Identification and structural characterization of anthrax toxin receptor 2 as the Clostridium perfringens NetF receptor

Abstract

Hemolysin β-pore-forming toxins (βPFTs) are key virulence factors of Clostridium perfringens, associated with severe diseases in humans and animals. Yet, the mechanisms by which Clostridium βPFTs recognize and engage specific target cells remain poorly understood. Here, we identify the cellular receptor for C. perfringens necrotizing enteritis toxin F (NetF), a recently discovered toxin implicated in severe enteritis in dogs and foals. We show that NetF binds to the same receptor as anthrax toxin, namely ANTXR2. Using cryo-electron microscopy, we determined the structure of the oligomeric NetF pre-pore as well as the transmembrane pore, both alone and in complex with the extracellular domain of ANTXR2. Unlike anthrax toxin, which binds to the apical MIDAS motif of ANTXR2 – as does the natural ANTXR2 ligand collagen type VI – NetF engages the receptor laterally, spanning both the von Willebrand A and the Ig-like domains. This interaction positions the toxin near the membrane, facilitating contact with membrane lipids and promoting transmembrane pore formation. Our findings uncover key principles of hemolysin βPFT-receptor recognition and advance our understanding of how pathogenic bacteria use these toxins to breach host defenses.

This study identifies ANTXR2 as the cellular receptor for Clostridium perfringens toxin NetF and determines its structure bound to the toxin using cryo-EM, revealing a distinct lateral binding mechanism that facilitates membrane pore formation.

Introduction

Clostridium perfringens is a versatile pathogen, causing wound infections, septicemia, enterotoxemia or enteritis in animals and humans^1–3. Especially, intestinal diseases in animals such as pigs, cattle, sheep, goats, chickens, dogs, and horses result in significant losses of livestock and companion animals worldwide. C. perfringens induces tissue damage via highly potent exotoxins of which hemolysin β-pore-forming toxins (βPFTs) represent the largest group, with currently 11 identified members^4,5. Two of these, C. perfringens beta-toxin (CPB) and necrotic enteritis toxin B (NetB), are essential virulence factors in necrotic enteritis in pigs and humans and poultry, respectively. Our knowledge about other C. perfringens hemolysin βPFTs, however, is limited. In 2015, a hemolysin βPFT called NetF was identified in C. perfringens isolates from dogs with hemorrhagic gastroenteritis and foals with necrotic enterocolitis⁶. Subsequent epidemiological evidence substantiated, that the toxin plays an important role in C. perfringens induced enteric disease in these animals^7–11. NetF molecular weight is 34.4 kDa, it shares 30%, 33% and 48% sequence identity with S. aureus α-hemolysin (Hla), CPB and NetB, respectively. It was shown to bind to the membrane of susceptible cells through interaction with an undefined sialylated receptor and to form oligomers with varying stoichiometries on liposomes¹².

Hemolysin βPFTs are mostly secreted by the bacteria as soluble monomers. They then bind to receptors on target cells, oligomerize into pre-pores and subsequently undergo conformational change to insert into the plasma membrane as ion permissive pores¹³. Thus, identifying membrane receptors and the structural basis of the interaction between the toxin, the receptor and the plasma membrane is key to understanding the function of these toxins. So far, only two receptors for clostridial hemolysin βPFTs have been identified: CD31 for C. perfringens beta-toxin (CPB) and ganglioside GM2 for delta-toxin^14,15. In addition, oligomeric pore-structures have only been determined for CPB and NetB^14,16,17, whereas precise structural information about clostridial toxin-receptor and toxin-membrane interaction is still lacking.

Here, we elucidate the structure of NetF in its oligomeric pre-pore and its membrane-inserted pore state. We identify anthrax toxin receptor 2 (ANTXR2), also known as capillary morphogenesis gene 2 (CMG2), as a receptor for NetF and show that the toxin can efficiently target human, bovine, equine, and canine ANTXR2. In addition, we determine the co-structure of NetF oligomer in complex with the extracellular part of ANTXR2, revealing insights into the interaction between a clostridial hemolysin βPFT and its membrane receptor.

Results

Structural characterization of NetF

Previous research demonstrated that NetF oligomerizes in different stoichiometries, forming complexes composed of six to nine protomers¹². We recombinantly expressed the toxin in Escherichia coli and confirmed that NetF binds and oligomerizes on synthetic membranes. Cholesterol increased membrane binding of NetF (Supplementary Fig. 1A–C). To optimize the conditions for cryo-EM characterization of NetF oligomeric form, we generated pores following two distinct methods: (1) incubation of NetF monomers with liposomes, leading to the formation of oligomeric pores, which were subsequently solubilized (Supplementary Fig. 1D), and (2) incubation of NetF monomers with pre-formed lipid nanodiscs (Supplementary Fig. 1E). Surprisingly, the latter led predominantly to nonameric pre-pores (Fig. 1A, B), while detergent-solubilized oligomers mostly adopted an octameric pore conformation (Fig. 1C, D), as monitored by cryo-EM. The structures of NetF pre-pore and pore represent a hemolysin-like oligomer organized into three distinct regions: the cap, the rim, and the stem (Fig. 1C, E). The high resolution cryo-EM maps enable precise atomic modeling of the backbone and most sidechains, except for the first eleven N-terminal amino acids, which are partially unstructured (Supplementary Fig. 2).

Fig. 1. Structure of NetF oligomers and structural changes required for pore formation.

A Cryo-EM map of NetF on MSP2N2 lipid nanodiscs containing DOPC:DOPG:Cholesterol. Upon NetF addition nine monomers bind and oligomerize into a pre-pore conformation. Consecutive protomers are colored by a pastel rainbow. Extra densities are shown in blue. B The structures of two consecutive protomers extracted from the pre-pore are shown in cartoon representation and colored as in A. Residues involved in protomer—protomer interaction are shown as ball and stick with the distances shown as red dashed lines. Two representations rotated by 180° are shown for clarity. C Cryo-EM map of NetF pores after oligomerization on DOPC:DOPG:Cholesterol liposomes and solubilization in DDM:CHS. Eight NetF monomers bind and oligomerize to form the membrane pore. Consecutive protomers in the oligomer are colored as in A with extra densities shown in blue. The position of the cap, rim and stem is shown. D The structure of a protomer extracted from the pore is shown in cartoon representation as in (B). Residues involved in inter-protomer interactions are shown in ball and stick representation with the inter-residue distance shown as red dashed lines. Two views are shown for clarity rotated by 180°. E Comparisons between the soluble NetF monomer (AlphaFold prediction) on the left, a protomer extracted from the NetF pre-pore (middle) and a protomer extracted from the pore (right). The position of the N-terminus is shown in orange and emphasized by a black arrow (top). While the monomer prediction shows the N-terminus part of a β-sheet of the cap domain (left) it moves upward from its position in the pre-pore (middle). This movement is accompanied by the pre-stem shifting towards the position previously occupied by the N-terminus (purple, emphasized by lower black arrow). Upon membrane insertion the pre-stem flips towards the membrane (purple and marked by the lower arrow). The N-terminus flips down to the position previously occupied by the pre-stem loop into the entrance cavity of the pore.

