key: cord-0874478-73kzlu82
authors: Lee, Hunsang; Beilhartz, Greg L.; Kucharska, Iga; Raman, Swetha; Cui, Hong; Lam, Mandy Hiu Yi; Liang, Huazhu; Rubinstein, John L.; Schramek, Daniel; Julien, Jean-Philippe; Melnyk, Roman A.; Taipale, Mikko
title: Recognition of Semaphorin Proteins by P. sordellii Lethal Toxin Reveals Principles of Receptor Specificity in Clostridial Toxins
date: 2020-06-25
journal: Cell
DOI: 10.1016/j.cell.2020.06.005
sha: dfa9e5a7a9e5cc3e0df0a5d3e54e04a993b48b16
doc_id: 874478
cord_uid: 73kzlu82

Pathogenic clostridial species secrete potent toxins that induce severe host tissue damage. Paeniclostridium sordellii lethal toxin (TcsL) causes an almost invariably lethal toxic shock syndrome associated with gynecological infections. TcsL is 87% similar to C. difficile TcdB, which enters host cells via Frizzled receptors in colon epithelium. However, P. sordellii infections target vascular endothelium, suggesting that TcsL exploits another receptor. Here, using CRISPR/Cas9 screening, we establish semaphorins SEMA6A and SEMA6B as TcsL receptors. We demonstrate that recombinant SEMA6A can protect mice from TcsL-induced edema. A 3.3 Å cryo-EM structure shows that TcsL binds SEMA6A with the same region that in TcdB binds structurally unrelated Frizzled. Remarkably, 15 mutations in this evolutionarily divergent surface are sufficient to switch binding specificity of TcsL to that of TcdB. Our findings establish semaphorins as physiologically relevant receptors for TcsL and reveal the molecular basis for the difference in tissue targeting and disease pathogenesis between highly related toxins.

A lethal bacterial toxin that plays a major role in toxic shock syndrome uses the host semaphorin proteins as receptors, and structural analysis shows a small binding interface that can be mutated to switch specificity between very different host receptors.

Paeniclostridium sordellii (also known as Clostridium sordellii) is an anaerobic gram-positive bacterium found in soil and in the gastrointestinal and vaginal tracts of animals and humans. P. sordellii is present in the rectal or vaginal tract of 3%-4% of women, but vaginal colonization rate after childbirth is as high as 29% (Aldape et al., 2016; Chong et al., 2016) . Although the majority of carriers are asymptomatic, pathogenic P. sordellii infections arise rapidly and are highly lethal. The origin of pathogenic P. sordellii strains is unclear, but most infections occur in women after childbirth, medically induced abortion, or miscarriage, leading to a toxic shock syndrome with almost 100% mortality within days (Aldape et al., 2016; Chong et al., 2016; Fischer et al., 2005; Ho et al., 2009) .

The primary cause of the high mortality associated with P. sordellii infections is the lethal toxin TcsL (Carter et al., 2011) , which belongs to the large clostridial toxin (LCT) family (Orrell et al., 2017) . LCTs enter the host cell by receptor-mediated endo-cytosis into acidified endosomes followed by pH-dependent pore formation and translocation into the cytoplasm (Papatheodorou et al., 2010; Pfeifer et al., 2003; Zhang et al., 2014) . After autoprocessing in the cytosol, the released cytotoxic glucosyltransferase enzymes potently modulate host cell function by inactivating small Rho-family GTPases by using uridine diphosphate (UDP)-glucose or UDP-N-acetylglucosamine as a co-substrate (Aktories and Barbieri, 2005; Aktories and Just, 1995; Reineke et al., 2007) .

Although all LCTs are highly similar at the sequence level, they differ in their tissue specificity and in their effects on cell morphology, physiology, and viability. TcsL is most closely related to the C. difficile cytotoxin TcdB, sharing almost 90% sequence similarity. TcdB is the causal virulence factor behind gastrointestinal diseases associated with C. difficile infections. TcdB binds Frizzled family receptors FZD1, FZD2, and FZD7 expressed in the colonic epithelium (Chen et al., 2018; Tao et al., 2016) , the primary site of C. difficile infection. In contrast, although present in the intestinal microbiota, P. sordellii does not infect or damage the colonic epithelium, suggesting that TcsL binds a different cell surface receptor. This is supported by earlier competition experiments with recombinant TcdB and TcsL in vitro , showing differential sensitivity of cell lines to TcsL and TcdB (Chaves-Olarte et al., 1997) . Although recent studies have begun to uncover receptors for other LCTs, the physiologically relevant TcsL receptors are still unknown. In addition, it is not understood at the molecular level how structurally and functionally similar LCTs can bind completely unrelated receptors.

Here, we employ genome-wide CRISPR/Cas9 screens to identify semaphorins SEMA6A and SEMA6B as the host cell receptors for TcsL. We show that recombinant SEMA6A ectodomain can protect lung endothelial cells from TcsL intoxication in vitro and mouse lungs from TcsL-induced edema in vivo. To further understand the molecular basis of LCT receptor specificity, we determine the cryogenic electron microscopy (cryo-EM) structure of TcsL bound to SEMA6A. Comparison of this structure to the TcsL-Frizzled2 structure (Chen et al., 2018) reveals that TcsL and TcdB bind their structurally unrelated receptors by using the same receptor-binding region. However, selective clustering of mutations explains how the two toxins have evolved distinct specificities. We formally demonstrate the role of the clustered mutations in the receptor-binding site by switching the receptor specificity of TcsL from SEMA6A to Frizzled. These results have broad implications for our understanding of host receptor pathogenic toxins evolution at the molecular level.

