the PVM and that the presence of the nutrient-permeable channel detected by patch clamping at the PVM correlates with nutrient uptake. Parasites relying only on EXP1 with the mutated catalytic site showed an intermediate sensitivity to the limited medium, indicating that the growth reduction in these parasites (Fig 4B) was also due to reduced nutrient acquisition with this version of EXP1, although there was no significant difference to the line expressing both, the mutated and the wild-type form of EXP1 (Fig 7E). Discussion Previous work led to the proposal that EXP1 is a GST of the MAPEG family that detoxifies hemoglobin byproducts and thereby protects malaria parasites from oxidative stress [16]. Furthermore, this work indicated that through this GST function, EXP1 can protect the parasite from the action of ART and that EXP1 transcription levels were elevated in ART-resistant parasites. Here, we show that the postulated GST activity of EXP1 is largely dispensable for parasite survival and that EXP1 protein levels did not influence ART susceptibility. Inhibition of hemoglobin catabolism by a protease inhibitor or the supplementation with reducing agents to protect from oxidative damage did not improve parasite growth when EXP1 was absent or its expression levels were growth limiting. Hence, although EXP1 may have GST activity in vitro, overall, our data indicate that this activity is not critical for blood stage growth. MAPEGs are a highly diverse and widely distributed family of proteins involved in the detoxification of metabolites and in glutathione and lipid metabolism [15]. It is possible that EXP1 derives from such proteins but has adopted other functions that are critical for parasite survival. We here provide functional evidence for a different role of EXP1 that is important for parasite survival. Apicomplexan parasites require an external supply of nutrients that reach the parasite by passive diffusion through a nonselective pore of the PVM [6]. Our patch-clamp experiments indicate that EXP1 is essential for the activity of a nutrient-permeable channel in the PVM. Recently, a different PVM protein, EXP2, was shown to be needed for this activity [7,9]. EXP2 oligomers also form the membrane-spanning pore of PTEX [3,8]. Hence, EXP2 is needed for both, the nutrient-permeable channel activity and for protein export, potentially as the pore of both activities [7]. In previous work, Gold and colleagues postulated that EXP2 is either part of a complex that mediates both protein transport and the nutrient-permeable channel function or that it is part of 2 molecularly distinct complexes, each serving only one of these functions [9]. Recent data showed that there is a large pool of EXP2 in trophozoites that is devoid of the PTEX component heat shock protein 101 (HSP101), lending support for the existence of 2 compositionally distinct EXP2 complexes [7]. However, targeting EXP2 cannot functionally distinguish these complexes. We here found that knocking out exp1 specifically affects the location of EXP2 in trophozoites and abolished the nutrient-permeable channel activity at the PVM but not protein export. We also show that EXP1 interacts with EXP2. Hence, EXP1 is a defining factor of this nutrient-permeable channel activity, likely through its interaction with EXP2 and by maintaining the correct localization of this protein. It is therefore likely that there are indeed 2 functionally and compositionally distinct pools of EXP2, one serving protein export (the PTEX complex) and one for the nutrient-permeable channel function (depending on EXP2 and EXP1). In further agreement with a PTEX-independent pool of EXP2, expression of EXP2 peaks in trophozoites, not in rings like the other PTEX components [7,34]. Interestingly, EXP1 has a very similar expression profile to EXP2 [40] (Fig 8A), congruent with a function together with the trophozoite expressed EXP2 in the nutrient-permeable channel activity (Fig 8B). Overall, this indicates that in rings, when protein export is a predominant requirement for the parasite, EXP2 exists foremost in PTEX. In trophozoites, when protein export is less important but nutrient acquisition for rapid growth is critical, most of EXP2 is needed together with EXP1 for nutrient uptake and derives from a pool expressed after production of PTEX. As the PTEX components (apart from EXP2) are not, or only poorly, expressed in trophozoites, residual PTEX complexes remaining from the ring stage are likely sufficient to accomplish the protein export needs in trophozoites. Hence, PTEX likely occupies only a minority of the EXP2 population in trophozoites (Fig 8B). This is supported by our finding that knocking out EXP1 specifically affected the location of EXP2 in trophozoites but not in rings. Overall, the current data support a model in which protein export and the nutrient-permeable channel function are accomplished by molecularly distinct complexes that share EXP2 and have reciprocal stage-specific expression peaks that temporally coincide with the needs of the respective parasite stages (Fig 8). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 8. Model of the EXP1-defined nutrient-permeable channel function of EXP2 distinct from the PTEX complex. (A) Transcript levels of EXP1, EXP2, and the PTEX components HSP101 and PTEX150 across the asexual intra-erythrocytic cycle. Values were obtained from a previous publication [40]. (B) Schematics of the PVM of wild-type (left) and ΔEXP1 parasites (right) during the ring (upper panel) and trophozoite stage (lower panel). Depicted molecules are explained in the box. Note that early expressed EXP2 is in light green and late expressed EXP2 is in dark green. In wild-type rings, EXP2 is predominantly associated with PTEX components to promote protein export, a major function of this stage. Less EXP2 is in the nutrient-permeable channel complex that depends on EXP1 but does not contain other PTEX components. The proportion of the abundance of the two EXP2 complexes is reversed in the trophozoite stage when most of protein export has been completed but rapid parasite growth demands more nutrients. PTEX150 and HSP101 are not expressed in this stage, but PTEX translocons left over from the ring stage make up a minor proportion of EXP2 complexes to maintain protein export activity. The nutrient-permeable channel function of EXP2 is negatively affected by loss of EXP1 either through indirect effects (indicated as aggregated EXP2, left part of PVM enlargement of ΔEXP1 KO) or a direct defect of the complex (“defective” nutrient-permeable channel, right part of ΔEXP1 KO). EXP1, exported protein 1; FKBPM, fragments per kilobase of exon model per million mapped reads; h.p.i., hours post invasion; HSP101, heat shock protein 101; KO, knockout; PTEX, Plasmodium translocon of exported proteins; PVM, parasitophorous vacuolar membrane. https://doi.org/10.1371/journal.pbio.3000473.g008 How loss of EXP1 affects the function of the nutrient-permeable channel activity remains to be determined. One possibility is that EXP1 has a role as part of the nutrient pore structure, and its absence could directly lead to a defective channel (Fig 8B). Alternatively, EXP1 may have a more indirect role. In this respect, it is interesting to note that EXP1 and the topologically related ETRAMPs form oligomeric arrays in the PVM and were hypothesized to compartmentalize different activities of the PVM [12]. The mislocalization of EXP2 could indicate that the EXP1 KO leads to EXP2 aggregation or otherwise negatively affects a functionally relevant spatial distribution of this protein in the membrane (Fig 8B). This effect could be specific for EXP2, as suggested by our IP data and the fact that ETRAMP4 did not show an altered distribution in the PVM of ΔEXP1 parasites. A more general effect on the state of PVM proteins cannot not be fully excluded. The EXP1 KO also affected the distribution of ETRAMP5 in the PVM. However, this could also mean that this ETRAMP is connected to EXP2 and EXP1. This possibility is supported by the fact that ETRAMP5 also appeared in the EXP2 interactome [31] and shows a similar expression profile to EXP1 and EXP2 [40]. Nevertheless, if this is the case, ETRAMP5 does not play a critical part, as we here show that it is not essential for growth of asexual blood stages. Consistent with a function in nutrient uptake of the parasite, EXP1 loss resembled previously observed starvation phenotypes [35, 36, 37], and EXP1 levels specifically influenced the capacity of parasites to grow in medium with low levels of amino acids. This supports a role of EXP1 in nutrient acquisition and indicates that the activity measured at the PVM indeed is critical for the parasite's nutrient supply. Such a role for EXP1 now allows us to specifically study the nutrient-permeable channel activity, independent from the protein export activity of EXP2. We confirmed previous data that EXP1 and EXP2 are stored in merozoites [33,34], indicating that not only PTEX but also the nutrient-permeable channel activity can be delivered to the PVM during invasion and thus be active in ring stages. This would allow access of nutrients to rings and would explain the delayed development of these stages in the EXP1 KO. Nevertheless, most ΔEXP1 rings eventually reached the young trophozoite stage, likely reflecting the lower dependence on exogenous nutrients up to this stage. The resemblance of a nutrient starvation phenotype did not derive from a restricted nutrient access to the infected RBC but rather is consistent with a loss of the nutrient permeability at the PVM in agreement with our patch-clamp observations. It is also noteworthy that the effect of the EXP1 KO was less pronounced on gametocytes, which are less metabolically active during their later development [41]. In conclusion, EXP1 is critical for the replication of P. falciparum in RBC, but this is independent of its postulated heme-detoxifying membrane GST activity. Rather, the evidence presented in this study supports the thesis that EXP1 is required for the EXP2-based nutrient-permeable channel activity of the PVM. Accordingly, the activity measured by patch-clamp indeed is important for the access of metabolites needed for parasite replication. These data clearly show that the function of EXP2 for nutrient uptake is functionally distinct from its role in protein transport. Materials and methods Plasmid constructs cloning For SLI-targeted gene disruption, the first 575 bp of the exp1 gene and the first 291 bp of the etramp5 gene without the start codon (including introns) were cloned into pSLI-TGD [20] using NotI and MluI (oligonucleotides and the plasmid constructs are listed in S1 Table). For conditional deletion of exp1, the first 575 bp of the exp1 gene were PCR amplified using primers to append a double myc tag, a first loxP site (not disrupting the open reading frame) and a recodonized T2A skip peptide. A codon changed exp1 gene was synthesized (Genscript) and PCR amplified with primers to add a second loxP site after the gene to obtain a second fragment. Both fragments were cloned into pSLI-3xHA, (derived from pSLI [20] by replacing GFP with 3xHA) using NotI and KpnI by Gibson cloning. This resulted in plasmid pSLI-exp1-loxP. Using the same strategy for the conditional excision of exp1, the first 986 bp (including introns) of the exp2 gene were amplified using primers to add a first loxP site and a recodonized T2A skip peptide to obtain a first fragment. A recodonized exp2 gene was synthesized (Genscript) and amplified with overlapping primers to append a second loxP site. Both PCR products were cloned into pSLI-exp1-loxP using NotI and XmaI by Gibson cloning. For complementation constructs, the recodonized exp1 gene was PCR amplified using primers to append the Ty sequence and cloned via XhoI and XmaI into p1xNLS-FRB-mCherry containing yDHODH as a resistance marker and different promoters (nmd3, sfa32, hsp86) driving expression of the expression cassette [20]. An AvrII site was (introduced with the primer) before the Ty sequence to allow for further cloning. This resulted in plasmid pEXP1comp. Mutated, chimeric, and deletion constructs were generated using overlapping primers detailed in S1 Table and cloned into pEXP1comp using XhoI and AvrII. mScarlet [42] was synthesized by GenScript and cloned together with SBP1 (using Gibson) into p1xNLS-FRB-mCherrynmd3 to generate pSBP1-mScarlet. In this vector, the XhoI and AvrII restriction sites were used to exchanged SBP1 with the first 219 bp of PF13_0191 [43] to obtain the soluble reporter in the PV (SP-mScarlet) or with recodonized exp1 to obtain the EXP1wt-mScarlet complementation construct. To obtain EXP2-GFP expressed under the nmd3 promoter, EXP2 was synthesized with a different codon usage (GenScript) and PCR amplified with overhanging sequences to fuse it with GFP and clone it into p1xNLS-FRB-mCherrynmd3 digested with XhoI and XmaI using Gibson cloning. P. falciparum culture and transfection Blood stages of P. falciparum parasites (strain 3D7) were cultured in human RBCs (O+) (transfusion blood, Universität Klinikum Eppendorf, Hamburg) or obtained from the NIH IRB-approved Research Donor Program in Bethesda. Cultures were maintained at 37°C in an atmosphere of 1% O2, 5% CO2, and 94% N2 and cultured using RPMI complete medium (Applichem, Darmstadt, Germany) containing 0.5% Albumax (Invitrogen, Carlsbad, CA) according to standard procedures [44]. For transfection of episomal constructs, Percoll-enriched synchronized mature schizonts were electroporated with 50 μg of plasmid DNA using a Nucleofector II (Lonza) [45]. Selection was done either with 4 nM WR99210 (Jacobus Pharmaceuticals, USA), 2 μg/ml Blasticidin S (Life Technologies, USA), or 0.9 μM DSM1 (BEI Resources; https://www.beiresources.org). For generation of stable integrant cell lines, parasites containing the episomal plasmids selected with WR were grown with 400 μg/ml G418 (Sigma-Aldrich, St. Louis, MO) to select for integrants carrying the desired genomic modification as described previously [20]. For SLI-TGD, a total of 6 independent 2-ml cultures containing the episomal plasmid were selected under G418. To confirm correct integration, genomic DNA from parasites selected under G418 was prepared with a QIAamp DNA Mini Kit and analyzed by PCR using primers specific for the 5' and 3' integration junctions of the exp1 locus and primers to detect the original locus. Generation of condΔEXP1 parasites and DiCre-mediated excision to obtain ΔEXP1 parasites The parasites containing the integrated pSLI-exp1-loxP construct were transfected with pSkip-Flox [20] using 2 μg/ml Blasticidin S to obtain a line expressing the DiCre fragments. To induce excision of the floxed copy of exp1 in the resulting cell line (condΔEXP1), the parasites were synchronized twice with 5% sorbitol with a time interval of 5 hours, after which the culture was split into 2 dishes of which one dish received rapalog (Clontech, Mountain View, CA) to a final concentration of 250 nM. The untreated dish served as control culture. The rapalog stock (500 mM in DMSO) was stored at −20°C and diluted 1:20 in RPMI as a working solution as described previously [20]. Parasites were cultured in the presence of rapalog for 48 hours. Cultures were synchronized with sorbitol at the beginning of the new cycle to obtain ring stages without EXP1. The parasites of this culture starting the cycle without EXP1 (termed ΔEXP1 parasites) were used for all experiments if not otherwise stated. For generation of EXP1-Ty-complementation and marker expression cell lines, the cell line condΔEXP1 was transfected with the corresponding yDHODH plasmids (see S1 Table) and selected with 0.9 DSMI μM (BEI Resources). Live cell imaging and confocal microscopy Fluorescence microscopy was done as previously described [46]. Parasites were incubated with 1 μg/ml DAPI in culture medium for 10 minutes to stain nuclei and analyzed using a Zeiss Axioscope M1 equipped with a 100X/1.4 numerical aperture oil immersion objective. A Hamamatsu Orca C4742-95 and the Zeiss Axiovision software were used for collecting images. Images were processed with Corel PHOTO-PAINT X6 (https://www.coreldraw.com). For counting of nuclei, ΔEXP1 and control parasites were stained with 1 μg/ml DAPI 36 to 44 hours post invasion (h.p.i.) in the second cycle on rapalog (ΔEXP1 parasites), and the number of nuclei was counted by 2 different analysts blinded to the identity of the sample. To quantify the localization of EXP2-GFP after depletion of EXP1, 3 different analysts counted cells (N = 25) scoring the number of cells with typical PVM localization (uninterrupted GFP signal surrounding more than 50% of the parasite) or cells with an aberrant localization (uninterrupted GFP signal surrounding less than 50% of the circumference of the parasite). The distribution of endogenous EXP2 was similarly scored by one analyst using IFA samples. Data were analyzed using Graph Pad Prism 6.07 (Graph Pad Software, https://www.graphpad.com). For time-lapse imaging, parasites were synchronized using 5% sorbitol, the culture was split into 2 dishes of which one received rapalog to 250 nM and a control without rapalog. After 48 hours, the resulting ΔEXP1 and control parasites were coated onto the bottom of a sterile, uncoated, hydrophobic, high, 35 mm μ-Dish (Ibidi) subdivided in 4 chambers using culture grade 0.5 mg/ml concanavalin A (Sigma) dissolved in dH2O as described previously [46]. Briefly, the concanavalin A (Sigma) was added to the dish surface for 10 minutes at 37°C, washed off using PBS, and the culture, resuspended in sterile PBS, was added and allowed to settle for 15 minutes, using the different chambers of the dish for ΔEXP1 and control parasites. Nonbound cells were washed off using DPBS, and prewarmed phenol red-free culture medium was added to the dish. Cells were viewed at 37°C using an Olympus FV1000 confocal microscope equipped with an Olympus Cellcubator. Using the multi-area time-lapse function of the Fluoview software and a motorized stage, at least 10 fields (containing 10–20 infected RBCs) were observed for each condition. Control and ΔEXP1 parasites were imaged simultaneously in the different chambers of the same dish for a period of 76 hours, and images were collected with a 1-hour interval. Focus was maintained using the Olympus ZDC autofocus system. An Olympus 60x/1.35 plan S apo oil immersion lens and Fluoview software version 1.7b was used. Parameters for image collection were usually 4–8 μs laser dwell time, 512 × 512 dpi, 16–32 z-stacks (0.38 μm step size), a zoom level of 3–5, and a 559 nm laser at 1%–5%. The time-lapse experiments were analyzed and processed in Imaris 7.7.2 (Bitplane). Image series were cropped with Image J (https://imagej.nih.gov.ij/), and single images were processed in Corel Photo-Paint X6 (https://www.coreldraw.com). To analyze the phenotype in ring stages of control and ΔEXP1 parasites, individual ring stages were scored in Imaris at every hour of the time-lapse experiment for ameboid shape, shape change compared to previous time point, position change in the RBC (more than half a cell diameter compared to previous time point), and hugging (defined as close apposition of parasite to RBC periphery), and the frequency was calculated for the total of time intervals examined for a given cell. Time to develop to trophozoite stage from start of the experiment was recorded by determining the time point when parasites contained a clear focus of hemozoin. Flow cytometry growth assays and Giemsa stages For flow cytometry (FC) growth curves, parasitemia was measured by FC and adjusted to 0.1%, and the parasites were divided in two 2-ml dishes (one with 250 nM and a control without rapalog). To follow the growth of the culture using Giemsa smears, the parasitemia was adjusted to 1%, and smears were collected after the intervals indicated. Medium was changed daily, and rapalog was added freshly every day. For the FC curves, the parasitemia was measured as previously described [20]: 20 μl resuspended parasite culture was incubated with dihydroethidium (Cayman Chemical, Ann Arbor, MI) and Hoechst (Cheomdex, Switzerland) at a final concentration of 4.5 μg/ml and 5 μg/ml, respectively, in RPMI for 20 minutes at room temperature protected from light. Before measuring, the cells were fixed with RPMI containing 0.003% glutaraldehyde. For every sample, 100,000 events were recorded using aLSRII flow cytometer (Beckton Dickinson), and parasitemia was determined with the FACS Diva software. For measuring the capacity to complement ΔEXP1 parasites, the growth of the cell lines with the complementation constructs was assayed over 5 days using the FC growth assay starting before excision of exp1 (cycle 0). The parasitemia was adjusted to 0.1% and divided in two 2-ml dishes, one with 250 nM rapalog and a control without rapalog. The parasitemia at day 5 was compared to that of the control. At least 4 independent replicas were analyzed for each cell line. To calculate the complementation activity for each construct, the level of growth (parasitemia at day 5 rapalog-treated/growth at day 5 unexcised) was compared to the percentage of complementation of the EXP1wtmid complementation construct. Data were analyzed with Graph Pad Prism version 6.07 (https://www.graphpad.com) and presented as mean ± SD. To assess growth in presence of reducing agents, 4-day FC growth assays were performed with matched synchronous ring stages starting with 0.1% parasitemia after one cycle ± rapalog (i.e., using parasites already starting without EXP1 and their matched controls). RPMI was supplemented with Trolox, ascorbic acid, N-acetylcysteine, and cysteine each at a final concentration of 100 μM. Every day, the parasitemia was measured by FC, and fresh medium with supplements was added. For examination of the effect of E64 on growth of ΔEXP1 parasites, ring synchronized parasites at a parasitemia of 1% after one cycle ± rapalog were grown overnight with and without 1 μM E64 (Sigma). The next day, cultures were thoroughly washed to remove E64 and further cultured without the inhibitor. In parallel, the same procedure was carried out with the same cell culture after one cycle ± rapalog but pretreated with 1 μM E64 for 2 hours prior to a pulse of 50 nM DHA (Adipogen, Switzerland). After 3 hours, DHA-treated cultures were washed extensively and further cultured without the drug. The parasitemia was measured each day for 72 hours. The survival rate was calculated as parasitemia of DHA treated or rapalog treated compared to parasitemia of the respective control ± E64. To evaluate sensitivity to low nutrient conditions, synchronous ring stages after 1 cycle ± rapalog were grown in complete amino acid–restricted medium and complete RPMI medium or medium containing 75 μM NaN3 (Sigma). To obtain amino acid restricted RPMI medium, complete medium was added in a 1/20 dilution to glucose and amino acid–free RPMI medium 1640 (US Biological). This resulted in a final concentration of 6 mM glucose and 1:20 of the concentration of every amino acid found in standard RPMI complete. Parasitemia was measured after 2 growth cycles using FC (day 5). Relative growth was calculated as parasitemia in restricted medium and NaN3 containing medium compared to parasitemia of respective control in complete medium without NaN3. To evaluate stage distribution of parasites in low amino acid concentration, synchronous ring stages of ΔEXP1 complementation lines were grown for 40 hours with and without rapalog. For this, schizonts that had been grown ± rapalog were Percoll (GE Healthcare, Sweden) purified and allowed to invade while shaking at 37°C at 750 rpm for 30 minutes in complete and amino acid–restricted medium and further cultured for 3 hours in the respective medium. Rings 0–3 h.p.i. were sorbitol synchronized, and the cultures were continued in the respective medium. Smears were collected after 18, 22, 26, and 30 h.p.i. Stages were counted microscopically, and percentage of ring and trophozoite (parasites containing a clear focus of hemozoin) stages was calculated for every time point. Production and purification of antisera Specific antisera to detect ETRAMP4 and ETRAMP10.1 were raised against the C-terminal domains produced as recombinant GST fusion proteins in Escherichia coli as described previously [10]. Rabbit antisera were raised commercially (Eurogentec) and purified over GST-sepharose columns (Genscript) containing the recombinant antigen crosslinked to the column according to established procedures using 30 mM DMP (Thermo Scientific) in 0.2 M Triethanolamine [47]. Briefly, the crude antiserum was diluted 1/10 in 1x TBS (20 mM TrisHCl [pH 7.0], 150 mM NaCl) containing 1% bovine serum albumin (BSA) and twice passed over the resin containing recombinant crosslinked GST to deplete antibodies binding GST. The flow through was collected and passed through a column containing resin with the corresponding recombinant GST fusion protein crosslinked to it. The antibodies on the resin were washed once with TBS containing 0.1%Triton-X-100, 5 times with TBS, 2 times with 0.1x TBS, and once with 0.1x TBS containing 0.1% Triton-X-100. Thereafter, the bound antibodies were eluted 10 times with 1 ml 0.1 M Glycine (pH 2.5), which was collected in tubes containing 25 μl of 1 M TrisHCl (pH 9.0). Dilutions of eluate 1 and 2 were used for all experiments. Animal handling and immunization at Eurogentec were carried out in accordance with good animal practices according to the Belgian national animal welfare regulations for Eurogentec SA, Seraing and approved by the ethics committee (CE/Sante/E/001) of the Centre d’Economie Rurale (CER Groupe, Marloie, Belgium). At the time of these immunizations, Eurogentec followed the European Union directive 86/609. IFAs IFAs to assess the location of the endogenously HA-tagged EXP1 or the TY-tagged complementation constructs were performed in suspension with Compound 2 [48]-stalled schizonts to differentiate protein located at the PPM from that located at the PVM. For this, trophozoite stages were treated with Compound 2 (1 μM) overnight, and arrested schizonts were harvested, washed in PBS, and fixed with 4% paraformaldehyde/0.0075% glutaraldehyde in PBS [49]. Cells were permeabilized with 0.1% Triton X-100 in PBS, blocked with 3% BSA in PBS, and incubated for 1 hour with primary antibodies: rat α-HA (Roche, Mannheim, Germany) (1:500), rabbit α-HA (Cell Signaling, USA) (1:500), mouse α-Ty (Sigma) (1:20,000), human α-MSP1 (PPM marker [1:1,000]) [50] diluted in 3% BSA in PBS. Cells were washed 3 times with PBS and incubated for 1 hour with Alexa 488 nm or Alexa 594 nm conjugated secondary antibodies specific for human, mouse, rabbit, or rat IgG (Invitrogen) diluted 1:2,000 in 3% BSA in PBS and containing 1 μg/ml DAPI. Cells were directly imaged after washing 5 times with PBS. For IFAs detecting PVM markers and exported proteins, the condΔEXP1, condΔEXP1+EXP2-GFP, condΔEXP2, and condΔEXP2+EXP1-Ty cell lines were grown for 48 hours on rapalog (250 nM) to obtain the corresponding ΔEXP1 or ΔEXP2 parasites. Rings in the second cycle were directly used or parasites were synchronized and allowed to develop to trophozoite stages. Cells were fixed and permeabilized as described above and incubated with rabbit α-SBP1 (C) (1:2,000) [31], rabbit α-KAHRP (1:500) (a kind gift of Prof. Brian Cooke), rabbit α-REX1 (1:10,000) [31], mouse α-REX2 (1:500) [51], mouse α-MSRP6 1:250 [43], mouse α-ETRAMP5 (1:500) [52], rabbit α-ETRAMP4 (1:500), rabbit α-ETRAMP10.1 (1:500), mouse α-GFP (Roche) (1:500), rabbit α-GFP (Thermo Fischer, USA) (1:500), mouse α-Ty (Sigma) (1:20,000), and mouse monoclonal 7.7 α-EXP2 (1:2,000). Cultures containing gametocytes were fixed in suspension as described above, air-dried as thin films on 10-well slides (Thermo Fischer), and fixed in 100% acetone for 30 minutes at room temperature. IFAs were labelled with mouse α-Pfs16 1:1,000 [53], rat α-Pfg377 1:1,000 [54], and rabbit α-spectrin 1:500 (Sigma). Staining of parasite membranes using lipid dyes Bodipy-TR-C5-ceramide (Invitrogen) staining was performed using a concentration of 2.5 μM (stock 5 μM) in RPMI as previously described [46] in ring and trophozoites of condΔEXP1 parasites and condΔEXP1 expressing EXP2-GFP after 1 cycle ± rapalog. For Lyso PC labelling, TopFluor LysoPC (Avanti Polar Lipids, Alabaster, AL) (1 mM stock in methanol) was resuspended in PBS to a final concentration of 20 μM, added to ring and trophozoite stages of control and ΔEXP1 parasites, and incubated for 15 minutes at 37°C. All microscopy images of the lipid dye stained parasites were recorded with the same acquisition settings and exposure time. Number of protrusions in each parasite were counted, and data were analyzed with Graph Pad Prism 6.07 (Graph Pad Software, http://www.graphpad.com). Electron microscopy Control and ΔEXP1 parasites were harvested 14 to 24 h.p.i. Cells were fixed with 2.5% glutaraldehyde (Electron Microscopy Sciences, USA) in 50 mM cacodylate buffer (pH 7.4) for 1 hour at room temperature. Cells were post fixed with 2% OsO4 in H2O (Electron Microscopy Sciences) for 40 minutes at 4°C in the dark, contrasted with 0.5% uranylacetate (Electron Microscopy Sciences) for 30 minutes at room temperature, and dehydrated through increasing concentrations of ethanol. Following embedding in epoxy resin (EPON) (Roth, Karlsruhe, Germany), 60 nm sections were generated with an Ultracut UC7 (Leica) and examined with a Tecnai Spirit transmission electron microscope (FEI), equipped with a LaB6 filament and operated at an acceleration voltage of 80 kV. Solubility assays of EXP1 constructs For saponin lysis to separate PV proteins from membrane-associated proteins, Percoll-enriched trophozoites (from 5–10 ml culture with a parasitemia of 5%–10%) of the cell lines expressing complementation Ty constructs were washed twice with PBS and incubated on ice for 10 minutes with 100 μl PBS containing a final concentration of 0.015% saponin (Sigma, Steinheim), followed by centrifugation at 16,000g for 5 minutes. The supernatant (containing PV and host cell soluble proteins) was transferred to a new tube and mixed with protease cocktail inhibitor (Roche) and 1 mM PMSF and reducing sodium dodecyl sulfate (SDS) sample buffer. The parasite pellet (containing membrane proteins and parasite proteins confined within the PPM) was washed once with DPBS and then resuspended in 100 μl of protein lysis buffer (0.5x PBS/4% SDS/0.5% Triton X-100) containing complete protease inhibitor cocktail. The pellet lysate was cleared using a centrifugation at 16,000g for 5 minutes, and the supernatant was transferred to a second tube, and reducing SDS sample buffer was added. Equivalent volumes were analyzed by SDS-polyacrylamide gel electrophoresis (PAGE) and western blotting. To assess leakage of PV proteins in the SP-mScarlet expressing ΔEXP1 parasites, the host cell cytosol and the parasite including the PV content was first separated by tetanolysin lysis as follows: Percoll-enriched trophozoites were generated from 5–10 ml of parasite culture (5%–10% parasitemia), washed with PBS, and incubated in 100 μl PBS containing 1 HU tetanolysin (Santa Cruz Biotechnology, USA) for 5 minutes at 37°C. The supernatant (containing soluble proteins from the host cell) was transferred to a new tube and mixed with protease cocktail inhibitor (Roche) and PMSF 1 mM and reducing SDS sample buffer. The parasite pellet was processed as described above for saponin lysis. For total parasites extracts, parasites were released from RBCs by incubation in 0.03% saponin in PBS for 10 minutes on ice followed by 3 washes with PBS. Proteins were then extracted with protein lysis buffer in the presence of protease cocktail inhibitor (Roche) and 1 mM PMSF. After centrifugation at 16,000g for 5 minutes, reducing SDS sample buffer was added to the supernatant, and the sample was analyzed by SDS-PAGE and immunoblotting. To test the membrane extractability of EXP2 after removal of EXP1, control and ΔEXP1 trophozoites were Percoll purified from 5–10 ml of parasite culture (5%–10% parasitemia), washed with PBS, and lysed in 100 μl 5 mM Tris-HCl (pH 8.0)/1 mM EDTA containing protease inhibitor cocktail (Roche) for 10 minutes on ice. Lysates were frozen at −80°C, thawed, and centrifuged 5 minutes at 16,000g. The resulting pellet was washed once with 5 mM Tris-HCl (pH 8.0)/1 mM EDTA and resuspended in 200 μl 5 mM Tris-HCl (pH 8.0)/1 mM EDTA. The suspension was divided into 4 tubes (50 μl each) and centrifuged 5 minutes at 16,000g. The 4 pellets were resuspended in one each of the following solutions: (1) 0.5 x PBS/4% SDS/0.5% Triton X-100 containing protease inhibitor cocktail (Roche) (corresponding to the total control); (2) 0.1 M Na2CO3 (pH 11.5); (3) 8 M urea/5 mM Tris-HCl (pH 8.0)/1 mM EDTA; and (4) Triton 1% in 1x PBS containing protease inhibitor cocktail (Roche). The samples were incubated on ice for 30 minutes, except for the first (total control), which was directly frozen. Tubes 2, 3, and 4 were centrifuged 5 minutes at 16,000g and the supernatant (extracted proteins) transferred into a fresh tube. The corresponding pellets were resuspended in 0.5x PBS/4% SDS/0.5% Triton X-100 containing protease inhibitor cocktail. Equivalent volumes of supernatant and pellet (or supernatant only for total control) were analyzed by SDS-PAGE and western blotting. For densitometric analyses, the intensity of EXP2 signal in supernatant and pellet was measured. A ratio supernatant/pellet EXP2 in every fraction was calculated and normalized to the ratio of the BIP signal. Formaldehyde in vivo cross linking In vivo cross linking was performed as described previously [12]. Parasite cultures (10 ml, 3%–5% parasitemia) were washed twice with PBS and split into 2 tubes. The cells were resuspended in PBS, and to one tube, formaldehyde (PFA) was added to a final concentration of 1%. The samples were incubated at 37°C for 30 minutes, and then Tris-HCl (pH 8.0) was added to 30 mM to quench the reaction. Both samples were centrifuged at 3,000g for 5 minutes followed by lysis in 10 ml of 10 mM Tris-HCl (pH 8.0) on ice for 1 hour. The sample was centrifuged at 5,000g for 15 minutes, and the pellet was washed 3 times in 1.5 ml ice-cold PBS with centrifugations at 16,000g. The layer on top of the pellet representing erythrocyte ghost membranes was removed, and the final pellet was resuspended in 2 volumes of protein lysis buffer and stored at −80°C. Equivalent volumes of cross-linked and non–cross-linked samples were analyzed by immunoblotting. Immunoblotting analyses Protein samples were resolved by SDS-PAGE and transferred to Amersham Protran membranes (GE Healthcare, Germany) in a tankblot device (Bio-Rad) using transfer buffer (0.192 M Glycine, 0.1% SDS, 25 mM Tris) with 20% methanol or 10 mM CAPS buffer (pH 11) without methanol. Membranes were blocked, and antibodies were diluted in PBS containing 5% skim milk. Washing steps were done with PBS. Primary antibodies were applied in the following dilutions: mouse α-Ty (Sigma), 1:20,000; rat α-HA (Roche), 1:1,000; rabbit α-HA (Cell Signaling), 1:1,000; mouse α-GFP (Roche), 1:1000; rabbit α-GFP (Thermo Fischer), 1:2,000; rat α-RFP (Chromotek, Germany), 1:1,000; rabbit α-SERA5, 1:2,000 [31] rabbit α-REX3, 1:2,000 [51]; rabbit α-SBP1(C),1:2,000 [31]; rabbit α-aldolase, 1:2,000 [31]; rabbit α-ETRAMP4, 1:500; and rabbit anti-BIP, 1:2,000 [55]. After 3 washes with PBS, horseradish peroxidase-conjugated secondary antibodies goat α-rat (Dianova, Hamburg, Germany) and goat α-mouse (Dianova; 1:3,000) and donkey α-rabbit (Dianova; 1:2,500) were incubated for 2 hours to overnight. Detection was done using enhanced chemiluminescence (Bio-Rad/Thermo Fischer), and signals were recorded with a ChemiDoc XRS imaging system (Bio-Rad). Densitometric analyses were performed with Image Lab software 5.2 (Bio-Rad). Intensity of Ty signal of EXP1wt-Ty constructs expressed under the different promoters was normalized to the BIP signal, and the ratio was compared to that of the EXP1wt-Tymid, which was set to 100%. Quantification of ROS Control and ΔEXP1 ring parasites were cultured overnight in the presence of 200 μM 5-ALA (Sigma) and further cultured in the presence of rapalog. The next day, after 2 washes with DPBS, the resulting trophozoite-stage parasites were incubated for 30 minutes with 5 μM CM2-DCFDA (Invitrogen) in DPBS at 37°C protected from light. Cells were washed twice with DPBS and further cultured in RPMI for 2 hours at 37°C under standard conditions. Parasites with similar size were imaged, and fluorescence was captured with the same acquisition settings to obtain comparable measurements of the fluorescence intensity. Fluorescence intensity (integrated density) was measured with Image J [56], and background was subtracted in each image. The data were analyzed with Graph Pad Prism version 6.07 (http://www.graphpad.com). To quantify the number of parasites exposed to oxidative stress, the parasites were incubated with CM2-DCFDA and further cultured in RPMI with no supplements or in the presence of diamide (100 μM) (Sigma) or Trolox (6-hydroxy-2,5,7,8-tetramethylchroman-2-carboxylic acid) (100 μM) (Sigma). Before analysis, the parasites were stained with 5 μg/ml Hoechst in RPMI for 20 minutes in the dark, and the number of cells that were CM2-DCDFA positive and 5-ALA positive was quantified by FC with an LSRII flow cytometer (BD Biosciences, Franklin Lakes, NJ). The percentage of cells exposed to oxidative stress was calculated as number of CM2-DCFDA and 5-ALA positive cells compared to the number of all 5-ALA positive cells. Mean fluorescence of CM2-DCDFA in the 5-ALA positive cells was estimated with Flow Jo 10 (https://www.flowjo.com), and data were analyzed with Graph Pad Prism 6.07 (http://www.graphpad.com). 