In the absence of X-ray data, the NetF soluble monomeric structure was predicted using AlphaFold3¹⁸ (Fig. 1E, Supplementary Fig. 3A). A comparison between the AlphaFold3 monomer prediction and a protomer extracted from the NetF pre-pore indicates that AlphaFold3 prediction is extremely accurate with a root-mean-square deviation (RMSD) of 1.9 Å across all amino acids. Although the overall prediction confidence score is high (pTM of 0.91), the N-terminal region, the pre-stem loop, and one loop of the rim domain display lower confidence (Supplementary Fig. 3A). The differences between the predicted monomer and the protomer, as well as comparison with other hemolysin βPFTs, suggest that multiple conformational changes are necessary for oligomerization (Fig. 1E). The oligomerization of NetF into pre-pore conformation involves extensive interactions (Supplementary Table 1 and 2), engaging a large area on each side of the cap and rim domains, measuring approximately 1400 Å2 and encompassing more than 50 residues (Fig. 1B, Supplementary Fig. 2B). The N-terminus, however, remains flexible and unresolved (residues 1–20). In the pre-pore conformation, the pre-stem loop is partially visible except for five residues (Fig. 1E). Notably, in contrast to the AlphaFold3 prediction of NetF monomer and to the monomer structure of other known hemolysins, the pre-stem of the pre-pore is moved upwards, away from its position on the cap domain. This structural rearrangement may play a critical role in pore formation, potentially priming the protein for membrane insertion.

The NetF pore structure reveals additional conformational changes required for membrane insertion. In the pre-pore, the pre-stem loop is folded against the cap domain, and its extraction during pore formation triggers a second rearrangement of the N-terminus. Although unresolved in both structures, the N-terminus position can be inferred from its orientation and diffuse density of the backbone and N-terminal His-tag (Fig. 2A–C). In the pre-pore, it points outward, but upon pre-stem loop unfolding, it shifts toward the cavity formed by the cap domain, loosely occupying the site of the pre-stem (Fig. 2B). A focused 3D classification of the upper part of the pre-pore and pore confirmed this difference (Fig. 2D). As the β-barrel forms, inter-protomer interactions expand by 50% to ~2100 Ų (Supplementary Table 1 and 2), involving 90 residues (Fig. 1B, Supplementary Fig. 2D). Notably, the cap and rim domains remain largely unchanged during this transition.

Fig. 2. Characterization of NetF–lipid interaction and the transmembrane pore.

A Cryo-EM map of NetF pre-pore showing extra densities (red) of hexa-histidine tagged N-terminus (N) and lipids bound to the rim domain (R). The density disk below originates from the nanodisc. Side view shows each protomer colored with extra densities in red. N-terminus density (N) appears above the oligomer while lipids bound to outer rim are visible. Median cut-through view reveals empty cap cavity and lipid densities at inner rim. B Cryo-EM map of NetF pore showing extra densities of lipids and detergents bound to rim domain (R) and β-barrel trans side (BT). Median cut-through view shows hexa-histidine tagged N-terminus (N) positioned in cap cavity center with density on inner cap domain. Lipid and detergent molecules (red) interact with rim domain inner and outer surfaces (R). Lipids also interact with β-barrel on trans cytoplasmic and outer cis sides. C NetF protomer ribbon model extracted from pre-pore (top) shows cap domain (orange), rim domain (pink), and pre-stem loop (green). Gray lipid densities bind rim domain outer (RO) and inner (RI) surfaces. Aromatic residues in rim and stem domains are red (ball-and-stick). Pore ribbon model (bottom) shows lipid densities (gray) on outer rim (RO), inner rim (RI), and aromatic ring on cis side of transmembrane β-barrel (BC). Aromatic residues involved in lipid interaction are red (ball-and-stick). D Class average following 3D classification focused on the shown region. Backbone of several N-terminal residues is visible in red. E NetF pore inner cavity (blue) calculated using HOLE. Large cis-side cavity forms from cap domain oligomerization. Arginine 113 forms tightest cis-side constriction (upper arrowhead). Middle arrowhead indicates aromatic belt formed by tyrosine 120 and phenylalanine 141. Second trans/cytoplasmic constriction forms from β-turns flipping inwards (lower arrowhead). Cavity dimensions shown left. (F) NetF pore radius plot from (E) showing two constriction sites: arginine 113 (cis side) and β-turn at lysine 130 (trans side).

The most significant rearrangement during pore formation is the refolding of the pre-stem loop into a 50 Å-long transmembrane β-barrel, characterized by a hydrophilic lumen and a hydrophobic exterior. The barrel has a diameter of approximately 10 Å, with two constrictions that narrow the lumen to 6.7 Å on the cis (i.e., extracellular) side and 8.2 Å on the trans (i.e., cytoplasmic) side (Fig. 2E, F). The lumen of the transmembrane pore is polar but contains no charged residues, except for an arginine residue (Arg 113) at the cis entrance and two lysine residues (Lys 130 and 131) on the trans side, both of which contribute to the constriction sites. While the Lys 130 points away from the lumen it bends the β-turn inwards creating a second constriction. An aromatic belt formed by residues Tyr 120 and Phe 141 is positioned on the cis side oriented towards the lipids, consistent with the typical features of transmembrane β-barrels (Fig. 2C, E).

NetF interacts with lipids through the rim and stem domains

The cryo-EM maps also reveal substantial extra densities surrounding the rim and stem domains of NetF oligomer, in particular in the pore conformation (Fig. 2A–C). Based on their shape and chemical environment, these densities correspond to bound lipid and detergent molecules. They are located around loops containing aromatic and charged residues: Tyr 74, Tyr 75, and Trp 254 bind lipids on the outer rim domain; Tyr 184, Trp 259, His 199 interact with lipids at the protomer-protomer interface. Furthermore, clear lipid/detergent densities are visible tightly interacting with the aromatic belt (Tyr 120 and Phe 141) as well as the middle and trans parts of the transmembrane β-barrel (Fig. 2B, C). It has been previously demonstrated for S. aureus α-hemolysin and related toxins that a loop in the rim domain interacts with lipid headgroups^19,20. The equivalent loop is however shortened in NetF and lacks the tryptophan residue required for lipid binding (Trp 179, Supplementary Fig. 3B), as also observed in C. perfringens β-toxin and NetB^17,21. The presence of NetF-bound lipids and detergents suggest that the rim domain is partially embedded in the membrane tightly interacting with lipids and cholesterol.

NetF resistance is associated with reduced ANTXR2 expression

We next set out to identify the cell surface receptor for NetF. To determine suitable cell lines, we performed viability assays on different human, murine, canine, equine, and bovine cells. NetF was toxic to human U937 and HFF-1 cells, canine MDCK and glial J3T cells, primary equine fibroblasts, and bovine endothelial BUcEC cells. In contrast, other cell lines from these species as well as two murine cell lines were resistant to NetF (Fig. 3A). We additionally determined that NetF rapidly leads to LDH release, shrinkage, and lysis in susceptible J3T cells (Supplementary Fig. 4A, B).

Fig. 3. ANTXR2 interacts with NetF and is essential for NetF cytotoxicity.