Genome-wide CRISPR/Cas9 identifies SEMA6A as a host factor needed for TcsL toxicity To identify potential cell-surface receptors for TcsL, we conducted a genome-wide CRISPR/Cas9 screen in Hap1 cells. We used the genome-wide TKOv3 guide RNA (gRNA) library that targets 18,053 genes with four different gRNAs per gene (Hart et al., 2017) . Cells were infected with the library and treated with 0.1 nM or 1 nM TcsL. Sequencing of gRNAs from the surviving population revealed only two genes that were significantly enriched in both screens (Figures 1A and 1B and Table S1 ). One was UDP-glucose pyrophosphorylase 2 (UGP2), which was also a top hit in previous CRISPR/Cas9 screens with two other LCTs, TcdA and TcdB (Tao et al., 2016 (Tao et al., , 2019 . UGP2 is needed for the synthesis of UDP-glucose, the sugar donor for the glucosylation activity of all LCTs (Aktories and Just, 1995) . The other hit was SEMA6A, encoding a transmembrane axon guidance molecule not previously linked to toxin function.

Notably, despite high sequence similarity between TcsL and TcdB, our screen did not identify Frizzled receptors, or CSPG4 or PVRL3, two other TcdB-associated receptors (LaFrance et al., 2015; Tao et al., 2016; Yuan et al., 2015) . Neither did we identify known receptors for C. perfringens TpeL or C. difficile TcdA, other related LCTs (Schorch et al., 2014; Tao et al., 2019) , although their receptors are expressed in Hap1 cells ( Figure S1 ).

We first validated screen results by knocking out UGP2 and SEMA6A in Hap1 cells. UGP2 KO and SEMA6A KO cells were high-ly resistant to TcsL ( Figure S1 ). Introduction of wild-type SEMA6A by lentiviral transduction to SEMA6A KO cells rendered the cells more sensitive to the toxin ( Figure S1 ). SEMA6A is a member of the semaphorin family, which in humans consists of twenty transmembrane and secreted proteins (Worzfeld and Offermanns, 2014) . The four human SEMA6 class proteins are SEMA6A, SEMA6B, SEMA6C, and SEMA6D ( Figure 1C ). SEMA6A is most closely related to SEMA6B with 69% sequence similarity and 53% identity, followed by SEMA6D (60% similarity, 46% identity) and SEMA6C (56% similarity, 40% identity). We generated knockout Hap1 cells for each SEMA6 family member and assayed their sensitivity to TcsL. SEMA6C KO and SEM-A6D KO cells were as sensitive to TcsL as control cells (50% growth inhibition [GI50]: 50 pM) ( Figure 1D ). In contrast, SEM-A6B KO cells were approximately 10-fold more resistant (GI50: 500 pM) than control cells, but not as resistant as SEMA6A KO cells (50-fold; GI50: 2.5 nM). Moreover, SEMA6A-SEMA6B double knockout cells were more resistant to TcsL than either knockout alone, suggesting that SEMA6A and SEMA6B act redundantly ( Figure 1D ). Consistent with this, HeLa cells that do not express SEMA6A or SEMA6B were highly resistant to TcsL ( Figure S1 ). Because SEMA6C and SEMA6D are expressed at low amounts in Hap1 and HeLa cells ( Figure S1 ), we further assessed their role in TcsL intoxication by ectopically expressing each SEMA6 in SEMA6A KO cells by lentiviral infection. In contrast to ectopic SEMA6A and SEMA6B expression, SEMA6C and SEMA6D did not render SEMA6A KO cells more sensitive to the toxin ( Figure 1E ). Thus, SEMA6A and SEMA6B but not SEMA6C or SEMA6D regulate cellular sensitivity to TcsL. These results are consistent with the model that both SEMA6A and SEMA6B can act as TcsL receptors, and sensitivity to the toxin is determined by the total amount of SEMA6A and SEMA6B.

Recombinant SEMA6A ectodomain protects cells and mouse lungs from TcsL toxicity We then expressed and purified the soluble recombinant extracellular domain (rECD) of SEMA6A and tested its effect on TcsL toxicity on Vero cells, a commonly used cell line for studying toxin function. Vero cells express SEMA6A but not SEMA6B ( Figure S1 ). SEMA6A rECD inhibited TcsL toxicity in a dosedependent manner ( Figure 2A) . Notably, SEMA6A rECD had a protective effect only when it was added to the cells before or simultaneously with TcsL ( Figure S1 ). When cells were pretreated for 1 h with TcsL, SEMA6A rECD had no effect on toxicity, suggesting that SEMA6A rECD must act before TcsL binds the cells. We also repeated the competition assay with soluble ectodomains of SEMA6B, SEMA6C, and SEMA6D. Consistent with our experiments in Hap1 cells (above), SEMA6B alleviated TcsL toxicity, whereas SEMA6C and SEMA6D had no effect (Figures 2B and 2C) . These results strongly suggest that TcsL can directly bind SEMA6A and SEMA6B and this interaction is needed for TcsL entry into the cell.