5-ALA uptake assay Synchronized ring stages after one cycle ± rapalog were cultured overnight in presence of 200 μM 5-ALA (Sigma). Parasites were stained with 1 μg/ml DAPI, and trophozoite stages were imaged using the same acquisition settings of the fluorescence microscope. Parasites were stained with Hoechst, and the PPIX-positive cells were detected with an LSRII flow cytometer (BD Biosciences) detecting PPIX emission with the 532 nm laser through a 605/40 nm bandpass filter and Hoechst emission with the 406 nm laser through a 440/40 nm bandpass filter to quantify the number of PPIX-positive and Hoechst-positive cells. Erythrocyte doublets were excluded using an FCS-A versus FCS-H display. Data were analyzed by BD FACS Diva software (BD Biosciences), FlowJo 10 (https://www.flowjo.com), and Graph Pad Prism 6v.07 (Graph Pad Software, http://www.graphpad.com). Electrophysiology To initiate the KO of condΔEXP1+SP-mScarlet parasites, late-stage–infected RBCs (65% Percoll interface) were left to infect new RBCs overnight. New rings were isolated (pellet of 65% Percoll), and 250 nM rapalog were added (day 1) to one culture dish; a second dish served as control without rapalog. On day 3, parasites were patch clamped. The control was cultured in parallel without the addition of rapalog. The PVM nutrient-permeable channel was detected in the PVM of parasites released from the host RBC after Percoll isolation and incubation in an isotonic high potassium buffer (140 mM KCl, 5 mM NaCl, 0.4 mM CaCl2, 0.4 mM MgCl2, 25 mM HEPES, 4.5 mg/ml glucose, 0.5% Albumax II, 66 nM phalloidin-Alexa488 [Invitrogen]) [57]. The phalloidin was added to visualize the attached host RBC and confirm release of the parasite. The parasites were transferred on the microscope in 150 mM NaCl, 5 mM KCl, 1.4 mM CaCl2, 1 mM MgCl2, 20 mM HEPES NaOH (pH 7.4), and 4.5 mg/ml glucose. The patch pipette (borosilicate glass) was pulled with a Model P80 (Sutter instrument) to 15–20 MΩ and filled with 155 mM CsCl, 1.4 mM CaCl2, 1 mM MgCl2, and 20 mM HEPES NaOH (pH 7.4). Electrophysiology data were recorded using an Axopatch 200B amplifier equipped with a CV203BU head stage (Molecular Devices, San Jose, CA). The signal was filtered at 10 kHz (8-pole Bessel) and digitized at 50 kHz using a Digidata 1550B (Molecular Devices). Error bars in the detection frequency bar graph were calculated in Excel (Microsoft); the P value was calculated in R (version 3.5.0, R Core Team) using the Fisher test function. Co-IP assays The cell line condΔEXP1 expressing EXP2-GFP was sorbitol synchronized, and ring-stage parasites were adjusted to approximately 5% parasitemia. The next day, late trophozoites were harvested and washed twice with DPBS. The culture was cross-linked with 0.5 mM dithiobis (succinimidylpropionate) (DSP, from a 20 mM stock in DMSO) (Pierce, USA) in DPBS for 30 minutes at room temperature, and the reaction was quenched with PBS containing 25 mM Tris-HCl. Cross-linked infected RBCs were purified in a Percoll gradient, washed twice with DPBS, and lysed with RIPA buffer (10 mM Tris HCl [pH 7.5], 150 mM NaCl, 0.1% SDS, 1% Triton) containing protease inhibitor cocktail (Roche) and 1 mM PMSF. After 2 freeze-thaw cycles, lysates were cleared by centrifugation at 16,000g for 10 minutes, and the supernatant was diluted 1:2 with RIPA buffer without detergents. The supernatants were incubated with 25 μl of mouse monoclonal anti-HA beads (Pierce, USA) or anti-GFP beads (Chromotek, Germany) for 3 hours at 4°C. Samples of input and post binding extracts were saved for immunoblot analysis. Beads were recovered by centrifugation and washed 5 times with RIPA buffer. Proteins were eluted in 50 μl 4x SDS sample buffer at 85°C for 5 minutes. Equal volumes of input post binding extract and bound fractions were subjected to western blot analysis. RSAs and determination of DHA IC50 RSAs were performed according to established procedures [30]. Briefly, synchronous ring stages of ΔEXP1 complementation lines were grown for 40 hours with and without rapalog. Kelch13C580Y [20] was analyzed as positive control for DHA resistance. Percoll-purified schizonts ± rapalog were allowed to invade fresh RBCs shaking at 37°C at 750 rpm for 30 minutes and further cultured for 3 hours. Rings 0 to 3 h.p.i. were obtained by sorbitol treatment. These rings were exposed to 350 nM DHA (Adipogen, Switzerland) for 6 hours alongside an untreated control. Following removal of DHA by thorough washing, parasites were cultured for 66 hours under standard conditions. The number of viable parasites was counted in 10,000 erythrocytes in Giemsa smears to calculate survival rate as parasitemia of ± rapalog DHA-treated compared to parasitemia of ± rapalog DHA-untreated cultures. For determination of IC50, the different ΔEXP1 complementation parasites were grown with and without rapalog for 48 hours. Ring stages were sorbitol synchronized, adjusted to a start parasitemia of approximately 1%, and cultured with increasing concentrations (0 to 50 nM) of DHA. The medium was changed after 24 hours, and fresh DHA was added. The parasitemia was measured by FC as described above after 48 hours, and the IC50 was calculated using GraphPadPrism version 6.07. Gametocyte induction in ΔEXP1 parasites ΔEXP1 and control parasites were sorbitol synchronized and grown in RPMI supplemented with 50 mM N-acetyl glucosamine (Serva) for 5 days without diluting the culture. Samples were collected first at day 3 for Giemsa smears and fixed for IFA in suspension or dried, and acetone-fixed for detection of early gametocytes with α-Pfs16. N-Ac-Gluc was removed after 5 days, and the parasites were further cultured. On day 8 after addition of N-Ac-Gluc, samples were collected for IFA labelled with Pfg377 to detect late gametocytes. RBC spectrin was labelled to count the number of gametocytes per 1,000 RBCs. Percentage of Pfs16- and Pfg377-positive cells in the rapalog-treated culture was compared to that in control parasites to calculate fold reduction of cells positive with the respective antigen. Data were analyzed by Graph Pad Prism version 6.07 (http://www.graphpad.com). Plasmid constructs cloning For SLI-targeted gene disruption, the first 575 bp of the exp1 gene and the first 291 bp of the etramp5 gene without the start codon (including introns) were cloned into pSLI-TGD [20] using NotI and MluI (oligonucleotides and the plasmid constructs are listed in S1 Table). For conditional deletion of exp1, the first 575 bp of the exp1 gene were PCR amplified using primers to append a double myc tag, a first loxP site (not disrupting the open reading frame) and a recodonized T2A skip peptide. A codon changed exp1 gene was synthesized (Genscript) and PCR amplified with primers to add a second loxP site after the gene to obtain a second fragment. Both fragments were cloned into pSLI-3xHA, (derived from pSLI [20] by replacing GFP with 3xHA) using NotI and KpnI by Gibson cloning. This resulted in plasmid pSLI-exp1-loxP. Using the same strategy for the conditional excision of exp1, the first 986 bp (including introns) of the exp2 gene were amplified using primers to add a first loxP site and a recodonized T2A skip peptide to obtain a first fragment. A recodonized exp2 gene was synthesized (Genscript) and amplified with overlapping primers to append a second loxP site. Both PCR products were cloned into pSLI-exp1-loxP using NotI and XmaI by Gibson cloning. For complementation constructs, the recodonized exp1 gene was PCR amplified using primers to append the Ty sequence and cloned via XhoI and XmaI into p1xNLS-FRB-mCherry containing yDHODH as a resistance marker and different promoters (nmd3, sfa32, hsp86) driving expression of the expression cassette [20]. An AvrII site was (introduced with the primer) before the Ty sequence to allow for further cloning. This resulted in plasmid pEXP1comp. Mutated, chimeric, and deletion constructs were generated using overlapping primers detailed in S1 Table and cloned into pEXP1comp using XhoI and AvrII. mScarlet [42] was synthesized by GenScript and cloned together with SBP1 (using Gibson) into p1xNLS-FRB-mCherrynmd3 to generate pSBP1-mScarlet. In this vector, the XhoI and AvrII restriction sites were used to exchanged SBP1 with the first 219 bp of PF13_0191 [43] to obtain the soluble reporter in the PV (SP-mScarlet) or with recodonized exp1 to obtain the EXP1wt-mScarlet complementation construct. To obtain EXP2-GFP expressed under the nmd3 promoter, EXP2 was synthesized with a different codon usage (GenScript) and PCR amplified with overhanging sequences to fuse it with GFP and clone it into p1xNLS-FRB-mCherrynmd3 digested with XhoI and XmaI using Gibson cloning. P. falciparum culture and transfection Blood stages of P. falciparum parasites (strain 3D7) were cultured in human RBCs (O+) (transfusion blood, Universität Klinikum Eppendorf, Hamburg) or obtained from the NIH IRB-approved Research Donor Program in Bethesda. Cultures were maintained at 37°C in an atmosphere of 1% O2, 5% CO2, and 94% N2 and cultured using RPMI complete medium (Applichem, Darmstadt, Germany) containing 0.5% Albumax (Invitrogen, Carlsbad, CA) according to standard procedures [44]. For transfection of episomal constructs, Percoll-enriched synchronized mature schizonts were electroporated with 50 μg of plasmid DNA using a Nucleofector II (Lonza) [45]. Selection was done either with 4 nM WR99210 (Jacobus Pharmaceuticals, USA), 2 μg/ml Blasticidin S (Life Technologies, USA), or 0.9 μM DSM1 (BEI Resources; https://www.beiresources.org). For generation of stable integrant cell lines, parasites containing the episomal plasmids selected with WR were grown with 400 μg/ml G418 (Sigma-Aldrich, St. Louis, MO) to select for integrants carrying the desired genomic modification as described previously [20]. For SLI-TGD, a total of 6 independent 2-ml cultures containing the episomal plasmid were selected under G418. To confirm correct integration, genomic DNA from parasites selected under G418 was prepared with a QIAamp DNA Mini Kit and analyzed by PCR using primers specific for the 5' and 3' integration junctions of the exp1 locus and primers to detect the original locus. Generation of condΔEXP1 parasites and DiCre-mediated excision to obtain ΔEXP1 parasites The parasites containing the integrated pSLI-exp1-loxP construct were transfected with pSkip-Flox [20] using 2 μg/ml Blasticidin S to obtain a line expressing the DiCre fragments. To induce excision of the floxed copy of exp1 in the resulting cell line (condΔEXP1), the parasites were synchronized twice with 5% sorbitol with a time interval of 5 hours, after which the culture was split into 2 dishes of which one dish received rapalog (Clontech, Mountain View, CA) to a final concentration of 250 nM. The untreated dish served as control culture. The rapalog stock (500 mM in DMSO) was stored at −20°C and diluted 1:20 in RPMI as a working solution as described previously [20]. Parasites were cultured in the presence of rapalog for 48 hours. Cultures were synchronized with sorbitol at the beginning of the new cycle to obtain ring stages without EXP1. The parasites of this culture starting the cycle without EXP1 (termed ΔEXP1 parasites) were used for all experiments if not otherwise stated. For generation of EXP1-Ty-complementation and marker expression cell lines, the cell line condΔEXP1 was transfected with the corresponding yDHODH plasmids (see S1 Table) and selected with 0.9 DSMI μM (BEI Resources). Live cell imaging and confocal microscopy Fluorescence microscopy was done as previously described [46]. Parasites were incubated with 1 μg/ml DAPI in culture medium for 10 minutes to stain nuclei and analyzed using a Zeiss Axioscope M1 equipped with a 100X/1.4 numerical aperture oil immersion objective. A Hamamatsu Orca C4742-95 and the Zeiss Axiovision software were used for collecting images. Images were processed with Corel PHOTO-PAINT X6 (https://www.coreldraw.com). For counting of nuclei, ΔEXP1 and control parasites were stained with 1 μg/ml DAPI 36 to 44 hours post invasion (h.p.i.) in the second cycle on rapalog (ΔEXP1 parasites), and the number of nuclei was counted by 2 different analysts blinded to the identity of the sample. To quantify the localization of EXP2-GFP after depletion of EXP1, 3 different analysts counted cells (N = 25) scoring the number of cells with typical PVM localization (uninterrupted GFP signal surrounding more than 50% of the parasite) or cells with an aberrant localization (uninterrupted GFP signal surrounding less than 50% of the circumference of the parasite). The distribution of endogenous EXP2 was similarly scored by one analyst using IFA samples. Data were analyzed using Graph Pad Prism 6.07 (Graph Pad Software, https://www.graphpad.com). For time-lapse imaging, parasites were synchronized using 5% sorbitol, the culture was split into 2 dishes of which one received rapalog to 250 nM and a control without rapalog. After 48 hours, the resulting ΔEXP1 and control parasites were coated onto the bottom of a sterile, uncoated, hydrophobic, high, 35 mm μ-Dish (Ibidi) subdivided in 4 chambers using culture grade 0.5 mg/ml concanavalin A (Sigma) dissolved in dH2O as described previously [46]. Briefly, the concanavalin A (Sigma) was added to the dish surface for 10 minutes at 37°C, washed off using PBS, and the culture, resuspended in sterile PBS, was added and allowed to settle for 15 minutes, using the different chambers of the dish for ΔEXP1 and control parasites. Nonbound cells were washed off using DPBS, and prewarmed phenol red-free culture medium was added to the dish. Cells were viewed at 37°C using an Olympus FV1000 confocal microscope equipped with an Olympus Cellcubator. Using the multi-area time-lapse function of the Fluoview software and a motorized stage, at least 10 fields (containing 10–20 infected RBCs) were observed for each condition. Control and ΔEXP1 parasites were imaged simultaneously in the different chambers of the same dish for a period of 76 hours, and images were collected with a 1-hour interval. Focus was maintained using the Olympus ZDC autofocus system. An Olympus 60x/1.35 plan S apo oil immersion lens and Fluoview software version 1.7b was used. Parameters for image collection were usually 4–8 μs laser dwell time, 512 × 512 dpi, 16–32 z-stacks (0.38 μm step size), a zoom level of 3–5, and a 559 nm laser at 1%–5%. The time-lapse experiments were analyzed and processed in Imaris 7.7.2 (Bitplane). Image series were cropped with Image J (https://imagej.nih.gov.ij/), and single images were processed in Corel Photo-Paint X6 (https://www.coreldraw.com). To analyze the phenotype in ring stages of control and ΔEXP1 parasites, individual ring stages were scored in Imaris at every hour of the time-lapse experiment for ameboid shape, shape change compared to previous time point, position change in the RBC (more than half a cell diameter compared to previous time point), and hugging (defined as close apposition of parasite to RBC periphery), and the frequency was calculated for the total of time intervals examined for a given cell. Time to develop to trophozoite stage from start of the experiment was recorded by determining the time point when parasites contained a clear focus of hemozoin. Flow cytometry growth assays and Giemsa stages For flow cytometry (FC) growth curves, parasitemia was measured by FC and adjusted to 0.1%, and the parasites were divided in two 2-ml dishes (one with 250 nM and a control without rapalog). To follow the growth of the culture using Giemsa smears, the parasitemia was adjusted to 1%, and smears were collected after the intervals indicated. Medium was changed daily, and rapalog was added freshly every day. For the FC curves, the parasitemia was measured as previously described [20]: 20 μl resuspended parasite culture was incubated with dihydroethidium (Cayman Chemical, Ann Arbor, MI) and Hoechst (Cheomdex, Switzerland) at a final concentration of 4.5 μg/ml and 5 μg/ml, respectively, in RPMI for 20 minutes at room temperature protected from light. Before measuring, the cells were fixed with RPMI containing 0.003% glutaraldehyde. For every sample, 100,000 events were recorded using aLSRII flow cytometer (Beckton Dickinson), and parasitemia was determined with the FACS Diva software. For measuring the capacity to complement ΔEXP1 parasites, the growth of the cell lines with the complementation constructs was assayed over 5 days using the FC growth assay starting before excision of exp1 (cycle 0). The parasitemia was adjusted to 0.1% and divided in two 2-ml dishes, one with 250 nM rapalog and a control without rapalog. The parasitemia at day 5 was compared to that of the control. At least 4 independent replicas were analyzed for each cell line. To calculate the complementation activity for each construct, the level of growth (parasitemia at day 5 rapalog-treated/growth at day 5 unexcised) was compared to the percentage of complementation of the EXP1wtmid complementation construct. Data were analyzed with Graph Pad Prism version 6.07 (https://www.graphpad.com) and presented as mean ± SD. To assess growth in presence of reducing agents, 4-day FC growth assays were performed with matched synchronous ring stages starting with 0.1% parasitemia after one cycle ± rapalog (i.e., using parasites already starting without EXP1 and their matched controls). RPMI was supplemented with Trolox, ascorbic acid, N-acetylcysteine, and cysteine each at a final concentration of 100 μM. Every day, the parasitemia was measured by FC, and fresh medium with supplements was added. For examination of the effect of E64 on growth of ΔEXP1 parasites, ring synchronized parasites at a parasitemia of 1% after one cycle ± rapalog were grown overnight with and without 1 μM E64 (Sigma). The next day, cultures were thoroughly washed to remove E64 and further cultured without the inhibitor. In parallel, the same procedure was carried out with the same cell culture after one cycle ± rapalog but pretreated with 1 μM E64 for 2 hours prior to a pulse of 50 nM DHA (Adipogen, Switzerland). After 3 hours, DHA-treated cultures were washed extensively and further cultured without the drug. The parasitemia was measured each day for 72 hours. The survival rate was calculated as parasitemia of DHA treated or rapalog treated compared to parasitemia of the respective control ± E64. To evaluate sensitivity to low nutrient conditions, synchronous ring stages after 1 cycle ± rapalog were grown in complete amino acid–restricted medium and complete RPMI medium or medium containing 75 μM NaN3 (Sigma). To obtain amino acid restricted RPMI medium, complete medium was added in a 1/20 dilution to glucose and amino acid–free RPMI medium 1640 (US Biological). This resulted in a final concentration of 6 mM glucose and 1:20 of the concentration of every amino acid found in standard RPMI complete. Parasitemia was measured after 2 growth cycles using FC (day 5). Relative growth was calculated as parasitemia in restricted medium and NaN3 containing medium compared to parasitemia of respective control in complete medium without NaN3. To evaluate stage distribution of parasites in low amino acid concentration, synchronous ring stages of ΔEXP1 complementation lines were grown for 40 hours with and without rapalog. For this, schizonts that had been grown ± rapalog were Percoll (GE Healthcare, Sweden) purified and allowed to invade while shaking at 37°C at 750 rpm for 30 minutes in complete and amino acid–restricted medium and further cultured for 3 hours in the respective medium. Rings 0–3 h.p.i. were sorbitol synchronized, and the cultures were continued in the respective medium. Smears were collected after 18, 22, 26, and 30 h.p.i. Stages were counted microscopically, and percentage of ring and trophozoite (parasites containing a clear focus of hemozoin) stages was calculated for every time point. Production and purification of antisera Specific antisera to detect ETRAMP4 and ETRAMP10.1 were raised against the C-terminal domains produced as recombinant GST fusion proteins in Escherichia coli as described previously [10]. Rabbit antisera were raised commercially (Eurogentec) and purified over GST-sepharose columns (Genscript) containing the recombinant antigen crosslinked to the column according to established procedures using 30 mM DMP (Thermo Scientific) in 0.2 M Triethanolamine [47]. Briefly, the crude antiserum was diluted 1/10 in 1x TBS (20 mM TrisHCl [pH 7.0], 150 mM NaCl) containing 1% bovine serum albumin (BSA) and twice passed over the resin containing recombinant crosslinked GST to deplete antibodies binding GST. The flow through was collected and passed through a column containing resin with the corresponding recombinant GST fusion protein crosslinked to it. The antibodies on the resin were washed once with TBS containing 0.1%Triton-X-100, 5 times with TBS, 2 times with 0.1x TBS, and once with 0.1x TBS containing 0.1% Triton-X-100. Thereafter, the bound antibodies were eluted 10 times with 1 ml 0.1 M Glycine (pH 2.5), which was collected in tubes containing 25 μl of 1 M TrisHCl (pH 9.0). Dilutions of eluate 1 and 2 were used for all experiments. Animal handling and immunization at Eurogentec were carried out in accordance with good animal practices according to the Belgian national animal welfare regulations for Eurogentec SA, Seraing and approved by the ethics committee (CE/Sante/E/001) of the Centre d’Economie Rurale (CER Groupe, Marloie, Belgium). At the time of these immunizations, Eurogentec followed the European Union directive 86/609. IFAs IFAs to assess the location of the endogenously HA-tagged EXP1 or the TY-tagged complementation constructs were performed in suspension with Compound 2 [48]-stalled schizonts to differentiate protein located at the PPM from that located at the PVM. For this, trophozoite stages were treated with Compound 2 (1 μM) overnight, and arrested schizonts were harvested, washed in PBS, and fixed with 4% paraformaldehyde/0.0075% glutaraldehyde in PBS [49]. Cells were permeabilized with 0.1% Triton X-100 in PBS, blocked with 3% BSA in PBS, and incubated for 1 hour with primary antibodies: rat α-HA (Roche, Mannheim, Germany) (1:500), rabbit α-HA (Cell Signaling, USA) (1:500), mouse α-Ty (Sigma) (1:20,000), human α-MSP1 (PPM marker [1:1,000]) [50] diluted in 3% BSA in PBS. Cells were washed 3 times with PBS and incubated for 1 hour with Alexa 488 nm or Alexa 594 nm conjugated secondary antibodies specific for human, mouse, rabbit, or rat IgG (Invitrogen) diluted 1:2,000 in 3% BSA in PBS and containing 1 μg/ml DAPI. Cells were directly imaged after washing 5 times with PBS. For IFAs detecting PVM markers and exported proteins, the condΔEXP1, condΔEXP1+EXP2-GFP, condΔEXP2, and condΔEXP2+EXP1-Ty cell lines were grown for 48 hours on rapalog (250 nM) to obtain the corresponding ΔEXP1 or ΔEXP2 parasites. Rings in the second cycle were directly used or parasites were synchronized and allowed to develop to trophozoite stages. Cells were fixed and permeabilized as described above and incubated with rabbit α-SBP1 (C) (1:2,000) [31], rabbit α-KAHRP (1:500) (a kind gift of Prof. Brian Cooke), rabbit α-REX1 (1:10,000) [31], mouse α-REX2 (1:500) [51], mouse α-MSRP6 1:250 [43], mouse α-ETRAMP5 (1:500) [52], rabbit α-ETRAMP4 (1:500), rabbit α-ETRAMP10.1 (1:500), mouse α-GFP (Roche) (1:500), rabbit α-GFP (Thermo Fischer, USA) (1:500), mouse α-Ty (Sigma) (1:20,000), and mouse monoclonal 7.7 α-EXP2 (1:2,000). Cultures containing gametocytes were fixed in suspension as described above, air-dried as thin films on 10-well slides (Thermo Fischer), and fixed in 100% acetone for 30 minutes at room temperature. IFAs were labelled with mouse α-Pfs16 1:1,000 [53], rat α-Pfg377 1:1,000 [54], and rabbit α-spectrin 1:500 (Sigma). Staining of parasite membranes using lipid dyes Bodipy-TR-C5-ceramide (Invitrogen) staining was performed using a concentration of 2.5 μM (stock 5 μM) in RPMI as previously described [46] in ring and trophozoites of condΔEXP1 parasites and condΔEXP1 expressing EXP2-GFP after 1 cycle ± rapalog. For Lyso PC labelling, TopFluor LysoPC (Avanti Polar Lipids, Alabaster, AL) (1 mM stock in methanol) was resuspended in PBS to a final concentration of 20 μM, added to ring and trophozoite stages of control and ΔEXP1 parasites, and incubated for 15 minutes at 37°C. All microscopy images of the lipid dye stained parasites were recorded with the same acquisition settings and exposure time. Number of protrusions in each parasite were counted, and data were analyzed with Graph Pad Prism 6.07 (Graph Pad Software, http://www.graphpad.com). Electron microscopy Control and ΔEXP1 parasites were harvested 14 to 24 h.p.i. Cells were fixed with 2.5% glutaraldehyde (Electron Microscopy Sciences, USA) in 50 mM cacodylate buffer (pH 7.4) for 1 hour at room temperature. Cells were post fixed with 2% OsO4 in H2O (Electron Microscopy Sciences) for 40 minutes at 4°C in the dark, contrasted with 0.5% uranylacetate (Electron Microscopy Sciences) for 30 minutes at room temperature, and dehydrated through increasing concentrations of ethanol. Following embedding in epoxy resin (EPON) (Roth, Karlsruhe, Germany), 60 nm sections were generated with an Ultracut UC7 (Leica) and examined with a Tecnai Spirit transmission electron microscope (FEI), equipped with a LaB6 filament and operated at an acceleration voltage of 80 kV. Solubility assays of EXP1 constructs For saponin lysis to separate PV proteins from membrane-associated proteins, Percoll-enriched trophozoites (from 5–10 ml culture with a parasitemia of 5%–10%) of the cell lines expressing complementation Ty constructs were washed twice with PBS and incubated on ice for 10 minutes with 100 μl PBS containing a final concentration of 0.015% saponin (Sigma, Steinheim), followed by centrifugation at 16,000g for 5 minutes. The supernatant (containing PV and host cell soluble proteins) was transferred to a new tube and mixed with protease cocktail inhibitor (Roche) and 1 mM PMSF and reducing sodium dodecyl sulfate (SDS) sample buffer. The parasite pellet (containing membrane proteins and parasite proteins confined within the PPM) was washed once with DPBS and then resuspended in 100 μl of protein lysis buffer (0.5x PBS/4% SDS/0.5% Triton X-100) containing complete protease inhibitor cocktail. The pellet lysate was cleared using a centrifugation at 16,000g for 5 minutes, and the supernatant was transferred to a second tube, and reducing SDS sample buffer was added. Equivalent volumes were analyzed by SDS-polyacrylamide gel electrophoresis (PAGE) and western blotting. To assess leakage of PV proteins in the SP-mScarlet expressing ΔEXP1 parasites, the host cell cytosol and the parasite including the PV content was first separated by tetanolysin lysis as follows: Percoll-enriched trophozoites were generated from 5–10 ml of parasite culture (5%–10% parasitemia), washed with PBS, and incubated in 100 μl PBS containing 1 HU tetanolysin (Santa Cruz Biotechnology, USA) for 5 minutes at 37°C. The supernatant (containing soluble proteins from the host cell) was transferred to a new tube and mixed with protease cocktail inhibitor (Roche) and PMSF 1 mM and reducing SDS sample buffer. The parasite pellet was processed as described above for saponin lysis. For total parasites extracts, parasites were released from RBCs by incubation in 0.03% saponin in PBS for 10 minutes on ice followed by 3 washes with PBS. Proteins were then extracted with protein lysis buffer in the presence of protease cocktail inhibitor (Roche) and 1 mM PMSF. After centrifugation at 16,000g for 5 minutes, reducing SDS sample buffer was added to the supernatant, and the sample was analyzed by SDS-PAGE and immunoblotting. To test the membrane extractability of EXP2 after removal of EXP1, control and ΔEXP1 trophozoites were Percoll purified from 5–10 ml of parasite culture (5%–10% parasitemia), washed with PBS, and lysed in 100 μl 5 mM Tris-HCl (pH 8.0)/1 mM EDTA containing protease inhibitor cocktail (Roche) for 10 minutes on ice. Lysates were frozen at −80°C, thawed, and centrifuged 5 minutes at 16,000g. The resulting pellet was washed once with 5 mM Tris-HCl (pH 8.0)/1 mM EDTA and resuspended in 200 μl 5 mM Tris-HCl (pH 8.0)/1 mM EDTA. The suspension was divided into 4 tubes (50 μl each) and centrifuged 5 minutes at 16,000g. The 4 pellets were resuspended in one each of the following solutions: (1) 0.5 x PBS/4% SDS/0.5% Triton X-100 containing protease inhibitor cocktail (Roche) (corresponding to the total control); (2) 0.1 M Na2CO3 (pH 11.5); (3) 8 M urea/5 mM Tris-HCl (pH 8.0)/1 mM EDTA; and (4) Triton 1% in 1x PBS containing protease inhibitor cocktail (Roche). The samples were incubated on ice for 30 minutes, except for the first (total control), which was directly frozen. Tubes 2, 3, and 4 were centrifuged 5 minutes at 16,000g and the supernatant (extracted proteins) transferred into a fresh tube. The corresponding pellets were resuspended in 0.5x PBS/4% SDS/0.5% Triton X-100 containing protease inhibitor cocktail. Equivalent volumes of supernatant and pellet (or supernatant only for total control) were analyzed by SDS-PAGE and western blotting. For densitometric analyses, the intensity of EXP2 signal in supernatant and pellet was measured. A ratio supernatant/pellet EXP2 in every fraction was calculated and normalized to the ratio of the BIP signal. Formaldehyde in vivo cross linking In vivo cross linking was performed as described previously [12]. Parasite cultures (10 ml, 3%–5% parasitemia) were washed twice with PBS and split into 2 tubes. The cells were resuspended in PBS, and to one tube, formaldehyde (PFA) was added to a final concentration of 1%. The samples were incubated at 37°C for 30 minutes, and then Tris-HCl (pH 8.0) was added to 30 mM to quench the reaction. Both samples were centrifuged at 3,000g for 5 minutes followed by lysis in 10 ml of 10 mM Tris-HCl (pH 8.0) on ice for 1 hour. The sample was centrifuged at 5,000g for 15 minutes, and the pellet was washed 3 times in 1.5 ml ice-cold PBS with centrifugations at 16,000g. The layer on top of the pellet representing erythrocyte ghost membranes was removed, and the final pellet was resuspended in 2 volumes of protein lysis buffer and stored at −80°C. Equivalent volumes of cross-linked and non–cross-linked samples were analyzed by immunoblotting. Immunoblotting analyses Protein samples were resolved by SDS-PAGE and transferred to Amersham Protran membranes (GE Healthcare, Germany) in a tankblot device (Bio-Rad) using transfer buffer (0.192 M Glycine, 0.1% SDS, 25 mM Tris) with 20% methanol or 10 mM CAPS buffer (pH 11) without methanol. Membranes were blocked, and antibodies were diluted in PBS containing 5% skim milk. Washing steps were done with PBS. Primary antibodies were applied in the following dilutions: mouse α-Ty (Sigma), 1:20,000; rat α-HA (Roche), 1:1,000; rabbit α-HA (Cell Signaling), 1:1,000; mouse α-GFP (Roche), 1:1000; rabbit α-GFP (Thermo Fischer), 1:2,000; rat α-RFP (Chromotek, Germany), 1:1,000; rabbit α-SERA5, 1:2,000 [31] rabbit α-REX3, 1:2,000 [51]; rabbit α-SBP1(C),1:2,000 [31]; rabbit α-aldolase, 1:2,000 [31]; rabbit α-ETRAMP4, 1:500; and rabbit anti-BIP, 1:2,000 [55]. After 3 washes with PBS, horseradish peroxidase-conjugated secondary antibodies goat α-rat (Dianova, Hamburg, Germany) and goat α-mouse (Dianova; 1:3,000) and donkey α-rabbit (Dianova; 1:2,500) were incubated for 2 hours to overnight. Detection was done using enhanced chemiluminescence (Bio-Rad/Thermo Fischer), and signals were recorded with a ChemiDoc XRS imaging system (Bio-Rad). Densitometric analyses were performed with Image Lab software 5.2 (Bio-Rad). Intensity of Ty signal of EXP1wt-Ty constructs expressed under the different promoters was normalized to the BIP signal, and the ratio was compared to that of the EXP1wt-Tymid, which was set to 100%. Quantification of ROS Control and ΔEXP1 ring parasites were cultured overnight in the presence of 200 μM 5-ALA (Sigma) and further cultured in the presence of rapalog. The next day, after 2 washes with DPBS, the resulting trophozoite-stage parasites were incubated for 30 minutes with 5 μM CM2-DCFDA (Invitrogen) in DPBS at 37°C protected from light. Cells were washed twice with DPBS and further cultured in RPMI for 2 hours at 37°C under standard conditions. Parasites with similar size were imaged, and fluorescence was captured with the same acquisition settings to obtain comparable measurements of the fluorescence intensity. Fluorescence intensity (integrated density) was measured with Image J [56], and background was subtracted in each image. The data were analyzed with Graph Pad Prism version 6.07 (http://www.graphpad.com). To quantify the number of parasites exposed to oxidative stress, the parasites were incubated with CM2-DCFDA and further cultured in RPMI with no supplements or in the presence of diamide (100 μM) (Sigma) or Trolox (6-hydroxy-2,5,7,8-tetramethylchroman-2-carboxylic acid) (100 μM) (Sigma). Before analysis, the parasites were stained with 5 μg/ml Hoechst in RPMI for 20 minutes in the dark, and the number of cells that were CM2-DCDFA positive and 5-ALA positive was quantified by FC with an LSRII flow cytometer (BD Biosciences, Franklin Lakes, NJ). The percentage of cells exposed to oxidative stress was calculated as number of CM2-DCFDA and 5-ALA positive cells compared to the number of all 5-ALA positive cells. Mean fluorescence of CM2-DCDFA in the 5-ALA positive cells was estimated with Flow Jo 10 (https://www.flowjo.com), and data were analyzed with Graph Pad Prism 6.07 (http://www.graphpad.com). 