A Viability of human, murine, equine, canine, and bovine cell lines after incubation with indicated doses of NetF (24 h, 37 °C), in % to untreated control cells. Data are represented as of 3 biological replicates and each performed in technical triplicates (n = 3) ± SD. B Volcano plots depicting gene-expression profiles of NetF-resistant U937 and J3T cells compared to the respective NetF-sensitive wild type polyclonal cell population performed by edgeR (version 3.19). Differential expression analysis was performed using the edgeR exact test (two-sided, negative binomial distribution). P values were adjusted for multiple comparisons using the Benjamini-Hochberg FDR correction. ANTXR2 stood out as the most relevant down-regulated hit present in both experiments. C TIDE analysis showing high percentage of frameshift and in frame mutations introduced in the ANTXR2 gene after sgRNA1- and sgRNA2-mediated ANTXR2 knockout in U937 cells. D Viability of U937 cells targeted by sgRNA1- and sgRNA2-mediated ANTXR2 knockout after incubation with 1 µg/ml NetF (24 h, 37 °C), in % to untreated control cells. The empty lentiviral vector (empty) was used as control. Data are represented as means of 3 biological replicates and each performed in technical triplicates (n = 3) ± SD. One-way ANOVA, Šidák’s multiple comparison test, **** (p < 0.0001), *** (p = 0.0002). E Viability of HAP1 cells expressing mNeonGreen-tagged human ANTXR2, mouse ANTXR2, and human ANTXR1 after incubation with indicated NetF doses (24 h, 37 °C), in % to untreated control cells. Data are represented as means of 3 biological replicates and each performed in technical triplicates (n = 3) ± SD. F Westen blot analysis of co-IP experiments in HAP1 cells expressing human ANTXR2, mouse ANTXR2 and human ANTXR1, incubated with 10 µg/ml NetF. HAP1hANTXR2 cells without addition of NetF (–) were used as negative control. IP was performed using anti-mNeonGreen beads and showed enriched pull-down of NetF when coupled with hANTXR2. Immunoblot containing 50% eluate (IP) was performed on the same membrane with anti-His antibody to detect NetF and anti-mNeonGreen antibody (pulldown efficiency control). An additional immunoblot was performed with equal sample volumes and probed with anti- ANTXR2 antibody. Immunoblot containing 5% loading fraction (inputs) was performed on the same membrane using anti-His, anti-mNeonGreen, and anti-tubulin (loading control) antibodies. The figure shows one of three replicates of the same experiment.

A small fraction of cells in several susceptible cell lines consistently survived NetF exposure, regardless of the toxin concentration applied. We chose the human U937 and the canine J3T cell lines, which showed less than 10% viability levels after 24 h of toxin incubation (Fig. 3A), to evaluate whether surviving cells were NetF-resistant subpopulations. Following six passages under constant toxin pressure, we isolated resistant subpopulations of both cell lines (U937-R and J3T-R) (Supplementary Fig. 4C). Compared with the parental cell populations mRNA expression profiles of these resistant subclones showed four significantly downregulated genes in both U937-R and J3T-R cells: ANTXR2, PRDM8, CREB5, and ZNF790 (Fig. 3B and Supplementary Fig. 4D). ANTXR2 was the only gene encoding a transmembrane protein. PRDM8, encoding an intracellular histone methyltransferase, is a neighboring gene to ANTXR2 and is co-regulated with it in the same transcriptional unit²². CREB5 encodes a transcription factor that binds to cis-responsive elements (CRE) in the promoter region of the ANTXR2 transcriptional unit²². No link has been identified between ANTXR2 expression and zinc finger protein 790 (ZNF790), a nuclear protein predicted to be involved in transcriptional regulation^23,24. Consistent with cell viability data, qRT-PCR analyses in human, canine, and bovine cells showed that ANTXR2 mRNA levels were higher in susceptible than in non-susceptible cells (Supplementary Fig. 4E). Interestingly, despite showing resistance to NetF, ANTXR2 mRNA was detectable in murine endothelial cells.

ANTXR2 is one of two receptors for anthrax toxin protective antigen (PA). The other PA receptor is ANTXR1, also known as tumor endothelial marker 8 (TEM8)²⁵. ANTXR1 and ANTXR2 are single pass transmembrane proteins with two extracellular domains: an N-terminal von Willebrand A (vWA) domain containing a metal ion-dependent adhesion site (MIDAS) motif and a membrane proximal immunoglobulin (Ig) domain²⁶. Additionally, they contain a C-terminal cytoplasmic domain involved in ligand-mediated signaling. ANTXR2 binds collagen VI via the MIDAS motif and participates in extracellular matrix homeostasis. Our RNA-sequencing (RNA-seq) data indicated that ANTXR1 expression was unchanged in U937-R and J3T-R compared to their susceptible counterparts (Supplementary Data 1). These results led us to hypothesize that ANTXR2 serves as the membrane receptor for NetF.

ANTXR2 interacts with NetF and is essential for its cytotoxicity

To validate the role of ANTXR2 in NetF toxicity, we performed CRISPR-Cas9-mediated knockouts (KOs) of ANTXR2 in U937 cells using two single-guide RNAs (sgRNAs) which led to a high number of frameshift and/or in-frame mutations (Fig. 3C). Viability assays showed that sgRNA1- and sgRNA2-targeted cells became resistant to NetF treatment (Fig. 3D) even though NetF binding and oligomerization were still detectable by western blot analysis on U937-R (Supplementary Fig. 4F) as well as ANTXR2-KO U937 (Supplementary Fig. 4G). PA binding was markedly reduced in U937sgRNA1 and U937sgRNA2 cells compared to cells treated with a control empty vector (Supplementary Fig. 4H). This indicated successful depletion of ANTXR2 from the surface of U937 cells. These results were confirmed using HFF-1 cells where disruption of the ANTXR2 gene by either sgRNA1 or sgRNA2 led to a significant increase in cell viability upon NetF treatment (Supplementary Fig. 4I, J).

We next overexpressed mNeonGreen-tagged human ANTXR2 (hANTXR2), murine ANTXR2 (mANTXR2), and human ANTXR1 (hANTXR1) as a control in HAP1 cells. Expression, localization and functionality of the constructs were confirmed by western blot, PA binding assays (Supplementary Fig. 5A) and immunofluorescence (IF) (Supplementary Fig. 5B). Ectopic overexpression of hANTXR2 rendered HAP1 cells highly susceptible to NetF, unlike overexpression of hANTXR1 (Fig. 3E). Furthermore, mANTXR2 did not confer susceptibility to NetF, consistent with our observation that all tested murine cell lines were resistant.

To assess interaction between ANTXR2 and NetF, we performed co-immunoprecipitations (co-IPs) from HAP1 cells overexpressing hANTXR2, mANTXR2_, or hANTXR1 using anti-mNeonGreen-coated beads. We were able to co-precipitate NetF monomers and SDS-resistant oligomers in cells overexpressing human ANTXR2, while NetF co-precipitation was reduced in cells overexpressing mANTXR2 or hANTXR1 (Fig. 3F).