A primary target of TcsL during P. sordellii infection is the vascular endothelium of the lung (Geny et al., 2007) . Therefore, we used immortalized human lung microvascular cells (HULECs) as a physiologically relevant cell line to study the role of SEMA6A and SEMA6B in TcsL intoxication. Notably, HULECs express $4-fold higher amounts of SEMA6A and $11-fold higher ll amounts of SEMA6B than Hap1 cells ( Figure 3A ). We assayed the sensitivity of HULECs to four related clostridial toxins: TcdA and TcdB from C. difficile, TpeL from C. perfringens, and TcsL ( Figure 3B ). TpeL showed low toxicity, and TcdA and TcdB were several orders of magnitude less toxic to HULECs than what is reported for other cell types (Chaves-Olarte et al., 1997; Gupta et al., 2017) . Remarkably, the GI 50 of TcsL was just $50 femtomolar, indicating that $200 toxin molecules per cell are lethal to HULECs. Furthermore, recombinant ectodomain of mouse Sema6a fused to Fc fragment (rSema6a-Fc) but not mouse Sema6c ectodomain-Fc (rSema6c-Fc) could block TcsL-induced rounding of HULECs in a dose-dependent manner ( Figures 3C and 3D ). These results further suggest that SEMA6A and SEMA6B are the physiologically relevant TcsL receptors in endothelial cells.

We then addressed the role of semaphorins in vivo in a mouse model of TcsL intoxication. We first examined the expression of Sema6a and Sema6b in mouse lung tissue by immunohistochemistry. Both proteins were highly expressed in lung endothelium and pneumocytes ( Figure 3E ). Mice were injected intraperitoneally with a lethal dose of TcsL together with rSema6a-Fc, rSema6c-Fc, or bovine serum albumin (BSA) (n = 3 for each group). Mice co-injected with BSA or Sema6c-Fc rapidly developed symptoms of TcsL intoxication, including decreased (E) Hap1 SEMA6A KO cells were infected with lentiviruses expressing 3xFLAG-tagged SEMA6 family proteins and tested for TcsL sensitivity. Data (n = 3) are represented as mean ± standard deviation. Shown at the bottom, expression of SEMA6 proteins in infected cell lines was validated with western blotting. See also Figure S1 and Table S1 . mobility and signs of ataxia and dehydration. Within 4 h, all symptomatic mice had a buildup of fluid in the lungs ( Figure 3F ). Histopathologically, symptomatic mice showed edema surrounding pulmonary vessels ( Figure 3F ). In contrast, rSema6a-Fc protected the mice from TcsL-induced symptoms. Mice coinjected with rSema6a-Fc had no pleural effusion and did not show signs of edema after 4 h ( Figures 3F and 3G ). Altogether, these data support that SEMA6A and SEMA6B are the physiologically relevant receptors for TcsL and that this interaction can be inhibited in vivo.

Cryo-EM structure of the TcsL 1285-1804 -SEMA6A complex Having established SEMA6A and SEMA6B as the cellular receptors of TcsL, we next structurally characterized the interaction between TcsL and SEMA6A by using cryo-EM. We used a shortened TcsL fragment spanning amino acids (aa) 1285-1804 (TcsL 1285-1804 ), analogous to the TcdB fragment previously used to determine the TcdB-Frizzled2 complex structure (Chen et al., 2018) . In a biolayer interferometry assay with immobilized ligand, which measures the avidity of interactions, the TcsL 1285-1804 fragment bound recombinant SEMA6A ectodomain with nanomolar apparent affinity ( Figures 4A and S3) . Glutaraldehyde cross-linking was used to prevent dissociation of the TcsL 1285-1804 -SEMA6A complex during cryo-EM grid preparation, as described previously Kastner et al., 2008; Yan et al., 2015) . Initial analysis of the dataset revealed that approximately 50% of the SEMA6A dimers were bound to TcsL by using this sample preparation approach. Consequently, 2D and 3D classification resulted in two homogeneous datasets corresponding to the TcsL-SEMA6A complex (179,188 particle images) (Figures S2 and S3) and unliganded SEMA6A (281,207 particle images) ( Figures S2 and S3 ), which were used to produce final density maps to overall resolutions of 3.3 and 3.1 Å , respectively (Table S2 ). Local resolution of the TcsL-SEMA6A map ranged from 2.8 Å at the binding interface to > 30 Å at the C terminus of TcsL ( Figure S4 ). Indeed, 3D vari-ability analysis revealed continuous motion of the C and N termini of TcsL in contrast to a largely rigid SEMA6A dimer and TcsL-SEMA6A interface (Video S1). A molecular model was built for the well-resolved central domain of TcsL (residues 1400-1637), corresponding to approximately 40% of the TcsL construct ( Figure 4A ).

The structure of TcsL is highly similar to previously determined structures of other clostridial toxins, particularly TcdB (rootmean-square deviation [RMSD] = 1.3 Å ) (Chen et al., 2018 Chumbler et al., 2016; Simeon et al., 2019) ( Figure S5 ). Similarly, unliganded and TcsL-bound SEMA6A structures are in remarkable agreement ( Figure S5 ), indicating that TcsL binding is not accompanied by significant conformational changes in the receptor and that glutaraldehyde treatment had a negligible effect on the SEMA6A structure.