5-ALA uptake assay Synchronized ring stages after one cycle ± rapalog were cultured overnight in presence of 200 μM 5-ALA (Sigma). Parasites were stained with 1 μg/ml DAPI, and trophozoite stages were imaged using the same acquisition settings of the fluorescence microscope. Parasites were stained with Hoechst, and the PPIX-positive cells were detected with an LSRII flow cytometer (BD Biosciences) detecting PPIX emission with the 532 nm laser through a 605/40 nm bandpass filter and Hoechst emission with the 406 nm laser through a 440/40 nm bandpass filter to quantify the number of PPIX-positive and Hoechst-positive cells. Erythrocyte doublets were excluded using an FCS-A versus FCS-H display. Data were analyzed by BD FACS Diva software (BD Biosciences), FlowJo 10 (https://www.flowjo.com), and Graph Pad Prism 6v.07 (Graph Pad Software, http://www.graphpad.com). Electrophysiology To initiate the KO of condΔEXP1+SP-mScarlet parasites, late-stage–infected RBCs (65% Percoll interface) were left to infect new RBCs overnight. New rings were isolated (pellet of 65% Percoll), and 250 nM rapalog were added (day 1) to one culture dish; a second dish served as control without rapalog. On day 3, parasites were patch clamped. The control was cultured in parallel without the addition of rapalog. The PVM nutrient-permeable channel was detected in the PVM of parasites released from the host RBC after Percoll isolation and incubation in an isotonic high potassium buffer (140 mM KCl, 5 mM NaCl, 0.4 mM CaCl2, 0.4 mM MgCl2, 25 mM HEPES, 4.5 mg/ml glucose, 0.5% Albumax II, 66 nM phalloidin-Alexa488 [Invitrogen]) [57]. The phalloidin was added to visualize the attached host RBC and confirm release of the parasite. The parasites were transferred on the microscope in 150 mM NaCl, 5 mM KCl, 1.4 mM CaCl2, 1 mM MgCl2, 20 mM HEPES NaOH (pH 7.4), and 4.5 mg/ml glucose. The patch pipette (borosilicate glass) was pulled with a Model P80 (Sutter instrument) to 15–20 MΩ and filled with 155 mM CsCl, 1.4 mM CaCl2, 1 mM MgCl2, and 20 mM HEPES NaOH (pH 7.4). Electrophysiology data were recorded using an Axopatch 200B amplifier equipped with a CV203BU head stage (Molecular Devices, San Jose, CA). The signal was filtered at 10 kHz (8-pole Bessel) and digitized at 50 kHz using a Digidata 1550B (Molecular Devices). Error bars in the detection frequency bar graph were calculated in Excel (Microsoft); the P value was calculated in R (version 3.5.0, R Core Team) using the Fisher test function. Co-IP assays The cell line condΔEXP1 expressing EXP2-GFP was sorbitol synchronized, and ring-stage parasites were adjusted to approximately 5% parasitemia. The next day, late trophozoites were harvested and washed twice with DPBS. The culture was cross-linked with 0.5 mM dithiobis (succinimidylpropionate) (DSP, from a 20 mM stock in DMSO) (Pierce, USA) in DPBS for 30 minutes at room temperature, and the reaction was quenched with PBS containing 25 mM Tris-HCl. Cross-linked infected RBCs were purified in a Percoll gradient, washed twice with DPBS, and lysed with RIPA buffer (10 mM Tris HCl [pH 7.5], 150 mM NaCl, 0.1% SDS, 1% Triton) containing protease inhibitor cocktail (Roche) and 1 mM PMSF. After 2 freeze-thaw cycles, lysates were cleared by centrifugation at 16,000g for 10 minutes, and the supernatant was diluted 1:2 with RIPA buffer without detergents. The supernatants were incubated with 25 μl of mouse monoclonal anti-HA beads (Pierce, USA) or anti-GFP beads (Chromotek, Germany) for 3 hours at 4°C. Samples of input and post binding extracts were saved for immunoblot analysis. Beads were recovered by centrifugation and washed 5 times with RIPA buffer. Proteins were eluted in 50 μl 4x SDS sample buffer at 85°C for 5 minutes. Equal volumes of input post binding extract and bound fractions were subjected to western blot analysis. RSAs and determination of DHA IC50 RSAs were performed according to established procedures [30]. Briefly, synchronous ring stages of ΔEXP1 complementation lines were grown for 40 hours with and without rapalog. Kelch13C580Y [20] was analyzed as positive control for DHA resistance. Percoll-purified schizonts ± rapalog were allowed to invade fresh RBCs shaking at 37°C at 750 rpm for 30 minutes and further cultured for 3 hours. Rings 0 to 3 h.p.i. were obtained by sorbitol treatment. These rings were exposed to 350 nM DHA (Adipogen, Switzerland) for 6 hours alongside an untreated control. Following removal of DHA by thorough washing, parasites were cultured for 66 hours under standard conditions. The number of viable parasites was counted in 10,000 erythrocytes in Giemsa smears to calculate survival rate as parasitemia of ± rapalog DHA-treated compared to parasitemia of ± rapalog DHA-untreated cultures. For determination of IC50, the different ΔEXP1 complementation parasites were grown with and without rapalog for 48 hours. Ring stages were sorbitol synchronized, adjusted to a start parasitemia of approximately 1%, and cultured with increasing concentrations (0 to 50 nM) of DHA. The medium was changed after 24 hours, and fresh DHA was added. The parasitemia was measured by FC as described above after 48 hours, and the IC50 was calculated using GraphPadPrism version 6.07. Gametocyte induction in ΔEXP1 parasites ΔEXP1 and control parasites were sorbitol synchronized and grown in RPMI supplemented with 50 mM N-acetyl glucosamine (Serva) for 5 days without diluting the culture. Samples were collected first at day 3 for Giemsa smears and fixed for IFA in suspension or dried, and acetone-fixed for detection of early gametocytes with α-Pfs16. N-Ac-Gluc was removed after 5 days, and the parasites were further cultured. On day 8 after addition of N-Ac-Gluc, samples were collected for IFA labelled with Pfg377 to detect late gametocytes. RBC spectrin was labelled to count the number of gametocytes per 1,000 RBCs. Percentage of Pfs16- and Pfg377-positive cells in the rapalog-treated culture was compared to that in control parasites to calculate fold reduction of cells positive with the respective antigen. Data were analyzed by Graph Pad Prism version 6.07 (http://www.graphpad.com). Supporting information S1 Fig. Conditional EXP1 KO. (A) Schematic representation of the SLI strategy to obtain a cell line for DiCre-based conditional KO of exp1. Top shows endogenous locus and plasmid pSLIΔEXP1cond. Light yellow box: cell line after SLI. Light orange box: the cell line transfected with pSkipFlox [20] and after induction of DiCre. “loxP” indicates the loxP site; asterisks indicates in-frame stop codon. Cre 60–343, Cre 19–59: Cre fragments; arrows, primers P1, P2, P3, and P4. (B) PCR on gDNA of condΔEXP1 and 3D7 parasites using the primers indicated in (a) confirming: 5'Int, 5'integration; absence of 'original locus'; 3'Int, 3' integration. (C) FC growth curves of synchronous condΔEXP1 ring stage parasites grown ± rapalog over 5 days (addition of rapalog starting day 1). Blue arrow indicates start of cycle without EXP1 ('ΔEXP1 parasites'). One representative of n = 3 experiments is shown. (D) IFA images of condΔEXP1 trophozoites grown for 24 hours (first cycle) or 72 hours (second cycle) with and without rapalog (control) probed with α-HA to detect EXP1*-HA and α-myc for the truncated EXP1 stub in the control. Note that after excision (rapalog), the stub contains both, myc- and HA-tag, and it will be recognized by both antibodies (see panel A). Nuclei were stained with DAPI; scale bars: 5 μm. (E) Number of nuclei in DAPI-stained control and ΔEXP1 parasites (rapalog) 40 h.p.i. One representative of n = 3 independent biological replicas. (F) Long-term FC growth curve of synchronous ring control and ΔEXP1 parasites (rapalog) after depletion of EXP1 (blue arrow) at the times indicated. Blue box shows zoom of restricted to 20% parasitemia on the y-axis to show raise in the control in early time points. Orange box, PCR with primers P1 and P2 (see panel A) from gDNA of control and rapalog-treated ΔEXP1 parasites on day 9. Mean of n = 2 independent experiments. Error bars indicate SD. (G) Transmission electron microscopy images of control and ΔEXP1 parasites (rapalog) 18 to 24 h.p.i. showing hugging in ΔEXP1 parasites. Scale bar, 500 nm. (H) Left, live cell images of control and ΔEXP1 (rapalog) parasites labelled with TopFluor Lyso PC (Lyso PC). Yellow arrows, tubo-vesicular network. Graph: quantification of protrusions per cell in n = 22 control cells and n = 38 ΔEXP1 parasites from 2 independent experiments. (I) Upper panel, live cell images of control and ΔEXP1 parasites (rapalog) expressing SPmScarlet. DAPI, nuclei. Light blue arrow shows a bleb. Lower panel: immunoblot of protein extracts from RBCs infected with these parasites, permeabilized with tetanolysin and separated into SN (host cell cytosol) and P, pellet (parasite within PVM). α-REX3, control for host cell cytosol; α-BIP, loading control. (E, H) green lines indicate mean and error bars SD; two-tailed unpaired t test, P values are indicated. BSD, blasticidine deaminase; DIC, differential interference contrast; EXP1*, recodonized exp1; HA, triple hemagglutinin; hDHFR: human dihydrofolate reductase; L, linker; Neo: neomycin phosphotransferase; NLS: nuclear localization signal; asterisk, in frame stop codon; SP, signal peptide; T2A, skip peptide; TM, transmembrane domain. https://doi.org/10.1371/journal.pbio.3000473.s001 (PDF) S2 Fig. Genetic complementation of ΔEXP1 parasites. (A) Schematics of the complementation constructs expressed in condΔEXP1 parasites. Numbers refer to amino acids of the domains shown in the legend. (B) Relative activity of the complementation constructs. Except where otherwise indicated, constructs were expressed under the nmd3 (mid) promoter. Each data point (red dot) shows growth of rapalog-treated versus unexcised parasites at the end of a 5-day growth assay relative to the growth of the wt construct. Green lines indicate activity of EXP1wtnmd3 (mid) (set as 100%) and absence of activity (ΔEXP1) set as 0%; n ≥ 4 independent experiments per cell line. Error bars indicate SD. (C) Mean ± SD of relative growth versus unexcised (control) and mean of relative complementation versus EXP1wtmid (used in panel B and in the graphs in Figs 2 and 3); n numbers evident in (B). (D) Percentage of rings and trophozoites of tightly synchronous parasites of ΔEXP1 and complemented ΔEXP1 parasites at the time points indicated after invasion (after an initial cycle ± rapalog). Mean of n = 2 independent experiments. (E) Amino acid sequence of the central region (including the TM domain) of EXP1 and selected ETRAMPs from P. falciparum. Boxes show conserved G, S, and T rich regions. Hydrophobic residues, red; positively charged, pink; negatively charged, blue; polar (N, T, G, S, Q, H, and Y), green. (F) Left, alignment of the EXP1 TM region from different Plasmodium species. Asterisk, conserved and double dot, partially conserved residues; mutated G, yellow boxes; predicted TM in P. falciparum EXP1 is boxed. Right, helical wheel diagram of the PfEXP1 TM domain (numbered from 1 to 23). https://doi.org/10.1371/journal.pbio.3000473.s002 (PDF) S3 Fig. Localization and solubility of EXP1 complementation constructs. (A) IFA images of compound 2-arrested condΔEXP1 schizont stages expressing the complementation constructs indicated above each panel (α-HA detects EXP1*-HA; α-Ty1, complementing EXP1 copy; anti-RFP, EXP1mScarlet. α-MSP1, PPM). DAPI, nuclei. Scale bars: 5 μm. (B) Immunoblots of extracts of the cell lines shown in (a). Saponin was used to separate the parasite pellet (P) from the supernatant (SN) containing PV and host cell content. α-Ty1 detects the complementation constructs, anti-RFP, EXP1mScarlet and α-SBP1 was used to detect a membrane-associated control protein. DIC, differential interference contrast. https://doi.org/10.1371/journal.pbio.3000473.s003 (PDF) S4 Fig. Oxidative stress and protein export in ΔEXP1 parasites. (A) FC analysis of matching 5-ALA-treated control and ΔEXP1 parasites (rapalog) after incubation with CM-H2DCFDA in RPMI alone or supplemented with diamide or Trolox at the time points indicated. The percentage of cells with oxidative stress corresponds to the number of CM-H2DCFDA positive cells of the total number of 5-ALA-positive cells. Error bars, SD. n = 4 independent biological replicas. (B) Fluorescence of control and ΔEXP1 parasites analyzed in (a). Green line, mean; error bars, SD. n = 4 independent biological replicas. (C) IFA images of control and ΔEXP1 parasites (rapalog) probed with α-HA (EXP1*-HA), α-SBP1, α-REX1, α-REX2, and α-MSRP6. Size bars, 5 μm. h.p.i., hours post invasion. https://doi.org/10.1371/journal.pbio.3000473.s004 (PDF) S5 Fig. Localization of EXP2 in ΔEXP1 ring stages. (A) Live cell images of control and ΔEXP1 (rapalog) ring stages episomally expressing EXP2-GFPnmd3. (B) IFA images of control and ΔEXP1 ring stages (rapalog); α-HA detects EXP1*-HA, α-EXP2 detects endogenous EXP2. Nuclei were stained with DAPI. Scale bars: 5 μm. DIC, differential interference contrast. https://doi.org/10.1371/journal.pbio.3000473.s005 (PDF) S6 Fig. SLI-TGD of ETRAMP5 has no effect on parasite growth. (A) Schematic representation of SLI-TGD to disrupt etramp5. Features as in S1A Fig. (B) PCR on gDNA of ETR5-TGD and wild-type 3D7 parasites confirming: 5'Int, 5'integration; absence of 'original locus'; 3'Int, 3' integration. (C) Live cell images of young trophozoite and schizont stages of ETR5-TGD parasites (fluorescence shows the truncated GFP-tagged protein). DAPI, nuclei; scale bars: 5 μm. (D) Immunoblot of extracts of ETR5-TGD parasites separated into saponin supernatant (SN, containing PV and host cell content) and parasite pellet (P). α-GFP, detects truncated ETR5; α-BIP: control for the parasite pellet. Asterisk, protein degraded down to GFP; double asterisk, unskipped protein (first T2A, no unskipped product detected at the GFP-Neomycin junction). The truncated protein has no TM and is therefore found in the SN. (E) Left: FC 5-day growth curves of synchronous 3D7 and ETR5-TGD parasites. Mean of n = 3 independent biological replicas. Right: fold increase in parasitemia over 5 days for 3D7 and ETR5-TGD parasites measured by FC. Green line indicates mean and error bars SD, two-tailed unpaired t test; P value indicated. DIC, differential interference contrast; ns, not significant. https://doi.org/10.1371/journal.pbio.3000473.s006 (PDF) S7 Fig. Conditional deletion of exp2. (A) Schematic representation of the SLI strategy to obtain a cell line for DiCre-based conditional KO of exp2. Features as in S1A Fig. (B) PCR products from gDNA of condΔEXP2 and wild-type 3D7 parasites confirming: 5'Int, 5'integration; absence of 'original locus'; 3'Int, 3' integration. (C) Strategy to deplete EXP2 from the PVM using synchronized condΔEXP2 ring stages divided into a culture with and one without rapalog. Top: schematic: green boxes and blue line around the parasite signify PVM with EXP2. Mid: PCR with primers P1 and P2 from gDNA 24 hours and 48 hours after addition of rapalog. Original: PCR product for locus with intact exp2; excised: PCR product after excision of exp2. Bottom: western blot using α-HA to detect EXP2*-HA and α-BIP as loading control. Note that the small truncated fragment after excision of the functional copy becomes HA-tagged (see panel A) but is not detected (likely due to its small size and its instability leading to low abundance, see panel F). (D) Giemsa smears of synchronous ΔEXP2 parasites (rapalog) compared to the controls. Blue arrow indicates start of a new cycle without EXP2. (E) FC growth curves of synchronous ring stage condΔEXP2 parasites grown ± rapalog over 5 days. Blue arrow indicates start of cycle without EXP2. One representative of n = 3 experiments. (H) IFA images of control and ΔEXP2 parasites (rapalog) probed with α-HA, which detects full functional (control) or truncated inactivated (rapalog) EXP2-HA and SBP1 (α-SBP1) or REX1 (α-REX1). DAPI, nuclei. Scale bars: 5 μm. Note that the truncated inactive version of EXP2 is not well detected, likely because it is degraded. DIC, differential interference contrast. https://doi.org/10.1371/journal.pbio.3000473.s007 (PDF) S8 Fig. EXP1–EXP2 interaction analysis and analysis of 5-ALA–treated ΔEXP1 parasites. (A) Western blot of reciprocal co-IP experiment using α-GFP with extracts of the cell line condΔEXP1+EXP2-GFPnmd3 to pull down EXP2-GFP. α-HA detects EXP1*-HA (monomer: asterisk, dimer: double asterisk); α-SERP, soluble PV protein; α-aldolase, cytosolic parasite protein. Input (I): total lysate before IP; post IP (P): lysate after IP; Eluate (E). One representative of n = 3 independent experiments. (B) IFA images of EXP1-3xHAendo and EXP2-3xHAendo merozoites probed with α-HA and α-MSP1 (plasma membrane marker). Nuclei were stained with DAPI; scale bar 2 μm. (C) Immunoblot of protein extracts derived from ΔEXP1 (rapa) and control trophozoites. Saponin was used to separate parasite pellet (P) from the supernatant (SN) containing PV and host cell soluble proteins. α-EXP2 detects endogenous EXP2; α-SERP, a soluble PV soluble to control for proper PVM permeabilization; α-BIP, as loading control and α-HA (detecting EXP1*-HA) to show loss of EXP1. One representative of n = 2 experiments. (D) Left: immunoblot of protein extracts from ΔEXP1 (rapa) and control trophozoites fractionated in SN and P after hypotonic lysis and extraction with Na2CO3, urea (peripheral membrane proteins) and Triton x-100 (TX-100, integral membrane proteins). α-EXP2 detects endogenous EXP2 and α-BIP, a parasite-internal peripheral membrane protein. Right: densitometric analysis of EXP2 intensity in SN and P. The ratio SN/ P of the EXP2 signal was calculated for Na2CO3 and urea and normalized to the ratio of BIP. Green line: mean of n = 4 independent experiments; error bars, SD. P values were calculated with a two-tailed unpaired t test. (E) FC analysis of 5-ALA–treated control and ΔEXP1 parasites (rapalog) shows that the number of PSAC-positive cells correlates with the number of cells that reached the trophozoite stage, irrespective of whether parasites contained EXP1 or not. Left, gating for Hoechst/PPIX-positive cells (upper right quadrant). Right: quantification of PPIX-positive cells and percentage of trophozoites in the same cultures. Mean of n = 3 independent biological replicates; two-tailed unpaired t test; P values indicated. (F) Gating strategy for quantification of P. falciparum–infected RBC cells by FC. Left panel: forward versus side scatter (FSC versus SSC) gating to define population of RBCs and exclude debris. Mid panel, forward scatter height (FSC-H) versus forward scatter area (FSC-A) density plot to define single RBCs and exclude doublets. Right panelL DAPI (Hoechst) versus PE-A (dihydroethidium, DHE) density plot to distinguish infected RBCs from uninfected RBCs. DIC, differential interference contrast. https://doi.org/10.1371/journal.pbio.3000473.s008 (PDF) S1 Table. Oligonucleotides used in this study. https://doi.org/10.1371/journal.pbio.3000473.s009 (PDF) S1 Data. Excel file containing seperate sheets of the numerical data underlying the graphs of the main figures. https://doi.org/10.1371/journal.pbio.3000473.s010 (XLSX) S2 Data. Excel file containing seperate sheets of the numerical data underlying the graphs of the supporting information figures. https://doi.org/10.1371/journal.pbio.3000473.s011 (XLSX) S1 Raw images. Minimally cropped blots and gels shown in the main figures and supporting information figures. https://doi.org/10.1371/journal.pbio.3000473.s012 (PDF) Acknowledgments We are grateful to Marcel Deponte for critical reading of the manuscript and interpretation of oxidative stress data, to Ralf Krumkamp for assistance with statistical analysis, to Arlett Heiber for purification of ETRAMP antisera, and to Svetlana Glushakova for helpful discussions. We thank Pietro Alano for α-Pfs16 and α-Pfg377 antibodies, Michael Blackman for Compound 2 and α-MSP1 antibodies, Tim Gilberger for α-BIP antibodies, Brian Cooke for α-KAHRP antibodies, and Matthias Marti for Topfluor Lyso PC. Monoclonal antibody 7.7 (α-EXP2) was obtained from The European Malaria Reagent Repository (http://www.malariaresearch.eu). We thank Jacobus Pharmaceuticals for WR99210. DSM1 (MRA-1161) was obtained from MR4/BEI Resources, NIAID, NIH.
Spatiotemporal dynamics of odor responses in the lateral and dorsal olfactory bulbBaker, Keeley L.;Vasan, Ganesh;Gumaste, Ankita;Pieribone, Vincent A.;Verhagen, Justus V.
doi: 10.1371/journal.pbio.3000409pmid: 31532763
Introduction Odor processing is critical for finding food and mates and detecting predators and is therefore vital for survival. Consequently, it is not surprising that, in mice, up to 5% of the protein coding genome is dedicated to the approximately 1,000 different olfactory receptors (ORs) present within the nasal epithelium, expressed by olfactory sensory neurons (OSNs) [1]. Each OSN expresses a single OR, and neurons expressing the same receptor are confined to one of 4 zones within the nose, albeit randomly located within these zones [2]. In the olfactory bulb (OB), the OSNs with a given receptor type map onto 1 or 2 glomeruli, one on the medial and one on the lateral surface, creating a mirror-symmetric glomerular map that wraps around the OB [3–6]. ORs can recognize multiple odorants, and the molecular features of an odorant can activate multiple ORs [7]. Molecular features of odorants also preferentially activate different olfactory epithelial zones [8, 9]. Spatial maps of glomerular activation have highlighted the topography of chemical properties of odorants in the OB [10]. The activity patterns of glomeruli are altered by the functional group of an odorant and its polarity, molecular shape (cyclic or noncyclic compounds), carbon chain length, concentration, and ortho- or retronasal route of entry of these odors [11–15]. Many of these functional mapping measurements have been performed using optical imaging from the dorsal olfactory bulb (dOB), where only approximately 25% of glomeruli can be accessed [12, 16–19]. The other parts of the bulb are not readily accessible, and thus there are limited data from nondorsal glomeruli. Dorsal glomeruli receive input only from the dorsal recess (zone 1) within the nasal cavity. Optical imaging of the lateral olfactory bulb (lOB) reports glomerular activation data from zones (2–4) of the epithelium [15]. All of the measurements from nondorsal glomeruli, to date, have been performed using population and activity–non-specific and low-speed imaging techniques, including intrinsic optical imaging [20], whole OB 2-deoxyglucose (2-DG) [21, 22], and functional magnetic resonance imaging (fMRI) [11]. These methods demonstrate time-averaged activity responses to odors. For example, activity patterns using 2-DG are obtained after a 45-minute exposure to a single odorant per animal, as the spatial maps are obtained in ex vivo brain slices. The spatial representations of hydrocarbons have been examined in the lateral bulb in rats by removing the eye and utilizing intrinsic optical imaging, highlighting two key areas of activation within the lateral bulb [20]. Intrinsic imaging does not relay direct information on neural firing rate, and therefore this study did not report response dynamics. This study also did not image the dorsal bulb simultaneously with the lateral bulb. Intrinsic imaging itself lacks neuronal specificity, as it monitors the hemodynamic response due not only to OSN input but also, for example, mitral cell/tufted cell output [23]. Although these techniques have given great insight into the spatial processing of odors [24], they are limited by their temporal resolution. Although spatial bulbar odor maps are one aspect underlying odor perception, it is well established that the glomerular activation dynamics are also able to contribute to perception. Widefield calcium imaging affords high temporal resolution and the temporal glomerular dynamics of the dOB have been extensively described over the first respiratory cycle [25, 26], an important time window because a single sniff can be used for odor discrimination [27, 28]. The dynamic glomerular activation patterns unfolds over approximately 200 ms across the dorsal glomerular layer following inhalation during odor presentation [19]. Indeed, mice can discriminate glomerular input activity duration differences down to only 10 ms [29] and detect temporal optogenetic odor information down to 10 ms relative to the sniff cycle [30]. Furthermore, our lab has shown that mice are able to discriminate the temporal differences in optogenetic activation of spatially separated glomeruli across the dorsal bulb of only 13 ms, independently of sniff timing [31]. Mice were also shown to be able to discriminate optogenetic activation (“play back”) of dynamic imaged odor maps from the same maps rendered static, but with equal integrated optical power [31]. Although activity timing plays a key role in information processing, this temporal patterning has never been explored across the lateral bulb itself, although the temporal differences between the medial and lateral areas of the OB have been investigated using multichannel recordings. This study highlighted important temporal response differences across the OB that are associated with temporal activations at the epithelium [32]. Here, using a dual camera imaging approach for simultaneous recording of odor responses, we examined both the spatial and temporal odor patterning in the lOB in concert with the dOB and, in doing so, also uncover a novel mechanosensory response in the lateral bulb. Results Dual imaging of the dOB and lOB Here, we have developed a dual imaging approach (Fig 1A) to measure the glomerular activity of both the dorsal and lateral regions of the OB. We investigated OR sensory neuron input using floxed-GCaMP6f reporter mice crossed with OMP-Cre animals [12] (Fig 1B). In addition to implementing classical dorsal region windowing of the OB (Fig 1C top), we also successfully exposed the lateral region of the OB by unilateral enucleation and preparation of an optical window medial to the eye (Fig 1C bottom). Both cameras were synchronized for simultaneous fluorescent imaging of these two regions of the OB. These two orthogonal imaging macroscopes had overlapping imaging planes in the lateral region of the dOB (Fig 1D). Four glomeruli confirmed excellent agreement in the glomerular fluorescence responses imaged by the dorsal and lateral camera (Fig 1E). Using frame subtraction before and after the first odor inhalation, regions of interest (ROIs) were accumulated (i.e., an ROI that responds to at least one stimulus) across odors in both the lateral and dorsal images (Fig 1F), and their fluorescence (% ΔF/F) response traces were isolated and analyzed (Fig 1G). We examined approximately 1.5 times the number of glomeruli in the dOB as in the lOB (dorsal: 232, lateral: 149 across 6 animals; S1 Table). For the first time, we simultaneously imaged OSN calcium activity in the dorsal (dOB) and lateral (lOB) regions of the OB. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Dual imaging of the dOB and lOB in GCaMP6f reporter mice. (A) Schematic representation of the dual microscope imaging setup. Illustrating the plane of the dorsal and lateral images. Inset: schematic of the odor delivery assembly relative to the nostrils, illustrating the odor and vacuum flow channels. (B) Schematic of the mouse genotype. (C) Top: resting light image of the dOB. Bottom: resting light image of the lateral region of the OB. (D) Top: odor-induced activation map of the dOB. Arrows highlight 4 individual glomeruli. Bottom: odor-induced activation map of the lOB. Arrows point to the same 4 glomeruli as top. (E) Left: glomerular response traces from the 4 glomeruli imaged with the dorsal camera in response to 1% heptanone. Middle: responses from the 4 glomeruli imaged with the lateral camera in response to 1% heptanone. Right: glomerulus 4 imaged from the dorsal and lateral camera, overlaid. (F) Top: an example of the ROI selection on the dorsal bulb for 1 animal. Bottom: an example of the ROI selection for the lateral bulb for 1 animal. (G) Top: % ΔF/F responses for the dorsal ROIs in panel F (top) selection of first odor response is indicated (1 animal). Middle: % ΔF/F responses for the lateral ROIs in panel F (Bottom) (1 animal). Bottom: a sniffing trace from the same animal. Inhalation is the upward inflection (see Methods for details). Underlying data for this figure can be found in S1 Data. dOB, dorsal olfactory bulb; lOB, lateral olfactory bulb; OB, olfactory bulb; ROI, region of interest. https://doi.org/10.1371/journal.pbio.3000409.g001 Simultaneous odor mapping of the dOB and lOB OSNs expressing different receptors that recognize related odor molecules project to neighboring glomeruli in the OB [33]. Therefore, there is chemotopic organization of the glomerular responses represented in spatial odor patterns [10, 34]. Differences in the spatial activation patterns of glomeruli are thought to play a primary role in identifying odors [10, 35]. Using our dual imaging approach, we investigated spatial odor patterns of glomerular OSN input (OMP-GCaMP6f mice) responses over the first respiration cycle in both the dOB and lOB (Fig 2A 1 animal, 3 trials). Six odors—amyl acetate (AA), carvone, heptanol, heptanone, hexanal, and methyl valerate (MV)—at 1% (saturated vapor [s.v.]) concentration were used. These odors were chosen to represent different molecular groups, as well as to be represented in the lateral bulb (though without concern where in the lOB) [10, 21, 24]. The global spatial organization of activation of glomeruli in the anterio-posterior (A-P), dorso-ventral (D-V), or medio-lateral (M-L) dimensions was determined using linear correlations of glomerular response amplitudes with their spatial location (see Methods and S1A Fig for examples). All these analyses were based on images of 512 × 512 pixels, where lower pixel numbers along the x-axis represent more anterior regions in the dOB and lOB. A fixed set of ROIs, used in every analysis, was accumulated across odor maps for each mouse, where each ROI responded to at least one odorant. For laterality of the dOB, a low pixel number along the y-axis represented more medial locations of ROIs. Each hemi-bulb was analyzed separately, and only the hemi-bulb ipsilateral to the exposed lOB is shown here. For the lOB, a lower y-axis pixel number represented more dorsal regions. Correlation analysis of glomerular responses with spatial location provides a simple metric that highlighted the differences in global response patterns for each odor (%ΔF/F responses of all odors shown in S2 Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Simultaneous odor mapping of the dOB and lOB. (A) Odor-induced activation maps for AA, carvone, heptanol, heptanone, hexanal, and MV. Top row: the dorsal region of the OB. Bottom row: the lateral region of the OB (1 animal, average of 3 trials). (B) Spatial odor map correlations between response amplitudes and the location along each spatial dimension of all identified glomeruli. Left: dorsal region of the OB. Right: the lateral region of the OB. Error bars are SEM (AA, heptanol, and hexanal 5 animals; all others 6 animals). (C) Across-glomerular response pattern similarities, assessed by Pearson correlations across glomerular response amplitudes. Left: for the dOB. Right: for the lOB (AA, heptanol, and hexanal 5 animals; all others 6 animals). (D) Average z-scores of glomerular response amplitudes (relative to pre-odor breathing response amplitudes). z-Scores are organized relative to MV. Left represents the dorsal region of the OB. Right: the lateral region of the OB (1 animal, 3 trials, same trials as shown in 2A). Underlying data for this figure can be found in S1 Data. AA, amyl acetate; dOB, dorsal olfactory bulb; lOB, lateral olfactory bulb; MV, methyl valerate; OB, olfactory bulb. https://doi.org/10.1371/journal.pbio.3000409.g002 In the dOB, all odors except heptanone clustered around intermediate anterior and intermediate mid-lateral locations. Correlations of heptanone glomerular response amplitudes with their location along the A-P dimension demonstrate a posteriorly dominated OB response. AA and carvone had no clear dominance along the M-L dimension. Heptanone was most dominant in the lateral dOB and MV in the medial region of the dOB (Fig 2B left, n = 5–6 animals, S2A Table). The odorants chosen for this study consist of a range of key molecular features, and the activation patterns of glomeruli for these odors differed across the dOB in both the A-P and M-L dimension. Glomerular odor patterns in the lOB were all dominant in the dorsal and anterior regions, thus showing similar coarse spatial organization, in contrast to the differences in their spatial patterns of the dOB (Fig 2B right, n = 5–6 animals, S2B Table). No odor dominated the ventral or posterior region of the lOB. The contrasting activation patterns of these odorants between the dOB and lOB suggest that the receptors within the epithelial zones may be relaying different chemical information for the same odor. We have highlighted the different spatial activation patterns across odors; however, their patterns can be similar in glomerular activation as ORs can transduce multiple odors [7]. We further investigated the correlations of glomerular responses across odors within the dOB (Fig 2C left, n = 5–6 animals) and the lOB (Fig 2C right, n = 5–6 animals), irrespective of location (the across-glomerular response patterns). These correlation matrices looked rather similar in general, in particular low correlations with heptanol response patterns (−0.21 to 0.34 and 0.64) amid otherwise mostly intermediate-high correlations, but some differences were clear. Glomerular activation was strong in the posterior region of the dOB in response to heptanone and strong in the anterior region for hexanal. Nevertheless, fairly high correlations between their glomerular response patterns highlighted that these odors activate similar glomeruli, suggesting that similarities in their odor structure are also being conveyed across the dOB (S3A Table). AA highly correlated with all odors except heptanol, demonstrating that a large number of the same glomeruli are being activated across odors. In the lateral bulb, there was a narrower range of correlations between glomerular activations across odors (r = −0.06 to 0.74; S3B Table) than in the dOB (r = −0.21 to 0.86). The most similar response patterns in the dOB (AA-MV, r = 0.86) were much less similar in the lOB (r = 0.41, Fig 2C right). The two odors with a weak negative correlation in the dOB (heptanone and heptanol) are also weakly negatively correlated in the lOB. In contrast, carvone and heptanol are weakly correlated in the dOB and more correlated in the lOB. The degree of correlation similarity between the dOB and lOB appears to be odor dependent and highlights differences in overlapping receptor activation patterns for these odors when comparing the dOB (zone 1) and the lOB (zone 2–4). To compare the response amplitudes of individual glomeruli across odors, the responses were z-scored relative to the standard deviation (SD) and mean of the pre-odor breathing responses. Glomerular responses in Fig 2A were organized relative to MV from high to low z-scored odor responses for both the dOB (Fig 2D left, n = 1 animal, 3 trials) and lOB (Fig 2D right, n = 1 animal, 3 trials). This demonstrates many highly significant odor responses (z > 3) and further shows the basis of the correlations of the odor responses (Fig 2C), highlighting the similarity in glomerular activation patterns across odors. We have shown here the distinct spatial odor patterns of the dOB and lOB, where their different chemotopy offers an insight into the integration of OR activation across all epithelial zones. Dorsal and lateral glomeruli response dynamics Glomerular activation can evolve during a single odor sniff cycle [19, 36], and temporal glomerular responses can be used to facilitate odor coding [37]. The temporal dynamics of the glomerular responses within the first sniff after odor delivery were determined using T90 values (Methods and S1B Fig). T90 is the time from the start of inhalation to 90% of the peak amplitude. T90 was based on a double sigmoidal fit and is more robust than T50 and T20, due to the lower impact of sampling jitter. Orthonasal airflow within the nasal cavity develops from the central domain of the dorsal meatus, which is associated with dOB projections, to the medial and lateral recesses of the ethmoid turbinates, having OSN projections to the lOB [38]. We examined the temporal activation differences between the dOB and lOB across odors using average T90 responses from all glomeruli (S2A and S2B Fig). Temporal response latencies differed strongly across odors (P < 0.0001, F(5, 2108) = 131.1), which also varied by region (P < 0.0001, F(5, 2108) = 6.26; interaction). A small but significant difference was observed in the overall temporal responses between the dOB and the lOB (P < 0.0024; F(1, 2108) = 9.22; two-way ANOVA [odor × OB region]) (S1C Fig). In addition to average temporal responses, we also examined the temporal patterning across the OB spatial dimensions. The spatiotemporal dynamics of dorsal glomerular activation after odor presentation has a stereotypical progression from the posterior region of the dOB to the anterior region [19, 39]. We used linear correlations of T90 of glomeruli with their location along each spatial dimension (as performed in Fig 2B for response amplitude, here T90) to examine the global spatial organization of T90 of all the glomeruli. This spatiotemporal correlation analysis was performed on both the pre-odor response (breathing clean air) and the first sniff after odor presentation. This analysis would highlight any temporal dynamics associated with the mechanosensation of breathing compared to the odor delivered (S5A and S5B Table). AA and MV temporal patterns differed across dorsal glomeruli (Fig 3A, 1 animal, 3 trials, all odors are shown in S2 and S3 Figs). For both odors, the pre-odor temporal dynamics were in the posterior region of the dOB. Upon AA presentation, the spatiotemporal dynamics shift from posterior to anterior (pre odor versus odor: A-P P = 0.03, M-L P = 0.80, n = 5 mice) (Fig 3B left). Upon MV presentation, spatiotemporal dynamics showed a larger shift (Fig 3B right) (pre odor versus odor: A-P P = 0.0017, M-L P = 0.01, n = 6 mice) with faster responses at the posterior-lateral region and slower responses (higher T90 responses) at the anterior-medial region. Spatiotemporal dynamics were slower in the anterior-medial region for 5 of the 6 odors (Fig 3C). T90 responses in the posterior dOB and the anterior dOB strongly depend on dOB region (P < 0.0001, F(1, 1278) = 144.5) and odor (P < 0.0001, F(5, 1278) = 141.1; two-way ANOVA [odor × dOB region]) (S1D Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Temporal dynamics of the dOB and lOB. (A) Color-scaled T90 of responses by ROIs for AA and MV. Top: the dorsal region of the OB. Bottom: the lateral region of the OB (1 animal). (B) Comparison of spatiotemporal odor dynamics, i.e., correlation between T90s of glomeruli and their location along each dimension across the dOB of AA (n = 5 mice) and MV (n = 6 mice) from pre-odor (breathing response to clean air) to odor onset (1st odor response peak). Arrow represents the direction from pre-odor to odor on. Top: the dorsal region of the OB, data are represented in both the A-P directions and the M-L as shown in panel A. Bottom: the lateral region of the OB, data are represented in both the A-P directions and the D-V as shown in panel A. Error bars are SEM. (C) Spatiotemporal odor map dynamics for all odors. Top: the dorsal region of the OB. Bottom: the lateral region of the OB (AA, heptanol, and hexanal 5 animals; all others 6 animals). (D) T90 correlations with odor amplitude (ΔF/F). Left: the dorsal region of the OB. Right: the lateral region of the OB. Error bars are SEM (AA, heptanol, and hexanal 5 animals; all others 6 animals). (E) Left: correlation of glomerular response amplitudes to MV sorted from anterior to posterior in the dOB (r2 = 0.24, slope = −0.08 ± 0.02, 1 animal, 57 glomeruli). Right: correlation of glomerular T90 responses to MV sorted from anterior to posterior in the dOB (r2 = 0.18, slope = −0.0002 ± 0.0009, 1 animal, 57 glomeruli). (F) Left: amplitude of glomeruli represented on a color scale. Red indicates high responses and blue low. Right: T90 of glomeruli represented on a color scale. Red indicates the slowest responses and blue the fastest (1 animal [same as panel E], 57 glomeruli). Underlying data for this figure can be found in S1 Data. A-P, anterio-posterior; AA, amyl acetate; D-V, dorso-ventral; dOB, dorsal olfactory bulb; lOB, lateral olfactory bulb; M-L, medio-lateral; MV, methyl valerate; OB, olfactory bulb; ROI, region of interest. https://doi.org/10.1371/journal.pbio.3000409.g003 In the lOB, slower responses were observed in the ventral region for both AA and MV (pre odor versus odor: A-P P = 0.68, D-V P = 0.11, n = 5 mice; pre odor versus odor: A-P P = 0.57, D-V P = 0.05, n = 6 mice) (Fig 3A and Fig 3B bottom). Spatiotemporal dynamics were slower in the ventral region for 5 of the 6 odors (Fig 3C). Similar to the dOB, the responses in the dorsal and ventral lOB are strongly influenced by region (P < 0.0001, F(1, 818) = 91.33) and odor (P < 0.0001, F(5, 818) = 40.67; two-way ANOVA [odor × lOB region]) (S1E Fig). The spatiotemporal dynamics of the lOB odor responses, but not clean air breathing responses, hence progress in the D-V pattern whereby A-P shift was odor dependent. These were very different from the posterior-lateral to anterior-medial odor evoked dynamics simultaneously observed in the dOB. The dynamics are in line with the airflow progression throughout the nasal cavity and the corresponding zones within the bulb [15]. The spatiotemporal dynamics were odor dependent and were not present in pre-odor inhalation-based (mechanosensory) dynamics. Correlation analysis of odor response amplitudes and T90s across glomeruli, averaged across 5–6 mice, suggests that the spatiotemporal patterns were also not predictable by the response amplitude of each odor in the dOB (Fig 3D left, S6 Table). This was in line with previous studies that have demonstrated that response latencies are not exclusively determined by the response amplitude of the glomeruli [19, 39]. However, in the lOB odor response, amplitudes and T90s were negatively correlated for all odors (Fig 3D Right, S6 Table), suggesting that their interaction is predominant in the lOB (though noting absence of strong responses ventrally). The spatial representations of all odors in the lOB were tightly clustered in the dorsal region and thus consistently negatively correlated with T90. To show that stronger responses do not by necessity result in faster responses, as appears to be the case for the lOB (Fig 3D), we show in Fig 3E and 3F that, for MV, the highest response amplitudes were in the anterior region of the dOB, and this was also the region of slowest responses (1 animal, 3 trials, 57 glomeruli). We demonstrate here that the spatiotemporal dynamics emerge in the dOB and lOB only as a result of odor stimulation, yet this is not a meta-effect of OSN odor response amplitude nor mechanosensation per se. The temporal dynamics of the lOB differ from those of the dOB and may provide additional temporal information relating to the OR projections from the nasal cavity and the airflow and sorption patterns of odor intake. Glomerular odor concentration dependence At low odor concentrations, only ORs with the highest affinity for a given odor will respond. With increasing odor concentration, additional glomeruli with lower affinity for the odor are activated [12, 39]. This recruitment of glomeruli for OSN input in response to odor concentration has been widely studied in the dOB, but not in lOB [12, 16, 25, 40]. Therefore, we explored the level of glomerular recruitment in the lOB. We compared the glomerular response patterns of two odor concentrations (0.1% and 1% [s.v.]) (Fig 4A and 4B, blue dots indicating low response amplitude and red dots high response amplitudes). Recruitment of glomeruli by higher odor concentration (Fig 4C top) was determined by whether the response amplitude to the first odor sniff was significantly above pre-odor breathing response amplitudes (P < 0.01). The number of glomeruli activated by the 1% odor concentration was normalized to 100%, and we report for the 0.1% concentration the number of activated glomeruli relative to 1% s.v. (S7 Table). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Glomerular odor concentration dependence. (A) Color-scaled response amplitudes for chosen ROIs for 0.1% heptanone. Top: the dorsal region of the OB. Bottom: the lateral region of the OB. (B) Color-scaled responses amplitude for chosen ROIs for 1% heptanone. Top: the dorsal region of the OB. Bottom: the lateral region of the OB. (C) Percent of glomeruli that are significantly activated by 0.1% (AA, heptanol, hexanal, and MV 5 animals; all others 6 animals) compared to the higher concentration of 1% (AA, heptanol, and hexanal 5 animals; all others 6 animals). The number of responding glomeruli to the higher odor concentration has been normalized to 100%. Top: the dorsal region of the OB. Bottom: the lateral region of the OB. Statistics represent two-way ANOVA (odor × concentration) per OB region with Bonferroni’s multiple comparisons test for concentration effect. Error bars are SEM. *P < 0.05, **P < 0.01, ***P <0.001, ****P < 0.0001. (D) Average ΔF/F responses for odors comparing 0.1% and 1% concentration. Top: the dorsal region of the OB. Bottom: the lateral region of the OB. Statistics as in panel C. Additional statistics represent two-way ANOVA (odor × OB region) with Bonferroni’s multiple comparisons test for OB region effect at 1% s.v. ♦P < 0.05, ♦♦P < 0.01, ♦♦♦P <0.001, ♦♦♦♦P < 0.0001. (E) Odor-induced activation maps for 0.1% heptanone presentation in the dorsal region of the OB, arrows indicate 4 glomeruli. Left: during odor presentation. Right: after the odor is removed by a vacuum. Arrows indicate the same glomeruli as left. (F) Glomerular response traces for the 4 glomeruli in panel E. Vertical blue bars indicate response and reference frames to compute response maps. (G) Odor-induced activation maps for 0.1% heptanone presentation in the lateral region of the OB, arrows indicate 7 glomeruli. Left: during odor presentation. Arrows show 4 of the glomeruli. Right: after the end of odor presentation. Arrows represent 3 additional glomeruli. (H) Glomerular response traces for the 7 glomeruli in panel F. Glomeruli 5–8 (reds) respond during odor presentation, and glomeruli 9–11 respond when the odor is removed by vacuum (blues). I–L same as E–H but for 1% heptanone. Underlying data for this figure can be found in S1 Data. AA, amyl acetate; MV, metyl valerate; OB, olfactory bulb; ROI, region of interest. https://doi.org/10.1371/journal.pbio.3000409.g004 We confirmed that the higher odor concentration recruited additional glomeruli in both the dOB (P < 0.0001, F (1, 46) = 79.68, two-way ANOVA [odor × concentration]) and the lOB (P < 0.0001, F (1, 50) = 80.14). Recruitment did not consistently differ across odors (P = 0.43, F(5, 48) = 0.97, two-way ANOVA [odor × OB region]) or between the lOB and dOB (P = 0.74, F(1, 48) = 0.11) nor did it show a significant interaction between odor and OB region (P = 0.54, F(5, 48) = 0.81). We confirmed also that the higher odor concentration evoked stronger glomerular responses in both the dOB (P < 0.0001, F (1, 2540) = 929, two-way ANOVA [odor × concentration]) and the lOB (P < 0.0001, F (1, 1638) = 923.1), which depended on the odor (interaction: P < 0.0001, F (5, 2540) = 95.26 and P < 0.0001, F (5, 1638) = 123.5, respectively). Average response amplitudes of glomeruli were significantly larger in the dOB than lOB across all odors at 1% s.v. (P < 0.0001, F(1, 2108) = 89.32, two-way ANOVA [odor × OB region]; S8 Table), suggesting the dOB is generally more responsive than the lOB (Fig 4D). Amplitude significantly differed across odors (P < 0.0001, F(5, 2108) = 215), and the interaction between odor and region was significant (P < 0.0001, F(5, 2108) = 9.45). For 0.1% concentration, a small subset of glomeruli in the ventral region of the lOB responded after odor delivery. This activation of glomeruli after odor delivery was not observed in the dOB. We explored 4 glomeruli in response to 0.1% heptanone in the dOB (Fig 4E) alongside the %ΔF/F responses of these glomeruli (Fig 4F). In the lOB, we examined 7 glomeruli (Fig 4G) of which 4 responded to the odor and 3 responded post odor (Fig 4H). This phenomenon was not observed in trials using the higher 1% odor concentration. The same glomeruli were compared (Fig 4I and 4J), and no activation post odor was observed (Fig 4K and 4L). Comparison of all glomeruli across the 6 odors in the lOB determined that during trials presenting 0.1% odor, 50.6% of glomeruli (n = 149 glomeruli) had a higher amplitude in response to odor removal (just after vacuum onset) than to odor onset (just after vacuum was turned off). For 1% odor trials, only 9.6% of these glomeruli (n = 149 glomeruli) had higher amplitude response to odor removal than to odor onset. This demonstrates a subset of glomeruli that respond post odor primarily in low concentration presentations. The lOB displays unique mechanosensitive activation Previous studies have demonstrated that OSNs can sense two modalities, chemical and mechanical. Mechanical stimulation enhances the response of OSNs to weak stimulation of odorants [41, 42], and a loss of mechanosensation impairs phase coding in mitral/tufted cells [43]. During low odor concentrations, individual glomeruli in the lateral bulb were responsive at two key points during the trial, namely, during odor delivery and post odor during removal by a vacuum (Fig 4E–4H). We next investigated whether this glomerular response to post–odor-vacuum onset was due to odor removal (change in chemical environment) or due to air-flow–related pressure change (mechanosensation). We replaced odor delivery with clean air and used the same vacuum for air removal in this subexperiment in an additional 3 mice. From these trials, we show the fluorescent response (% ΔF/F) for two glomeruli (Fig 5A) over the 12-s trial period during air presentation and vacuum. One glomerulus was chosen from the dorsal region and one from the ventral region of the lOB (Fig 5B), a region we previously showed to be responsive to vacuum (Fig 4G and 4H). During the initial 0.5 L/min air delivery and 2.5 L/min vacuum period, both glomeruli displayed stereotypical breathing responses (Fig 5A) [44]. These were mostly absent in the ventral glomerulus during air delivery (i.e., vacuum off), while the dorsal glomerulus remained unchanged. When the vacuum flow was reintroduced, the ventral glomerulus was strongly activated. We investigated a range of clean air flow rates (0.5, 0.25, 0.1, 0.05, and 0.005 L/min) and vacuum flow rates (2.5, 1.25, and 0 L/min) to examine whether they were affecting the glomerular responses. The dorsal glomerulus remained unresponsive across clean air flow rates; however, the ventral glomerulus amplitude response changed with both the air and vacuum flow rate (Fig 5A). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. The lOB displays mechanosensitive activation beyond breathing. (A) Raw traces (% ΔF/F) of two glomerular breathing responses during presentation of clean air flow rates (0.5, 0.25, 0.1, 0.05, and 0.005 L/min) and vacuum rates (2.5, 1.25, and 0 L/min) and room air with no external flows (1 animal, 2 glomeruli). These are presented as examples of the trial stimulus; green represents a glomerulus in the dorsal region of the lOB, blue in the ventral (maps shown in panel B). Red bar illustrates the period in which only the air is presented (“Vacuum off”), and grey is when the vacuum is also turned on (“Vacuum on”); height indicates flow rates relative to other trials. Vertical blue bars indicate response and reference frames to compute response maps. (B1, vacuum off response; B2, vacuum on response; maps shown in panel B). (B) Activation map in response to inspiration during only clean air flow rate of 0.5 L/min (Left top) and additionally the vacuum 2.5 L/min (Left bottom). Arrows indicate the ROI displayed in panel A. B1 and B2 indicate the respective time points in panel A (0.5 L/min airflow: vacuum 2.5 L/min). Right: activation map of all glomeruli chosen to clean air flow rate of 0.5 L/min (Top) and vacuum 2.5 L/min (Bottom). (C) The z-scores of the lOB glomerular responses during only clean air flow (red) and also vacuum (grey), organized from dorsal to ventral, for different clean air flow rates (0.5, 0.25, 0.1, 0.05, and 0.005 L/min) and vacuum rates (2.5, 1.25, and 0 L/min) and room air (3 animals, 47 glomeruli). Linear correlation fits are indicated. (D) Top left: histogram of the ratio of vacuum off responses relative to room air breathing responses in the lateral bulb. Top right: histogram of the ratio of vacuum on breathing responses relative to room air breathing responses in the lateral bulb (3 animals, 47 glomeruli, total 329 glomerular responses across all flow rates). Number of responses (n) was divided into number of glomerular responses <1.6 and >1.6 times above the room air breathing response. Bottom: same as top but in the dorsal bulb. Underlying data for this figure can be found in S1 Data. lOB, lateral olfactory bulb; OB, olfactory bulb; ROI, region of interest. https://doi.org/10.1371/journal.pbio.3000409.g005 The z-scores of glomerular responses to both air and vacuum across the D-V spatial axis (Fig 5C) were used to examine whether there was a spatial organization of the lOB glomerular activation. During airflow (0.5 L/min), glomerular responses did not show a D-V dominance; however, during vacuum (2.5 L/min), large responses consistently corresponded with the ventral region of the lOB (S9 Table). When the air (0.005 L/min) or vacuum (0 L/min) flow rates were lowest, the responses of the ventral glomeruli were similar to that observed when the trial stimulus was breathing room air (“room air”). This was surprising, as their respective vacuum (2.5 L/min) and airflow (0.5 L/min) rates were high. This suggests that it was not a single stimulus that the ventral glomerulus is responding to but some nonlinear combination of both the clean air and vacuum flow rates. We show the equivalent data for the dOB in S5 Fig, where it is clear that dOB glomeruli lack this sensitivity. The magnitude of the glomerular responses relative to breathing room air was determined across all stimulus conditions (Fig 5D). During clean air presentation, most lOB glomeruli had a response below breathing amplitudes. However, during vacuum, there was a bimodal distribution of glomerular responses with an initial distribution similar to air, and an additional distribution up to 3 times the room air breathing response. This was not observed in the dOB. We next examined whether this response was related to flow rate or the change in pressure. Even though the flow rates are similar, the differential pressure change is not (S4A and S4B Fig), and therefore the flow rates cannot explain the consistently varying responses (Fig 6A). The vast difference in responsiveness of dOB and lOB glomeruli to these stimuli is clearly shown in S4C and S4D Fig. We show that the lOB responds positively to a negative pressure change (vacuum on) and negatively to a positive pressure change (vacuum off) (Fig 6C) primarily in the ventral lOB and more weakly in the dorsal lOB (Fig 6D). This, however, was not observed in the dOB (Fig 6E and 6F). In the lOB, a single third-order polynomial can accurately capture the glomerular responses across pressure change values, due to the large uneven order terms to capture the response asymmetry around x = 0. In the dOB, however, the responses can only adequately be captured by a separate fit for the negative and positive pressure changes. Our results show a unique set of glomeruli in the ventral region of the lOB that respond to the change in pressure of air and vacuum, a phenomenon not previously observed. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. The lOB is sensitive to differential pressure. (A) Mean z-scored OB responses to different flow rates in the ventral lOB. (B) Mean z-scored OB responses to different flow rates in the dOB. (C) Mean z-scored OB responses to different pressure changes in the lOB. (D) Mean z-scored OB responses to different pressure changes in the lOB. Left: the ventral lOB. Right: the dorsal lOB. (E) Mean z-scored OB responses to different pressure changes in the dOB. (F) Mean z-scored OB responses to different pressure changes in the dOB. Left: the anterior dOB. Right: the posterior dOB: 3 animals; dOB: 99 glomeruli; and lOB: 47 glomeruli. Graphs are reorganized from S4C and S4D Fig, all offset by respective z-scores for no–flow-change condition (0–0, room; see Methods). Only the lOB—and the vlOB in particular—shows a strong and positive response to an intermediary drop in pressure at the odor delivery tube. The dOB only shows mild response suppression, particularly by intermediate changes in pressure. Pressure unit is relative only. Underlying data for this figure can be found in S1 Data. dOB, dorsal olfactory bulb; lOB, lateral olfactory bulb; OB, olfactory bulb; vlOB, https://doi.org/10.1371/journal.pbio.3000409.g006 Dual imaging of the dOB and lOB Here, we have developed a dual imaging approach (Fig 1A) to measure the glomerular activity of both the dorsal and lateral regions of the OB. We investigated OR sensory neuron input using floxed-GCaMP6f reporter mice crossed with OMP-Cre animals [12] (Fig 1B). In addition to implementing classical dorsal region windowing of the OB (Fig 1C top), we also successfully exposed the lateral region of the OB by unilateral enucleation and preparation of an optical window medial to the eye (Fig 1C bottom). Both cameras were synchronized for simultaneous fluorescent imaging of these two regions of the OB. These two orthogonal imaging macroscopes had overlapping imaging planes in the lateral region of the dOB (Fig 1D). Four glomeruli confirmed excellent agreement in the glomerular fluorescence responses imaged by the dorsal and lateral camera (Fig 1E). Using frame subtraction before and after the first odor inhalation, regions of interest (ROIs) were accumulated (i.e., an ROI that responds to at least one stimulus) across odors in both the lateral and dorsal images (Fig 1F), and their fluorescence (% ΔF/F) response traces were isolated and analyzed (Fig 1G). We examined approximately 1.5 times the number of glomeruli in the dOB as in the lOB (dorsal: 232, lateral: 149 across 6 animals; S1 Table). For the first time, we simultaneously imaged OSN calcium activity in the dorsal (dOB) and lateral (lOB) regions of the OB. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Dual imaging of the dOB and lOB in GCaMP6f reporter mice. (A) Schematic representation of the dual microscope imaging setup. Illustrating the plane of the dorsal and lateral images. Inset: schematic of the odor delivery assembly relative to the nostrils, illustrating the odor and vacuum flow channels. (B) Schematic of the mouse genotype. (C) Top: resting light image of the dOB. Bottom: resting light image of the lateral region of the OB. (D) Top: odor-induced activation map of the dOB. Arrows highlight 4 individual glomeruli. Bottom: odor-induced activation map of the lOB. Arrows point to the same 4 glomeruli as top. (E) Left: glomerular response traces from the 4 glomeruli imaged with the dorsal camera in response to 1% heptanone. Middle: responses from the 4 glomeruli imaged with the lateral camera in response to 1% heptanone. Right: glomerulus 4 imaged from the dorsal and lateral camera, overlaid. (F) Top: an example of the ROI selection on the dorsal bulb for 1 animal. Bottom: an example of the ROI selection for the lateral bulb for 1 animal. (G) Top: % ΔF/F responses for the dorsal ROIs in panel F (top) selection of first odor response is indicated (1 animal). Middle: % ΔF/F responses for the lateral ROIs in panel F (Bottom) (1 animal). Bottom: a sniffing trace from the same animal. Inhalation is the upward inflection (see Methods for details). Underlying data for this figure can be found in S1 Data. dOB, dorsal olfactory bulb; lOB, lateral olfactory bulb; OB, olfactory bulb; ROI, region of interest. https://doi.org/10.1371/journal.pbio.3000409.g001 Simultaneous odor mapping of the dOB and lOB OSNs expressing different receptors that recognize related odor molecules project to neighboring glomeruli in the OB [33]. Therefore, there is chemotopic organization of the glomerular responses represented in spatial odor patterns [10, 34]. Differences in the spatial activation patterns of glomeruli are thought to play a primary role in identifying odors [10, 35]. Using our dual imaging approach, we investigated spatial odor patterns of glomerular OSN input (OMP-GCaMP6f mice) responses over the first respiration cycle in both the dOB and lOB (Fig 2A 1 animal, 3 trials). Six odors—amyl acetate (AA), carvone, heptanol, heptanone, hexanal, and methyl valerate (MV)—at 1% (saturated vapor [s.v.]) concentration were used. These odors were chosen to represent different molecular groups, as well as to be represented in the lateral bulb (though without concern where in the lOB) [10, 21, 24]. The global spatial organization of activation of glomeruli in the anterio-posterior (A-P), dorso-ventral (D-V), or medio-lateral (M-L) dimensions was determined using linear correlations of glomerular response amplitudes with their spatial location (see Methods and S1A Fig for examples). All these analyses were based on images of 512 × 512 pixels, where lower pixel numbers along the x-axis represent more anterior regions in the dOB and lOB. A fixed set of ROIs, used in every analysis, was accumulated across odor maps for each mouse, where each ROI responded to at least one odorant. For laterality of the dOB, a low pixel number along the y-axis represented more medial locations of ROIs. Each hemi-bulb was analyzed separately, and only the hemi-bulb ipsilateral to the exposed lOB is shown here. For the lOB, a lower y-axis pixel number represented more dorsal regions. Correlation analysis of glomerular responses with spatial location provides a simple metric that highlighted the differences in global response patterns for each odor (%ΔF/F responses of all odors shown in S2 Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Simultaneous odor mapping of the dOB and lOB. (A) Odor-induced activation maps for AA, carvone, heptanol, heptanone, hexanal, and MV. Top row: the dorsal region of the OB. Bottom row: the lateral region of the OB (1 animal, average of 3 trials). (B) Spatial odor map correlations between response amplitudes and the location along each spatial dimension of all identified glomeruli. Left: dorsal region of the OB. Right: the lateral region of the OB. Error bars are SEM (AA, heptanol, and hexanal 5 animals; all others 6 animals). (C) Across-glomerular response pattern similarities, assessed by Pearson correlations across glomerular response amplitudes. Left: for the dOB. Right: for the lOB (AA, heptanol, and hexanal 5 animals; all others 6 animals). (D) Average z-scores of glomerular response amplitudes (relative to pre-odor breathing response amplitudes). z-Scores are organized relative to MV. Left represents the dorsal region of the OB. Right: the lateral region of the OB (1 animal, 3 trials, same trials as shown in 2A). Underlying data for this figure can be found in S1 Data. AA, amyl acetate; dOB, dorsal olfactory bulb; lOB, lateral olfactory bulb; MV, methyl valerate; OB, olfactory bulb. https://doi.org/10.1371/journal.pbio.3000409.g002 In the dOB, all odors except heptanone clustered around intermediate anterior and intermediate mid-lateral locations. Correlations of heptanone glomerular response amplitudes with their location along the A-P dimension demonstrate a posteriorly dominated OB response. AA and carvone had no clear dominance along the M-L dimension. Heptanone was most dominant in the lateral dOB and MV in the medial region of the dOB (Fig 2B left, n = 5–6 animals, S2A Table). The odorants chosen for this study consist of a range of key molecular features, and the activation patterns of glomeruli for these odors differed across the dOB in both the A-P and M-L dimension. Glomerular odor patterns in the lOB were all dominant in the dorsal and anterior regions, thus showing similar coarse spatial organization, in contrast to the differences in their spatial patterns of the dOB (Fig 2B right, n = 5–6 animals, S2B Table). No odor dominated the ventral or posterior region of the lOB. The contrasting activation patterns of these odorants between the dOB and lOB suggest that the receptors within the epithelial zones may be relaying different chemical information for the same odor. We have highlighted the different spatial activation patterns across odors; however, their patterns can be similar in glomerular activation as ORs can transduce multiple odors [7]. We further investigated the correlations of glomerular responses across odors within the dOB (Fig 2C left, n = 5–6 animals) and the lOB (Fig 2C right, n = 5–6 animals), irrespective of location (the across-glomerular response patterns). These correlation matrices looked rather similar in general, in particular low correlations with heptanol response patterns (−0.21 to 0.34 and 0.64) amid otherwise mostly intermediate-high correlations, but some differences were clear. Glomerular activation was strong in the posterior region of the dOB in response to heptanone and strong in the anterior region for hexanal. Nevertheless, fairly high correlations between their glomerular response patterns highlighted that these odors activate similar glomeruli, suggesting that similarities in their odor structure are also being conveyed across the dOB (S3A Table). AA highly correlated with all odors except heptanol, demonstrating that a large number of the same glomeruli are being activated across odors. In the lateral bulb, there was a narrower range of correlations between glomerular activations across odors (r = −0.06 to 0.74; S3B Table) than in the dOB (r = −0.21 to 0.86). The most similar response patterns in the dOB (AA-MV, r = 0.86) were much less similar in the lOB (r = 0.41, Fig 2C right). The two odors with a weak negative correlation in the dOB (heptanone and heptanol) are also weakly negatively correlated in the lOB. In contrast, carvone and heptanol are weakly correlated in the dOB and more correlated in the lOB. The degree of correlation similarity between the dOB and lOB appears to be odor dependent and highlights differences in overlapping receptor activation patterns for these odors when comparing the dOB (zone 1) and the lOB (zone 2–4). To compare the response amplitudes of individual glomeruli across odors, the responses were z-scored relative to the standard deviation (SD) and mean of the pre-odor breathing responses. Glomerular responses in Fig 2A were organized relative to MV from high to low z-scored odor responses for both the dOB (Fig 2D left, n = 1 animal, 3 trials) and lOB (Fig 2D right, n = 1 animal, 3 trials). This demonstrates many highly significant odor responses (z > 3) and further shows the basis of the correlations of the odor responses (Fig 2C), highlighting the similarity in glomerular activation patterns across odors. We have shown here the distinct spatial odor patterns of the dOB and lOB, where their different chemotopy offers an insight into the integration of OR activation across all epithelial zones. Dorsal and lateral glomeruli response dynamics Glomerular activation can evolve during a single odor sniff cycle [19, 36], and temporal glomerular responses can be used to facilitate odor coding [37]. The temporal dynamics of the glomerular responses within the first sniff after odor delivery were determined using T90 values (Methods and S1B Fig). T90 is the time from the start of inhalation to 90% of the peak amplitude. T90 was based on a double sigmoidal fit and is more robust than T50 and T20, due to the lower impact of sampling jitter. Orthonasal airflow within the nasal cavity develops from the central domain of the dorsal meatus, which is associated with dOB projections, to the medial and lateral recesses of the ethmoid turbinates, having OSN projections to the lOB [38]. We examined the temporal activation differences between the dOB and lOB across odors using average T90 responses from all glomeruli (S2A and S2B Fig). Temporal response latencies differed strongly across odors (P < 0.0001, F(5, 2108) = 131.1), which also varied by region (P < 0.0001, F(5, 2108) = 6.26; interaction). A small but significant difference was observed in the overall temporal responses between the dOB and the lOB (P < 0.0024; F(1, 2108) = 9.22; two-way ANOVA [odor × OB region]) (S1C Fig). In addition to average temporal responses, we also examined the temporal patterning across the OB spatial dimensions. The spatiotemporal dynamics of dorsal glomerular activation after odor presentation has a stereotypical progression from the posterior region of the dOB to the anterior region [19, 39]. We used linear correlations of T90 of glomeruli with their location along each spatial dimension (as performed in Fig 2B for response amplitude, here T90) to examine the global spatial organization of T90 of all the glomeruli. This spatiotemporal correlation analysis was performed on both the pre-odor response (breathing clean air) and the first sniff after odor presentation. This analysis would highlight any temporal dynamics associated with the mechanosensation of breathing compared to the odor delivered (S5A and S5B Table). AA and MV temporal patterns differed across dorsal glomeruli (Fig 3A, 1 animal, 3 trials, all odors are shown in S2 and S3 Figs). For both odors, the pre-odor temporal dynamics were in the posterior region of the dOB. Upon AA presentation, the spatiotemporal dynamics shift from posterior to anterior (pre odor versus odor: A-P P = 0.03, M-L P = 0.80, n = 5 mice) (Fig 3B left). Upon MV presentation, spatiotemporal dynamics showed a larger shift (Fig 3B right) (pre odor versus odor: A-P P = 0.0017, M-L P = 0.01, n = 6 mice) with faster responses at the posterior-lateral region and slower responses (higher T90 responses) at the anterior-medial region. Spatiotemporal dynamics were slower in the anterior-medial region for 5 of the 6 odors (Fig 3C). T90 responses in the posterior dOB and the anterior dOB strongly depend on dOB region (P < 0.0001, F(1, 1278) = 144.5) and odor (P < 0.0001, F(5, 1278) = 141.1; two-way ANOVA [odor × dOB region]) (S1D Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Temporal dynamics of the dOB and lOB. (A) Color-scaled T90 of responses by ROIs for AA and MV. Top: the dorsal region of the OB. Bottom: the lateral region of the OB (1 animal). (B) Comparison of spatiotemporal odor dynamics, i.e., correlation between T90s of glomeruli and their location along each dimension across the dOB of AA (n = 5 mice) and MV (n = 6 mice) from pre-odor (breathing response to clean air) to odor onset (1st odor response peak). Arrow represents the direction from pre-odor to odor on. Top: the dorsal region of the OB, data are represented in both the A-P directions and the M-L as shown in panel A. Bottom: the lateral region of the OB, data are represented in both the A-P directions and the D-V as shown in panel A. Error bars are SEM. (C) Spatiotemporal odor map dynamics for all odors. Top: the dorsal region of the OB. Bottom: the lateral region of the OB (AA, heptanol, and hexanal 5 animals; all others 6 animals). (D) T90 correlations with odor amplitude (ΔF/F). Left: the dorsal region of the OB. Right: the lateral region of the OB. Error bars are SEM (AA, heptanol, and hexanal 5 animals; all others 6 animals). (E) Left: correlation of glomerular response amplitudes to MV sorted from anterior to posterior in the dOB (r2 = 0.24, slope = −0.08 ± 0.02, 1 animal, 57 glomeruli). Right: correlation of glomerular T90 responses to MV sorted from anterior to posterior in the dOB (r2 = 0.18, slope = −0.0002 ± 0.0009, 1 animal, 57 glomeruli). (F) Left: amplitude of glomeruli represented on a color scale. Red indicates high responses and blue low. Right: T90 of glomeruli represented on a color scale. Red indicates the slowest responses and blue the fastest (1 animal [same as panel E], 57 glomeruli). Underlying data for this figure can be found in S1 Data. A-P, anterio-posterior; AA, amyl acetate; D-V, dorso-ventral; dOB, dorsal olfactory bulb; lOB, lateral olfactory bulb; M-L, medio-lateral; MV, methyl valerate; OB, olfactory bulb; ROI, region of interest. https://doi.org/10.1371/journal.pbio.3000409.g003 In the lOB, slower responses were observed in the ventral region for both AA and MV (pre odor versus odor: A-P P = 0.68, D-V P = 0.11, n = 5 mice; pre odor versus odor: A-P P = 0.57, D-V P = 0.05, n = 6 mice) (Fig 3A and Fig 3B bottom). Spatiotemporal dynamics were slower in the ventral region for 5 of the 6 odors (Fig 3C). Similar to the dOB, the responses in the dorsal and ventral lOB are strongly influenced by region (P < 0.0001, F(1, 818) = 91.33) and odor (P < 0.0001, F(5, 818) = 40.67; two-way ANOVA [odor × lOB region]) (S1E Fig). The spatiotemporal dynamics of the lOB odor responses, but not clean air breathing responses, hence progress in the D-V pattern whereby A-P shift was odor dependent. These were very different from the posterior-lateral to anterior-medial odor evoked dynamics simultaneously observed in the dOB. The dynamics are in line with the airflow progression throughout the nasal cavity and the corresponding zones within the bulb [15]. The spatiotemporal dynamics were odor dependent and were not present in pre-odor inhalation-based (mechanosensory) dynamics. Correlation analysis of odor response amplitudes and T90s across glomeruli, averaged across 5–6 mice, suggests that the spatiotemporal patterns were also not predictable by the response amplitude of each odor in the dOB (Fig 3D left, S6 Table). This was in line with previous studies that have demonstrated that response latencies are not exclusively determined by the response amplitude of the glomeruli [19, 39]. However, in the lOB odor response, amplitudes and T90s were negatively correlated for all odors (Fig 3D Right, S6 Table), suggesting that their interaction is predominant in the lOB (though noting absence of strong responses ventrally). The spatial representations of all odors in the lOB were tightly clustered in the dorsal region and thus consistently negatively correlated with T90. To show that stronger responses do not by necessity result in faster responses, as appears to be the case for the lOB (Fig 3D), we show in Fig 3E and 3F that, for MV, the highest response amplitudes were in the anterior region of the dOB, and this was also the region of slowest responses (1 animal, 3 trials, 57 glomeruli). We demonstrate here that the spatiotemporal dynamics emerge in the dOB and lOB only as a result of odor stimulation, yet this is not a meta-effect of OSN odor response amplitude nor mechanosensation per se. The temporal dynamics of the lOB differ from those of the dOB and may provide additional temporal information relating to the OR projections from the nasal cavity and the airflow and sorption patterns of odor intake. Glomerular odor concentration dependence At low odor concentrations, only ORs with the highest affinity for a given odor will respond. With increasing odor concentration, additional glomeruli with lower affinity for the odor are activated [12, 39]. This recruitment of glomeruli for OSN input in response to odor concentration has been widely studied in the dOB, but not in lOB [12, 16, 25, 40]. Therefore, we explored the level of glomerular recruitment in the lOB. We compared the glomerular response patterns of two odor concentrations (0.1% and 1% [s.v.]) (Fig 4A and 4B, blue dots indicating low response amplitude and red dots high response amplitudes). Recruitment of glomeruli by higher odor concentration (Fig 4C top) was determined by whether the response amplitude to the first odor sniff was significantly above pre-odor breathing response amplitudes (P < 0.01). The number of glomeruli activated by the 1% odor concentration was normalized to 100%, and we report for the 0.1% concentration the number of activated glomeruli relative to 1% s.v. (S7 Table). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Glomerular odor concentration dependence. (A) Color-scaled response amplitudes for chosen ROIs for 0.1% heptanone. Top: the dorsal region of the OB. Bottom: the lateral region of the OB. (B) Color-scaled responses amplitude for chosen ROIs for 1% heptanone. Top: the dorsal region of the OB. Bottom: the lateral region of the OB. (C) Percent of glomeruli that are significantly activated by 0.1% (AA, heptanol, hexanal, and MV 5 animals; all others 6 animals) compared to the higher concentration of 1% (AA, heptanol, and hexanal 5 animals; all others 6 animals). The number of responding glomeruli to the higher odor concentration has been normalized to 100%. Top: the dorsal region of the OB. Bottom: the lateral region of the OB. Statistics represent two-way ANOVA (odor × concentration) per OB region with Bonferroni’s multiple comparisons test for concentration effect. Error bars are SEM. *P < 0.05, **P < 0.01, ***P <0.001, ****P < 0.0001. (D) Average ΔF/F responses for odors comparing 0.1% and 1% concentration. Top: the dorsal region of the OB. Bottom: the lateral region of the OB. Statistics as in panel C. Additional statistics represent two-way ANOVA (odor × OB region) with Bonferroni’s multiple comparisons test for OB region effect at 1% s.v. ♦P < 0.05, ♦♦P < 0.01, ♦♦♦P <0.001, ♦♦♦♦P < 0.0001. (E) Odor-induced activation maps for 0.1% heptanone presentation in the dorsal region of the OB, arrows indicate 4 glomeruli. Left: during odor presentation. Right: after the odor is removed by a vacuum. Arrows indicate the same glomeruli as left. (F) Glomerular response traces for the 4 glomeruli in panel E. Vertical blue bars indicate response and reference frames to compute response maps. (G) Odor-induced activation maps for 0.1% heptanone presentation in the lateral region of the OB, arrows indicate 7 glomeruli. Left: during odor presentation. Arrows show 4 of the glomeruli. Right: after the end of odor presentation. Arrows represent 3 additional glomeruli. (H) Glomerular response traces for the 7 glomeruli in panel F. Glomeruli 5–8 (reds) respond during odor presentation, and glomeruli 9–11 respond when the odor is removed by vacuum (blues). I–L same as E–H but for 1% heptanone. Underlying data for this figure can be found in S1 Data. AA, amyl acetate; MV, metyl valerate; OB, olfactory bulb; ROI, region of interest. https://doi.org/10.1371/journal.pbio.3000409.g004 We confirmed that the higher odor concentration recruited additional glomeruli in both the dOB (P < 0.0001, F (1, 46) = 79.68, two-way ANOVA [odor × concentration]) and the lOB (P < 0.0001, F (1, 50) = 80.14). Recruitment did not consistently differ across odors (P = 0.43, F(5, 48) = 0.97, two-way ANOVA [odor × OB region]) or between the lOB and dOB (P = 0.74, F(1, 48) = 0.11) nor did it show a significant interaction between odor and OB region (P = 0.54, F(5, 48) = 0.81). We confirmed also that the higher odor concentration evoked stronger glomerular responses in both the dOB (P < 0.0001, F (1, 2540) = 929, two-way ANOVA [odor × concentration]) and the lOB (P < 0.0001, F (1, 1638) = 923.1), which depended on the odor (interaction: P < 0.0001, F (5, 2540) = 95.26 and P < 0.0001, F (5, 1638) = 123.5, respectively). Average response amplitudes of glomeruli were significantly larger in the dOB than lOB across all odors at 1% s.v. (P < 0.0001, F(1, 2108) = 89.32, two-way ANOVA [odor × OB region]; S8 Table), suggesting the dOB is generally more responsive than the lOB (Fig 4D). Amplitude significantly differed across odors (P < 0.0001, F(5, 2108) = 215), and the interaction between odor and region was significant (P < 0.0001, F(5, 2108) = 9.45). For 0.1% concentration, a small subset of glomeruli in the ventral region of the lOB responded after odor delivery. This activation of glomeruli after odor delivery was not observed in the dOB. We explored 4 glomeruli in response to 0.1% heptanone in the dOB (Fig 4E) alongside the %ΔF/F responses of these glomeruli (Fig 4F). In the lOB, we examined 7 glomeruli (Fig 4G) of which 4 responded to the odor and 3 responded post odor (Fig 4H). This phenomenon was not observed in trials using the higher 1% odor concentration. The same glomeruli were compared (Fig 4I and 4J), and no activation post odor was observed (Fig 4K and 4L). Comparison of all glomeruli across the 6 odors in the lOB determined that during trials presenting 0.1% odor, 50.6% of glomeruli (n = 149 glomeruli) had a higher amplitude in response to odor removal (just after vacuum onset) than to odor onset (just after vacuum was turned off). For 1% odor trials, only 9.6% of these glomeruli (n = 149 glomeruli) had higher amplitude response to odor removal than to odor onset. This demonstrates a subset of glomeruli that respond post odor primarily in low concentration presentations. The lOB displays unique mechanosensitive activation Previous studies have demonstrated that OSNs can sense two modalities, chemical and mechanical. Mechanical stimulation enhances the response of OSNs to weak stimulation of odorants [41, 42], and a loss of mechanosensation impairs phase coding in mitral/tufted cells [43]. During low odor concentrations, individual glomeruli in the lateral bulb were responsive at two key points during the trial, namely, during odor delivery and post odor during removal by a vacuum (Fig 4E–4H). We next investigated whether this glomerular response to post–odor-vacuum onset was due to odor removal (change in chemical environment) or due to air-flow–related pressure change (mechanosensation). We replaced odor delivery with clean air and used the same vacuum for air removal in this subexperiment in an additional 3 mice. From these trials, we show the fluorescent response (% ΔF/F) for two glomeruli (Fig 5A) over the 12-s trial period during air presentation and vacuum. One glomerulus was chosen from the dorsal region and one from the ventral region of the lOB (Fig 5B), a region we previously showed to be responsive to vacuum (Fig 4G and 4H). During the initial 0.5 L/min air delivery and 2.5 L/min vacuum period, both glomeruli displayed stereotypical breathing responses (Fig 5A) [44]. These were mostly absent in the ventral glomerulus during air delivery (i.e., vacuum off), while the dorsal glomerulus remained unchanged. When the vacuum flow was reintroduced, the ventral glomerulus was strongly activated. We investigated a range of clean air flow rates (0.5, 0.25, 0.1, 0.05, and 0.005 L/min) and vacuum flow rates (2.5, 1.25, and 0 L/min) to examine whether they were affecting the glomerular responses. The dorsal glomerulus remained unresponsive across clean air flow rates; however, the ventral glomerulus amplitude response changed with both the air and vacuum flow rate (Fig 5A). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. The lOB displays mechanosensitive activation beyond breathing. (A) Raw traces (% ΔF/F) of two glomerular breathing responses during presentation of clean air flow rates (0.5, 0.25, 0.1, 0.05, and 0.005 L/min) and vacuum rates (2.5, 1.25, and 0 L/min) and room air with no external flows (1 animal, 2 glomeruli). These are presented as examples of the trial stimulus; green represents a glomerulus in the dorsal region of the lOB, blue in the ventral (maps shown in panel B). Red bar illustrates the period in which only the air is presented (“Vacuum off”), and grey is when the vacuum is also turned on (“Vacuum on”); height indicates flow rates relative to other trials. Vertical blue bars indicate response and reference frames to compute response maps. (B1, vacuum off response; B2, vacuum on response; maps shown in panel B). (B) Activation map in response to inspiration during only clean air flow rate of 0.5 L/min (Left top) and additionally the vacuum 2.5 L/min (Left bottom). Arrows indicate the ROI displayed in panel A. B1 and B2 indicate the respective time points in panel A (0.5 L/min airflow: vacuum 2.5 L/min). Right: activation map of all glomeruli chosen to clean air flow rate of 0.5 L/min (Top) and vacuum 2.5 L/min (Bottom). (C) The z-scores of the lOB glomerular responses during only clean air flow (red) and also vacuum (grey), organized from dorsal to ventral, for different clean air flow rates (0.5, 0.25, 0.1, 0.05, and 0.005 L/min) and vacuum rates (2.5, 1.25, and 0 L/min) and room air (3 animals, 47 glomeruli). Linear correlation fits are indicated. (D) Top left: histogram of the ratio of vacuum off responses relative to room air breathing responses in the lateral bulb. Top right: histogram of the ratio of vacuum on breathing responses relative to room air breathing responses in the lateral bulb (3 animals, 47 glomeruli, total 329 glomerular responses across all flow rates). Number of responses (n) was divided into number of glomerular responses <1.6 and >1.6 times above the room air breathing response. Bottom: same as top but in the dorsal bulb. Underlying data for this figure can be found in S1 Data. lOB, lateral olfactory bulb; OB, olfactory bulb; ROI, region of interest. https://doi.org/10.1371/journal.pbio.3000409.g005 The z-scores of glomerular responses to both air and vacuum across the D-V spatial axis (Fig 5C) were used to examine whether there was a spatial organization of the lOB glomerular activation. During airflow (0.5 L/min), glomerular responses did not show a D-V dominance; however, during vacuum (2.5 L/min), large responses consistently corresponded with the ventral region of the lOB (S9 Table). When the air (0.005 L/min) or vacuum (0 L/min) flow rates were lowest, the responses of the ventral glomeruli were similar to that observed when the trial stimulus was breathing room air (“room air”). This was surprising, as their respective vacuum (2.5 L/min) and airflow (0.5 L/min) rates were high. This suggests that it was not a single stimulus that the ventral glomerulus is responding to but some nonlinear combination of both the clean air and vacuum flow rates. We show the equivalent data for the dOB in S5 Fig, where it is clear that dOB glomeruli lack this sensitivity. The magnitude of the glomerular responses relative to breathing room air was determined across all stimulus conditions (Fig 5D). During clean air presentation, most lOB glomeruli had a response below breathing amplitudes. However, during vacuum, there was a bimodal distribution of glomerular responses with an initial distribution similar to air, and an additional distribution up to 3 times the room air breathing response. This was not observed in the dOB. We next examined whether this response was related to flow rate or the change in pressure. Even though the flow rates are similar, the differential pressure change is not (S4A and S4B Fig), and therefore the flow rates cannot explain the consistently varying responses (Fig 6A). The vast difference in responsiveness of dOB and lOB glomeruli to these stimuli is clearly shown in S4C and S4D Fig. We show that the lOB responds positively to a negative pressure change (vacuum on) and negatively to a positive pressure change (vacuum off) (Fig 6C) primarily in the ventral lOB and more weakly in the dorsal lOB (Fig 6D). This, however, was not observed in the dOB (Fig 6E and 6F). In the lOB, a single third-order polynomial can accurately capture the glomerular responses across pressure change values, due to the large uneven order terms to capture the response asymmetry around x = 0. In the dOB, however, the responses can only adequately be captured by a separate fit for the negative and positive pressure changes. Our results show a unique set of glomeruli in the ventral region of the lOB that respond to the change in pressure of air and vacuum, a phenomenon not previously observed. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. The lOB is sensitive to differential pressure. (A) Mean z-scored OB responses to different flow rates in the ventral lOB. (B) Mean z-scored OB responses to different flow rates in the dOB. (C) Mean z-scored OB responses to different pressure changes in the lOB. (D) Mean z-scored OB responses to different pressure changes in the lOB. Left: the ventral lOB. Right: the dorsal lOB. (E) Mean z-scored OB responses to different pressure changes in the dOB. (F) Mean z-scored OB responses to different pressure changes in the dOB. Left: the anterior dOB. Right: the posterior dOB: 3 animals; dOB: 99 glomeruli; and lOB: 47 glomeruli. Graphs are reorganized from S4C and S4D Fig, all offset by respective z-scores for no–flow-change condition (0–0, room; see Methods). Only the lOB—and the vlOB in particular—shows a strong and positive response to an intermediary drop in pressure at the odor delivery tube. The dOB only shows mild response suppression, particularly by intermediate changes in pressure. Pressure unit is relative only. Underlying data for this figure can be found in S1 Data. dOB, dorsal olfactory bulb; lOB, lateral olfactory bulb; OB, olfactory bulb; vlOB, https://doi.org/10.1371/journal.pbio.3000409.g006 Discussion Here, we explore both the spatial and temporal patterning of both the chemosensory and mechanosensory input responses to the lOB and dOB in tandem for the first time. This is achieved through simultaneous imaging and expands upon the spatial and temporal information that has been extensively gleaned only from the dOB. We have not only examined how odor information is presented at the lOB but also how this compares to the dOB. In addition, we have also uncovered a unique mechanosensory response—unlike breathing—that is, interestingly, completely unique to the lOB. Spatial and temporal patterning in the OB is influenced by a number of factors such as the molecular features of the odor that enters the nasal cavity [15], the OR distribution across the olfactory epithelium (OE) [5], and the sorption of the odor into the mucus layer of the nose [13, 45, 46]. The spatial patterns of odors have been excellently described throughout the entire glomerular layer using 2-DG [20, 21, 47] and fMRI [11]. Although high–spatial-resolution maps using intrinsic and calcium imaging have primarily focused on the dOB [25, 40, 48], intrinsic imaging has previously been used to explore the lateral bulb [20, 21, 47]. Here, we expand this knowledge by investigating the presynaptic glomerular responses to odors with various molecular groups (ester, terpenoid, alcohol, ketone, alkyl aldehyde, methyl ester) simultaneously in the dOB and lOB, seamlessly evaluating a greater area of the glomerular sheet. Previously, spatial odor maps of the lOB, visualized with intrinsic imaging, defined 2 key areas of activation: one in the dorsal region and one in the ventral region [20]. It was concluded that odorants carvone, hexanal, and heptanone all had primary activation in the more ventral region, which is contrasted by our data presented here. Differences between calcium and intrinsically imaged spatial odor maps have previously been observed [40], and comparisons of calcium and intrinsic imaging have highlighted the temporal response differences between the two methods. Firstly, the half-maximum time of an intrinsic response is up to 6 times slower than calcium, and secondly, intrinsic signals saturate quickly, potentially misrepresenting glomerular odor responses for other neuronal processes [23]. Using intrinsic imaging of the lOB, nearly all odors were mapped to the ventral region, which in our study is where we observe a mechanosensory response. It is possible that within that study, the same mechanosensory responses were being observed, which could not be resolved from odor activation because the technique lacked temporal resolution. Therefore, we suggest that this current study, affording a higher temporal resolution, more accurately represents the spatial odor patterns in the lOB. Not only were we able to define the spatial odor patterns of the lOB, but we were also able to compare these to the dOB. Comparisons using the dual imaging approach has highlighted key differences between the spatial representations of odors in the dOB and lOB. For example, we highlighted that in the dOB the spatial maps of heptanone and MV were very different, while in the lOB they were very similar. On the dorsal surface, it is well established that odorants with similar molecular features group together [25, 40], and chemotopic organization has been shown in the lateral bulb [47]. Studies have also suggested that there only exists a coarse chemotopic organization with the dorsal bulb [49]. Carbon chain length has been shown to shift spatial maps in the dOB using intrinsic imaging [48]; however, using calcium imaging, little effect was observed [40]. Using fMRI, carbon chain length has been shown to shift spatial patterns in the lOB [11]. Previous spatial mapping of the lOB using intrinsic imaging has shown that the hydrocarbon skeleton is represented in the lOB [20]. In our case, heptanone, a ketone, and MV, an ester, only differ by one carbon in length. In line with these suggestions, the spatial maps of heptanone and MV may differ greatly dorsally because of differing functional groups but map similarly laterally because their carbon chain length is similar, although we admit our odor array is neither sufficiently large nor systematically diverse to strongly address the issue of chemotopy. Thus, the integration of information across the epithelium zones may dramatically increase the amount of information processed for a given odor and the subsequent differences between odors. A key consideration is the zonal distribution of OR classes across the zones in the OE. Class 1 ORs are expressed dorsally in zone 1, Class II ORs in zones 1–4, and trace amine-associated receptors (TAARs) mostly dorsally but some ventrally [50, 51]. Zone 1 subsequently maps to the medial part of the dOB, the dorsal TAARs to the caudiomedial part of the dOB, and the Class II ORs in zone 1–4 to any of the other regions of the glomerular layer [50]. The lOB likely only contains axonal terminals from OSNs with Class II receptors which are organized such that more dorsal lOB glomeruli receive afferents from more medial zone 1 OE, versus more ventral lOB glomeruli from lateral zone 4 OE [52]. The odorants chosen here must activate some ORs from both classes to some extent as odor responses are observed in both the dOB and the lOB for all odors. Odor information processing has strongly been linked to the perception of temporal information related to odor presentation. Firstly, the temporal responses of mitral/tufted cells have been shown to lock to the sniff cycle, and behaviorally mice can detect temporal optogenetic odor information down to 10 ms relative to the sniff cycle [30]. However, this is not the sole source of temporal information as optogenetic timing differences across glomeruli can be perceived down to 13 ms irrespective of sniff cycle [31]. Temporal patterning across the OB may be a result of zonation within the OE. OSN projections to the dOB and lOB express receptors in different locations within the nasal cavity. The comparison across both regions of the bulb affords us the ability to not only look at the OR activations from the dorsal recess but all areas of the epithelium throughout an inhalation. The dorsal recess experiences higher odor concentrations and air flows compared to other areas of the epithelium, which may contribute to differing temporal patterns [53]. Multichannel recordings have described different temporal responses between mirror glomeruli in the medial OB and lOB [32, 54]. The ventral-medial glomeruli respond faster than the D-L glomeruli, which in our study show the fastest responses of the lOB overall. We observed that the odor responses were slower in the lOB compared to the dOB in line with airflow patterns. We examined the temporal patterning across glomeruli, and it is well established that the dOB displays stereotypical wave of glomerular responses from the posterior-lateral region to the anterior-medial region [19, 55]. We extend this knowledge by also examining the temporal responses in the lateral region. The lateral bulb glomerular responses temporally vary along the D-V axis rather than the roughly A-P axis that is observed in the dOB. Previous optical imaging studies have highlighted that glomerular temporal patterns are locked to respiration [39]. This temporal progression is in line with the sequence of airflow across the epithelium [38]. We found that this temporal pattern is absent in clean air respiration and is not consistently explained by regional differences in amplitude of the odors. We hence demonstrate that the lOB and dOB each have unique global temporal patterning. In situ hybridization mapping of OSN OR projections from the nasal epithelium to the bulb illustrates a D-V arrangement of glomeruli that correlates with the dorsomedial/ventrolateral axis in the epithelium [52], in line with the airflow patterns that develop in the nasal cavity [38]. Considering that the temporal differences in activation of glomeruli across the dOB can be perceived by mice [31], it remains an open question whether the difference in temporal activation in the lOB is providing additional timing information, although this is likely for complex naturalistic odor mixtures. It has been demonstrated previously that OSNs have dual functions, responding to both chemical and mechanical stimuli [41, 42], suggesting that these cells are providing airflow information in addition to odor detection. Respiration has long been known to modulate bulbar activity [56, 57] and perhaps serves as a reference for the temporal activity of glomeruli in response to odor. Mice can detect odor stimulation relative to the phase of the sniff cycle [30, 58], and mechanosensation is important for phase coding of odor identity [43]. We have demonstrated that the ventral lOB displays unusual sensitivity to mechanosensation. Although it is possible that the artificial flow rates have created a pressure change that is not normally present in the nose, it is likely these responses may highlight a region of the bulb specifically monitoring pressure change in the nasal cavity. We have considered the possibility that odorants could play a role in these responses, as the vacuum may have drawn room odorants over the nares. Indeed, c-fos experiments have highlighted the ventral lOB responses to urine [59, 60]. Furthermore, TAARs are exquisitely sensitive, and while most project dorsally, Taar 6, 7a, 7b, and 7d may project ventrally [50]. However, every precaution was taken to ensure that the imaging setup was very clean for imaging. A room vacuum was placed above the imaging setup to continually cycle room air to prevent the accumulation of odors, and a high-efficiency particulate air (HEPA) and active carbon filter was also running constantly in the room during imaging. Although we cannot eliminate a role for odor completely, if indeed these responses were odor responses, they would be present in all experiments with vacuum, which isn’t the case. We observed this mechanosensitive response only when both airflow and vacuum are present and not determined by airflow but the differential pressure created between both. This suggests that the response is not simply a result of either positive or negative pressure within the nose but the transition point between the two as in respiration. It further suggests some inhibitory mechanism of odor transduction onto this mechanosensory phenomenon. The receptor projections to the ventral region of the lOB originate from zone 4 within the OE. Unilateral nares occlusion leads to an increase in expression level of ORs specifically in this zone, suggesting that the loss of mechanosensation may have changed their expression level [61]. Interestingly, high airflow rates in the nasal cavity enhances the sensitivity to low odor concentrations but not high odor concentrations [62]. The sorption of odors into the mucosal layer is also enhanced by flow rate [63]. We show that the temporal patterning of odor responses in the lOB is slowest toward the ventral region, and potentially this mechanosensory response in the ventral lOB is providing the reference point for respiratory phase transitions or playing a role in enhancing the sensitivity to low odor concentrations. Overall, these data have expanded our knowledge about both the spatial and temporal information about odorants across the dorsal and lateral presynaptic glomerular layer and highlighted a unique mechanosensory response unique to the lateral bulb. Methods Ethics statement All procedures were performed in accordance with protocols approved by the Pierce Animal Care and Use Committee (PACUC) JV1-2016 and JV1-2019. These procedures are in agreement with the National Institutes of Health Guide for the Care and Use of Laboratory Animals (8th edition). Surgery Nine OMP-GCaMP6f mice (generated by crossing OMP-Cre [Jax Stock #006668] with GCaMP6f floxed transgenic mice [Jax Stock #024105]) aged 12 to 20 wk, both males and females, were used. They were anaesthetized using isoflurane (4% for induction, 1.5%–2.5% for maintenance). Anesthetic maintenance was monitored using the pedal withdrawal reflex and supplemented as needed. Core body temperature of the animal was maintained at approximately 37°C with a thermostatically controlled heating pad. Post surgery, the animals were placed in their home cage on a heating pad. Carprofen (5 mg/kg, sc) was administered prior to surgery and Buprenorphine (50 μg/kg, IM) at the start of surgery. Mice received supplemental carprofen 24 h post surgery, and weight was monitored for the duration of the experiment. Animals were placed in a stereotaxic holder, and the animals were prepared using aseptic procedures. For exposure of the dOB, the skin was removed, and the underlying bone was thinned using a dental drill. For exposure of the lOB, an enucleation of the left eye was performed, and the upper and lower eyelids were removed. The bone overlying the lateral portion of the OB was thinned. A seamless covering of cyanoacrylate was applied to both the dorsal and lateral windows at once. A head cap was secured using cyanoacrylate and dental cement for stability during imaging. Animals were given at least 24 hr of recovery post surgery before imaging to reduce any OB inflammation as a result of windowing. Imaging All imaging was carried out between 24 and 72 hr post surgery after full recovery. Imaging was performed under ketamine:dexdomitor (100:0.5 mg/kg, i.p, 25% boosters). Atropine (0.03 mg/kg, i.p.) was administered at the start of imaging and every 2 hr hereafter. Eye lubricant was used throughout (Lubrifresh P.M. lubricant eye ointment). Simultaneous recordings of the dOB and lOB were made using 2 identical setups consisting of 2 Hamamatsu ORCA Flash 4.0 LT sCMOS cameras (Hamamatsu, Japan) at a frame rate of 30 Hz and with 4 × 4 binning to 512 × 512 pixels. Two high-power LED 470 nm (Thorlabs, Newark, NJ) was driven by a T-Cube LED Driver (LEDD1B, Thorlabs, Newark, NJ). The custom-made tandem-lens type [64] was used at a 2.7 magnification (FOV: 5 × 5 mm). Imaging lenses were prime Nikon F-mount (ccd lens: 135 mm f/3.5, used at f/8; object lens: 50 mm f/3.5, used at f/3.5). Custom code written in LabView (National Instruments) controlled simultaneous image acquisition using both sCMOS cameras and timing controls for the light source and odor delivery. Sniffing and odor presentation data were acquired simultaneously through a National Instruments data acquisition device. Each imaging session consisted of manually triggered trials with intertrial intervals of ≥2 min. Each trial consisted of 12 s of imaging in which an odor was presented in one 3 s pulse, using a custom-built multichannel auto-switching flow dilution olfactometer [54] with dedicated lines for each odor to avoid cross-contamination. Odorants were presented orthonasally to the animal at concentrations of either 0.1% or 1% s.v. Saturation was maintained by a flow (0.5 or 5 mL/min for 0.1% or 1% s.v., respectively) of filtered high-purity nitrogen (Airgas, NI ISP300, <0.1 ppm THC, H2O, and O2 contaminants) passing through passivated stainless steel spargers (IDEX health and science, A-243, 2 μm inlet filter) in PFA vials (Savillex 200-30-12) connected to the nose chamber (Fig 1A) via an air-dilution manifold. Odors were diluted using clean air (Airgas, AI UZ300, ultra-zero grade, <1 ppm THC, CO2, and CO contaminants) at a flow rate of 499.5 or 495 mL/min for 0.1% or 1% s.v., respectively, for a constant combined air-nitrogen-odor flow rate of 500 mL/min into the nose chamber. The nose chamber consisted of a 1 × 0.5 × 1” (WxHxD) Teflon block with two 5-mm ID channels 10 mm from the front of the block. This allowed connection of a 1/8” OD Teflon tube for diluted odor flow into one side and a vacuum connection (4 mm ID, 8 mm OD Tygon) for outflow on the opposing side. Flow of odorants was continuous and was removed via the vacuum (2.5 L/min) that was switched off for odorant delivery. A central channel of 6 mm ID connected the orthogonal odor-vacuum stream to the frontally placed nares. The tip of the mouse’s nose (including just the nares, OD 2 mm, Fig 1A) was placed just inside the chamber, whereby there was approximately 2 mm of space surrounding the entire nose for unrestricted flow. Odorants (Sigma-Aldrich) used were heptanone, hexanal, AA, carvone, MV, and heptanol (stored under nitrogen in the dark). Mice were freely breathing, which was continuously measured by a piezoelectric strip positioned on the animal's thorax. Three OMP-GCaMP6f animals that had no prior exposure to odors were used for the mechanosensory air experiments. The clean (medical grade) air presented to the animal was not delivered via the multichannel auto-switching flow dilution olfactometer but was carbon filtered and delivered directly from a mass flow controller. Data analysis Custom code written in LabView was used to extract the fluorescence traces from each trial. Frame subtraction was performed by selecting video frames just before and after odor presentation. This presented an image that highlighted regions that responded to odor stimulation. Multiple ROIs that resembled glomeruli were manually selected per mouse. This process was repeated for all trials, and additional ROIs were selected to accommodate all glomeruli that responded in a particular mouse for all odors and all concentrations. This generated an accumulated list of ROIs that could directly compare the responsiveness of all glomeruli across odors. The ROIs were used to extract mean fluorescence intensity traces from time series images of all trials. Graphpad Prism (version 7; GraphPad Software, CA) was use to generate plots and for statistical analysis. All data are presented as mean ± SEM. Data overview For each trial, the optical imaging response traces (one .txt file for all ROIs per trial), ROI location (of the diagonally opposed corners of the rectangular area; no more than 40 per imaged OB area [lateral/dorsal] per animal; a .txt file), and sniffing and odor presentation timing traces (a .txt file) were analyzed in Matlab (R2018a, The Mathworks, MA) in batch mode. The total data set consisted of 6 animals with up to 48 trials each (6 odors × 2 conc × 3 repeats) and up to 80 ROIs (S1 Table). All data were referenced to the very stable OB imaging sampling times (virtually jitter-free 30.00 fps), to which odor and sniffing data were resampled and then shifted for proper alignment, followed by truncation. Alignment was verified by comparing an optically imaged LED driven in parallel by the odor-on vacuum valve with that of the odor valve .txt file. The offset between the imaged data and odor/breathing data did not vary between trials. Spatiotemporal analysis of response patterns The start of inhalation with the piezo sensor (once bandpass filtered) was a sharp downward deflection after a relatively shallow downward slope, verified by co-imaging of the thorax movement of a mouse. The timing thereof was determined by band-pass filtering (4th order, 1–10 Hz, zero phase shift) of the z-scored piezo voltage signal, followed by peak detection, from which the onset could reliably (1 sample jitter, 33.0 ms) be determined. To determine the peak response amplitudes (% ΔF/F) and temporal parameters (including T90, see S1B Fig) of the glomerular responses, we used a custom algorithm that fitted the optical signals from each ROI to a double sigmoid function as described previously [44, 65]. The analysis allowed robust and objective measurement of response timing. Briefly, the signal from each ROI, after each identified inhalation, was band-pass filtered (second-order Butterworth, 0.1–7 Hz) followed by fourth-order Daubechies wavelet decomposition, soft thresholding of the coefficients at level 3, and then reconstruction. The onset time was defined based on the time of peak in the product of the first and the second derivatives of the optical signal. Starting at this time, each response was fitted (least-squares curve fitting) with a double-sigmoid function (a sigmoid rise followed by a sigmoid fall). The time of the peak of this response was defined as the peak in this fitted response function rather than the peak of the raw optical signal. Next, we identified the inhalation onset time for each trial that evoked the first dorsal OB response during presentation of odor (“odor on response”; odor on from 3.4–6.4 s) by finding the largest response within the series of fitted responses averaged across ROIs for the period 2.8–4.5 s. We similarly determined the inhalation onset for largest mean lOB response (as we did not find any odor “off” responses in the dOB) per trial after the odor was turned off, between 6.0 and 9.0 s. These two inhalations were used to analyze the spatiotemporal responses for both the dOB and lOB (T10, T50, T90, Tpeak, and peak % ΔF/F) across glomeruli and the bulbs. Global odor on response maps (Fig 2B) were established by correlating (Matlab function “corr”) for each mouse and odor the peak % ΔF/F of glomeruli with their location along each of their spatial dimensions (dOB: A-P and M-L [“laterality”: distance—in pixels—from midline]; lOB: A-P and D-V). Global maps of odor on response dynamics (Fig 3A–3C) were similarly made for T90 instead of response amplitude. The resulting correlation coefficients were subsequently averaged across mice. For pre-odor responses, the amplitudes and dynamics of each ROI were first averaged across all responses per trial, prior to making correlations (Fig 3B and 3C). Correlations of specific odor maps similarities (Fig 2C) were made by correlating odor on % ΔF/F responses from all identified ROIs across odors for dOB and lOB separately. z-Scores of each odor response were calculated for each trial by subtracting from an odor-evoked response the mean of the trial’s pre-odor responses and dividing this difference by the SD of the pre-odor responses: Recruitment (Fig 4C) of glomeruli was quantified by first assessing whether each glomerular odor on response (peak % ΔF/F) was significantly larger than its own baseline breathing peak response amplitudes. Each odor on/off response was z-scored relative to its own response amplitudes prior to odor onset (see Fig 2D). A glomerulus responded (was “recruited”) when the likelihood was <1% (i.e., P < 0.01). The odor off histograms (Fig 5D) were based on the off response amplitude (peak % ΔF/F) of each glomerulus. This was divided by the off response amplitude (peak % ΔF/F) during breathing trials with no air or vacuum. As no stimulus was delivered for the breathing trial, the off response was the taken as the peak response at the same timepoint at in the stimulus trials. The histograms are based on all stimulus-delivered trials combined. We attempted to record intranasal pressure changes via a nasal cannula, but we were unable to reliably detect any changes unless the nose was in much closer proximity to the odor delivery tube than during the imaging experiments. Hence, to reliably measure the relative pressure change for the vacuum on and vacuum off responses (S4A Fig), a tube (31 mm long, 6.0 mm OD, 3.0 mm ID) was connected to the positive sensing orifice of a Buxco (TRD 5700) pressure sensor. The opposite edge of the tube was placed just inside the odor delivery tube (ID 7.1 mm) used during the in vivo study, thereby leaving a wide gap, like the in vivo setup. Buxco sensor output was band-pass filtered (0–1 kHz), amplified (1000×), and the resulting voltage was recorded for steady-state air and vacuum flow rate combinations (S4A Fig, and calculated changes in B). Dynamic conditions settled to steady-state values nearly instantaneously. To explore pressure change-OB response functions, the mean of the z-scored response across all glomeruli (dOB: 99; lOB: 47; n = 3 mice) was calculated. Out of a total of 1,584 dOB responses (16 flow conditions × 99 glomeruli) and 752 lOB responses (16 × 47), we removed 4 outliers (z-score of z-score was < −4 or > 4, based on distribution of all z-scores for vacuum on and off separately and dOB and lOB separately) from each dataset (S4C and S4D Fig). For plotting of the functions (Fig 6), the responses were uniformly offset by the mean of the respective z-scores for no flow change condition (“0–0 [room]”; z = 0.9 for dOB and z = 1.5 for lOB), which had a positive bias due to our sniff-response–selecting algorithm picking the largest response averaged across glomeruli within a defined response period. Ethics statement All procedures were performed in accordance with protocols approved by the Pierce Animal Care and Use Committee (PACUC) JV1-2016 and JV1-2019. These procedures are in agreement with the National Institutes of Health Guide for the Care and Use of Laboratory Animals (8th edition). Surgery Nine OMP-GCaMP6f mice (generated by crossing OMP-Cre [Jax Stock #006668] with GCaMP6f floxed transgenic mice [Jax Stock #024105]) aged 12 to 20 wk, both males and females, were used. They were anaesthetized using isoflurane (4% for induction, 1.5%–2.5% for maintenance). Anesthetic maintenance was monitored using the pedal withdrawal reflex and supplemented as needed. Core body temperature of the animal was maintained at approximately 37°C with a thermostatically controlled heating pad. Post surgery, the animals were placed in their home cage on a heating pad. Carprofen (5 mg/kg, sc) was administered prior to surgery and Buprenorphine (50 μg/kg, IM) at the start of surgery. Mice received supplemental carprofen 24 h post surgery, and weight was monitored for the duration of the experiment. Animals were placed in a stereotaxic holder, and the animals were prepared using aseptic procedures. For exposure of the dOB, the skin was removed, and the underlying bone was thinned using a dental drill. For exposure of the lOB, an enucleation of the left eye was performed, and the upper and lower eyelids were removed. The bone overlying the lateral portion of the OB was thinned. A seamless covering of cyanoacrylate was applied to both the dorsal and lateral windows at once. A head cap was secured using cyanoacrylate and dental cement for stability during imaging. Animals were given at least 24 hr of recovery post surgery before imaging to reduce any OB inflammation as a result of windowing. Imaging All imaging was carried out between 24 and 72 hr post surgery after full recovery. Imaging was performed under ketamine:dexdomitor (100:0.5 mg/kg, i.p, 25% boosters). Atropine (0.03 mg/kg, i.p.) was administered at the start of imaging and every 2 hr hereafter. Eye lubricant was used throughout (Lubrifresh P.M. lubricant eye ointment). Simultaneous recordings of the dOB and lOB were made using 2 identical setups consisting of 2 Hamamatsu ORCA Flash 4.0 LT sCMOS cameras (Hamamatsu, Japan) at a frame rate of 30 Hz and with 4 × 4 binning to 512 × 512 pixels. Two high-power LED 470 nm (Thorlabs, Newark, NJ) was driven by a T-Cube LED Driver (LEDD1B, Thorlabs, Newark, NJ). The custom-made tandem-lens type [64] was used at a 2.7 magnification (FOV: 5 × 5 mm). Imaging lenses were prime Nikon F-mount (ccd lens: 135 mm f/3.5, used at f/8; object lens: 50 mm f/3.5, used at f/3.5). Custom code written in LabView (National Instruments) controlled simultaneous image acquisition using both sCMOS cameras and timing controls for the light source and odor delivery. Sniffing and odor presentation data were acquired simultaneously through a National Instruments data acquisition device. Each imaging session consisted of manually triggered trials with intertrial intervals of ≥2 min. Each trial consisted of 12 s of imaging in which an odor was presented in one 3 s pulse, using a custom-built multichannel auto-switching flow dilution olfactometer [54] with dedicated lines for each odor to avoid cross-contamination. Odorants were presented orthonasally to the animal at concentrations of either 0.1% or 1% s.v. Saturation was maintained by a flow (0.5 or 5 mL/min for 0.1% or 1% s.v., respectively) of filtered high-purity nitrogen (Airgas, NI ISP300, <0.1 ppm THC, H2O, and O2 contaminants) passing through passivated stainless steel spargers (IDEX health and science, A-243, 2 μm inlet filter) in PFA vials (Savillex 200-30-12) connected to the nose chamber (Fig 1A) via an air-dilution manifold. Odors were diluted using clean air (Airgas, AI UZ300, ultra-zero grade, <1 ppm THC, CO2, and CO contaminants) at a flow rate of 499.5 or 495 mL/min for 0.1% or 1% s.v., respectively, for a constant combined air-nitrogen-odor flow rate of 500 mL/min into the nose chamber. The nose chamber consisted of a 1 × 0.5 × 1” (WxHxD) Teflon block with two 5-mm ID channels 10 mm from the front of the block. This allowed connection of a 1/8” OD Teflon tube for diluted odor flow into one side and a vacuum connection (4 mm ID, 8 mm OD Tygon) for outflow on the opposing side. Flow of odorants was continuous and was removed via the vacuum (2.5 L/min) that was switched off for odorant delivery. A central channel of 6 mm ID connected the orthogonal odor-vacuum stream to the frontally placed nares. The tip of the mouse’s nose (including just the nares, OD 2 mm, Fig 1A) was placed just inside the chamber, whereby there was approximately 2 mm of space surrounding the entire nose for unrestricted flow. Odorants (Sigma-Aldrich) used were heptanone, hexanal, AA, carvone, MV, and heptanol (stored under nitrogen in the dark). Mice were freely breathing, which was continuously measured by a piezoelectric strip positioned on the animal's thorax. Three OMP-GCaMP6f animals that had no prior exposure to odors were used for the mechanosensory air experiments. The clean (medical grade) air presented to the animal was not delivered via the multichannel auto-switching flow dilution olfactometer but was carbon filtered and delivered directly from a mass flow controller. Data analysis Custom code written in LabView was used to extract the fluorescence traces from each trial. Frame subtraction was performed by selecting video frames just before and after odor presentation. This presented an image that highlighted regions that responded to odor stimulation. Multiple ROIs that resembled glomeruli were manually selected per mouse. This process was repeated for all trials, and additional ROIs were selected to accommodate all glomeruli that responded in a particular mouse for all odors and all concentrations. This generated an accumulated list of ROIs that could directly compare the responsiveness of all glomeruli across odors. The ROIs were used to extract mean fluorescence intensity traces from time series images of all trials. Graphpad Prism (version 7; GraphPad Software, CA) was use to generate plots and for statistical analysis. All data are presented as mean ± SEM. Data