NetF targets ANTXR2 across different animal species

Since NetF was identified in canine and equine enteritis, we tested whether ANTXR2 ortholog expression sensitizes HAP1 cells to NetF. We ectopically expressed these orthologs as C-terminally mNeonGreen-tagged proteins (Fig. 4A). Equine and bovine ANTXR2 localized correctly to the plasma membrane (Supplementary Fig. 6A) and conferred susceptibility (Fig. 4B), whereas canine ANTXR2 (cANTXR2) was largely retained intracellularly (Supplementary Fig. 6A), preventing assessment. To overcome this, we expressed cANTXR2 in HEK293 cells alongside hANTXR2, mANTXR2, and hANTXR1 as controls. In this cell line, cANTXR2 was expressed at the cell surface and rendered HEK293 cells highly susceptible to NetF (EC50: 1–2 ng/ml), while mANTXR2-expressing cells were susceptible only at much higher NetF concentrations (EC50: 125 ng/ml) (Supplementary Fig. 6B, C, Fig. 4C). These results demonstrate that NetF toxicity depends on the surface expression of ANTXR2 and that it interacts with equine, bovine and canine ANTXR2. Interestingly, we noticed that high concentration of NetF formed oligomers on cells overexpressing ANTXR2, including mANTXR2 (Supplementary Fig. 6D). This is reminiscent of our observations in HAP1 cells (Fig. 3F) and U937 cells (Supplementary Fig. 4F, G).

Fig. 4. NetF interacts with ANTXR2 orthologs of many animal species.

A Western blot analysis for expression levels of HAP1 cells transfected with various mammalian ANTXR2 orthologs. Immunoblots were probed on the same membrane with anti-mNeonGreen and anti-tubulin (loading control) antibodies. An additional western blot was performed with equal sample volumes and probed with anti-ANTXR2 antibody. B Viability of HAP1 cells expressing different mammalian ANTXR2 orthologs after incubation with 1 µg/ml NetF (24 h, 37 °C), in % to untreated control cells. Data are represented as means of 3 biological replicates and each performed in technical triplicates (n = 3). One-way ANOVA, Sidak’s multiple comparison test, **** (p < 0.0001). C Viability of HEK293 cells expressing different mammalian ANTXR2 orthologs after incubation with 1 µg/ml NetF (24 h, 37 °C), in % to untreated control cells. Data are represented as means of 3 biological replicates and each performed in technical triplicates (n = 3) ± SD. D Schematic drawing of the extracellular and transmembrane regions of recombinant ANTXR2 WT and chimeric molecules used for ectopic expression in mammalian cells. Human: blue, mouse: red, vWA: von Willebrand A domain Ig: Ig domain. E Western blot analysis for expression levels of HAP1 cells expressing human-mouse ANTXR2 chimeric constructs. Immunoblots were probed on the same membrane with anti-mNeonGreen and anti-tubulin (loading control) antibodies. An additional western blot was performed with equal sample volumes and probed with anti-ANTXR2 antibody. F Viability of HAP1 cells expressing indicated chimeric constructs (Supplementary Fig. 7A, B) after incubation with 1 µg/ml NetF (24 h, 37 °C), in % to untreated control cells. Data are represented as means of 3 biological replicates and each performed in technical triplicates (n = 3) ± SD. One-way ANOVA, Sidak’s multiple comparison test, **** (p < 0.0001).

Given the high sequence identity of human and murine ANTXR2, we were intrigued that mANTXR2 did not constitute an efficient receptor. Since mANTXR2 differs the most from other mammalian orthologs in its Ig-like domain (Supplementary Fig. 7A), we generated chimeric proteins between hANTXR2 and mANTXR2 (Fig. 4D), expressed the constructs in HAP1 cells and confirmed their expression and correct cellular localization by western blot and IF (Fig. 4E, Supplementary Fig. 6D). Replacing the Ig-like domain of the human protein with the corresponding murine domain significantly decreased susceptibility of HAP1 cells to NetF (Fig. 4F). Conversely, replacing the murine Ig-like domain with the human Ig-like domain in mANTXR2 rendered cells susceptible to NetF. This suggests that, unlike for PA binding, the Ig-like domain of ANTXR2 is involved in NetF interaction and that differences between the murine and human protein in this domain affect this interaction.

Structural characterization of NetF and ANTXR2 interaction

We overexpressed the extracellular region of hANTXR2 (termed extANTXR2) in Expi293 cells, incorporating a C-terminal FLAG and Hisx6 tag (Supplementary Fig. 8A). Addition of purified extANTXR2 to pre-formed NetF oligomers in detergent resulted in receptor binding with varying stoichiometries, most commonly two to four extANTXR2 molecules per NetF pore, as observed by cryo-EM (Supplementary Fig. 8). Since binding on consecutive protomers was commonly observed without apparent steric hindrance, we hypothesize that, in vivo, the ANTXR2-NetF protomer stoichiometry within a NetF oligomer could reach up to a 1:1 ratio. Particle classification of the consensus structure (Fig. 5A, Supplementary Fig. 8E) revealed subsets with ANTXR2 molecules bound that could achieve sufficiently high resolution for detailed structural characterization of the interaction (Supplementary Fig. 8F). Symmetry expansion and focused refinement on a single NetF protomer with bound ANTXR2 further improved the resolution of the receptor binding site (Fig. 5C).

Fig. 5. Structural characterization of the NetF ANTXR2 interaction.

A Top and side views of the consensus initial model obtained from initial processing of NetF:ANTXR2 sample. The density of NetF is marked as the inner density while the position of ANTXR2 density is marked by black arrows. B Cryo-EM map of NetF protomer bound to ANTXR2. NetF density is shown in gray while for ANTXR2 the vWA domain is shown in blue and the Ig-like domain in yellow. Extra densities corresponding to the NetF N-terminus and the glycosylation of Asp260 (arrowhead) are shown in red. Two representations rotated by 180° are shown for clarity. The red dotted rectangle highlights protein—protein contacts of the N-terminus of ANTXR2 with NetF. C The interface between NetF and ANTXR2 is shown in cartoon representation. Residues involved in the interface are shown as ball-and-stick representation and the distances are shown as red dashed lines. NetF is shown in pink while for ANTXR2 the vWA domain is shown in blue and the Ig-like domain is shown in yellow. The tyrosine patch at the interface is shown as transparent surface representation in red. Two views rotated by 180° are shown for clarity.

The 2.1–3.12 Å resolution map shows that NetF interacts with ANTXR2 over a contact area of approximately 1400 Ų, involving binding sites on both VWA and Ig-like domains (Fig. 5C, Supplementary Table 1 and 3). A single ANTXR2 molecule interacts with only a single NetF protomer. More than 40 amino acids from each protomer contribute to the interfacial region, while ten residues on ANTXR2 and 11 residues on a NetF protomer establish ten hydrogen bonds and nine salt bridges (Fig. 5C, Supplementary Table 1 and 3). Additionally, seven ANTXR2 residues and nine NetF residues are engaged in hydrophobic interactions. Note that some residues are involved in more than one type of interaction. Overall, 15 ANTXR2 residues participate in direct interactions, eight in the vWA domain and seven in the Ig-like domain (Supplementary Fig. 7A, black boxes). Sequence comparison of mammalian ANTXR2 orthologs showed that five patches contain differences between the human and murine sequences, either within the boxed regions or directly adjacent to them (patch S1 in vWA domain, and patches S2 to S5 in the Ig-like domain). To evaluate which of these mutations may explain the resistance to NetF of murine lines expressing ANTXR2, we introduced each of these murine substitutions individually into hANTXR2 in silico and predicted the resulting complexes with AlphaFold3 (Supplementary Data 2). Three substitutions had little or no effect on the predicted interaction. In contrast, two changes in the Ig-like domain corresponding to Q192 in human versus S192 in mouse (patch S2) and E201 in human versus K201 in mouse (patch S3) reduced the predicted interaction confidence (ipTM 0.83 to below 0.7). The E201K substitution produced an inverted orientation of ANTXR2 relative to NetF. Introducing both substitutions together further decreased the ipTM (below 0.6) and also yielded flipped configuration. These results support the chimera experiments (Fig. 4F) and indicate that, among the sequence differences present at the interface, the substitutions at S2 and S3 are the primary contributors to the reduced NetF sensitivity of murine ANTXR2.