SEMA6A interacts with TcsL at two predominant sites (Figure 4B) , burying a total surface area of 1,236 Å 2 . The major contact is through a short a-helix (residues 101-110) between blades 1 and 2, in a characteristic ''extrusion'' of semaphorin family beta-propellers (Janssen et al., 2010; Love et al., 2003; Nogi et al., 2010) . The other site is in the flexible loop between beta sheets 3B and 3C (Asp189, Phe190, Leu191, and Ile193) ( Figures 4B and S6) . Notably, the TcsL-binding interface is where SEMA6A's cognate ligand Plexin A2 also binds ( Figure 4C ) (Janssen et al., 2010; Nogi et al., 2010) . Indeed, over half of the TcsL-binding interface in SEMA6A is shared with Plexin A2 (Figures 4C and S6 ). Another notable similarity between Plexin A2 and TcsL is their semaphorin specificity: both bind SEMA6A and SEMA6B but not SEMA6C or SEMA6D (Worzfeld and Offermanns, 2014) . In the SEMA6A-TcsL structure, two SEMA6A residues (Arg108 and Ile193) that engage in contact with TcsL are identical in SEMA6B but not conserved in SEMA6C and SEMA6D ( Figure S6 ), likely explaining the specificity of TcsL. Ile193 forms critical contacts also with Plexin A2 (Figure S6) , suggesting that the semaphorin specificity of TcsL and Plexin A2 are based on similar molecular principles.

On TcsL, the SEMA6A interaction interface is located on a convex edge of the delivery domain ( Figure 5A ). The highly ll hydrophobic interface is mainly formed by residues located in an antiparallel beta sheet (residues 1466-1511), flanked by an a-helix (1433-1438) on one side and a loop (1596-1601) on the other. Strikingly, this location on C. difficile cytotoxin TcdB is also the one used to bind Frizzled receptors, which are structurally unrelated to semaphorins (Chen et al., 2018 ) ( Figure 5A ). Nevertheless, both toxins bind their receptors in a similar manner. The hydrophobic nature of this pocket is conserved in TcsL but the amino acid identities are not ( Figure 5A ). Moreover, the interaction of TcdB with FZD2 is bridged by a palmitoleic acid (PAM) moiety, which is buried in a hydrophobic pocket of TcdB (Fig-ure 5B ) (Chen et al., 2018) . In TcsL, the pocket buries the hydrophobic side chain of SEMA6A Met109 in an analogous manner to the palmitoleic acid moiety in the TcdB-FZD2 structure (Figure 5B and Table S3 ). In fact, although none of the six residues that in TcdB contact PAM through hydrophobic interactions are conserved in TcsL, all six corresponding residues in TcsL form similar hydrophobic contacts with SEMA6A Met109 (Figure 5B and Table S3 ). We experimentally validated the TcsL-SEMA6A interaction surface by first generating and purifying a variant of TcsL (TcsL 4mut ) that carried four mutations in core interacting residues Article (Cys1433Asp, Ile1434Lys, Ala1486Ser, and Tyr1596Arg) (shown in Figure 5A ). TcsL Cys1433, Ile1434, and Ala1486 form key hydrophobic interactions in the pocket accommodating SEMA6A Met109, whereas TcsL Tyr1596 interacts with several residues in the flexible loop of SEMA6A through hydrophobic interactions and through a hydrogen bond with SEMA6A Phe190 backbone amide ( Figure 5A and Table S3 ). TcsL 4mut was highly soluble and retained its inositol hexaphosphate (IP6)-dependent autoprocessing activity in vitro, suggesting that it is properly folded ( Figure 5C ). However, TcsL 4mut was 2,500-fold less toxic in Vero cells than the wild-type TcsL, indicating that the four interface residues are physiologically relevant ( Figure 5C ). We then validated the relevance of SEMA Met109 in TcsL toxicity. We introduced 3xFLAG-tagged wild-type SEMA6A or Met109Asp mutant by lentiviral infection into SEMA6A/6B double knockout Hap1 cells and assessed their effect on TcsL toxicity. In contrast to wild-type SEMA6A, the Met109Asp mutant did not make SEMA6A/6B double knockout cells more sensitive to TcsL, although both mutants were expressed at similar levels ( Figure 5D ).

Our results establish that TcsL and TcdB have evolved to bind different host receptors through the same interacting region. Consistent with this, the surface is highly diverged between the two toxins: of the 25 SEMA6A-interacting residues in TcsL, only five are identical and nine are similar compared with those in TcdB. Conversely, of the 19 residues in TcdB that interact with FZD2 or PAM, only three are identical and eight are similar in TcsL. In stark contrast, 76% of non-interacting surface residues in TcdB are identical in TcsL (p < 0.0001, Fisher's exact test), indicating that the divergence is specific to the receptorbinding surface. This surface might represent a more general receptor-binding site in clostridial toxins, given that the divergence extends to all members of the family ( Figure S6 ). In particular, the five-stranded beta sheet that forms the base of the hydrophobic pocket for SEMA6A Met109 in TcsL and for the palmitoleic acid moiety in the FZD2-TcdB complex is highly variable between LCTs ( Figure 6A and Figure S6 ). We directly tested the role of the evolutionarily divergent surface in receptor specificity with recombinant proteins. As expected, TcsL 1285-1804 interacted with SEMA6A, and TcdB 1285-1804 formed a stable complex with FZD7 in size exclusion chromatography ( Figure 6B ). However, TcsL did not interact with FZD7, consistent with the extensive divergence of the receptor-binding interface ( Figure 6B) . We then generated a TcsL frizzled-binding domain (FBD) variant TcsL(FBD) 1285-1804 that had a TcdB-like interface, by changing 15 TcsL residues in the receptor-binding interface to those in TcdB ( Figure 6C ). Remarkably, these mutations were sufficient to switch TcsL binding specificity: the TcsL 1285-1804 (FBD) hybrid protein robustly interacted with FZD7 but no longer bound SEMA6A ( Figure 6D ). Thus, selective clustering of mutations on a single surface of the delivery domain of TcdB and TcsL explains the divergent receptor specificity of the two toxins.