overview For each trial, the optical imaging response traces (one .txt file for all ROIs per trial), ROI location (of the diagonally opposed corners of the rectangular area; no more than 40 per imaged OB area [lateral/dorsal] per animal; a .txt file), and sniffing and odor presentation timing traces (a .txt file) were analyzed in Matlab (R2018a, The Mathworks, MA) in batch mode. The total data set consisted of 6 animals with up to 48 trials each (6 odors × 2 conc × 3 repeats) and up to 80 ROIs (S1 Table). All data were referenced to the very stable OB imaging sampling times (virtually jitter-free 30.00 fps), to which odor and sniffing data were resampled and then shifted for proper alignment, followed by truncation. Alignment was verified by comparing an optically imaged LED driven in parallel by the odor-on vacuum valve with that of the odor valve .txt file. The offset between the imaged data and odor/breathing data did not vary between trials. Spatiotemporal analysis of response patterns The start of inhalation with the piezo sensor (once bandpass filtered) was a sharp downward deflection after a relatively shallow downward slope, verified by co-imaging of the thorax movement of a mouse. The timing thereof was determined by band-pass filtering (4th order, 1–10 Hz, zero phase shift) of the z-scored piezo voltage signal, followed by peak detection, from which the onset could reliably (1 sample jitter, 33.0 ms) be determined. To determine the peak response amplitudes (% ΔF/F) and temporal parameters (including T90, see S1B Fig) of the glomerular responses, we used a custom algorithm that fitted the optical signals from each ROI to a double sigmoid function as described previously [44, 65]. The analysis allowed robust and objective measurement of response timing. Briefly, the signal from each ROI, after each identified inhalation, was band-pass filtered (second-order Butterworth, 0.1–7 Hz) followed by fourth-order Daubechies wavelet decomposition, soft thresholding of the coefficients at level 3, and then reconstruction. The onset time was defined based on the time of peak in the product of the first and the second derivatives of the optical signal. Starting at this time, each response was fitted (least-squares curve fitting) with a double-sigmoid function (a sigmoid rise followed by a sigmoid fall). The time of the peak of this response was defined as the peak in this fitted response function rather than the peak of the raw optical signal. Next, we identified the inhalation onset time for each trial that evoked the first dorsal OB response during presentation of odor (“odor on response”; odor on from 3.4–6.4 s) by finding the largest response within the series of fitted responses averaged across ROIs for the period 2.8–4.5 s. We similarly determined the inhalation onset for largest mean lOB response (as we did not find any odor “off” responses in the dOB) per trial after the odor was turned off, between 6.0 and 9.0 s. These two inhalations were used to analyze the spatiotemporal responses for both the dOB and lOB (T10, T50, T90, Tpeak, and peak % ΔF/F) across glomeruli and the bulbs. Global odor on response maps (Fig 2B) were established by correlating (Matlab function “corr”) for each mouse and odor the peak % ΔF/F of glomeruli with their location along each of their spatial dimensions (dOB: A-P and M-L [“laterality”: distance—in pixels—from midline]; lOB: A-P and D-V). Global maps of odor on response dynamics (Fig 3A–3C) were similarly made for T90 instead of response amplitude. The resulting correlation coefficients were subsequently averaged across mice. For pre-odor responses, the amplitudes and dynamics of each ROI were first averaged across all responses per trial, prior to making correlations (Fig 3B and 3C). Correlations of specific odor maps similarities (Fig 2C) were made by correlating odor on % ΔF/F responses from all identified ROIs across odors for dOB and lOB separately. z-Scores of each odor response were calculated for each trial by subtracting from an odor-evoked response the mean of the trial’s pre-odor responses and dividing this difference by the SD of the pre-odor responses: Recruitment (Fig 4C) of glomeruli was quantified by first assessing whether each glomerular odor on response (peak % ΔF/F) was significantly larger than its own baseline breathing peak response amplitudes. Each odor on/off response was z-scored relative to its own response amplitudes prior to odor onset (see Fig 2D). A glomerulus responded (was “recruited”) when the likelihood was <1% (i.e., P < 0.01). The odor off histograms (Fig 5D) were based on the off response amplitude (peak % ΔF/F) of each glomerulus. This was divided by the off response amplitude (peak % ΔF/F) during breathing trials with no air or vacuum. As no stimulus was delivered for the breathing trial, the off response was the taken as the peak response at the same timepoint at in the stimulus trials. The histograms are based on all stimulus-delivered trials combined. We attempted to record intranasal pressure changes via a nasal cannula, but we were unable to reliably detect any changes unless the nose was in much closer proximity to the odor delivery tube than during the imaging experiments. Hence, to reliably measure the relative pressure change for the vacuum on and vacuum off responses (S4A Fig), a tube (31 mm long, 6.0 mm OD, 3.0 mm ID) was connected to the positive sensing orifice of a Buxco (TRD 5700) pressure sensor. The opposite edge of the tube was placed just inside the odor delivery tube (ID 7.1 mm) used during the in vivo study, thereby leaving a wide gap, like the in vivo setup. Buxco sensor output was band-pass filtered (0–1 kHz), amplified (1000×), and the resulting voltage was recorded for steady-state air and vacuum flow rate combinations (S4A Fig, and calculated changes in B). Dynamic conditions settled to steady-state values nearly instantaneously. To explore pressure change-OB response functions, the mean of the z-scored response across all glomeruli (dOB: 99; lOB: 47; n = 3 mice) was calculated. Out of a total of 1,584 dOB responses (16 flow conditions × 99 glomeruli) and 752 lOB responses (16 × 47), we removed 4 outliers (z-score of z-score was < −4 or > 4, based on distribution of all z-scores for vacuum on and off separately and dOB and lOB separately) from each dataset (S4C and S4D Fig). For plotting of the functions (Fig 6), the responses were uniformly offset by the mean of the respective z-scores for no flow change condition (“0–0 [room]”; z = 0.9 for dOB and z = 1.5 for lOB), which had a positive bias due to our sniff-response–selecting algorithm picking the largest response averaged across glomeruli within a defined response period. Supporting information S1 Fig. Glomerular dynamics across dOB and lOB. (A) Amplitude responses of glomeruli for heptanone (left) and hexanal (right) and their position across the A-P dimension in the dOB, demonstrating their linear correlations, as shown in Fig 2B. B Generic fluorescence trace of a glomerulus displaying the determination of the T90 of a response. (C) The average T90 responses over all glomeruli for each odor in the dorsal bulb (AA, heptanol, and hexanal 5 animals; all others 6 animals) and lOB (AA, heptanol, and hexanal 5 animals; all others 6 animals). (D) Comparison of the T90 responses of glomeruli in the anterior and posterior dOB. (E) Comparison of the T90 responses of glomeruli in the dorsal and ventral lOB. Statistics represent two-way ANOVA (odor × OB region) with Bonferroni’s multiple comparisons test. ♦ denotes statistical significance between dorsal and lateral T90 for all odors. Error bars are SEM. ♦ P < 0.05, ♦♦ P < 0.01, ♦♦♦ P < 0.001, ♦♦♦♦ P < 0.0001. Underlying data for this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3000409.s001 (TIFF) S2 Fig. Glomerular peak responses across dOB and lOB. (A) Color-scaled ΔF/F responses for ROIs for all odors. ΔF/F are scaled between 0 and 43% (1 animal). (B) Color-scaled ΔF/F responses for ROIs for all odors. ΔF/F are scaled to each odors maximum value (1 animal). Underlying data for this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3000409.s002 (TIFF) S3 Fig. Glomerular dynamics across dOB and lOB. (A) Color-scaled T90 responses for ROIs for all odors. T90 are scaled between 0 and 350 ms (1 animal). (B) Color-scaled T90 responses for ROIs for all odors. T90 are scaled to each odors maximum value (1 animal). Underlying data for this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3000409.s003 (TIFF) S4 Fig. Glomerular responses in dOB and lOB versus air flow conditions. (A) The differential pressure change of the individual (positive) clean air and (negative) vacuum flow rates and the differential pressure change of them combined (see Methods). Arrows demonstrate a relatively large reduction of the pressure drop of both clean air and vacuum compared to vacuum alone. (B) A comparison of net flow rate and differential pressure at all air and vacuum flow rates. (C) The z-score values of all glomeruli in response to either vacuum on or vacuum off at all flow rates in the dOB. (D) Same as C but in the lOB, 3 animals, dOB: 99 glomeruli, lOB: 47 glomeruli. Underlying data for this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3000409.s004 (TIFF) S5 Fig. The dOB lacks the pressure sensitivity that the lOB displays. The z-scores of the dOB glomerular responses during only clean air flow (red) and also vacuum (grey), organized from anterior to posterior, for different clean air flow rates (0.5, 0.25, 0.1, 0.05, and 0.005 L/min) and vacuum rates (2.5, 1.25, and 0 L/min) and room air (3 animals, 47 glomeruli). Linear correlation fits are indicated, 3 animals, 99 glomeruli. Underlying data for this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3000409.s005 (TIFF) S1 Table. Animal number, glomerular number, and trial number for both the dOB and lOB for 0.1, 1% odor concentration and air. https://doi.org/10.1371/journal.pbio.3000409.s006 (TIFF) S2 Table. Global correlation values of odor on responses across spatial dimensions. (A) In the dOB, A-P, and M-L dimensions. (B) In the lOB, A-P and D-L dimensions. https://doi.org/10.1371/journal.pbio.3000409.s007 (TIFF) S3 Table. Pearson correlation values of glomerular responses for all odors. (A) In the dOB. (B) In the lOB. https://doi.org/10.1371/journal.pbio.3000409.s008 (TIFF) S4 Table. Average of T90 responses for all glomeruli in the dOB and lOB. https://doi.org/10.1371/journal.pbio.3000409.s009 (TIFF) S5 Table. Linear correlations of T90 with location along each spatial dimension prior to odor presentation (pre-odor breathing responses) and upon odor presentation (post odor). (A) In the dorsal bulb. (B) In the lateral bulb. https://doi.org/10.1371/journal.pbio.3000409.s010 (TIFF) S6 Table. Correlation between the glomerular response amplitudes and the T90 responses of these glomeruli in the dOB and the lOB. https://doi.org/10.1371/journal.pbio.3000409.s011 (TIFF) S7 Table. The percentage of glomeruli significantly activated by 0.1% odors normalized relative to 1% in the dOB and lOB. https://doi.org/10.1371/journal.pbio.3000409.s012 (TIFF) S8 Table. The average amplitudes of all glomeruli responding to 0.1 and 1% for all odors. (A) In the dOB. (B) In the lOB. https://doi.org/10.1371/journal.pbio.3000409.s013 (TIFF) S9 Table. The slope and significance values for the linear regressions of the z-scores for glomerular responses and their location along the D-V dimension of the lOB in response to air and vacuum. https://doi.org/10.1371/journal.pbio.3000409.s014 (TIFF) S1 Data. https://doi.org/10.1371/journal.pbio.3000409.s015 (XLSX) Acknowledgments We appreciate the technical support of the John B. Pierce Shop. We thank Dr. Gordon Shepherd, Dr. Doug Storace, and Dr. Ruaidhri Jackson for helpful comments on the manuscript.
Control over single-cell distribution of G1 lengths by WNT governs pluripotencyJang, Jiwon;Han, Dasol;Golkaram, Mahdi;Audouard, Morgane;Liu, Guojing;Bridges, Daniel;Hellander, Stefan;Chialastri, Alex;Dey, Siddharth S.;Petzold, Linda R.;Kosik, Kenneth S.
doi: 10.1371/journal.pbio.3000453pmid: 31557150
Introduction How a pluripotent population decides among fate choices is a highly relevant question of intense interest. Although growth conditions can deterministically direct the population-level outcome of stem cells once differentiation begins, the contribution of single-cell variation within the stem cell population [1–12] to the collective fate decision is poorly understood. In the absence of distinct cell types that establish stable variation across the population, stem cell variation arises from dynamic physiological events such as the cell cycle, changes in the microenvironment, and stochasticity that together establish a dynamic equilibrium. From this perspective, individual cells transit between distinct metastable states while maintaining the overall structure of the population [5,13–15]. Underlying stem cell heterogeneity is the influence of variations in the activity of signaling pathways at the single-cell level such as Wingless-INT (WNT), Bone morphogenetic protein (BMP), Nodal growth differentiation factor (NODAL), and Fibroblast growth factor (FGF) that can confer transient lineage biases to pluripotent stem cell subpopulations [16,17]. Highly variable gene expression patterns among single cells arise from the particularly permissive and dynamic chromatin structure of stem cells [18]. At the population level, well-defined robust behavior emerges from stochastic dynamics at the single-cell level [19]. Pluripotent stem cells have a relatively low percentage of G1-phase cells because of a shortened duration of G1 [20–22]. This finding has been well documented in several ways [23–25], including by the use of the fluorescent ubiquitination–based cell-cycle indicator (FUCCI) system [26]. Pluripotent stem cells initiate differentiation from G1 phase [22,25,27–30]. This finding links heterogeneous gene expression and cell-cycle progression as shown by RNA sequencing (RNA-seq) of FUCCI-labeled G1 cells [31]. Thus, G1 phase–constrained gene expression may be a possible mechanism for “lineage priming,” particularly in light of G1-associated up-regulation of the epigenetic mark, 5-hydroxymethylcytosine (5-hmC), thought to have a role in gene activation [31–33] and cyclin D–dependent transcription [34]. How single-cell variation influences population behavior and provides the many developmental options available to human embryonic stem cells (hESCs) has been addressed according to a theoretical framework in which single cells have statistical properties that increase the potential of the population [35]. Here, we show that hESCs have high cell-to-cell variation in absolute G1 length, and with increased single-cell variation, a population bias toward neuroectoderm (NE) emerged well before the onset of differentiation. Thus, G1 length distribution patterns of a stem cell population represent a probability density curve that can predict differentiation outcome as a predominantly NE or mesendoderm (ME) population. An hESC population with a short and narrow distribution of G1 lengths was biased toward predominantly ME, whereas a long and wide distribution of G1 lengths biased the pluripotent stem cells toward both NE and ME lineages. WNT is centrally positioned in decisions regarding pluripotency because it can control both self-renewal and differentiation, a property that is likely related to its control over G1 length [36–39]. The control that WNT exerts over G1 length results in a stable, yet dynamic, population. Lineage priming occurs by reducing WNT levels, which promotes differentiation and skews the single-cell distribution of stem cell G1 lengths toward longer time intervals and higher single-cell variation. Furthermore, the effect of G1 lengthening on downstream gene expression operates through increased 5-hmC modification. We propose that the high variation in gene expression across single stem cells represents a large parameter space that can be collapsed to a single-cell G1 length distribution curve, which predicts the population differentiation potential to NE or ME. Results Cell-to-cell variation in G1 and S/G2/M lengths in an hESC population To analyze absolute lengths of each phase of the cell cycle at a single-cell level, we used the FUCCI reporter [26]. Based on cell cycle–dependent degradation of chromatin licensing and DNA replication factor 1 (CDT1) and geminin DNA replication inhibitor (GEMININ) proteins, cells show no color in early G1, red in mid/late G1, and green in S/G2/M (S1A Fig). FUCCI hESCs showed high expression of pluripotency genes and activated markers for NE and ME upon differentiation (S1B and S1C Fig). Time-lapse images were taken from the FUCCI hESCs grown in mTeSR1 medium for 24–48 h with 10-min intervals (S1 and S2 Movies). Compared with previous reports [25,31], one obvious difference we found in our data is that early G1 phase is very short—less than one frame (10 min) in most cells (S1D Fig and S3 Movie). This discrepancy is likely due to the use of different fluorescent proteins. Pauklin and colleagues [25] used monomeric Kusabira orange 2 (mKO2) fluorescent protein fused to CDT1 (for G1 phase), whereas we used mcherry. The maturation half-time of mKO2 is much longer (1.2 h) than mcherry’s (15 min) [40,41], which could explain a higher percentage of early G1 cells (no color) in their results. To confirm these results, we derived three independent clonal lines from FUCCI hESCs and observed that almost all cells have a no-color phase shorter than 10 min (S1E Fig). These findings suggest that defining early G1 phase by the absence of fluorescent signal could be influenced by differences in maturation time of fluorescence proteins. The short maturation of mcherry–CDT1 fusion proteins more accurately represents the duration of the entire G1 phase and demonstrates that early G1 contributes very little to overall G1 length. Individual hESCs showed high variation in the lengths of G1 and S/G2/M phases, with coefficients of variation (CVs) 36.6% and 18.6%, respectively (Fig 1A). G1 lengths ranged from 4 h to 10 h, and S/G2/M lengths ranged from 20 h to 40 h (Fig 1B). Interestingly, G1 lengths inversely correlated with S/G2/M length (p < 0.0001) in the same cell (Fig 1C), suggesting that cells with short G1 tend to have long S/G2/M and vice versa. This inverse correlation suggests that individual cells conserve total cell-cycle length through compensatory adjustments of cell-cycle phases. Recently, a compensatory relationship between S- and G2-phase lengths was observed in chordate ascidians during gastrulation [42]. Overall, these results suggest that hESCs have high single-cell heterogeneity in G1 and S/G2/M lengths under self-renewing conditions. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Single-cell heterogeneity in G1 and S/G2/M lengths in self-renewing hESCs. (A) Absolute cell-cycle length of each cell-cycle state in H9 hESCs grown in mTeSR1 medium (n = 114 for G1 and n = 60 for S/G2/M). (B) Histograms for G1 and S/G2/M lengths. (C) Inverse relationship between G1 and S/G2/M lengths in individual cells (n = 30). Both G1 and S/G2/M lengths were measured from the same cell. Data in (C) are a subset of those in (A). Data were collected and pooled from three independent experiments. Underlying data can be found in S1 Data. hESC, human embryonic stem cell. https://doi.org/10.1371/journal.pbio.3000453.g001 Distribution patterns of single-cell G1 length determine differentiation outcomes of stem cell populations G1 phase has the potential to make cell fate decisions in response to extracellular cues, whereas cells in S/G2/M phase do not respond to differentiation cues [25,30,31,43]. The different response of each cell-cycle state to differentiation cues is mechanistically supported by evidence that epigenetic modifications are regulated in a cell cycle–dependent manner [31,44–46]. Recent papers nicely showed that gene activation–related epigenetic markers such as 5-hydroxymethylcytosine and histone 3 lysine 4 trimethylation (H3K4me3) accumulated in lineage-specific genes as the G1 phase progresses [31,45,46]. Therefore, we hypothesized that individual hESCs differ in their differentiation potential based upon their absolute G1 lengths. To broach this question, we compared H9 hESCs in two well-defined feeder-free stem cell media, Essential 8 (E8) and mTeSR1. These two different media conditions gave us a unique opportunity to study the correlation between G1 length and differentiation potential in an identical cell line. Consistent with the robust ability of both media to maintain pluripotency, H9 cells showed high expression levels of key pluripotency genes (Octamer-binding transcription factor 4 [OCT4], NANOG, Sex determining region Y-box 2 [SOX2]) under both conditions (S2A Fig). To measure the differentiation potential, we spontaneously differentiated H9 cells in hESC medium (see Materials and methods) without FGF2. In this differentiation condition, H9 cells in E8 exhibited highly biased differentiation toward ME, whereas H9 cells in mTeSR1 differentiated more equally to both NE and ME lineages (Fig 2A). The mRNA levels of lineage markers showed consistent patterns with the flow cytometry data (S2B Fig). Lineage-specific differentiation protocols, dual SMAD inhibition for NE differentiation, and FGF2 and BMP4 for ME differentiation also confirmed that hESCs differ in their differentiation propensity depending on the conditions under which they were grown before differentiation (S2C Fig) [47,48]. Both media conditions (mTeSR1 and E8) showed, as expected, similar cell-cycle patterns, with a low percentage of cells in G1 and a high percentage of cells in S/G2/M (Fig 2B, S2D Fig). These results were further confirmed by using 5-bromo-2′-deoxyuridine (BrdU)/7-amino-actinomycin D (7-AAD), which clearly separates S phase from the others (S2E Fig). When absolute G1 length was measured from single cells, however, a dramatic difference emerged, suggesting that mTeSR1 and E8 represent two distinct distributions (p < 2.2 × 10−16 by the Kolmogorov-Smirnov test and p < 2.2 × 10−16 by the Mann-Whitney U test). Cells grown in E8 showed a loss of the long G1 population and consequently a decreased mean G1 length (Fig 2C). Biological replicates displayed similar mean and variation (S2F Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Single-cell distribution patterns of absolute G1 length determine differentiation propensity of hESC populations. (A) Immunoflow cytometry of PAX6 and GATA6 in H9 cells differentiated for 8 d by FGF2 deprivation (n = 4). (B) Propidium iodide staining analysis of H9 cells either in E8 or in mTeSR1 media and human dermal fibroblasts (n = 4 for E8 and mTeSR1, n = 3 for fibroblasts). (C) Histograms for G1 length of H9 cells grown either in E8 or in mTeSR1 (n = 112 for E8 and n = 114 for mTeSR1 pooled from three independent experiments); U test: p-value < 2.2 × 10−16, KS test: p-value < 2.2 × 10−16. (D-F) qPCR analysis of lineage markers in differentiated day 8 H9 cells overexpressing p21 (D), CDK4R24C (E), or CDK6R31C (F) (n = 4). Ctrl represents cells transduced with empty lentiviral vectors and treated with the same dose of doxycycline (1 μg/ml). Transgene expression was turned off at the onset of differentiation by doxycycline withdrawal. Error bars represent SD. *p < 0.01 (Student t test). Underlying data can be found in S1 Data. CDK, Cyclin-dependent kinase; Ctrl, control; Diff. differentiated; E8, Essential 8; FSC, forward scatter; FGF, Fibroblast growth factor; FL, fluorescence intensity; GATA6, GATA binding protein 6; HAND1, Heart and neural crest derivatives expressed 1; hESC, human embryonic stem cell; KS, Kolmogorov-Smirnov; OTX1, Orthodenticle homeobox 1; PAX6, Paired box 6; qPCR, quantitative PCR; Undiff., undifferentiated; ZBTB16, Zinc finger and BTB domain containing 16. https://doi.org/10.1371/journal.pbio.3000453.g002 Because the G1/S checkpoint plays a key role in determining G1 length, we measured the expression of cyclins. Consistent with a short average G1 length, H9 cells in E8 medium showed higher expression of cyclins D and E, which facilitate the G1/S transition (S2G and S2H Fig), but not cyclins A and B (S2G and S2H Fig). Given the ME-biased potential of E8 in the context of our spontaneous differentiation condition, we assumed that the population of cells with long G1 might be responsible for the NE lineage preference. To further validate this idea, we compared three well-defined human induced pluripotent stem cell (hiPSC) lines (S3A Fig) [49]. As high variability in differentiation potential was reported in iPSC lines [50–52], the three hiPSC lines have a dramatic difference in differentiation propensity (S3B Fig). Consistent with the hESC data (Fig 2C), highly neurogenic hiPSC1 contained a substantial population of cells with longer G1 compared with hiPSC2 and hiPSC3 (S3C Fig). However, all hiPSC lines maintained similar relative proportions of cells in each cell-cycle state (S3D Fig), emphasizing that relative cell-cycle length is not related to the differentiation potential of human pluripotent stem cells (hPSCs). Rather, variation in length among the same number of cells in G1 establishes differentiation potential. To test the functional relationship between absolute G1 length and cell fates, we modulated G1 length in self-renewing H9 cells by overexpressing p21 or constitutively active Cyclin-dependent kinases (CDK4R24C and CDK6R31C) (S4A Fig). We used a doxycycline-dependent lentiviral system to control the transgene expression. Overexpression of p21 increased the average length of G1, whereas CDK4R24C and CDK6R31C reduced it (S4B Fig). The modulation of absolute G1 length did not affect pluripotency gene expression and the relative cell-cycle patterns (S4A, S4C and S4D Fig), which suggests that hESCs can tolerate high variation in G1 length. Transgene expression was turned off at the onset of differentiation to exclude any potential effect of the transgene on differentiation (S4E Fig). Transient increase of G1 length by p21 further promoted NE derivation at the expense of ME differentiation (Fig 2D). These results were confirmed by abemaciclib, a potent and selective chemical inhibitor of CDK4 and CDK6. Abemaciclib treatment for 18 h before differentiation phenocopied the effect of p21 overexpression (S4F Fig). Transient decrease of G1 length by CDK4R24C and CDK6R31C showed the opposite effect (Fig 2E and 2F). Furthermore, increased G1 length by p21 overexpression in H9 cells grown in E8 medium was sufficient to improve neurogenic potential to the level of those in mTeSR1 at the expense of ME lineage derivation (S4G Fig), suggesting that the difference in differentiation propensity between the two media conditions can be attributed to G1 length. Collectively, these results suggest that G1 length biases the differentiation potential of self-renewing hESCs, and thus, the G1 length distribution patterns determine population fates upon differentiation. Asymmetric sister cell G1 duration contributes to heterogeneity in single-cell G1 length Cells grown in mTeSR1 have not only an increased mean G1 length but also a greater CV than those of E8 (Fig 3A). The larger variation of G1 length in a highly neurogenic stem cell population was also observed in hiPSC lines (S3C Fig). The CV of hiPSC1 is significantly higher than hiPSC2 and 3 (p = 0.0001284737 for hiPSC1 versus hiPSC2 and p = 0.01884844 for hiPSC1 versus hiPSC3) [53]. Because the stem cell populations with larger variation also had higher mean G1 length, we analyzed the relationship between G1 length and variation. Single-cell G1 length data of cells grown in mTeSR1 were divided into cells with G1 lengths longer or shorter than 6 h. The 6-h cutoff was chosen because most of the cells in E8 had G1 lengths less than 6 h. Interestingly, cells with G1 < 6 h showed a similar level of G1 length variation compared with those in E8, whereas cells with G1 > 6 h had higher variation (Fig 3B). These data suggest that G1 length is related to single-cell variation. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Relationship between G1 length and single-cell variation. (A) Stdev, mean, and CV of the G1 length distributions shown in Fig 2C. (B) Stdev, mean, and CV of the mTeSR1 distribution (Fig 2C) divided by 6-h cutoff of G1 length. (C) Correlation plot of G1 lengths between sister cells (n = 55 from three independent experiments). (D) Ring plot showing the difference in G1 lengths between sister cells (n = 55 from three independent experiments). (E) Ratio of symmetric and asymmetric sister cell G1 durations in H9 cells grown either in E8 or in mTeSR1 (n = 56 for E8 and n = 55 for mTeSR1 pooled from three independent experiments). (F) Ratio of symmetric and asymmetric sister cell G1 durations in the mTeSR1 distribution divided by 6-h cutoff of G1 length (n = 55 pooled from three independent experiments). Underlying data can be found in S1 Data. CV, coefficient of variation; E8, Essential 8; Stdev, standard deviation. https://doi.org/10.1371/journal.pbio.3000453.g003 Next, we sought the source of single-cell variation in G1 length. Interestingly, G1 lengths between sister cells showed a good correlation (Fig 3C), suggesting that sister cells tend to have similar G1 lengths. However, cells that do not share G1 length between sisters are frequent (Fig 3D). Therefore, we hypothesized that single-cell variation in a population might arise from asymmetry of G1 length between sister cells. To test this hypothesis, we measured the difference of G1 length between sister cells (ΔG1) and divided ΔG1 by mean G1 length of sister cells (<G1>). We set ΔG1/<G1> less than 0.2 as symmetric sister cell G1 duration and ΔG1/<G1> greater than 0.2 as asymmetric sister cell G1 duration. Consistent with the large variation of G1 length, H9 cells in mTeSR1 showed a higher percentage of asymmetric sister cell G1 duration than those cells grown in E8 (Fig 3E). This pattern was also observed with sliding cutoffs (S5 Fig), suggesting that the selection of a specific cutoff does not determine the pattern. When the mTeSR1 data were divided into two groups, G1 < 6 h and G1 > 6 h, a higher percentage of asymmetric sister cell G1 duration was observed in cells with G1 > 6 h (Fig 3F). Cells with G1 < 6 h showed a similar percentage of asymmetric sister cell G1 duration compared with those in E8. Overall, these results suggest that long G1 length is related to increased asymmetric sister cell G1 duration and a larger single-cell variation in G1 length. WNT/β-catenin pathway controls G1 length distribution patterns To understand the controls over single-cell G1 length variation, we investigated an upstream regulator of G1 length in hESCs, the WNT/β-catenin pathway. This pathway directly controls the expression of many cell-cycle genes and plays a key role in embryonic stem cell (ESC) self-renewal [37–39,54,55]. The down-regulation of WNT/β-catenin drives mouse ESC (mESC) from naïve to primed pluripotency [56]. Furthermore, the unequal distribution of WNT/β-catenin pathway proteins during cell division induces asymmetric division of hESCs [57]. This evidence points to WNT/β-catenin pathway as a strong candidate to control G1 length variation in hESCs. To test this hypothesis, we first analyzed the endogenous activity of WNT/β-catenin pathway in H9 cells grown in either E8 or mTeSR1 by measuring a level of nuclear β-catenin proteins. Higher levels of nuclear β-catenin proteins, but not total proteins, were observed in E8 than in mTeSR1 (Fig 4A and 4B), which was confirmed by the expression of WNT/β-catenin target genes (Fig 4C). A WNT reporter, TOP-flash, also showed a higher mean green fluorescent protein (GFP) intensity in E8 than in mTeSR1 (S6A and S6B Fig). hiPSC2 and hiPSC3 with short G1 also showed higher expression of WNT/β-catenin target genes than hiPSC1 (S6C Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Population WNT levels determine distribution patterns of single-cell G1 length. (A and B) Western blot of β-catenin in nuclear (n = 3) and total (n = 4) fractions of H9 cells. (C) qPCR analysis of WNT target genes in H9 cells grown either in E8 or in mTeSR1 (n = 6). (D) Histograms for G1 length of H9 cells grown in mTeSR1 and treated with recombinant human WNT3A proteins (100 ng/ml) (n = 114 for mTeSR1 and n = 104 for mTeSR1 + WNT3A pooled from three independent experiments); U test: p-value < 2.2 × 10−16, KS test: p-value = 2.492 × 10−11. (E) Histograms for G1 length of H9 cells grown in E8 and treated with IWP-2 (5 μM) (n = 112 for E8 and n = 104 for E8 + IWP-2 pooled from three independent experiments); U test: p-value = 8.282 × 10−12, KS test: p-value = 1.167 × 10−9. Underlying data can be found in S1 Data. AXIN2, Axis inhibition protein 2; C-MYC, c-myc proto-oncogene; CV, coefficient of variation; E8, Essential 8; KS, Kolmogorov-Smirnov; N-MYC, n-myc proto-oncogene; n.s., not significant; OCT4, Octamer-binding transcription factor 4; qPCR, quantitative PCR; Stdev, standard deviation; WNT, Wingless-INT. https://doi.org/10.1371/journal.pbio.3000453.g004 Consistently, WNT activation by recombinant WNT3A proteins was sufficient to shift the G1 length distribution of mTeSR1 to that of E8 without affecting the expression of pluripotency genes and relative cell-cycle patterns (Fig 4D, S6D and S6E Fig). WNT activation induced direct binding of β-catenin around genomic loci of cyclins D1 and E2, which likely increased the expression of those genes and thereby contributed to G1 length shortening (S6F and S6G Fig) [58]. We also confirmed the WNT effect on G1 length in three clonal FUCCI lines (S6H Fig). Furthermore, WNT3A treatment reduced asymmetric sister cell G1 duration and single-cell variation (Fig 4D, S6I Fig). WNT activation also shifted the G1 distribution pattern of hiPSC1 toward those of hiPSC2 and hiPSC3 (S6J Fig). Furthermore, inhibition of WNT production by IWP-2, a Porcupine O-Acyltransferase (PORCN) inhibitor, increased the proportion of cells with longer G1 (Fig 4E, S6K Fig). Collectively, these results suggest that WNT/β-catenin pathway controls the single-cell distribution of G1 length in hESCs. Exponential relationship between WNT level and G1 length is captured by a Poisson model To gain a better insight of WNT control over G1 length, we analyzed the quantitative relationship between WNT level and G1 length. G1 length was measured with increasing doses of recombinant WNT3A proteins. Axis inhibition protein 2 (AXIN2) expression was used as a marker for endogenous WNT activity because AXIN2 expression is a general indicator of WNT/β-catenin pathway activity [59]. AXIN2 expression showed a linear correlation with the amount of WNT3A proteins added (Fig 5A); hence, WNT3A proteins were not present in saturation conditions. In striking contrast, we observed an exponential relationship between AXIN2 expression and G1 length (Fig 5B). Based on this observation, we pursued a regression model to better understand the nature of G1 length distribution in hESCs (for model details, see Materials and methods). G1 lengths can be modeled as a sequence of exponentially distributed intracellular events. It was reported that a gamma distribution, or a shifted gamma distribution, provides good fits for these types of distributions [60–62]. In particular, plotting G1 length by shifting its value to the origin, such that G1* = G1 − min(G1), we can, to high accuracy, describe G1* with a generalized Poisson distribution (which is a special case of a gamma distribution). We start our analysis by illustrating that G1* ~ Poisson(μ), where μ = <G1*> is the average of G1*. Then, by regressing the experimental data using a Poisson regression as follows: where g(.) = log is called “link function,” covariate x is defined as AXIN2 expression for G1* as a response variable (i.e., μ). We used a q-q plot to examine the validity of our regression model (S7A Fig). The above Poisson regression model trained by the hESC data with various doses of WNT3A proteins (Fig 5B) results in approximate model parameters (adjusted R-squared = 0.86) (Fig 5C). Next, this model was tested on all data, which shows a reasonable agreement with the experimental data (adjusted R-squared = 0.84) (Fig 5D). Our analysis showed that each G1 length distribution closely followed a Poisson distribution (Fig 5D, S7A and S7B Fig). Hence, we propose a Poisson distribution for G1* as well as an exponential relation between <G1*> and WNT level as follows: where , Г: gamma function ∴ Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. A Poisson model predicts single-cell distribution patterns of G1 length based on population WNT levels. (A) Correlation between AXIN2 expression and the dose of recombinant human WNT3A proteins added to H9 cells in mTeSR1 (n = 4). (B) Relationship between average G1 length and population AXIN2 expression in H9 cells treated with various doses of recombinant human WNT3A proteins (n = 114 for 0 ng/ml, n = 42 for 10 ng/ml, n = 48 for 20 ng/ml, and n = 104 for 100 ng/ml pooled from two to three independent experiments). Error bars represent SEM. (C) Poisson regression trained with the data in Fig 5B. (D) Poisson regression analysis for combined hESCs and hiPSCs. G1 length is average of multiple single-cell G1 lengths. AXIN2 expression was measured by qPCR (n = 4). Underlying data can be found in S1 Data. AXIN2, Axis inhibition protein 2; E8, Essential 8; GAPDH, glyceraldehyde 3-phosphate dehydrogenase; GLM, generalized linear model; hESC, human embryonic stem cell; hiPSC, human induced pluripotent stem cell; qPCR, quantitative PCR; WNT, Wingless-INT. https://doi.org/10.1371/journal.pbio.3000453.g005 The equation recapitulates the key features of our experimental data in which low WNT levels result in increased mean G1 length and higher CV (, see Materials and methods) and, thus, link long G1 length to high single-cell variation. Taken together, these data suggest that single-cell distribution patterns of G1 length can be predicted by our Poisson model based on given population WNT levels, highlighting the important role of WNT in establishing G1 length distribution patterns of stem cell populations. G1 length–driven 5-hmC accumulation underlies NE differentiation potential of hESCs Finally, we sought the molecular mechanism for G1 length control over differentiation outcomes of hESC populations. SMADs are important regulators of hESC differentiation. Recently, it was reported that SMAD2/3 activity changes during the cell cycle [25]. Therefore, we analyzed SMAD2/3 activity by measuring nuclear levels of SMAD2/3 proteins. Despite different G1 lengths, H9 cells showed no significant difference in SMAD2/3 activity under mTeSR1 or E8 media conditions (S8A Fig). Furthermore, G1 shortening by CDK4R24C or CDK6R31C did not affect nuclear shuttling of SMAD2/3 proteins (S8B Fig), suggesting that absolute G1 length is not related to SMAD2/3 activity. The 5-hmC converted from DNA methylation (5-methylcystosine [5-mC]) by Ten-eleven translocation (TET) family proteins plays a role in global DNA demethylation [63,64]. The 5-hmC levels in promoters and enhancers correlate with open chromatin structures and gene activation [65–67]. Recently, it was reported that global 5-hmC levels increase during G1 phase in hESCs and that genomic accumulation of 5-hmC is a time-dependent process [31,68]. Based on this evidence, we asked whether 5-hmC could be a key mediator linking G1 length distributions to differentiation outcomes of hESC populations by priming lineage genes. First, we obtained genome-wide 5-hmC profiles in hESCs grown in mTeSR1 at a base resolution level using a unique DNA modification–dependent restriction endonuclease AbaSI (S9A Fig) [68]. The 5-hmC levels were measured across whole gene bodies (including 1 kb upstream and downstream of genes). hESC-specific and lineage-specific genes are defined by gene expression patterns during early hESC differentiation using our published RNA-seq data [69]. Interestingly, when quantitatively comparing 5-hmC levels of lineage-specific genes with total background, we observed significantly higher 5-hmC levels for lineage-specific genes (U test p-value < 2.2 × 10−16) (Fig 6A). Furthermore, lineage-specific genes showed as high 5-hmC levels as hESC-specific genes even though the expression of lineage-specific genes is much lower than hESC-specific genes in hESCs (U test p-value < 2.2 × 10−16) (Fig 6A, S9B Fig). These data suggest a potential role of 5-hmC in priming lineage-specific gene activation upon hESC differentiation. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. G1 length–driven 5-hmC accumulation is necessary for NE differentiation. (A) Empirical cumulative distribution of 5-hmC levels measured across whole gene bodies (including 1 kb upstream and downstream of genes) in H9 hESCs grown in mTeSR1 (U test: p-value < 2.2 × 10−16 for Lineage-specific versus Total, p-value < 2.2 × 10−16 for hESC-specific versus Total, and p-value = 0.04956 for Lineage-specific versus hESC-specific). (B) Immunofluorescence assay for 5-hmC in H9 cells grown either in mTeSR1 or in E8 (n = 187 for mTeSR1 and n = 202 for E8 from three independent experiments). (C and D) Immunofluorescence assay for 5-hmC in H9 cells grown in mTeSR1 overexpressing either CDK6R31C (C) or p21 (D) (panel C: n = 222 for Ctrl and n = 207 for CDK6R31C pooled from three independent experiments, panel D: n = 187 for Ctrl and n = 194 for p21 pooled from three independent experiments). (E) qPCR analysis of lineage markers in H9 cells transfected with TET1 siRNA and then differentiated for 8 d (n = 4). Error bars represent SD. *p < 0.01 (Student t test). Underlying data can be found in S1 Data. 5-hmC, 5-hydroxymethylcytosine; CDK, Cyclin-dependent kinase; Ctrl, control; Diff., differentiated; E8, Essential 8; GATA6, GATA binding protein 6; HAND1, Heart and neural crest derivatives expressed 1; hESC, human embryonic stem cell; ME, mesendoderm; NC, negative control; NE, neuroectoderm; OTX1, Orthodenticle homeobox 1; PAX6, Paired box 6; qPCR, quantitative PCR; RNA prep., RNA preparation; siRNA, small interfering RNA; TET, Ten-eleven translocation; tf., transfection; Total, total protein coding genes; ZBTB16, Zinc finger and BTB domain containing 16. https://doi.org/10.1371/journal.pbio.3000453.g006 Next, to test whether the length of the G1 phase could influence global levels of 5-hmC in hESCs, we compared global 5-hmC levels of hESCs grown either in E8 or mTeSR1. As expected, mTeSR1 showed a significantly higher level of 5-hmC than E8, consistent with the longer G1 lengths observed in mTeSR1 (Fig 6B). However, there was no significant difference in the expression of TET genes (S9C Fig), which catalyzes conversion of 5-mC to 5-hmC [64]. These results imply that G1 length is tightly linked with 5-hmC levels. To further validate this, we showed that shortening G1 length by expressing constitutively active CDK6R31C decreased 5-hmC levels without affecting 5-mC levels (Fig 6C, S9D Fig). Furthermore, G1-phase lengthening by p21 overexpression increased 5-hmC levels (Fig 6D). Given the neurogenic potential of hESC populations with long single-cell G1 lengths, we tested whether 5-hmC accumulation is essential for NE derivation. To reduce global 5-hmC levels, we used TET1 small interfering RNAs (siRNAs) because TET1 is most abundant in hESCs among TET family genes (S9E–S9G Fig). Decreasing global 5-hmC levels significantly suppressed NE derivation upon differentiation (Fig 6E, S9H and S9I Fig). However, ME differentiation was not affected or even promoted by 5-hmC reduction (Fig 6E, S9H and S9I Fig). Overall, these results suggest G1 length–driven 5-hmC accumulation as a molecular mechanism for NE differentiation potential of hESC populations. Cell-to-cell variation in G1 and S/G2/M lengths in an hESC population To analyze absolute lengths of each phase of the cell cycle at a single-cell level, we used the FUCCI reporter [26]. Based on cell cycle–dependent degradation of chromatin licensing and DNA replication factor 1 (CDT1) and geminin DNA replication inhibitor (GEMININ) proteins, cells show no color in early G1, red in mid/late G1, and green in S/G2/M (S1A Fig). FUCCI hESCs showed high expression of pluripotency genes and activated markers for NE and ME upon differentiation (S1B and S1C Fig). Time-lapse images were taken from the FUCCI hESCs grown in mTeSR1 medium for 24–48 h with 10-min intervals (S1 and S2 Movies). Compared with previous reports [25,31], one obvious difference we found in our data is that early G1 phase is very short—less than one frame (10 min) in most cells (S1D Fig and S3 Movie). This discrepancy is likely due to the use of different fluorescent proteins. Pauklin and colleagues [25] used monomeric Kusabira orange 2 (mKO2) fluorescent protein fused to CDT1 (for G1 phase), whereas we used mcherry. The maturation half-time of mKO2 is much longer (1.2 h) than mcherry’s (15 min) [40,41], which could explain a higher percentage of early G1 cells (no color) in their results. To confirm these results, we derived three independent clonal lines from FUCCI hESCs and observed that almost all cells have a no-color phase shorter than 10 min (S1E Fig). These findings suggest that defining early G1 phase by the absence of fluorescent signal could be influenced by differences in maturation time of fluorescence proteins. The short maturation of mcherry–CDT1 fusion proteins more accurately represents the duration of the entire G1 phase and demonstrates that early G1 contributes very little to overall G1 length. Individual hESCs showed high variation in the lengths of G1 and S/G2/M phases, with coefficients of variation (CVs) 36.6% and 18.6%, respectively (Fig 1A). G1 lengths ranged from 4 h to 10 h, and S/G2/M lengths ranged from 20 h to 40 h (Fig 1B). Interestingly, G1 lengths inversely correlated with S/G2/M length (p < 0.0001) in the same cell (Fig 1C), suggesting that cells with short G1 tend to have long S/G2/M and vice versa. This inverse correlation suggests that individual cells conserve total cell-cycle length through compensatory adjustments of cell-cycle phases. Recently, a compensatory relationship between S- and G2-phase lengths was observed in chordate ascidians during gastrulation [42]. Overall, these results suggest that hESCs have high single-cell heterogeneity in G1 and S/G2/M lengths under self-renewing conditions. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Single-cell heterogeneity in G1 and S/G2/M lengths in self-renewing hESCs. (A) Absolute cell-cycle length of each cell-cycle state in H9 hESCs grown in mTeSR1 medium (n = 114 for G1 and n = 60 for S/G2/M). (B) Histograms for G1 and S/G2/M lengths. (C) Inverse relationship between G1 and S/G2/M lengths in individual cells (n = 30). Both G1 and S/G2/M lengths were measured from the same cell. Data in (C) are a subset of those in (A). Data were collected and pooled from three independent experiments. Underlying data can be found in S1 Data. hESC, human embryonic stem cell. https://doi.org/10.1371/journal.pbio.3000453.g001 Distribution patterns of single-cell G1 length determine differentiation outcomes of stem cell populations G1 phase has the potential to make cell fate decisions in response to extracellular cues, whereas cells in S/G2/M phase do not respond to differentiation cues [25,30,31,43]. The different response of each cell-cycle state to differentiation cues is mechanistically supported by evidence that epigenetic modifications are regulated in a cell cycle–dependent manner [31,44–46]. Recent papers nicely showed that gene activation–related epigenetic markers such as 5-hydroxymethylcytosine and histone 3 lysine 4 trimethylation (H3K4me3) accumulated in lineage-specific genes as the G1 phase progresses [31,45,46]. Therefore, we hypothesized that individual hESCs differ in their differentiation potential based upon their absolute G1 lengths. To broach this question, we compared H9 hESCs in two well-defined feeder-free stem cell media, Essential 8 (E8) and mTeSR1. These two different media conditions gave us a unique opportunity to study the correlation between G1 length and differentiation potential in an identical cell line. Consistent with the robust ability of both media to maintain pluripotency, H9 cells showed high expression levels of key pluripotency genes (Octamer-binding transcription factor 4 [OCT4], NANOG, Sex determining region Y-box 2 [SOX2]) under both conditions (S2A Fig). To measure the differentiation potential, we spontaneously differentiated H9 cells in hESC medium (see Materials and methods) without FGF2. In this differentiation condition, H9 cells in E8 exhibited highly biased differentiation toward ME, whereas H9 cells in mTeSR1 differentiated more equally to both NE and ME lineages (Fig 2A). The mRNA levels of lineage markers showed consistent patterns with the flow cytometry data (S2B Fig). Lineage-specific differentiation protocols, dual SMAD inhibition for NE differentiation, and FGF2 and BMP4 for ME differentiation also confirmed that hESCs differ in their differentiation propensity depending on the conditions under which they were grown before differentiation (S2C Fig) [47,48]. Both media conditions (mTeSR1 and E8) showed, as expected, similar cell-cycle patterns, with a low percentage of cells in G1 and a high percentage of cells in S/G2/M (Fig 2B, S2D Fig). These results were further confirmed by using 5-bromo-2′-deoxyuridine (BrdU)/7-amino-actinomycin D (7-AAD), which clearly separates S phase from the others (S2E Fig). When absolute G1 length was measured from single cells, however, a dramatic difference emerged, suggesting that mTeSR1 and E8 represent two distinct distributions (p < 2.2 × 10−16 by the Kolmogorov-Smirnov test and p < 2.2 × 10−16 by the Mann-Whitney U test). Cells grown in E8 showed a loss of the long G1 population and consequently a decreased mean G1 length (Fig 2C). Biological replicates displayed similar mean and variation (S2F Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Single-cell distribution patterns of absolute G1 length determine differentiation propensity of hESC populations. (A) Immunoflow cytometry of PAX6 and GATA6 in H9 cells differentiated for 8 d by FGF2 deprivation (n = 4). (B) Propidium iodide staining analysis of H9 cells either in E8 or in mTeSR1 media and human dermal fibroblasts (n = 4 for E8 and mTeSR1, n = 3 for fibroblasts). (C) Histograms for G1 length of H9 cells grown either in E8 or in mTeSR1 (n = 112 for E8 and n = 114 for mTeSR1 pooled from three independent experiments); U test: p-value < 2.2 × 10−16, KS test: p-value < 2.2 × 10−16. (D-F) qPCR analysis of lineage markers in differentiated day 8 H9 cells overexpressing p21 (D), CDK4R24C (E), or CDK6R31C (F) (n = 4). Ctrl represents cells transduced with empty lentiviral vectors and treated with the same dose of doxycycline (1 μg/ml). Transgene expression was turned off at the onset of differentiation by doxycycline withdrawal. Error bars represent SD. *p < 0.01 (Student t test). Underlying data can be found in S1 Data. CDK, Cyclin-dependent kinase; Ctrl, control; Diff. differentiated; E8, Essential 8; FSC, forward scatter; FGF, Fibroblast growth factor; FL, fluorescence intensity; GATA6, GATA binding protein 6; HAND1, Heart and neural crest derivatives expressed 1; hESC, human embryonic stem cell; KS, Kolmogorov-Smirnov; OTX1, Orthodenticle homeobox 1; PAX6, Paired box 6; qPCR, quantitative PCR; Undiff., undifferentiated; ZBTB16, Zinc finger and BTB domain containing 16. https://doi.org/10.1371/journal.pbio.3000453.g002 Because the G1/S checkpoint plays a key role in determining G1 length, we measured the expression of cyclins. Consistent with a short average G1 length, H9 cells in E8 medium showed higher expression of cyclins D and E, which facilitate the G1/S transition (S2G and S2H Fig), but not cyclins A and B (S2G and S2H Fig). Given the ME-biased potential of E8 in the context of our spontaneous differentiation condition, we assumed that the population of cells with long G1 might be responsible for the NE lineage preference. To further validate this idea, we compared three well-defined human induced pluripotent stem cell (hiPSC) lines (S3A Fig) [49]. As high variability in differentiation potential was reported in iPSC lines [50–52], the three hiPSC lines have a dramatic difference in differentiation propensity (S3B Fig). Consistent with the hESC data (Fig 2C), highly neurogenic hiPSC1 contained a substantial population of cells with longer G1 compared with hiPSC2 and hiPSC3 (S3C Fig). However, all hiPSC lines maintained similar relative proportions of cells in each cell-cycle state (S3D Fig), emphasizing that relative cell-cycle length is not related to the differentiation potential of human pluripotent stem cells (hPSCs). Rather, variation in length among the same number of cells in G1 establishes differentiation potential. To test the functional relationship between absolute G1 length and cell fates, we modulated G1 length in self-renewing H9 cells by overexpressing p21 or constitutively active Cyclin-dependent kinases (CDK4R24C and CDK6R31C) (S4A Fig). We used a doxycycline-dependent lentiviral system to control the transgene expression. Overexpression of p21 increased the average length of G1, whereas CDK4R24C and CDK6R31C reduced it (S4B Fig). The modulation of absolute G1 length did not affect pluripotency gene expression and the relative cell-cycle patterns (S4A, S4C and S4D Fig), which suggests that hESCs can tolerate high variation in G1 length. Transgene expression was turned off at the onset of differentiation to exclude any potential effect of the transgene on differentiation (S4E Fig). Transient increase of G1 length by p21 further promoted NE derivation at the expense of ME differentiation (Fig 2D). These results were confirmed by abemaciclib, a potent and selective chemical inhibitor of CDK4 and CDK6. Abemaciclib treatment for 18 h before differentiation phenocopied the effect of p21 overexpression (S4F Fig). Transient decrease of G1 length by CDK4R24C and CDK6R31C showed the opposite effect (Fig 2E and 2F). Furthermore, increased G1 length by p21 overexpression in H9 cells grown in E8 medium was sufficient to improve neurogenic potential to the level of those in mTeSR1 at the expense of ME lineage derivation (S4G Fig), suggesting that the difference in differentiation propensity between the two media conditions can be attributed to G1 length. Collectively, these results suggest that G1 length biases the differentiation potential of self-renewing hESCs, and thus, the G1 length distribution patterns determine population fates upon differentiation. Asymmetric sister cell G1 duration contributes to heterogeneity in single-cell G1 length Cells grown in mTeSR1 have not only an increased mean G1 length but also a greater CV than those of E8 (Fig 3A). The larger variation of G1 length in a highly neurogenic stem cell population was also observed in hiPSC lines (S3C Fig). The CV of hiPSC1 is significantly higher than hiPSC2 and 3 (p = 0.0001284737 for hiPSC1 versus hiPSC2 and p = 0.01884844 for hiPSC1 versus hiPSC3) [53]. Because the stem cell populations with larger variation also had higher mean G1 length, we analyzed the relationship between G1 length and variation. Single-cell G1 length data of cells grown in mTeSR1 were divided into cells with G1 lengths longer or shorter than 6 h. The 6-h cutoff was chosen because most of the cells in E8 had G1 lengths less than 6 h. Interestingly, cells with G1 < 6 h showed a similar level of G1 length variation compared with those in E8, whereas cells with G1 > 6 h had higher variation (Fig 3B). These data suggest that G1 length is related to single-cell variation. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Relationship between G1 length and single-cell variation. (A) Stdev, mean, and CV of the G1 length distributions shown in Fig 2C. (B) Stdev, mean, and CV of the mTeSR1 distribution (Fig 2C) divided by 6-h cutoff of G1 length. (C) Correlation plot of G1 lengths between sister cells (n = 55 from three independent experiments). (D) Ring plot showing the difference in G1 lengths between sister cells (n = 55 from three independent experiments). (E) Ratio of symmetric and asymmetric sister cell G1 durations in H9 cells grown either in E8 or in mTeSR1 (n = 56 for E8 and n = 55 for mTeSR1 pooled from three independent experiments). (F) Ratio of symmetric and asymmetric sister cell G1 durations in the mTeSR1 distribution divided by 6-h cutoff of G1 length (n = 55 pooled from three independent experiments). Underlying data can be found in S1 Data. CV, coefficient of variation; E8, Essential 8; Stdev, standard deviation. https://doi.org/10.1371/journal.pbio.3000453.g003 Next, we sought the source of single-cell variation in G1 length. Interestingly, G1 lengths between sister cells showed a good correlation (Fig 3C), suggesting that sister cells tend to have similar G1 lengths. However, cells that do not share G1 length between sisters are frequent (Fig 3D). Therefore, we hypothesized that single-cell variation in a population might arise from asymmetry of G1 length between sister cells. To test this hypothesis, we measured the difference of G1 length between sister cells (ΔG1) and divided ΔG1 by mean G1 length of sister cells (<G1>). We set ΔG1/<G1> less than 0.2 as symmetric sister cell G1 duration and ΔG1/<G1> greater than 0.2 as asymmetric sister cell G1 duration. Consistent with the large variation of G1 length, H9 cells in mTeSR1 showed a higher percentage of asymmetric sister cell G1 duration than those cells grown in E8 (Fig 3E). This pattern was also observed with sliding cutoffs (S5 Fig), suggesting that the selection of a specific cutoff does not determine the pattern. When the mTeSR1 data were divided into two groups, G1 < 6 h and G1 > 6 h, a higher percentage of asymmetric sister cell G1 duration was observed in cells with G1 > 6 h (Fig 3F). Cells with G1 < 6 h showed a similar percentage of asymmetric sister cell G1 duration compared with those in E8. Overall, these results suggest that long G1 length is related to increased asymmetric sister cell G1 duration and a larger single-cell variation in G1 length. WNT/β-catenin pathway controls G1 length distribution patterns To understand the controls over single-cell G1 length variation, we investigated an upstream regulator of G1 length in hESCs, the WNT/β-catenin pathway. This pathway directly controls the expression of many cell-cycle genes and plays a key role in embryonic stem cell (ESC) self-renewal [37–39,54,55]. The down-regulation of WNT/β-catenin drives mouse ESC (mESC) from naïve to primed pluripotency [56]. Furthermore, the unequal distribution of WNT/β-catenin pathway proteins during cell division induces asymmetric division of hESCs [57]. This evidence points to WNT/β-catenin pathway as a strong candidate to control G1 length variation in hESCs. To test this hypothesis, we first analyzed the endogenous activity of WNT/β-catenin pathway in H9 cells grown in either E8 or mTeSR1 by measuring a level of nuclear β-catenin proteins. Higher levels of nuclear β-catenin proteins, but not total proteins, were observed in E8 than in mTeSR1 (Fig 4A and 4B), which was confirmed by the expression of WNT/β-catenin target genes (Fig 4C). A WNT reporter, TOP-flash, also showed a higher mean green fluorescent protein (GFP) intensity in E8 than in mTeSR1 (S6A and S6B Fig). hiPSC2 and hiPSC3 with short G1 also showed higher expression of WNT/β-catenin target genes than hiPSC1 (S6C Fig). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Population WNT levels determine distribution patterns of single-cell G1 length. (A and B) Western blot of β-catenin in nuclear (n = 3) and total (n = 4) fractions of H9 cells. (C) qPCR analysis of WNT target genes in H9 cells grown either in E8 or in mTeSR1 (n = 6). (D) Histograms for G1 length of H9 cells grown in mTeSR1 and treated with recombinant human WNT3A proteins (100 ng/ml) (n = 114 for mTeSR1 and n = 104 for mTeSR1 + WNT3A pooled from three independent experiments); U test: p-value < 2.2 × 10−16, KS test: p-value = 2.492 × 10−11. (E) Histograms for G1 length of H9 cells grown in E8 and treated with IWP-2 (5 μM) (n = 112 for E8 and n = 104 for E8 + IWP-2 pooled from three independent experiments); U test: p-value = 8.282 × 10−12, KS test: p-value = 1.167 × 10−9. Underlying data can be found in S1 Data. AXIN2, Axis inhibition protein 2; C-MYC, c-myc proto-oncogene; CV, coefficient of variation; E8, Essential 8; KS, Kolmogorov-Smirnov; N-MYC, n-myc proto-oncogene; n.s., not significant; OCT4, Octamer-binding transcription factor 4; qPCR, quantitative PCR; Stdev, standard deviation; WNT, Wingless-INT. https://doi.org/10.1371/journal.pbio.3000453.g004 Consistently, WNT activation by recombinant WNT3A proteins was sufficient to shift the G1 length distribution of mTeSR1 to that of E8 without affecting the expression of pluripotency genes and relative cell-cycle patterns (Fig 4D, S6D and S6E Fig). WNT activation induced direct binding of β-catenin around genomic loci of cyclins D1 and E2, which likely increased the expression of those genes and thereby contributed to G1 length shortening (S6F and S6G Fig) [58]. We also confirmed the WNT effect on G1 length in three clonal FUCCI lines (S6H Fig). Furthermore, WNT3A treatment reduced asymmetric sister cell G1 duration and single-cell variation (Fig 4D, S6I Fig). WNT activation also shifted the G1 distribution pattern of hiPSC1 toward those of hiPSC2 and hiPSC3 (S6J Fig). Furthermore, inhibition of WNT production by IWP-2, a Porcupine O-Acyltransferase (PORCN) inhibitor, increased the proportion of cells with longer G1 (Fig 4E, S6K Fig). Collectively, these results suggest that WNT/β-catenin pathway controls the single-cell distribution of G1 length in hESCs. Exponential relationship between WNT level and G1 length is captured by a Poisson model To gain a better insight of WNT control over G1 length, we analyzed the quantitative relationship between WNT level and G1 length. G1 length was measured with increasing doses of recombinant WNT3A proteins. Axis inhibition protein 2 (AXIN2) expression was used as a marker for endogenous WNT activity because AXIN2 expression is a general indicator of WNT/β-catenin pathway activity [59]. AXIN2 expression showed a linear correlation with the amount of WNT3A proteins added (Fig 5A); hence, WNT3A proteins were not present in saturation conditions. In striking contrast, we observed an exponential relationship between AXIN2 expression and G1 length (Fig 5B). Based on this observation, we pursued a regression model to better understand the nature of G1 length distribution in hESCs (for model details, see Materials and methods). G1 lengths can be modeled as a sequence of exponentially distributed intracellular events. It was reported that a gamma distribution, or a shifted gamma distribution, provides good fits for these types of distributions [60–62]. In particular, plotting G1 length by shifting its value to the origin, such that G1* = G1 − min(G1), we can, to high accuracy, describe G1* with a generalized Poisson distribution (which is a special case of a gamma distribution). We start our analysis by illustrating that G1* ~ Poisson(μ), where μ = <G1*> is the average of G1*. Then, by regressing the experimental data using a Poisson regression as follows: where g(.) = log is called “link function,” covariate x is defined as AXIN2 expression for G1* as a response variable (i.e., μ). We used a q-q plot to examine the validity of our regression model (S7A Fig). The above Poisson regression model trained by the hESC data with various doses of WNT3A proteins (Fig 5B) results in approximate model parameters (adjusted R-squared = 0.86) (Fig 5C). Next, this model was tested on all data, which shows a reasonable agreement with the experimental data (adjusted R-squared = 0.84) (Fig 5D). Our analysis showed that each G1 length distribution closely followed a Poisson distribution (Fig 5D, S7A and S7B Fig). Hence, we propose a Poisson distribution for G1* as well as an exponential relation between <G1*> and WNT level as follows: where , Г: gamma function ∴ Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. A Poisson model predicts single-cell distribution patterns of G1 length based on population WNT levels. (A) Correlation between AXIN2 expression and the dose of recombinant human WNT3A proteins added to H9 cells in mTeSR1 (n = 4). (B) Relationship between average G1 length and population AXIN2 expression in H9 cells treated with various doses of recombinant human WNT3A proteins (n = 114 for 0 ng/ml, n = 42 for 10 ng/ml, n = 48 for 20 ng/ml, and n = 104 for 100 ng/ml pooled from two to three independent experiments). Error bars represent SEM. (C) Poisson regression trained with the data in Fig 5B. (D) Poisson regression analysis for combined hESCs and hiPSCs. G1 length is average of multiple single-cell G1 lengths. AXIN2 expression was measured by qPCR (n = 4). Underlying data can be found in S1 Data. AXIN2, Axis inhibition protein 2; E8, Essential 8; GAPDH, glyceraldehyde 3-phosphate dehydrogenase; GLM, generalized linear model; hESC, human embryonic stem cell; hiPSC, human induced pluripotent stem cell; qPCR, quantitative PCR; WNT, Wingless-INT. https://doi.org/10.1371/journal.pbio.3000453.g005 The equation recapitulates the key features of our experimental data in which low WNT levels result in increased mean G1 length and higher CV (, see Materials and methods) and, thus, link long G1 length to high single-cell variation. Taken together, these data suggest that single-cell distribution patterns of G1 length can be predicted by our Poisson model based on given population WNT levels, highlighting the important role of WNT in establishing G1 length distribution patterns of stem cell populations. G1 length–driven 5-hmC accumulation underlies NE differentiation potential of hESCs Finally, we sought the molecular mechanism for G1 length control over differentiation outcomes of hESC populations. SMADs are important regulators of hESC differentiation. Recently, it was reported that SMAD2/3 activity changes during the cell cycle [25]. Therefore, we analyzed SMAD2/3 activity by measuring nuclear levels of SMAD2/3 proteins. Despite different G1 lengths, H9 cells showed no significant difference in SMAD2/3 activity under mTeSR1 or E8 media conditions (S8A Fig). Furthermore, G1 shortening by CDK4R24C or CDK6R31C did not affect nuclear shuttling of SMAD2/3 proteins (S8B Fig), suggesting that absolute G1 length is not related to SMAD2/3 activity. The 5-hmC converted from DNA methylation (5-methylcystosine [5-mC]) by Ten-eleven translocation (TET) family proteins plays a role in global DNA demethylation [63,64]. The 5-hmC levels in promoters and enhancers correlate with open chromatin structures and gene activation [65–67]. Recently, it was reported that global 5-hmC levels increase during G1 phase in hESCs and that genomic accumulation of 5-hmC is a time-dependent process [31,68]. Based on this evidence, we asked whether 5-hmC could be a key mediator linking G1 length distributions to differentiation outcomes of hESC populations by priming lineage genes. First, we obtained genome-wide 5-hmC profiles in hESCs grown in mTeSR1 at a base resolution level using a unique DNA modification–dependent restriction endonuclease AbaSI (S9A Fig) [68]. The 5-hmC levels were measured across whole gene bodies (including 1 kb upstream and downstream of genes). hESC-specific and lineage-specific genes are defined by gene expression patterns during early hESC differentiation using our published RNA-seq data [69]. Interestingly, when quantitatively comparing 5-hmC levels of lineage-specific genes with total background, we observed significantly higher 5-hmC levels for lineage-specific genes (U test p-value < 2.2 × 10−16) (Fig 6A). Furthermore, lineage-specific genes showed as high 5-hmC levels as hESC-specific genes even though the expression of lineage-specific genes is much lower than hESC-specific genes in hESCs (U test p-value < 2.2 × 10−16) (Fig 6A, S9B Fig). These data suggest a potential role of 5-hmC in priming lineage-specific gene activation upon hESC differentiation. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. G1 length–driven 5-hmC accumulation is necessary for NE differentiation. (A) Empirical cumulative distribution of 5-hmC levels measured across whole gene bodies (including 1 kb upstream and downstream of genes) in H9 hESCs grown in mTeSR1 (U test: p-value < 2.2 × 10−16 for Lineage-specific versus Total, p-value < 2.2 × 10−16 for hESC-specific versus Total, and p-value = 0.04956 for Lineage-specific versus hESC-specific). (B) Immunofluorescence assay for 5-hmC in H9 cells grown either in mTeSR1 or in E8 (n = 187 for mTeSR1 and n = 202 for E8 from three independent experiments). (C and D) Immunofluorescence assay for 5-hmC in H9 cells grown in mTeSR1 overexpressing either CDK6R31C (C) or p21 (D) (panel C: n = 222 for Ctrl and n = 207 for CDK6R31C pooled from three independent experiments, panel D: n = 187 for Ctrl and n = 194 for p21 pooled from three independent experiments). (E) qPCR analysis of lineage markers in H9 cells transfected with TET1 siRNA and then differentiated for 8 d (n = 4). Error bars represent SD. *p < 0.01 (Student t test). Underlying data can be found in S1 Data. 5-hmC, 5-hydroxymethylcytosine; CDK, Cyclin-dependent kinase; Ctrl, control; Diff., differentiated; E8, Essential 8; GATA6, GATA binding protein 6; HAND1, Heart and neural crest derivatives expressed 1; hESC, human embryonic stem cell; ME, mesendoderm; NC, negative control; NE, neuroectoderm; OTX1, Orthodenticle homeobox 1; PAX6, Paired box 6; qPCR, quantitative PCR; RNA prep., RNA preparation; siRNA, small interfering RNA; TET, Ten-eleven translocation; tf., transfection; Total, total protein coding genes; ZBTB16, Zinc finger and BTB domain containing 16. https://doi.org/10.1371/journal.pbio.3000453.g006 Next, to test whether the length of the G1 phase could influence global levels of 5-hmC in hESCs, we compared global 5-hmC levels of hESCs grown either in E8 or mTeSR1. As expected, mTeSR1 showed a significantly higher level of 5-hmC than E8, consistent with the longer G1 lengths observed in mTeSR1 (Fig 6B). However, there was no significant difference in the expression of TET genes (S9C Fig), which catalyzes conversion of 5-mC to 5-hmC [64]. These results imply that G1 length is tightly linked with 5-hmC levels. To further validate this, we showed that shortening G1 length by expressing constitutively active CDK6R31C decreased 5-hmC levels without affecting 5-mC levels (Fig 6C, S9D Fig). Furthermore, G1-phase lengthening by p21 overexpression increased 5-hmC levels (Fig 6D). Given the neurogenic potential of hESC populations with long single-cell G1 lengths, we tested whether 5-hmC accumulation is essential for NE derivation. To reduce global 5-hmC levels, we used TET1 small interfering RNAs (siRNAs) because TET1 is most abundant in hESCs among TET family genes (S9E–S9G Fig). Decreasing global 5-hmC levels significantly suppressed NE derivation upon differentiation (Fig 6E, S9H and S9I Fig). However, ME differentiation was not affected or even promoted by 5-hmC reduction (Fig 6E, S9H and S9I Fig). Overall, these results suggest G1 length–driven 5-hmC accumulation as a molecular mechanism for NE differentiation potential of hESC populations. Discussion Whereas relative G1 length as reflected in the ratio of S/G2/M phase to G0/G1 phase by propidium iodide fluorescence reflects the population [21,30], absolute G1 length distributions across single cells reflect stem cell duality by establishing a differentiation bias while maintaining self-renewal properties. The maintenance of the stem cell population is buffered against variation in single-cell G1 lengths, whereas, on the other hand, this variation is exploited to generate multiple fates from a stem cell population. Furthermore, we found that single-cell variation in G1 length can be determined by the environmental WNT level—i.e., WNT level in the media—thereby providing an example of how a uniform influence on the entire population can affect single-cell variation. Modeling revealed that population WNT levels account for the G1 length distributions remarkably well, which suggests that in culture, WNT can be an important environmental factor in the establishment of the G1 length distribution. Given the complexity in environmental factors, however, we note that WNT is not likely to be the only factor controlling G1 length and that other signaling molecules potentially cooperate on G1 length regulation. Reduced levels of environmental WNT increase asymmetric sister cell G1 duration and G1 length variation in a stem cell population. Stochasticity of the WNT control over G1 length and “finite number effect” could underlie this observation [70]. G1 length distribution can represent the collective action of multiple genes. Because the effect of WNT levels on G1 length distributions is an exponential function, the steep decline represented by this function will create a sharp boundary of cell fates along a falling gradient of WNT levels. Given the pleiotropic functions of WNT in development, G1 length control is likely one of many effects induced by WNT. Consistent with extensive cross talk between signaling pathways, we observed that WNT activation increases Transforming growth factor Beta (TGF-β)–SMAD2/3 activity in hESCs (S8C Fig). However, this observation seems to be independent of G1 length because G1 length modulation had no effect on SMAD2/3 activity (S8A and S8B Fig). Therefore, we suggest that WNT activation affects the differentiation propensity of a stem cell population by shifting G1 length distributions, whereas G1 length–independent functions of WNT also contribute to stem cell fate decisions. Increased G1 length after differentiation has been observed in ESCs and neural stem cells. Furthermore, modulation of G1 length affects stem cell differentiation [21,23,71]. We found that absolute G1 lengths among hESCs bias the differentiation potential of the population. Transient modulation of absolute G1 lengths before differentiation was sufficient to influence differentiation outcome of hESCs. Our model differs from, but does not exclude, the one by Pauklin and colleagues [25]. Pauklin and colleagues [25] used a FUCCI reporter to sort hESCs into early and late G1 cells and found that late G1 cells predominantly differentiate into NE, whereas early G1 cells become ME. By taking a “snapshot” of captured cells, this model does not address the dynamic nature of the population from one generation to the next. Our model explains these experimental results well, given that sorted late G1 populations would be enriched for long G1 cells. Furthermore, the boundary between early and late G1 phases as described by Pauklin and colleagues [25] depends upon the maturation time of fluorescent proteins and, therefore, upon the specific fluorescent proteins used in a reporter system. In contrast, our model adopts an explicit value, absolute length of the G1 phase in single cells. Our single-cell data from time-lapse imaging capture important biological features underlying stem cell differentiation potential, which cannot be seen in a population analysis. For example, Pauklin and colleagues [25] fail to explain dramatic differences in differentiation potential between hESCs grown either in mTeSR1 or in E8 medium with a similar percentage of cells in G1 phase. In contrast, our work identified the important difference in single-cell distributions of absolute G1 length that are responsible for differentiation competency of stem cell populations. We propose that the absolute lengths of G1 phase establish bias and contribute to the differentiation fates of hPSCs. Therefore, hPSC lines containing predominantly short G1 cells cannot efficiently contribute to NE even though they have a similar number of cells in G1 phase. hPSC lines containing a variable subpopulation of long-G1 cells are capable of deriving NE upon differentiation. These results clearly show that stem cell populations with longer G1 lengths favor NE lineage over ME. On the other hand, the question of differentiation bias in a single cell is more difficult for many reasons; among them is the likely change in G1 length (and consequently bias) from generation to generation. Challenges for imaging over weeks while tracking cell location make this technically difficult. From a cell population perspective, we showed that a G1 length–dependent accumulation of 5-hmC favors NE gene activation and that this epigenetic