Notably, the N-terminus of ANTXR2 wraps around NetF protomer on one side of the cap/rim domain interface, while Leu 274 and Asn 275, located on a loop on the opposite side, are inserted in a groove between the same domains of NetF (Fig. 5B, C). A short loop of ANTXR2 (Gly 243—Asn 250) which contains one of the two putative glycosylation sites (Asn250) is flexible and not well resolved in the EM map. Its position is however too far from the NetF protomer to be directly involved in binding. The glycosylation state of Asn 250 cannot be confidently determined, while a clear density at the second putative glycosylation site, Asn 260, confirms the presence of a glycan on the side of ANTXR2 facing away from the NetF-binding interface (Fig. 5C). Overall, these data indicate that the two glycosylation sites Asn 250 and Asn 260 are not directly involved in the interaction with NetF. One particularity of the NetF structure is the presence of 15 exposed tyrosine residues on its surface with 9 of them on the outer cap and rim surface and 6 of them located in the contact site (Fig. 5C). Hydrogen bonds and cation – pi interactions are part of the contact site. Six more tyrosine residues are involved in membrane interaction. This distribution suggests a role in receptor binding, potential oligomer stabilization, and membrane anchoring, highlighting tyrosine-rich regions as key structural features of NetF.

Discussion

In this study, we elucidate the structure of NetF in its oligomeric pre-pore and its membrane-inserted pore state. We show that it strongly interacts with lipids in a way reminiscent to the binding of Hla and other βPFTs of the hemolysin sub-family^17,19,20. In addition, we identify ANTXR2, as a receptor for NetF. By determining the co-structure of the NetF oligomeric pore bound to the extracellular domain of ANTXR2 we unravel the detailed structural basis of the interaction a hemolysin βPFT and its membrane receptor.

Our results show that although anthrax PA and NetF share the same cellular receptor, their interaction with ANTXR2 markedly differs. PA binds ANTXR2 at the apical portion of the vWA domain, engaging the MIDAS motif, similarly to ANTXR2 interaction with collagen VI^27,28. In contrast, NetF interacts laterally with both the Ig-like and vWA domains of ANTXR2, making contacts across a broader surface area but without engaging with the MIDAS motif. Additionally, NetF interacts with membrane lipids through its rim domain, while also engaging with the membrane-proximal part of ANTXR2. During receptor binding, the toxin rim domain primarily interacts with the receptor Ig-like domain, while the cap domain interacts with the vWA domain.

Another key difference that distinguishes NetF from anthrax PA is the length of their β-barrels. NetF forms a ~ 50 Å-long β-barrel pore, matching the thickness of the lipid bilayer. In contrast, PA has an extended β-barrel ( ~ 110 Å) because it binds at the distal region of ANTXR2, requiring it to span both the entire extracellular region of ANTXR2 ( ~ 50 Å) and the lipid bilayer²⁹.

Prior to our study, the only reported structure of a hemolysin βPFT in complex with the extracellular portion of its receptor was that of S. aureus LukGH bound to CD11b/CD18³⁰. Interestingly, LukGH interacts with the most distal part of CD11b, approximately 200-250 Å away from the membrane when the receptor is fully extended. Given that its β-barrel length is similar to that of NetF, LukGH pore formation would require either receptor bending or toxin detachment from the receptor. In contrast, NetF binds to the membrane-proximal Ig-like domain of ANTXR2, enabling direct membrane interaction and pore insertion. Interesting, like NetF, CPB binds the membrane-proximal Ig-like domain of its receptor, CD31²¹, suggesting a conserved binding pattern among clostridial hemolysin βPFTs. Additionally, the ANTXR2 structure obtained in this study includes the complete extracellular domain with its posttranslational modifications.

While we elucidated the binding mechanism of NetF, its role in disease development remains to be determined. We show that NetF effectively targets ANTXR2 of various mammals, including dogs and horses, where NetF-producing C. perfringens strains are linked to enteric disease¹⁰.

Since anthrax toxin induces immune suppression and vascular collapse^31–34, primarily by targeting ANTXR2³⁵, NetF may play a similar role in canine and equine enteritis. However, unlike anthrax PA, NetF does not use ANTXR1 as a receptor.

NetF oligomerization alone is not sufficient for toxicity. We observe robust oligomer formation on protein-free liposomes and at the surface of NetF-resistant cell lines, yet without inducing cell death. Although the molecular basis of this discrepancy is beyond the scope of this study, one possibility is that the high protein content of cellular membranes may limit the ability of NetF oligomers to engage and insert into the bilayer, such that ANTXR2 might act as a docking scaffold stabilizing the oligomer during membrane insertion.

While our data show different binding specificities towards different species—particularly murine—the residues identified in the protein-protein interaction appear to be mostly conserved. Specifically, out of the 15 ANTXR2 residues that directly interact with NetF, 11 are invariant across all species considered in this study. Only one residue shows poor conservation. We identified three contact residues that differ in mouse. One is a conservative change (glutamate replaced by aspartate) within the VWA domain. In agreement with this minor physicochemical difference, the equivalent human–mouse domain substitution did not affect susceptibility to NetF. The two other differences are located in the Ig-like domain: a leucine to proline substitution within a weakly conserved contact site, and an adjacent asparagine to ser substitution. AlphaFold predictions indicate that neither of these two substitutions is critical for the interaction. Further AlphaFold predictions indicated that a glutamine-to-serine substitution, and more prominently an aspartate-to-lysine substitution, positioned adjacent to or one residue away from a contact residue, could reduce the interaction strength. This decrease may stem from local changes in polarity or charge that subtly alter the geometry or electrostatics of the binding surface.

In conclusion, by revealing the extensive molecular interaction of NetF with the vWA and Ig-like domains of its receptor ANTXR2, we enhance our understanding of receptor-mediated targeting by hemolysin βPFTs and provide unprecedented insight into the specific amino acid interactions driving toxin-receptor binding.

Methods

Cell culture

HEK293FT, HFF-1, EpH4, bEnd.3, equine fibroblasts, J3T-bg, MDCK, and BoMac cell lines were cultured in DMEM medium (Gibco, 41965039) supplemented with 10% fetal calf serum FCS (BioConcept, 2-01F00-I) and 2 mM L-Glutamine (Gibco, 25030081). HeLa and ECaNep cell lines were cultured in MEM medium (Gibco, 21090022), supplemented with 10% FCS and 2 mM L-Glutamine. U937 cells were cultured in RPMI 1640 medium (Gibco, 11875093) supplemented with 10% FCS and 2 mM L-Glutamine. HAP1 cells were cultured in IMDM medium with GlutaMAX™ (Gibco, 31980022) supplemented with 10% FCS. BUcEC cells were cultured in PriGrow C medium (Abm, TM100) supplemented with 2% FCS. All cell lines were cultured in the presence of penicillin-streptomycin (Gibco, 10378016) and grown at 37 °C in an atmosphere containing 5% CO2.