Identifying the cellular receptors of bacterial toxins that are the primary determinants of human diseases is of major importance both for elucidating the fundamental strategies used by pathogens to cause disease and for informing the development of therapeutics to block their uptake. Our discovery and structural characterization of semaphorins as P. sordellii lethal toxin receptors provides a potential therapeutic avenue for these devastating infections and reveals how evolution has sculpted the receptor-binding interfaces of large clostridial toxins to recognize distinct host receptors.

SEMA6 family proteins coordinate axon repulsion and attraction during development (Suto et al., 2007; Tawarayama et al., 2010; Xu et al., 2000) , and as such their role in toxin entry in endothelial cells might seem unexpected. However, semaphorins regulate several other cellular processes outside the central nervous system (Neufeld et al., 2012; Worzfeld and Offermanns, 2014) . For example, SEMA6A promotes angiogenesis through vascular endothelial growth factor (VEGF) signaling in vascular endothelial cells, and SEMA6B positively regulates endothelial cell proliferation (Kigel et al., 2011; Segarra et al., 2012) . It seems that P. sordellii and TcsL have evolved to exploit the nonneuronal semaphorin function to their own advantage during infection. In canonical semaphorin-plexin signaling, semaphorins act as ligands for plexins either through cell-cell contacts or in cis. However, signaling can also occur through the reverse route where semaphorins are the receptors that induce downstream signaling events (Haklai-Topper et al., 2010; Jongbloets and Pasterkamp, 2014; Perez-Branguli et al., 2016) . Our cryo-EM structure of the TcsL-SEMA6A complex shows that the binding site of TcsL on SEMA6A significantly overlaps with the native Plexin A2 binding site (Janssen et al., 2010; Nogi et al., 2010) . It is highly unlikely that both ligands could simultaneously bind SEMA6A because of steric clashes. However, disruption of the plexin-semaphorin axis is unlikely to contribute to acute toxicity of TcsL. For example, Hap1 and Vero cells that are highly sensitive to the toxin do not express the Plexin A2 or A4 ( Figure S1 ), and we did not identify known downstream regulators of reverse SEMA6A signaling, such as Abl or Mena (Perez-Branguli et al., (B) Composite cryo-EM map of the Tcsl-SEMA6A complex. SEMA6A monomers are colored pink and yellow, TcsL domain resolved to medium and high resolution (< 7 Å ) is colored dark blue. Low-pass filtered density (10 Å ) of the TcsL protein is shown in light blue. Insets I and II show contact residues between SEMA6A (pink) and TcsL (blue). (C) Atomic model built into the SEMA6A-TcsL map. Cryo-EM density of SEMA6A-TcsL is shown as gray mesh with the model built shown as sticks (pink for SEMA6A and blue for TcsL). (D) Comparison of Plexin A2-SEMA6A (PDB: 3OKY) (Janssen et al., 2010) and TcsL-SEMA6A binding interactions. M109 of SEMA6A interacts with hydrophobic residues of Plexin A2, including L407 and V398 (left). TcsL buries M109 of SEMA6A in a binding pocket containing several hydrophobic residues (middle). An overlay of TcsL and Plexin A2 binding surfaces (shown in blue and purple, respectively) reveals a subset of SEMA6A residues participating in binding to both protein ligands (red). See also Figures S2-S5 . Article 2016), in our CRISPR/Cas9 screens. Future work can address whether disruption of this signaling axis signaling modulates TcsL toxicity in vivo. SEMA6A and SEMA6B add to the list of proteins that have been identified as receptors or host factors for large clostridial toxins, alongside Frizzled, CSPG4, and Nectin3 for TcdB; LRP1 for TpeL; and sulfated glycosaminoglycans and LDLR for TcdA (La-France et al., 2015; Schorch et al., 2014; Tao et al., 2016 Tao et al., , 2019 Yuan et al., 2015) . Given the high degree of conservation between all LCTs, we cannot exclude the possibility that TcsL also binds additional receptors on host cells. However, we have not identified additional factors in CRISPR/Cas9 screens in cells that do not express SEMA6A or SEMA6B. Moreover, if such receptors exist, they must be distinct from those of other LCTs.

TcsL shares almost 90% similarity with C. difficile cytotoxin TcdB, which binds Frizzled family receptors in the colonic epithelium (Chen et al., 2018; Tao et al., 2016) . Previous work had suggested that TcsL and TcdB use different receptors (Chaves-Olarte et al., 1997) , which is consistent with their distinct target tissues during infection. Given the sequence conservation, one might have expected that their receptors are also related to each other. This was not the case: semaphorins and Frizzled receptors are structurally and evolutionarily unrelated. However, the cryo-EM structure of the TcsL-SEMA6A complex revealed that TcsL and TcdB use exactly the same region to interact with their cognate receptors. Comparison of the structures of TcsL and TcdB bound to their cognate receptors provides an elegant molecular explanation for the distinct receptor specificity: selective clustering of adaptive mutations in the receptor-binding interface. The high sequence diversity in this region between all clostridial toxins indicates that this interface is likely a more general receptor-binding site for these toxins. Indeed, the receptor-binding site of TpeL, a more distantly related clostridial toxin, has been mapped to a region that includes the hypervariable site (Schorch et al., 2014) .