modification is therefore a likely contributor to population bias. However, G1 length likely affects many cellular processes, and 5-hmC is probably one of many mechanisms potentially linking G1 phase to stem cell fates. Overall, our findings suggest that, in a cell culture system, WNT-driven G1 length distributions play a key role in determining differentiation potential of stem cell populations. However, the use of this mechanism at the organismal level remains an open question. During development, tissue-specific stem cells change their differentiation potential depending on the context and needs. Dynamic control over G1 length distribution patterns could be an efficient mechanism to regulate differentiation potential of tissue-specific stem cells in dynamic embryonic environments. Materials and methods Cell culture H9 cells were maintained in feeder-free conditions on Matrigel (Corning) either in mTeSR1 (Stem Cell Technologies) or E8 (ThermoFisher Scientific) medium. H9 cells were authenticated by short tandem repeat analysis. hiPSC lines (provided by Dr. Yoav Gilad at University of Chicago) were grown on Matrigel in E8 medium [49]. Cells were passaged every 5–6 d by ReLeSR (Stem Cell Technologies). Cells were passaged every 4–5 d by Accutase (ThermoFisher Scientific). HEK293T cells and human dermal fibroblasts were maintained in DMEM supplemented with 10% fetal bovine serum (FBS). Differentiation of hESCs and hiPSCs hESCs and hiPSCs were induced to differentiate on Matrigel in hESC culture medium (DMEM/F12, 15% knockout serum replacement, MEM nonessential amino acid solution, and 0.1 mM β-mercaptoethanol) without FGF2. For NE-directed differentiation, H9 cells were differentiated on Matrigel with hESC culture medium containing Noggin (200 ng/ml) and SB431542 (10 μM) [48]. For ME-directed differentiation, H9 cells were induced to differentiate in mTeSR1 medium containing 10 ng/ml BMP4 [47]. Generation of FUCCI cell lines and cell cycle–length measurement H9 hESCs and hiPSCs were transduced with a lentiviral vector expressing the FUCCI reporter (mCherry-hCDT1-T2A-mAG-hGEMININ). Transduced cells were selected by 300 μg/ml neomycin treatment. Cells expressing the FUCCI reporter were imaged by Olympus DSU confocal microscope for 24–48 h with 10-min interval in a chamber with 5% CO2 and 37 °C temperature. Duration of each cell-cycle state was measured by manual tracking with the following criteria: G1, from red on to green on; S/G2/M, from red off to green off. To minimize tracking bias, we selected cells randomly regardless of their position in the colony. Derivation of clonal FUCCI hESC lines H9 hESCs were transduced with a lentiviral vector expressing the FUCCI reporter and maintained with 300 μg/ml neomycin. After two rounds of passaging, cells were dissociated with Accutase (ThermoFisher Scientific), FACS sorted using FUCCI-driven GFP signals, and plated onto Matrigel-coated 96-well plates, which were preincubated with CloneR (Stem Cell Technologies) at a density of one cell per well. After 1 wk of media refreshing according to CloneR manufacturer’s protocol, clonal positive wells were selected with the following criteria: (1) only one colony, (2) round/compact colony shape, (3) robust GFP or mCherry expressions across the colony. Each of the clonal FUCCI lines were expanded by at least four rounds of passaging and then used for further experiments. Immunofluorescence Cells were fixed with 4% paraformaldehyde and permeabilized with 0.25% Triton X-100, followed by blocking with 10% FBS in PBS for 1 h. Samples were stained with primary antibodies for OCT4 (Santa Cruz Biotechnology, sc-5279), NANOG (R&D Systems, AF1997), SOX2 (Millipore, AB5603), PAX6 (Santa Cruz Biotechnology, sc-81649), and GATA6 (Cell Signaling Technology, 5851) overnight at 4 °C. Secondary antibody staining was performed for 1 h at room temperature with Alexa Fluor 488-donkey anti-goat IgG, Alexa Fluor 555-donkey anti-mouse IgG, and Alexa Fluor 555-donkey anti-rabbit IgG (ThermoFisher Scientific). For 5-hmC staining, cells were treated with 1.5 M HCl for 30 min at room temperature after 4% paraformaldehyde fixation (antibody: Active Motif, 39769). For 5-mC detection, cells were fixed with ice-cold 70% ethanol for 5 min, followed by 1.5 M HCl for 30 min at room temperature. Samples were blocked with 5% FBS and 0.3% Triton X-100 in PBS for 1 h and then stained with anti-5-mC antibody (Cell Signaling Technology, 28692) diluted in PBS with 1% BSA and 0.3% Triton X-100 overnight at 4 °C. Secondary antibody staining was performed for 1 h at room temperature with Alexa Fluor 555-donkey anti-rabbit IgG (ThermoFisher Scientific). All images were taken using Olympus IX71 fluorescence microscope. Western blotting RIPA lysis buffer (Millipore) was used to lyse cells in the presence of protease inhibitor cocktail (Roche). Protein concentration was measured using Pierce BCA Protein Assay Kit (ThermoFisher Scientific). Same amounts of protein were resolved by 10% SDS-PAGE, followed by transfer to nitrocellulose membranes (GE Healthcare Life Sciences). PBST (0.1% Tween 20 in PBS) containing 5% skim milk was used to block the membranes. Immunoblotting was performed overnight at 4 °C with antibodies for OCT4 (Santa Cruz Biotechnology, sc-9081), cyclin A2 (Cell Signaling Technology, #4656), cyclin B2 (Santa Cruz Biotechnology, sc-28303), cyclin D1 (Millipore, 04–221), cyclin E2 (Santa Cruz Biotechnology, sc-28351), CDK4 (Cell Signaling Technology, 12790), CDK6 (Cell Signaling Technology, 3136), p21 (Cell Signaling Technology, 2947), β-catenin (Millipore, 04–958), Smad2/3 (R&D systems, AF3797), and β-actin (Sigma-Aldrich, A5441). The membranes were stained with secondary antibodies Alexa Flour 680-goat anti-mouse IgG (ThermoFisher Scientific) and IRDye 800CW-goat anti-rabbit IgG (LI-COR) for 1 h at room temperature. Protein bands were visualized with LI-COR Odyssey Imaging System (LI-COR). Immunoflow cytometry Cells were fixed and permeabilized with Fixation/Permeabilization Solution Kit (BD Biosciences). Samples were stained with primary antibodies for PAX6 and GATA6 for 30 min at room temperature, followed by secondary antibody staining with Alexa Fluor 647-donkey anti-rabbit IgG and Alexa Fluor 647-donkey anti-mouse IgG (ThermoFisher Scientific) for 30 min at room temperature. Samples were analyzed by BD Accuri C6 flow cytometry (BD Biosciences). Propidium iodide staining Cells were dissociated into single cells by Accutase (Stem Cell Technologies). Dissociated cells were fixed with cold EtOH for 1 h on ice, followed by RNase treatment. Samples were stained with propidium iodide and analyzed by BD Accuri C6 flow cytometry (BD Biosciences). BrdU/7-AAD staining BrdU/7-AAD staining was conducted using APC BrdU Flow Kit (BD Pharmingen) according to the manufacturer’s protocol. Briefly, H9 cells were treated with BrdU (10 μM) for 15 min, washed with PBS twice, and dissociated into single cells by Accutase. In total, 106 cells per sample were then fixed, permeabilized, and treated with DNaseI (30 μg per 106 cells) for 1 h. Samples were stained with fluorochrome-conjugated anti-BrdU antibody for 20 min and washed, followed by resuspension in 7-AAD solution. At least 10,000 cells per each sample were analyzed by BD Accuri C6 flow cytometry (BD Biosciences). Quantitative real-time PCR (qPCR) Total RNAs were extracted using TRIzol Reagent (ThermoFisher Scientific), followed by reverse transcription with SuperScript III First-Strand Synthesis System (ThermoFisher Scientific). qPCR was performed with Power SYBR Green PCR Master Mix (Applied Biosystems) using QuantStudio 12K Flex Real-Time PCR System. GAPDH was used as a normalization control. siRNA transfection H9 cells were transfected with TET1 siRNAs (ThermoFisher Scientific HSS129586, HSS129587) (100 nM) with RNAiMAX (ThermoFisher Scientific) according to the manufacturer’s protocol. Lentiviral production and concentration Lentiviral vector plasmids were transfected into HEK293T cells with packaging plasmids psPAX2 (Addgene #12260) and pMD2.G (Addgene #12259). Supernatants were collected 48 h posttransfection and filtered through a 0.45-μm filter. Viral supernatants were concentrated by ultracentrifugation. Lentiviral vector cloning To generate a FUCCI lentiviral vector, mCherry-hCDT1-mAG-hGeminin was PCR amplified from the FUCCI reporter (a gift from A. Miyawaki at Brain Science Institute, RIKEN, Wako, Japan) and cloned into pWPI-hPLK2WT-neo (Addgene #35385), replacing hPLK2WT. P21, CDK4R24C, and CDK6R31C were PCR amplified from cDNA library of human dermal fibroblasts, pBABE-hygro CDK4 R24C (Addgene #11254), and pcDNA3.1-mouse cdk6 R31C (Addgene #75171), respectively. PCR products were cloned into a doxycycline inducible lentiviral vector expressing eGFP-puro. Whole-genome 5-hmC sequencing (Aba-seq) Samples were prepared as described in scAba-seq with minor modifications [68]. In total, approximately 50,000 cells were suspended in 10 μL of lysis buffer (100 μg Qiagen Protease, 0.04% Triton X-100, 10× NEB CutSmart buffer) and incubated at 50 °C for 15 h, 75 °C for 20 min, and 80 °C for 5 min. In total, 10 μL of 10× NEB CutSmart buffer, 10 U NEB T4-BGT, and 2.5× NEB UDP-Glucose was added to glucosylate 5-hmC sites in the genome, and the sample was incubated at 37 °C for 16 h. Next, 10 μL of protease mix (100 μg Qiagen Protease, 1× NEB CutSmart) was added and incubated at 50 °C for 5 h, 75 °C for 20 min, and 80 °C for 5 min. In all, 10 μL of digestion mix (10 U AbaSI, 1× NEB CutSmart) was added and incubated at 25 °C for 2 h and 65 °C for 20 min. Next, 1 μL of 0.5 μM ds-adaptor was added, followed by 9 μL of ligation mix (2,000 U T4 DNA Ligase, 1× T4-ligase buffer, 3.33 mM ATP), and each sample was incubated at 16 °C for 16 h. The ds-adaptor sequences are described in scAba-seq [68]. DNA cleanup was then performed with 0.825× Agencourt Ampure XP beads and eluted in 25 μL of nuclease-free water. The samples were vacuum centrifuged to a volume of 8 μL. In all, 12 μL of in vitro transcription mix was added (2 μL of each ribonucleotide, 2 μL of T7 buffer, 2 μL T7 enzyme mix) and incubated at 37 °C for 13 h. Library preparation starting from amplified RNA was performed as described in the CEL-Seq2 protocol with the following minor modifications [72]. RNA was fragmented by adding 5 μL of fragmentation buffer (200 mM Tris-acetate [pH 8.1], 500 mM KOAc, 150 mM MgOAc) at 94 °C for 2 min and immediately placed on ice. In total, 2.5 μL of fragmentation stop buffer (0.5 M EDTA) was added to quench the reaction. Next, RNA cleanup was performed using 0.825× Agencourt RNAClean beads and finally resuspended in 22 μL of nuclease-free water. Thereafter, all library preparation steps were performed as described in the CEL-Seq2 protocol [72]. The 5-hmC data analysis pipeline is described in scAba-seq [68]. The 5-hmC site annotation was performed using HOMER software [73]. The 5-hmC level per transcript was quantified by the total number of 5-hmC sites per bin size (gene body and 1 kb upstream and downstream of each transcript) and then normalized with the total number of CG sites for the same bin size. Normalized 5-hmC levels of different transcript isoforms were summed as the 5-hmC level for an individual gene for the following comparison. Lineage genes were obtained based on their expression and significantly up-regulated in either NE or ME from our previously published paper [69]. Likewise, significantly higher expression of genes in hESCs versus both NE and ME were considered as hESC-specific genes. Genome-wide 5-hmC data are available in Gene Expression Omnibus (GEO): GSE113236. Poisson model We start our analysis by proposing a Poisson regression for G1 length as response variable and Axin2 expression as model covariate. For our G1 length to follow a Poisson distribution, var(G1) = <G1> must hold true, but this was not a valid assumption for either hESC or hiPSC. Therefore, we equally shift all G1 length values toward the origin because a Poisson distribution P(x|μ) should be defined for all x > 0. We define G1* as shifted G1 toward the origin G1* = G1 − min(G1), where min(G1) is a single constant obtained for G1 values of the whole data set {hiPSC, hESC}. As a result, the transformed G1 (i.e., G1*) approximately has the mentioned Poisson property var(G1*) ≈ <G1*>. We then used a generalized linear model with g(.) = log(.) as link function (assuming that WNT levels in single cells are independent of each other and their measurements follow a normal distribution) as follows: where g(.) = log is called “link function,” resulting in approximate model variables . The higher-order terms (e.g., β2x2) cannot explain a significant variation in the observations and are neglected to avoid overfitting (p > 0.1). We compare the quantiles log(G1*) with the quantiles of a normal distribution to assess the assumptions of our model in a q-q plot (S7A Fig). This analysis demonstrates that a normality assumption for ƞ is reasonable. Next, we use a Box-Cox transformation (S7B Fig) to show that the log transformation of G1* results in the closest to normal distribution among all of the power transformations (λmax = 0.124~0 and thus, log transformation is the best candidate among all power transformations). Given G1* ~ Poisson(μ*) (and thus, μ* = var(G1*) = <G1*>), the CV of G1 can be found as follows: where σ and σ* are the standard deviations of G1 and G1*. As a result, we conclude that the G1 distribution depends on WNT levels such that high WNT levels reduce the average G1 in a cell population and a population expressing higher WNT levels corresponds to having more homogenous G1 length distributions. Statistical analysis When two groups were compared, a two-tailed Student t test was used to determine statistical significance. p-Values of less than 0.05 were considered significant. When comparing two distributions of G1 length, the Kolmogorov-Smirnov test and the Mann-Whitney U test were used. Cell culture H9 cells were maintained in feeder-free conditions on Matrigel (Corning) either in mTeSR1 (Stem Cell Technologies) or E8 (ThermoFisher Scientific) medium. H9 cells were authenticated by short tandem repeat analysis. hiPSC lines (provided by Dr. Yoav Gilad at University of Chicago) were grown on Matrigel in E8 medium [49]. Cells were passaged every 5–6 d by ReLeSR (Stem Cell Technologies). Cells were passaged every 4–5 d by Accutase (ThermoFisher Scientific). HEK293T cells and human dermal fibroblasts were maintained in DMEM supplemented with 10% fetal bovine serum (FBS). Differentiation of hESCs and hiPSCs hESCs and hiPSCs were induced to differentiate on Matrigel in hESC culture medium (DMEM/F12, 15% knockout serum replacement, MEM nonessential amino acid solution, and 0.1 mM β-mercaptoethanol) without FGF2. For NE-directed differentiation, H9 cells were differentiated on Matrigel with hESC culture medium containing Noggin (200 ng/ml) and SB431542 (10 μM) [48]. For ME-directed differentiation, H9 cells were induced to differentiate in mTeSR1 medium containing 10 ng/ml BMP4 [47]. Generation of FUCCI cell lines and cell cycle–length measurement H9 hESCs and hiPSCs were transduced with a lentiviral vector expressing the FUCCI reporter (mCherry-hCDT1-T2A-mAG-hGEMININ). Transduced cells were selected by 300 μg/ml neomycin treatment. Cells expressing the FUCCI reporter were imaged by Olympus DSU confocal microscope for 24–48 h with 10-min interval in a chamber with 5% CO2 and 37 °C temperature. Duration of each cell-cycle state was measured by manual tracking with the following criteria: G1, from red on to green on; S/G2/M, from red off to green off. To minimize tracking bias, we selected cells randomly regardless of their position in the colony. Derivation of clonal FUCCI hESC lines H9 hESCs were transduced with a lentiviral vector expressing the FUCCI reporter and maintained with 300 μg/ml neomycin. After two rounds of passaging, cells were dissociated with Accutase (ThermoFisher Scientific), FACS sorted using FUCCI-driven GFP signals, and plated onto Matrigel-coated 96-well plates, which were preincubated with CloneR (Stem Cell Technologies) at a density of one cell per well. After 1 wk of media refreshing according to CloneR manufacturer’s protocol, clonal positive wells were selected with the following criteria: (1) only one colony, (2) round/compact colony shape, (3) robust GFP or mCherry expressions across the colony. Each of the clonal FUCCI lines were expanded by at least four rounds of passaging and then used for further experiments. Immunofluorescence Cells were fixed with 4% paraformaldehyde and permeabilized with 0.25% Triton X-100, followed by blocking with 10% FBS in PBS for 1 h. Samples were stained with primary antibodies for OCT4 (Santa Cruz Biotechnology, sc-5279), NANOG (R&D Systems, AF1997), SOX2 (Millipore, AB5603), PAX6 (Santa Cruz Biotechnology, sc-81649), and GATA6 (Cell Signaling Technology, 5851) overnight at 4 °C. Secondary antibody staining was performed for 1 h at room temperature with Alexa Fluor 488-donkey anti-goat IgG, Alexa Fluor 555-donkey anti-mouse IgG, and Alexa Fluor 555-donkey anti-rabbit IgG (ThermoFisher Scientific). For 5-hmC staining, cells were treated with 1.5 M HCl for 30 min at room temperature after 4% paraformaldehyde fixation (antibody: Active Motif, 39769). For 5-mC detection, cells were fixed with ice-cold 70% ethanol for 5 min, followed by 1.5 M HCl for 30 min at room temperature. Samples were blocked with 5% FBS and 0.3% Triton X-100 in PBS for 1 h and then stained with anti-5-mC antibody (Cell Signaling Technology, 28692) diluted in PBS with 1% BSA and 0.3% Triton X-100 overnight at 4 °C. Secondary antibody staining was performed for 1 h at room temperature with Alexa Fluor 555-donkey anti-rabbit IgG (ThermoFisher Scientific). All images were taken using Olympus IX71 fluorescence microscope. Western blotting RIPA lysis buffer (Millipore) was used to lyse cells in the presence of protease inhibitor cocktail (Roche). Protein concentration was measured using Pierce BCA Protein Assay Kit (ThermoFisher Scientific). Same amounts of protein were resolved by 10% SDS-PAGE, followed by transfer to nitrocellulose membranes (GE Healthcare Life Sciences). PBST (0.1% Tween 20 in PBS) containing 5% skim milk was used to block the membranes. Immunoblotting was performed overnight at 4 °C with antibodies for OCT4 (Santa Cruz Biotechnology, sc-9081), cyclin A2 (Cell Signaling Technology, #4656), cyclin B2 (Santa Cruz Biotechnology, sc-28303), cyclin D1 (Millipore, 04–221), cyclin E2 (Santa Cruz Biotechnology, sc-28351), CDK4 (Cell Signaling Technology, 12790), CDK6 (Cell Signaling Technology, 3136), p21 (Cell Signaling Technology, 2947), β-catenin (Millipore, 04–958), Smad2/3 (R&D systems, AF3797), and β-actin (Sigma-Aldrich, A5441). The membranes were stained with secondary antibodies Alexa Flour 680-goat anti-mouse IgG (ThermoFisher Scientific) and IRDye 800CW-goat anti-rabbit IgG (LI-COR) for 1 h at room temperature. Protein bands were visualized with LI-COR Odyssey Imaging System (LI-COR). Immunoflow cytometry Cells were fixed and permeabilized with Fixation/Permeabilization Solution Kit (BD Biosciences). Samples were stained with primary antibodies for PAX6 and GATA6 for 30 min at room temperature, followed by secondary antibody staining with Alexa Fluor 647-donkey anti-rabbit IgG and Alexa Fluor 647-donkey anti-mouse IgG (ThermoFisher Scientific) for 30 min at room temperature. Samples were analyzed by BD Accuri C6 flow cytometry (BD Biosciences). Propidium iodide staining Cells were dissociated into single cells by Accutase (Stem Cell Technologies). Dissociated cells were fixed with cold EtOH for 1 h on ice, followed by RNase treatment. Samples were stained with propidium iodide and analyzed by BD Accuri C6 flow cytometry (BD Biosciences). BrdU/7-AAD staining BrdU/7-AAD staining was conducted using APC BrdU Flow Kit (BD Pharmingen) according to the manufacturer’s protocol. Briefly, H9 cells were treated with BrdU (10 μM) for 15 min, washed with PBS twice, and dissociated into single cells by Accutase. In total, 106 cells per sample were then fixed, permeabilized, and treated with DNaseI (30 μg per 106 cells) for 1 h. Samples were stained with fluorochrome-conjugated anti-BrdU antibody for 20 min and washed, followed by resuspension in 7-AAD solution. At least 10,000 cells per each sample were analyzed by BD Accuri C6 flow cytometry (BD Biosciences). Quantitative real-time PCR (qPCR) Total RNAs were extracted using TRIzol Reagent (ThermoFisher Scientific), followed by reverse transcription with SuperScript III First-Strand Synthesis System (ThermoFisher Scientific). qPCR was performed with Power SYBR Green PCR Master Mix (Applied Biosystems) using QuantStudio 12K Flex Real-Time PCR System. GAPDH was used as a normalization control. siRNA transfection H9 cells were transfected with TET1 siRNAs (ThermoFisher Scientific HSS129586, HSS129587) (100 nM) with RNAiMAX (ThermoFisher Scientific) according to the manufacturer’s protocol. Lentiviral production and concentration Lentiviral vector plasmids were transfected into HEK293T cells with packaging plasmids psPAX2 (Addgene #12260) and pMD2.G (Addgene #12259). Supernatants were collected 48 h posttransfection and filtered through a 0.45-μm filter. Viral supernatants were concentrated by ultracentrifugation. Lentiviral vector cloning To generate a FUCCI lentiviral vector, mCherry-hCDT1-mAG-hGeminin was PCR amplified from the FUCCI reporter (a gift from A. Miyawaki at Brain Science Institute, RIKEN, Wako, Japan) and cloned into pWPI-hPLK2WT-neo (Addgene #35385), replacing hPLK2WT. P21, CDK4R24C, and CDK6R31C were PCR amplified from cDNA library of human dermal fibroblasts, pBABE-hygro CDK4 R24C (Addgene #11254), and pcDNA3.1-mouse cdk6 R31C (Addgene #75171), respectively. PCR products were cloned into a doxycycline inducible lentiviral vector expressing eGFP-puro. Whole-genome 5-hmC sequencing (Aba-seq) Samples were prepared as described in scAba-seq with minor modifications [68]. In total, approximately 50,000 cells were suspended in 10 μL of lysis buffer (100 μg Qiagen Protease, 0.04% Triton X-100, 10× NEB CutSmart buffer) and incubated at 50 °C for 15 h, 75 °C for 20 min, and 80 °C for 5 min. In total, 10 μL of 10× NEB CutSmart buffer, 10 U NEB T4-BGT, and 2.5× NEB UDP-Glucose was added to glucosylate 5-hmC sites in the genome, and the sample was incubated at 37 °C for 16 h. Next, 10 μL of protease mix (100 μg Qiagen Protease, 1× NEB CutSmart) was added and incubated at 50 °C for 5 h, 75 °C for 20 min, and 80 °C for 5 min. In all, 10 μL of digestion mix (10 U AbaSI, 1× NEB CutSmart) was added and incubated at 25 °C for 2 h and 65 °C for 20 min. Next, 1 μL of 0.5 μM ds-adaptor was added, followed by 9 μL of ligation mix (2,000 U T4 DNA Ligase, 1× T4-ligase buffer, 3.33 mM ATP), and each sample was incubated at 16 °C for 16 h. The ds-adaptor sequences are described in scAba-seq [68]. DNA cleanup was then performed with 0.825× Agencourt Ampure XP beads and eluted in 25 μL of nuclease-free water. The samples were vacuum centrifuged to a volume of 8 μL. In all, 12 μL of in vitro transcription mix was added (2 μL of each ribonucleotide, 2 μL of T7 buffer, 2 μL T7 enzyme mix) and incubated at 37 °C for 13 h. Library preparation starting from amplified RNA was performed as described in the CEL-Seq2 protocol with the following minor modifications [72]. RNA was fragmented by adding 5 μL of fragmentation buffer (200 mM Tris-acetate [pH 8.1], 500 mM KOAc, 150 mM MgOAc) at 94 °C for 2 min and immediately placed on ice. In total, 2.5 μL of fragmentation stop buffer (0.5 M EDTA) was added to quench the reaction. Next, RNA cleanup was performed using 0.825× Agencourt RNAClean beads and finally resuspended in 22 μL of nuclease-free water. Thereafter, all library preparation steps were performed as described in the CEL-Seq2 protocol [72]. The 5-hmC data analysis pipeline is described in scAba-seq [68]. The 5-hmC site annotation was performed using HOMER software [73]. The 5-hmC level per transcript was quantified by the total number of 5-hmC sites per bin size (gene body and 1 kb upstream and downstream of each transcript) and then normalized with the total number of CG sites for the same bin size. Normalized 5-hmC levels of different transcript isoforms were summed as the 5-hmC level for an individual gene for the following comparison. Lineage genes were obtained based on their expression and significantly up-regulated in either NE or ME from our previously published paper [69]. Likewise, significantly higher expression of genes in hESCs versus both NE and ME were considered as hESC-specific genes. Genome-wide 5-hmC data are available in Gene Expression Omnibus (GEO): GSE113236. Poisson model We start our analysis by proposing a Poisson regression for G1 length as response variable and Axin2 expression as model covariate. For our G1 length to follow a Poisson distribution, var(G1) = <G1> must hold true, but this was not a valid assumption for either hESC or hiPSC. Therefore, we equally shift all G1 length values toward the origin because a Poisson distribution P(x|μ) should be defined for all x > 0. We define G1* as shifted G1 toward the origin G1* = G1 − min(G1), where min(G1) is a single constant obtained for G1 values of the whole data set {hiPSC, hESC}. As a result, the transformed G1 (i.e., G1*) approximately has the mentioned Poisson property var(G1*) ≈ <G1*>. We then used a generalized linear model with g(.) = log(.) as link function (assuming that WNT levels in single cells are independent of each other and their measurements follow a normal distribution) as follows: where g(.) = log is called “link function,” resulting in approximate model variables . The higher-order terms (e.g., β2x2) cannot explain a significant variation in the observations and are neglected to avoid overfitting (p > 0.1). We compare the quantiles log(G1*) with the quantiles of a normal distribution to assess the assumptions of our model in a q-q plot (S7A Fig). This analysis demonstrates that a normality assumption for ƞ is reasonable. Next, we use a Box-Cox transformation (S7B Fig) to show that the log transformation of G1* results in the closest to normal distribution among all of the power transformations (λmax = 0.124~0 and thus, log transformation is the best candidate among all power transformations). Given G1* ~ Poisson(μ*) (and thus, μ* = var(G1*) = <G1*>), the CV of G1 can be found as follows: where σ and σ* are the standard deviations of G1 and G1*. As a result, we conclude that the G1 distribution depends on WNT levels such that high WNT levels reduce the average G1 in a cell population and a population expressing higher WNT levels corresponds to having more homogenous G1 length distributions. Statistical analysis When two groups were compared, a two-tailed Student t test was used to determine statistical significance. p-Values of less than 0.05 were considered significant. When comparing two distributions of G1 length, the Kolmogorov-Smirnov test and the Mann-Whitney U test were used. Supporting information S1 Fig. H9 hESC line expressing the FUCCI reporter. (A) Representative image of FUCCI H9 cells in each cell-cycle state from multiple independent experiments. (B) qPCR analysis of pluripotency genes in FUCCI-expressing H9 cells (n = 4). (C) qPCR analysis of lineage markers in FUCCI H9 cells differentiated for 8 d by FGF2 deprivation (n = 4). (D) Representative images of FUCCI H9 cells undergoing cell division. (E) Representative images of clonal FUCCI lines undergoing cell division. The duration of no-color phase was measured by live-cell imaging (n = 30 for each clone). Error bars represent SD. *p < 0.01 (Student t test). Underlying data can be found in S2 Data. FGF, Fibroblast growth factor; FUCCI, fluorescent ubiquitination–based cell-cycle indicator; hESC, human embryonic stem cell; qPCR, quantitative PCR. https://doi.org/10.1371/journal.pbio.3000453.s001 (TIF) S2 Fig. G1 length distribution predicts differentiation outcomes of hESCs. (A) qPCR analysis of pluripotency genes in H9 cells grown either in E8 or in mTeSR1 (n = 4). (B) qPCR analysis of lineage markers in H9 cells differentiated by FGF2 deprivation for 7 d (n = 4). (C) qPCR analysis of lineage markers in H9 cells differentiated for 5 d to NE (dual SMAD inhibition) or ME (FGF2&BMP4) lineage (n = 4). (D) FUCCI reporter in H9 cells grown either in E8 or in mTeSR1. Representative images were shown from three independent experiments. (E) BrdU/7-AAD staining and flow cytometry analysis of H9 cells grown either in E8 or in mTeSR1 media (n = 3). (F) G1 length data of biological replicates in Fig 2C. (G and H) qPCR (G) and western blot (H) analyses of cyclins in H9 cells grown either in E8 or in mTeSR1 (n = 4 for qPCR and n = 3~4 for western blot). Error bars represent SD. *p < 0.01 (Student t test). Underlying data can be found in S2 Data. 7-AAD, 7-amino-actinomycin D; BMP, Bone morphogenetic protein; BrdU, 5-bromo-2′-deoxyuridine; E8, Essential 8; FGF, Fibroblast growth factor; FUCCI, fluorescent ubiquitination–based cell-cycle indicator; hESC, human embryonic stem cell; ME, mesendoderm; NE, neuroectoderm; qPCR, quantitative PCR. https://doi.org/10.1371/journal.pbio.3000453.s002 (TIF) S3 Fig. G1 length distribution predicts differentiation outcomes of hiPSCs. (A) Immunofluorescence of pluripotency genes in hiPSC lines grown in E8 medium. Representative images were shown from three independent experiments. (B) qPCR analysis of lineage markers in hiPSC lines differentiated for 7 d by FGF2 deprivation (n = 4). (C) Histograms for G1 length of hiPSC lines (n = 100 for hiPSC1, n = 120 for hiPSC2, and n = 108 for hiPSC3 pooled from two to three independent experiments); U test: p-value = 2.242 × 10−14 for hiPSC1 versus hiPSC2, p-value = 3.395 × 10−9 for hiPSC1 versus hiPSC3; KS test: p-value = 3.064 × 10−14 for hiPSC1 versus 2, p-value = 3.209 × 10−7 for hiPSC1 versus 3. (D) Propidium iodide staining analysis of hiPSC lines (n = 3). Error bars represent SD. *p < 0.01 (Student t test). Underlying data can be found in S2 Data. E8, Essential 8; FGF, Fibroblast growth factor; hiPSC, human induced pluripotent stem cell; KS, Kolmogorov-Smirnov; qPCR, quantitative PCR. https://doi.org/10.1371/journal.pbio.3000453.s003 (TIF) S4 Fig. Modulation of G1 length affects differentiation propensity of stem cell populations. (A) Western blot of p21, CDK4, CDK6, and OCT4 in H9 cells expressing p21, CDK4R24C, or CDK6R31C. Representative images were shown from two independent experiments. (B) Average G1 length of FUCCI H9 cells expressing p21, CDK4R24C, or CDK6R31C (n = 54 for ctrl, n = 43 for p21, n = 52 for CDK4R24C, and n = 54 for CDK6R31C pooled from two independent experiments), &p < 0.01 (U test); §p < 0.01 (KS test). (C) qPCR analysis of pluripotency genes in H9 cells expressing p21, CDK4R24C, or CDK6R31C (n = 4). (D) Propidium iodide staining analysis of H9 cells expressing p21, CDK4R24C, or CDK6R31C (n = 3). (E) Dox-induced transgene expression and shutdown after Dox withdrawal in H9 cells transduced with p21, CDK4R24C, or CDK6R31C lentiviral vectors. Relative protein levels were analyzed by western blot assay (n = 3). (F) Immunofluorescence assay for PAX6 and GATA6 in H9 cells treated with Abemaciclib (0.5 μM) or vehicle (DMSO) for 18 h and then differentiated for 8–10 d. Representative images were shown from two independent experiments. (G) qPCR analysis of lineage markers in differentiated day 8 H9 cells overexpressing p21 (n = 4). Transgene expression was turned off at the onset of differentiation by Dox withdrawal. Error bars represent SD. #p < 0.05, *p < 0.01 (Student t test). Underlying data can be found in S2 Data. CDK, Cyclin-dependent kinase; ctrl, control; Dox, doxycycline; FUCCI, fluorescent ubiquitination–based cell-cycle indicator; GATA6, GATA binding protein 6; KS, Kolmogorov-Smirnov; OCT4, Octamer-binding transcription factor 4; PAX6, Paired box 6; qPCR, quantitative PCR. https://doi.org/10.1371/journal.pbio.3000453.s004 (TIF) S5 Fig. Ratio of asymmetric sister cell G1 duration in H9 cells grown either in E8 or in mTeSR1. Difference in G1 length between sister cells (ΔG1) was divided by mean G1 length (<G1>) between sister cells. Asymmetric sister cell G1 duration was defined by ΔG1/<G1> values with various cutoffs (n = 56 for E8 and n = 55 for mTeSR1 pooled from three independent experiments). Underlying data can be found in S2 Data. E8, Essential 8. https://doi.org/10.1371/journal.pbio.3000453.s005 (TIF) S6 Fig. WNT/β-catenin pathway controls G1 length distribution patterns. (A) Validation of TOP-flash reporter by recombinant WNT3A treatment. Representative images were shown from three independent experiments. (B) TOP-flash activity in H9 cells grown either in E8 or in mTeSR1 (n = 3). (C) qPCR analysis of WNT target genes in hiPSC lines grown in E8 (n = 4). (D) Propidium iodide staining analysis of H9 cells in mTeSR1 and treated with recombinant human WNT3A proteins (100 ng/ml) (n = 3). (E) qPCR analysis of pluripotency genes in H9 cells grown in mTeSR1 and treated with recombinant human WNT3A proteins (100 ng/ml) (n = 3~4). (F) Analysis of ChIP-seq peaks on the genomic loci of cyclins D1 and E2 with or without WNT3A treatment (GSE64758). Arrows represent genes. (G) qPCR analysis of cyclins D and E in H9 cells grown in mTeSR1 and treated with recombinant human WNT3A proteins (100 ng/ml) (n = 4). (H) G1 length of three clonal FUCCI lines treated with recombinant WNT3A (100 ng/ml) (n = 30 for each sample). *U test p-value < 0.0001. (I) Ratio of symmetric and asymmetric sister cell G1 durations in H9 cells grown in mTeSR1 and treated with recombinant human WNT3A proteins (100 ng/ml) (cutoff: ΔG1/<G1> = 0.2, n = 55 for untreated and n = 52 for WNT3A pooled from three independent experiments). (J) Histograms for G1 length of hiPSC1 grown in E8 and treated with recombinant human WNT3A proteins (100 ng/ml) (n = 100 for hiPSC1 pooled from three independent experiments and n = 86 for hiPSC1 + WNT3A pooled from two independent experiments); U test: p-value < 2.2 × 10−16, KS test: p-value < 2.2 × 10−16. (K) qPCR analysis for AXIN2 expression in H9 cells treated with 5 μM IWP-2 (n = 4). Error bars represent SD. *p < 0.01 (Student t test). Underlying data can be found in S2 Data. ChIP-seq, chromatin immunoprecipitation followed by sequencing; E8, Essential 8; FUCCI, fluorescent ubiquitination–based cell-cycle indicator; hiPSC, human induced pluripotent stem cell; KS, Kolmogorov-Smirnov; MFI, mean fluorescence intensity; qPCR, quantitative PCR; WNT, Wingless-INT. https://doi.org/10.1371/journal.pbio.3000453.s006 (TIF) S7 Fig. Validation of the Poisson model. (A) q-q plots comparing the quantiles of log(G1*) with the quantiles of a normal distribution. (B) Box-Cox transformation of G1*. Underlying data can be found in S2 Data. https://doi.org/10.1371/journal.pbio.3000453.s007 (TIF) S8 Fig. G1 length modulation does not affect SMAD activity in hESCs. (A) Western blot of SMAD2/3 in nuclear and total fractions of H9 cells grown either in E8 or in mTeSR1 (n = 4). (B) Western blot of SMAD2/3 in nuclear and total fractions of H9 cells overexpressing CDK4R24C or CDK6R31C (n = 3). (C) Western blot of SMAD2/3 in nuclear and total fractions of H9 cells treated with 100 ng/ml of WNT3A for 24 h (n = 3). Error bars represent SD. *p < 0.01 (Student t test). Underlying data can be found in S2 Data. CDK, Cyclin-dependent kinase; E8, Essential 8; hESC, human embryonic stem cell; WNT, Wingless-INT. https://doi.org/10.1371/journal.pbio.3000453.s008 (TIF) S9 Fig. G1 length–driven global 5-hmC levels affect differentiation propensity of stem cell populations. (A) Pie plot for annotated genome categories of 5-hmC sites detected in H9 hESCs grown in mTeSR1. (B) Gene expression patterns of hESC-specific and lineage-specific genes in H9 hESCs (U test: p-value < 2.2 × 10−16) (GSE69982). (C) qPCR analysis of TET1, TET2, and TET3 in H9 cells grown either in E8 or mTeSR1 (n = 4). (D) Immunofluorescence assay for 5-mC in H9 cells grown in mTeSR1 expressing CDK6R31C. Representative images were shown from three independent experiments. (E) Gene expression levels of TET1, TET2, and TET3 in hESCs (GSE69982). (F) qPCR analysis of TET1 in H9 cells transfected with siRNAs (n = 3). (G) Immunofluorescence assay for 5-hmC in H9 cells transfected with siRNAs. Representative images were shown from three independent experiments. (H) Immunofluorescence assay for PAX6 and GATA6 in H9 cells transfected with siRNAs and then differentiated for 9 d. Representative images were shown from three independent experiments. (I) qPCR analysis of lineage markers in H9 cells grown in mTeSR1, transfected with TET1 siRNA #2, and then differentiated for 8 d without FGF2 (n = 4). Error bars represent SD. *p < 0.01 (Student t test). Underlying data can be found in S2 Data. 5-hmC, 5-hydroxymethylcytosine; 5-mC, 5-methylctosine; CDK, Cyclin-dependent kinase; E8, Essential 8; FGF, Fibroblast growth factor; GATA6, GATA binding protein 6; hESC, human embryonic stem cell; PAX6, Paired box 6; qPCR, quantitative PCR; siRNA, small interfering RNA; TET, ten-eleven translocation. https://doi.org/10.1371/journal.pbio.3000453.s009 (TIF) S1 Table. qPCR primers used in this study. qPCR, quantitative PCR. https://doi.org/10.1371/journal.pbio.3000453.s010 (DOCX) S1 Data. Numerical values used in figures. https://doi.org/10.1371/journal.pbio.3000453.s011 (XLSX) S2 Data. Numerical values used in Supporting information figures. https://doi.org/10.1371/journal.pbio.3000453.s012 (XLSX) S1 Movie. Time-lapse imaging of FUCCI hESCs in mTeSR1. FUCCI, fluorescent ubiquitination–based cell-cycle indicator; hESC, human embryonic stem cell. https://doi.org/10.1371/journal.pbio.3000453.s013 (AVI) S2 Movie. Tracked FUCCI hESCs in mTeSR1. FUCCI, fluorescent ubiquitination–based cell-cycle indicator; hESC, human embryonic stem cell. https://doi.org/10.1371/journal.pbio.3000453.s014 (AVI) S3 Movie. FUCCI hESCs undergoing M to G1 phase transition. FUCCI, fluorescent ubiquitination–based cell-cycle indicator; hESC, human embryonic stem cell. https://doi.org/10.1371/journal.pbio.3000453.s015 (AVI) Acknowledgments We thank the Gilad lab at University of Chicago for hiPSC lines and Sahand Hormoz at California Institute of Technology and Tau-Mu Yi at University of California Santa Barbara for insightful comments.