Plasmids

All plasmids used in this study are listed in Supplementary Table 4. psPAX2 was a gift from Didier Trono (Addgene plasmid # 12260; http://n2t.net/addgene:12260; RRID:Addgene_12260). pMD2.G was a gift from Didier Trono (Addgene plasmid # 12259; http://n2t.net/addgene:12259; RRID:Addgene_12259). LentiCRISPR v2 was a gift from Feng Zhang (Addgene plasmid # 52961; http://n2t.net/addgene:52961; RRID:Addgene_52961)³⁶. pHR-SFFV_3C-Twin-Strep was a gift from A. Radu Aricescu (Addgene plasmid # 113900; http://n2t.net/addgene:113900; RRID:Addgene_113900)³⁷. pMSP2N2 was a gift from Stephen Sligar (Addgene plasmid # 29520; http://n2t.net/addgene:29520; RRID:Addgene_29520)³⁸.

Primers and sgRNAs

All primers and sgRNAs used in this study were ordered from Microsynth AG (Balgach, Switzerland) and are listed in Supplementary Table 5.

Recombinant protein expression and purification

Hexahistidine tagged NetF in pET-19b was expressed in E. Coli BL21(DE3). 50–100 ng plasmid was transformed in 50 µl bacteria by heat shock (45 s at 42 °C) and plated on selection plates with 100 µg/ml ampicillin. A single colony was used to inoculate a starting pre-culture of 100 ml LB with ampicillin. The pre-culture was allowed to grow overnight at 37 °C. Cultures for protein expression were started by adding 5 ml of the pre-culture into 500 ml LB with ampicillin and grown at 37 °C until OD600 of 0.6–0.8, when temperature was decreased to 18 °C. Once the culture reached 18 °C protein expression was induced by the addition of 100 µM IPTG. Expression was carried out overnight, followed by centrifugation. Bacterial pellets were resuspended in 50 mM Tris pH 8.0, 500 mM NaCl, 20 mM Imidazole, 2 mM β-mercaptoethanol supplemented with EDTA-free protease inhibitors (Sigma), lysozyme, 5 mM MgCl₂ and benzonase. Cells were broken by three passages through an LM10 microfluidizer (Microfluidics) and the cell lysate was cleared by centrifugation at 25,000 x g for one hour at 4 °C. The supernatant was loaded on a 5 ml HisTrap column (Cytiva, 29000594) and eluted using a 0–500 mM imidazole gradient. Fractions containing NetF were pooled and dialyzed against 20 mM Tris pH 8, 150 mM NaCl, 1 mM Dithiothreitol. To obtain pure NetF monomers, the sample was run on a Superdex 200 Increase 10/300 GL (Cytiva) size exclusion chromatography column. The purity of the sample was checked by SDS-PAGE. The extracellular domain of ANTXR2 (amino acids 36–282) was cloned for expression in Expi293 cells (Thermo Fischer) with a N-terminal secretion signal and a C-terminal hexahistidine tag followed by a FLAG tag. The vector was a kind gift of Sylvia Ho. For transfection Expi293 cells were grown in Expi293 expression medium (Gibco). Cells were seeded at 2×10⁶ cells/ml in fresh medium and allowed to grow overnight at 37 °C, 8% CO₂. 126 µg of plasmid was added to 5 ml Opti-MEM medium (Gibco, 51985-042) and vortexed. Polyethylenimine (Polysciences) was added at final concentration of 75 µg/ml. The mix was incubated at room temperature for 15–20 min and added to the cells. Cells were placed in the incubator and aliquots were removed daily to check protein expression levels and viability. After 72 h, soy hydrolysate 1.6 ml (50x stock), glucose (Millipore, 49159) 0.54 ml (45% stock) and fresh media (17.86 ml) were added as supplements and the cells were grown for another four days before harvesting. The supernatant from 60 ml culture was collected and incubated with 1 ml of anti-FLAG M2 magnetic beads (Sigma) at 4 °C overnight. Subsequently, the beads were washed three times with TBS (50 mM Tris pH 7.5, 50 mM NaCl) followed by three elution’s with 1 ml TBS with 100 µg/ml FLAG peptide (Sigma). The different fractions were inspected by SDS-PAGE and the fractions containing ANTXR2 were pulled and concentrated to 0.3 mg/ml (Supplementary Fig. 8).

Preparation of NetF oligomers

Liposomes and MSP2N2 nanodiscs were prepared as previously described^6,39. Briefly the respective lipids were mixed in chloroform and dried under nitrogen followed by one hour under high vacuum. The lipid film was solubilized in 50 mM Tris pH 7.5, 500 mM NaCl for liposomes and 20 mM Hepes pH 8, 100 mM KCl, 100 mM Na cholate for nanodiscs. Liposomes were extruded through a 100 nm filter (Avanti) followed by incubation with NetF (1:166 molar ratio). Oligomerization on liposomes was checked by SDS-PAGE and the oligomers were solubilized in DDM:CHS 1:0.1%. MSP2N2 was expressed in E. coli and purified on HisTrap column (Cytiva) as described above. Nanodiscs were formed by detergent removal with biobeads (biorad) followed by dialysis against 20 mM Hepes pH 7.5, 100 mM KCl overnight. NetF was added directly to preformed nanodiscs and checked by cryo-EM (1:10 molar ratio).

Quantification of oligomerization from SDS-PAGE

Quantification was performed using Fiji⁴⁰. Briefly, a rectangular box was drawn to enclose the oligomer bands. The same rectangle was then used for each band of all gels making sure that all monomeric and oligomeric signals are considered. The sum of the signals from the oligomeric and monomeric NetF was normalized to 100% in each lane. The violin plots were created in python using seaborn⁴¹ and matplotlib⁴² (Supplementary Fig. 1).

Sample preparation for cryo-EM

All samples for cryo-EM were vitrified using an FEI Vitrobot 4 according to the manufacturer’s instructions. The grids (Quantifoil 2/1 or 1.2/1.3 with 2 nm C) were glow discharged prior to usage. After vitrification, the grids were screened on an FEI Tecnai F20 equipped with a Falcon III detector and Gatan 626 cryo-holder. For high resolution acquisition, the grids were clipped and loaded directly into a FEI Titan Krios G4 equipped with a Falcon 4i detector and Selectris energy filter with a 20 eV slit.