The striking diversity in the receptor-binding specificity of clostridial toxins is reminiscent of receptor-binding proteins of several viruses. For example, closely related coronaviruses can bind evolutionarily unrelated host receptors through highly variable loops in the spike protein (Wong et al., 2017; Wu et al., 2020) . Thus, in a similar manner to rapidly evolving viral proteins, a sculptable receptor-binding interface in large clostridial toxins provides an evolutionarily flexible platform that might allow pathogenic clostridial species to adapt to and attack novel host organisms and tissues.

Detailed methods are provided in the online version of this paper and include the following: 

Genome-wide CRISPR screen 50 million Hap1 cells were infected with TKOv3 library at 200-fold library coverage at multiplicity of infection (MOI) of 0.3. The infected cells were selected with 2 mg/mL puromycin for 3 days (T0) and maintained at a minimum of 100-fold gRNA library coverage at any time point. Cells were passaged until the fourth day (T4) to allow sufficient time for protein turnover. 7 million cells were seeded into two 10 cm plates per condition and allowed to recover overnight after trypsinization. At T5, TcsL was added at a final concentration of 0.1 nM or 1 nM, and the cells were further incubated for 48 h. At T7, cells were washed with 1xPBS and allowed to repopulate in normal growth media (IMDM/10% FBS). The untreated population of cells (negative controls) were passaged every three days in parallel. The medium of toxin treated cells were replenished every 3 days. Once surviving colonies were visible, cells were trypsinized and re-seeded to facilitate repopulation. Once the toxin treated cells reached 100% confluence, the cells were collected with the corresponding untreated population. Genomic DNA was extracted from the frozen pellets using QIAamp DNA Blood Maxi Kit (QIAGEN) following the manufacturer's protocol.

The gRNA sequences were PCR amplified from the extracted genomic DNA. Each amplified sample was then barcoded and processed on Illumina Next-seq high-output mode at a read depth of at least 5 million reads per sample. MAGeCK software (Li et al., 2014) was used to generate rankings for positively enriched genes. (Pettersen et al., 2004) to generate a starting model, followed by manual rebuilding using Coot (Emsley et al., 2010) . All models were refined using the phenix.real_space_refine with secondary structure and geometry restraints. The final models were evaluated by MolProbity . Statistics of the map reconstruction and model refinement are presented in Table S2 .

TcsL intoxication experiments were conducted in triplicate unless otherwise noted. All figures show arithmetic means of replicate experiments and error bars denote standard deviations. Fisher's exact test was used to assess the difference between receptorbinding domain conservation and the non-receptor binding surface residues.

Cell 182, 1-12.e1-e7, July 23, 2020 e7

Please cite this article in press as: Lee et al Figure S1 . Validation of SEMA6A and SEMA6B as host factors required for TcsL intoxication, related to Figure 1 and Figure 2 (A) Expression of SEMA6 family genes, the cognate SEMA6A/6B ligands Plexin A2 and Plexin A4, and known clostridial toxin receptors and host cell factors in Hap1 and HeLa cells based on Human Protein Atlas (proteinatlas.org).

(legend continued on next page) ll Article

PHENIX: a comprehensive Python-based system for macromolecular structure solution

Bacterial cytotoxins: targeting eukaryotic switches

Monoglucosylation of low-molecular-mass GTP-binding Rho proteins by clostridial cytotoxins

A novel murine model of Clostridium sordellii myonecrosis: Insights into the pathogenesis of disease

TcsL is an essential virulence factor in Clostridium sordellii ATCC 9714

Clostridium difficile differ with respect to enzymatic potencies, cellular substrate specificities, and surface binding to cultured cells

MolProbity: all-atom structure validation for macromolecular crystallography

Structural basis for recognition of frizzled proteins by Clostridium difficile toxin B

Structure of the full-length Clostridium difficile toxin B

Vaginal and Rectal Clostridium sordellii and Clostridium perfringens Presence Among Women in the United States

Crystal structure of Clostridium difficile toxin A

Structural basis of assembly of the human T cell receptor-CD3 complex

Features and development of Coot

Fatal toxic shock syndrome associated with Clostridium sordellii after medical abortion

Clostridium sordellii lethal toxin kills mice by inducing a major increase in lung vascular permeability

UCSF ChimeraX: Meeting modern challenges in visualization and analysis

Functional defects in Clostridium difficile TcdB toxin uptake identify CSPG4 receptor-binding determinants

Cis interaction between Semaphorin6A and Plexin-A4 modulates the repulsive response to Sema6A

Evaluation and Design of Genome-Wide CRISPR/SpCas9 Knockout Screens. G3 (Bethesda)

Undiagnosed cases of fatal Clostridium-associated toxic shock in Californian women of childbearing age

Structural basis of semaphorin-plexin signalling

cryo-SPARC: algorithms for rapid unsupervised cryo-EM structure determination

Structure-guided design fine-tunes pharmacokinetics, tolerability, and antitumor profile of multispecific frizzled antibodies

Autocatalytic cleavage of Clostridium difficile toxin B

Alignment of cryo-EM movies of individual particles by optimization of image translations

Semaphorin 6A regulates angiogenesis by modulating VEGF signaling

Selection and characterization of ultrahigh potency designed ankyrin repeat protein inhibitors of C. difficile toxin B

Interactions between plexin-A2, plexin-A4, and semaphorin 6A control lamina-restricted projection of hippocampal mossy fibers

Addressing preferred specimen orientation in single-particle cryo-EM through tilting

Frizzled proteins are colonic epithelial receptors for C. difficile toxin B

Sulfated glycosaminoglycans and low-density lipoprotein receptor contribute to Clostridium difficile toxin A entry into cells

Roles of semaphorin-6B and plexin-A2 in lamina-restricted projection of hippocampal mossy fibers

Improvement of cryo-EM maps by density modification (Biochemistry)

An improved cryogen for plunge freezing

An extracellular biochemical screen reveals that FLRTs and Unc5s mediate neuronal subtype recognition in the retina

Receptor-binding loops in alphacoronavirus adaptation and evolution

Semaphorins and plexins as therapeutic targets

A new coronavirus associated with human respiratory disease in China

The transmembrane protein semaphorin 6A repels embryonic sympathetic axons

Structure of a yeast spliceosome at 3.6-angstrom resolution

Chondroitin sulfate proteoglycan 4 functions as the cellular receptor for Clostridium difficile toxin B

Translocation domain mutations affecting cellular toxicity identify the Clostridium difficile toxin B pore

CRISPR screen validation The gRNAs targeting SEMA6A, SEMA6B, SEMA6C, SEMA6D and UGP2 were chosen from TKOv3 library and were cloned into pX459 (Addgene plasmid #62988; a gift from Feng Zhang) and lentiCRISPRv2 (Addgene plasmid #52961

One day post-transfection, the medium was changed to medium containing 2 mg/mL puromycin and was further selected for 3 days. Cells were washed and moved to a 10 cm plate with fresh growth medium with no antibiotics. Wild-type and knock-out cells were seeded on a 96-well plate a day before toxin application at < 40% confluency. Toxins were serially diluted in 1xPBS with 10% glycerol before applying to cells. Cells were incubated with toxins for 24 to 48 h. Cell viability was measured either using AlamarBlue dye (Invitrogen) or CellTiter-Glo reagent (Promega) following the manufacturer's protocol. For ectopic expression of SEMA6A, SEMA6B, SEMA6C, and SEMA6D, full-length genes were cloned into a pLenti6.2 plasmid with a C-terminal FLAG and V5 tags. Lentiviruses were packaged by transfecting the lentiviral plasmid with VSV-G and psPAX packaging plasmids into HEK293T cells. Virus-containing media was collected three days post-transfection. The media containing packaged virus was added to HAP1-SEMA6A knock-out cells in a 1:10 ratio and selected on 10 mg/mL blasticidin for 7 to 10 days

Site-directed mutagenesis The mutation in SEMA6A was generated with oligonucleotides containing the mutation of interest and a standard site-directed mutagenesis protocol, followed by digestion of the template plasmid with DpnI (NEB). Mutations in TcsL(1285-1804) and TcsL were made by synthesizing the mutant fragment and cloning by NebBuilder (NEB)

96-well plate (Corning) and incubated overnight to attach. For toxicity experiments, TcsL was added at the indicated concentrations and incubated for 24 or 48 h (indicated in the figures) at 37 C, 5% CO 2 . Nanoluciferase levels were measured using the Nano-Glo Luciferase Assay System (Promega) as per the manufacturer's instructions. For SEMA6 competition experiments, recombinant SEMA6 proteins were added immediately before TcsL at the indicated concentrations and the assay proceeded as above. Cell viability experiments with Vero cells (without Nluc reporter) were conducted exactly as above, except 10 ml PrestoBlue reagent was added to all wells

To express and isolate recombinant TcsL, transformed B. megaterium was inoculated into LB containing tetracycline and grown to an A 600 of 0.7, followed by overnight xylose induction at 37 C. Bacterial pellets were collected, resuspended with 20 mM Tris pH 8, 0.5 M NaCl, and passed twice through an EmulsiFlex C3 microfluidizer (Avestin) at 15,000 psi, then clarified by centrifuging at 18,000 x g for 20 min. TcsL was purified by nickel affinity chromatography followed by anion exchange chromatography using HisTrap FF and Hi-Trap Q columns (GE Healthcare), respectively. Fractions containing TcsL were verified by SDS-PAGE, then pooled and diafiltered with a 100,000 MWCO ultrafiltration device (Corning) into 20 mM Tris pH 7.5, 150 mM NaCl. Finally, glycerol was added to 15% v/v, the protein concentration was estimated by A 280 (using extinction coefficient 300205), and stored at À80 C. TcsL fragments (1285-1804) and all mutants thereof were synthesized (IDT) and cloned into pET Champion vector with an N-terminal 6xHIS-SUMO-Strep-TEV sequence and a C-terminal TEV-6xHIS. Positive clones were verified by sequencing. NiCo21 (DE3) competent E. coli (NEB) were transformed and inoculated in LB media with kanamycin and grown to an A 600 of 0.6. Protein expression was induced by the addition of 1 mM IPTG for 4 h at 23 C. Cells were pelleted and protein was purified similarly to TcsL, except after nickel affinity chromatography, they were passed through a size exclusion column and eluted into 20 mM Tris pH 8, 150 mM NaCl. Mouse Sema6A-Fc and Sema6c-Fc expression constructs were a gift from Woj Wojtowicz (Addgene plasmids 72163 and 72167, respectively) and human SEMA6A was gene-synthesized (GeneArt)