Single particle acquisition, reconstruction, atomic model building and analysis

Data was acquired at ×165,000 magnification corresponding to a pixel size of 0.73 Å2. Automatic data acquisition was set up using EPU automatic mode as detailed in Supplementary Fig. 9 and Supplementary Fig. 10 and saved as eer stacks with a total dose of ~40e^-/Å2. The movies were converted to tiff stacks using relion⁴³ and processed in cryosparc⁴⁴. Briefly, motion correction⁴⁵ and patch CTF⁴⁶ was performed followed by template picker. The automatic picks were visually inspected, the particles were extracted and 2D classified. Following several rounds of 2D classification a subset of the particles was used to generate an initial model followed by homogeneous refinement. In the case of the ANTXR2 complex with NetF, after a 3D refinement procedure, the dataset was subjected to 3D classification, and all particles belonging to classes exhibiting receptor density were subsequently pooled (Supplementary Fig. 8). Symmetry expansion was performed followed by a new round of 3D classification with a focused mask around one NetF protomer and ANTXR2 density. The signal was subtracted from the dataset and local refinement was performed to obtain the map of ANTXR2 bound to a NetF protomer. ModelAngelo⁴⁷ was used for automated model building followed by Phenix⁴⁸ dock, rebuild and real-space refine and manual inspection and corrections in Coot⁴⁹. To calculate the pore diameter and generate the Fig. 2D we used HOLE⁵⁰ and hole-cmm⁵¹ to obtain a ChimeraX-compatible model. In order to better resolve NetF flexible N-terminal region, 3D classification without alignment was performed in cryosparc with a cylindrical mask covering that region of the pore and the pre-pore. Four classes were requested with a filtering resolution of 2 Å.

Interface measurements and salt bridge identification were performed PDBePISA⁵². H-bonds were identified with PDBePISA or with ChimeraX built-in function. Hydrophobic interactions at the interface were identified in ChimeraX by selecting side-chain carbon atoms from hydrophobic residues and computing inter-chain contacts within 4 Å. Details about data collection, refinement, and validation statistics are displayed in Supplementary Table 6

Cell viability assays

NetF-induced cytotoxicity was evaluated using a Resazurin-based viability assay. 50,000 U937 cells in suspension were plated per well of a 96 well plate. Fibroblasts (HFF-1, equine and canine primary fibroblasts) were grown to 100% confluency in a 96 well plate. All other cell lines were grown to 50% confluency in a 96 well plate. All cells were then incubated with NetF for 24 h. Resazurin salt (Sigma, R7017) was diluted in PBS and added to cells at a 0.002% final concentration, incubated for 2 h at 37 °C and fluorescent signal intensity was quantified using an Hidex Sense microplate reader (Hidex Oy; 544/590 nm Ex/Em). Signal intensity of NetF-treated cells was normalized to that of untreated control cells to determine relative viability.

Time-course cytotoxicity assay

Cells were seeded at a density of 1 × 10⁵ cells/mL and grown to approximately 70% confluency in a 96-well plate. At the start of the experiment, the culture medium was replaced with fresh medium containing 10 ng/mL NetF. At each indicated time point incubation at 37 °C, the supernatant was collected, transferred to a new 96-well plate and stored at 4 °C. After the final time point, NetF cytotoxicity was measured by using Lactate dehydrogenase (LDH) assay kit (Promega, G1780) according to manufacturer recommendations. Absorbance at 490 nm was measured using a microplate reader as mentioned above. The percentage of NetF cytotoxicity was calculated using the following formula: % NetF Cytotoxicity = Experimental value—Negative control (media-only)/Triton control x 100. Data were analyzed and plotted with Graphpad Prism 9.5.1. For visualization of cytotoxic effects, cells were seeded at the same density in 12-well plate. After each time point of exposure to 10 ng/mL NetF, cells were washed once with PBS, fixed with 4% PFA/PBS for 10 min at RT, permeabilized with 0.2% Triton X-100/PBS for 10 min at RT and stained with Trypan Blue for 2 min at RT. After staining, cells were washed once with PBS, imaged by light microscopy and photographed using a Nikon DS-Fi3.

RNA-seq

NetF-resistant U937 and J3T subclones (U937-R, J3T-R) were obtained by keeping WT cells under continuous NetF pressure. U937 and J3T cell lines were cloned by limiting dilution to obtain single-cell-derived populations. Single-cell clonal populations were treated with 1 µg/ml NetF for 24 h. Surviving cells were then cultured in medium freshly supplemented with 1 µg/ml NetF for the following 14 days to obtain complete NetF-resistant cell populations. Total RNA of WT and NetF-resistant cells was isolated using a RNeasy Mini Kit (Qiagen, 74104) following the manufacturer’s instruction. RNA purity was assessed by spectrophotometry (NanoDrop) and RNA integrity was assessed with a 2100 BioAnalyzer (Agilent). PolyA selected mRNAs were sequenced at Novogene Cambridge Sequencing Center. Differential expression analysis was performed using the edgeR R package (3.22.5). P values were adjusted using the Benjamini & Hochberg method. Corrected P value of 0.05 and absolute fold change of 2 were set as the threshold for significantly differential expression.

Reverse transcription and RNA quantification

Total RNAs were isolated using Qiagen RNeasy Kit and 1 µg RNA was used for first strand cDNA synthesis using GoScript Reverse Transcriptase (Promega, A5003) with random primers. Real-time PCR was performed using FastStart Universal SYBR Green Master (Roche, 4913850001) in applied biosystems 7500 real-time PCR system. ANTXR2 expression was estimated in susceptible cells relative to resistance cells. GAPDH was used for normalisation of the samples.

Generation of ANTXR2 knockout cell lines

U937 and HFF-1 cells were transduced with lentiviral vectors expressing Cas9 and sgRNAs targeting ANTXR2 (sgRNA1 and sgRNA2; Supplementary Table 4). sgRNAs were cloned into lentiCRISPR v2 and transformed into E. coli Endura cells (LGC Biosearch Technologies, 60242-1). Cells were grown at 30 °C and plasmids were purified and Sanger sequenced with the U6 forward primer (5′-GGGCAGGAAGAGGGCCTAT-3′). Lentivirus was produced in HEK293FT cells by co-transfection of 22.5 μg transfer vector, 20 μg psPAX2, and 2 μg pMD2.G using calcium phosphate precipitation. After 5 min incubation with 50 μL of 2.5 M CaCl₂, the DNA was mixed with 2× HEPES-buffered saline and added dropwise to 2 × 10⁶ HEK293FT cells plated the previous day in 100 mm dishes. Growth medium was replaced 16 h post-transfection. Lentivirus-containing medium was collected at 40 h and 64 h post-transfection, filtered through 0.45 μm cellulose filters (TPP, 99745), and stored at –80 °C. For transduction, cells were seeded in T25 flask at a density yielding ~20% confluency in 4 ml viral supernatant with 8 μg/ml Polybrene (Millipore, TR 1003-G) and incubated for 24 h Puromycin (Gibco, A11138-03) selection was initiated 24 h post-transduction (2 μg/ml for U937, 1 μg/ml for HFF-1) and maintained for 3 days. Knockout efficiency was assessed using the TIDE web tool⁵³. Genomic DNA of each cell line was extracted using the DNeasy Blood & Tissue Kit (Qiagen, 69504), following manufacturers instructions, and target loci were PCR-amplified using primers flanking the sgRNA sites. Amplicons were purified (NucleoSpin Gel and PCR Clean-up, Macherey-Nagel, 740609.50) and Sanger sequenced (Microsynth, Switzerland).