Recombinant FZD7 was expressed and purified as previously described

The protein was eluted with an increasing gradient of imidazole with a maximum concentration of 500 mM, in a buffer containing 20 mM Tris pH 8.0, 500 mM NaCl and 5% (v/v) glycerol. This step was followed by gel filtration chromatography (Superdex 200 Increase, GE Healthcare) in 20 mM Tris pH 8.0, 150 mM NaCl buffer. To test the formation of toxin-receptor complexes

After 10-30 mg/mL of His-tagged TcsL1285-1804 was immobilized on Ni-NTA sensors, the baseline was established for 60 s. Subsequently, the loaded biosensors were dipped into wells containing serial dilutions of human SEMA6A, to determine the rate of association. Sensors were then dipped back into kinetics buffer to establish the dissociation rate. The curves were fitted to a 1:1 binding model and the apparent dissociation constant (K D ) was evaluated using Forté Bio's Data Analysis software 9

GE Healthcare) in 20 mM HEPES pH 7.0 and 50 mM NaCl. Fractions containing the complex were concentrated to 0.12 mg/mL and proteins were crosslinked by addition of 0.05% (v/v) glutaraldehyde (Sigma Aldrich) and incubated at room temperature for 60 min. The reaction was stopped by addition of Tris-HCl pH 7.0 to a final concentration of 50 mM. Subsequently, the complex was concentrated to 0.9 mg/mL, spun down for 30 min at 14

Motion correction was performed with Patch Motion algorithm and CTF parameters were estimated from the average of aligned movie frames with Patch CTF. 3,896,638 particle images were selected by template matching and individual particle images were corrected for beam-induced motion using local motion algorithm (Rubinstein and Brubaker, 2015) within cry-oSPARC v2. Ab initio structure determination and classification revealed that $50% particle images corresponded to a SEMA6A dimer and the remaining particles to a SEMA6 dimer-TcsL complex. The overwhelming majority of particles for the SEMA6A-TcsL complex had one TcsL molecule bound to the SEMA6A dimer, with no evidence of two TcsL molecules bound to the dimer that could be clearly identified for this cross-linked sample. Since the TcsL-SEMA6A particles exhibited partially preferred orientation on vitrified cryo-EM grids, we collected an additional dataset of 2,843 images under 40 tilt (Lyumkis

Recognition of Semaphorin Proteins by P. sordellii Lethal Toxin Reveals Principles of Receptor Specificity in Clostridial Toxins

Sensitivity of SEMA6A and UGP2 knockout cells to TcsL. SEMA6A and UGP2 knockout cells were generated with CRISPR/Cas9. SEMA6A-3xFLAG was ectopically expressed in SEMA6A knockout cells by lentiviral infection. Data (n = 3) are represented as mean ± standard deviation. Right, SEMA6A expression in wild-type Hap1 cells, SEMA6A KO cells

Hap1 and HeLa cells were treated with increasing concentrations of TcsL and cell viability measured 24 h later. Data (n = 3) are represented as mean ± standard deviation. Inset, expression of SEMA6A in Vero, Hap1, and HeLa cells was assessed by western blot

SEMA6A ectodomain protects Vero cells from TcsL toxicity only when added simultaneously with the toxin. SEMA6A and TcsL were added to Vero cells simultaneously or after 1-min or 1 h pre-incubation. Alternatively, TcsL was added for 1 h prior to treatment with SEMA6A. Data (n = 2) are represented as mean ± standard deviation

The authors would like to thank Dr. Amy Tong, Mr. Christopher Mogg, Mr. Syed Haider, and Ms. Melissa Geng for their help with this project. We are grateful to Dr. Samir Benlekbir for help with cryo-EM data collection and for advice regarding specimen preparation. M.T. and H. 

Lead Contact Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Mikko Taipale (mikko.taipale@utoronto.ca).

Plasmids generated in this study are available from Addgene or from the authors with a Material Transfer Agreement. Figure S6 . Conservation of SEMA6A and TcsL interface residues in semaphoring and large clostridial toxin families, related to Figure 6 A, Sequence alignment of SEMA6 family proteins. Residues conserved in all four SEMA6 family proteins are denoted in light blue. SEMA6A residues forming contacts with TcsL are indicated as red circles, and those interacting with Plexin A2 are indicated as blue circles. The two TcsL-interacting residues that differ between SEMA6A/SEMA6B and SEMA6C/SEMA6D are colored orange. Secondary structure elements with numbered beta-propeller blades are shown below the alignment. (B) The receptor-binding surface of TcsL and TcdB is highly divergent between all clostridial toxins. Partial sequence alignment of six known large clostridial toxins. Secondary structure elements and the consensus sequence (at least 4/6 identical residues) are shown above the alignment. Amino acids are colored based on their biophysical properties (ClustalX coloring) if at least 4/6 residues are similar in each column. Red boxes indicate interface residues in TcsL/SEMA6A or TcdB/Fzd2 complexes. The gray box highlights the evolutionarily divergent beta sheet in the receptor-binding interface. (C) Alignment entropy was calculated as a 20-aa moving window along the alignment of six known clostridial toxins. Red bars indicate the location of TcsL/ SEMA6A interface residues.ll Article