Generation of transgenic HAP1 and HEK293 cells

C-terminally mNeonGreen-tagged constructs were generated by cloning ANTXR2 and TEM8 variants (Supplementary Table 5) into a lentiviral transfer vector under the control of a CMV promoter, with a puromycin resistance cassette for selection. Murine, canine, equine, and bovine ANTXR2 sequences, as well as chimeric and truncated constructs, were derived from human and murine templates using standard cloning procedures. Site-directed mutagenesis was used to introduce point mutations and small region deletions. Primers used for all constructs are listed in Supplementary Table 5. All plasmids were sequence-verified. Lentivirus was produced in HEK293FT cells using a second-generation packaging system and calcium phosphate transfection. Viral supernatants were collected at 40 and 64 h post-transfection, filtered as indicated above, and used to transduce HAP1 and HEK293 cells. Puromycin selection (0.8 μg/ml for HAP1, 1.5 μg/ml for HEK293) was applied for 3 days. Due to low basal expression in HAP1 cells, transduced populations were FACS-sorted to isolate the top 5% mNeonGreen-expressing cells. Monoclonal HAP1 populations expressing selected constructs (Supplementary Table 7) were further isolated by limiting dilution.

Immunoblotting

Cells were lysed in RIPA buffer (10 mM Tris-HCl pH 7.5, 150 mM NaCl, 0.5 mM EDTA, 0.1% SDS, 1 % Triton X-100, 1% Na-Deoxycholate) supplemented with protease inhibitor cocktail (Roche, 11836170001), and incubated for 30 min on ice. Lysates were clarified by centrifugation (14,000 x g, 15 min, 4 °C), mixed with reducing sample buffer to 2% SDS final concentration, and boiled for 5 min at 95 °C. Proteins were resolved by SDS-PAGE and transferred on 0.45 mm nitrocellulose membranes (Thermo Scientific, 88018) using a Trans-Blot Turbo system (Bio-Rad). Membranes were blocked in 2% BSA (Roche, 10735078001)/PBS for 1 h at room temperature and incubated overnight at 4 °C with primary antibodies: rabbit anti-PA (1:2,000), mouse anti-β-tubulin (1:5,000), mouse anti-His-tag (1:1,000), rabbit anti-mNeonGreen (1:2,000), and mouse anti-ANTXR2 (1:2,000). Secondary antibodies swine anti-rabbit HRP and rabbit anti-mouse HRP were used at a 1:5000 dilution. IRDye 800CW donkey anti-mouse IgG was used at a 1:20000 dilution. Western blots with HRP were developed with Pierce™ ECL system (Thermo Scientific, 32209) and blots were scanned using Azure Biosystem gel imager. Further information about the antibodies can be found in the reporting summary.

Co-IP assays

Cells were grown to confluency. For PA binding assays, lysates were diluted in 400 µl dilution buffer (10 mM Tris-HCl pH 7.5, 150 mM NaCl, 0.5 mM EDTA, protease inhibitors). Cells were incubated on ice with 1 µg/ml PA83 and 0.1 µg/ml lethal factor for 30 min. Afterwards, cells were lysed immediately or after 60 min incubation at 37 °C to promote PA83 cleavage into PA63. For ANTXR2-NetF binding assays, cells were incubated with 10 µg/ml NetF for 30 min on ice, followed by 15 min incubation at 37 °C. All cells, including those detached due to NetF-induced toxicity, were collected for lysis.

Cells were washed with ice-cold PBS and lysed in RIPA buffer on ice for 30 min. Lysates were diluted with 400 μl dilution buffer (10 mM Tris-HCl pH 7.5, 150 mM NaCl, 0.5 mM EDTA, protease inhibitors) to reduce detergent concentrations. Clarified lysates (14,000 ×  g, 15 min, 4 °C for PA binding assays; 5000 x g, 10 min, 4 °C for ANTXR2-NetF binding assays) were either used directly for immunoblotting or incubated overnight with anti-mNeonGreen magnetic beads (Chromotek, ntmak-20) according to the manufacturer’s instructions. Beads were washed three times with ice-cold buffer (10 mM Tris-HCl pH 7.5, 150 mM NaCl, 0.05% NP-40 substitute, 0.5 mM EDTA), resuspended in reducing sample buffer adjusted to 2% SDS, and boiled for 5 min at 95 °C. Beads were removed before performing immunoblotting as described above.

Fluorescence microscopy

Cells were seeded in 96-well imaging microplates (Agilent, 204626-100), grown to ~50% confluency and fixed in 4% paraformaldehyde in PBS for 10 min at room temperature. After a PBS wash, cells were permeabilized in 0.2% Triton X-100 in PBS for 10 min, then incubated with 100 nM rhodamine-phalloidin (Cytoskeleton, PHDR1) for 30 min. Following three PBS washes, DNA was stained with 0.1 µg/mL DAPI (Invitrogen) for 5 min. Cells were rinsed once in PBS and stored in PBS at 4 °C until imaging. Images were acquired using a Nikon Ti2 Cicero spinning disc confocal microscope (40X objective). Further image analysis was performed using Fiji⁴⁰.

Surface biotinylation

Biotinylation of cell surface proteins were performed using EZ-Link Sulfo-NHS-Biotin (Thermo Scientific, A39256) according to manufacturer’s recommendations. Briefly, cells were cultured in T25 flasks and the expression of ANTXR2-mNeonGreen was verified under the fluorescence microscope before surface biotinylation. Cells were scraped and lysed in PBS containing 0.5% NP-40 and Protease inhibitor cocktail EDTA-free and immunoprecipitation was performed using Dynabeads Streptavidin Magnetic Beads (Thermo Fisher Scientific, 112005D). Biotinylated proteins were eluted from beads by boiling them in SDS-PAGE loading buffer and immunoblotting was performed with anti-mNeonGreen antibodies.

Quantification and statistical analysis

Statistics were performed using GraphPad Prism 6. Unless otherwise specified, one-way ANOVA followed by Šidák’s multiple comparisons test was performed to assess pairwise group differences, with the family-wise error rate controlled at 5% (α = 0.05).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

H.P. and B.Z. conceptualized the study, supervised the research, and secured funding. C.W., F.C., I.I., and A.N. designed and performed experiments and analyzed the data. F.F. and J.F. performed experiments. L.A. performed experiments under the supervision of F.G.v.d.G. F.C., I.I., and H.P. wrote the initial draft with input from C.W. F.C., I.I., H.P., and B.Z. substantially revised and refined the manuscript. All authors reviewed and approved the final version.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Data availability

The RNA-seq data generated in this study have been deposited in the European Nucleotide Archive under accession PRJEB78819. The cryo-EM maps and their respective models generated in this study have been deposited are available in the worldwide Protein Data Bank and the EM Data Bank under accession 9RSM (NetF prepore model) EMD-54221 (NetF prepore map) 9RSU (NetF pore model) EMD-54226 (NetF pore map) 9RT2 (NetF-ANTXR2 C4 model) 9RT4 (NetF-ANTXR2 local refinement focused model) EMD-54238 (NetF-ANTXR2 C4 map) EMD-54245 (NetF-ANTXR2 local refinement focused map). The AlphaFold predictions used in this study are available as additional data in the supplemental material and supplemental source data file. Source Data are provided as a Source Data file. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-69526-6.