Proteomic analysis of the host–pathogen interface in experimental cholera

Abstract

The microbial cell surface is a site of critical microbe–host interactions that often control infection outcomes. Defining the set of host proteins present at this interface has been challenging. Here we used a surface-biotinylation approach coupled to quantitative mass spectrometry to identify and quantify both bacterial and host proteins present on the surface of diarrheal fluid-derived Vibrio cholerae in an infant rabbit model of cholera. The V. cholerae surface was coated with numerous host proteins, whose abundance were driven by the presence of cholera toxin, including the C-type lectin SP-D. Mice lacking SP-D had enhanced V. cholerae intestinal colonization, and SP-D production shaped both host and pathogen transcriptomes. Additional host proteins (AnxA1, LPO and ZAG) that bound V. cholerae were also found to recognize distinct taxa of the murine intestinal microbiota, suggesting that these host factors may play roles in intestinal homeostasis in addition to host defense.

Interactions between host and microbial factors at the pathogen cell surface can determine infection outcomes. Comprehensive profiling of the bacterial cell-surface proteome or the host proteins that bind the pathogen during infection has been challenging, partly due to the difficulty of obtaining sufficient quantities of in vivo pathogen cells. Recent advances in tandem mass tag-mass spectrometry (TMT-MS), and the development of cell-impermeable protein labeling reagents^1-4, indicate that new approaches to monitor the proteome of the microbial surface in vivo, including at the pathogen–host interface during infection, should be feasible.

The Gram-negative bacterium V. cholerae causes the severe diarrheal disease cholera, which remains an important threat to global public health. The hallmark symptom of cholera is large quantities (up to 20 l per day in severe cases) of watery diarrhea containing up to 10⁹ colony forming units (CFU) ml⁻¹ of V. cholerae⁵. The production of choleric diarrhea is thought to promote the pathogen’s dissemination and subsequent transmission to naïve hosts. In humans, orogastric inoculation of purified cholera toxin (CT) is sufficient to trigger cholera-like diarrhea⁶. CT is an AB₅-type toxin that is secreted by V. cholerae in the small intestine (SI)^7-9, the site of pathogen colonization.

Cholera is restricted to humans, but several animal models have been developed to study V. cholerae intestinal colonization and diarrheal disease¹⁰. The most frequently used animals are suckling mice, but this model yields relatively small amounts of diarrheal fluid. Instead, for larger-scale investigations, suckling rabbits are more commonly used, where orogastric inoculation with V. cholerae leads to robust SI colonization and a disease that closely mimics severe human cholera¹¹. As in humans¹², V. cholerae intestinal colonization in infant rabbits depends on the pathogen surface-exposed toxin-coregulated pilus. Infant rabbits develop large volumes of CT-dependent watery diarrheal fluid, and CT is also sufficient to induce diarrhea in these animals¹¹. In infant rabbits, diarrheal fluid accumulates in the cecum (roughly 0.5–1 ml per animal) before excretion. The fluid contains a high density (10⁹–10¹⁰ CFU) of V. cholerae, and thus provides a relatively pure source of in vivo organisms that has been leveraged for various high-throughput investigations, including RNA-sequencing (RNA-seq) analyses of the pathogen’s in vivo transcriptome¹³ and Tn-Seq analyses of its genetic requirements for in vivo growth^14-16.

Limited analyses of the V. cholerae proteome in vivo^17,18 have been reported. Previous efforts using activity-based protein profiling defined the active serine hydrolases in diarrheal fluid of infant rabbits¹⁹. This study suggested that secreted V. cholerae proteases are capable of degrading both host proteases in diarrheal fluid as well as host intestinal lectins bound to V. cholerae such as intelectin¹⁹. The latter observation raised the possibility that the pathogen is bound and/or targeted by specific host proteins as it traverses the gastrointestinal tract.

Here we used the infant rabbit model of cholera and TMT-MS to define the bulk proteome of choleric diarrhea. Analyses of surface-labeled diarrheal fluid-derived V. cholerae cells enabled identification of both pathogen and host proteins present at this interface. Notably, we discovered that CT accounts for nearly the complete set of >600 host proteins identified in diarrheal fluid. One of the most abundant CT-dependent proteins, surfactant protein D (SP-D), was found to directly bind V. cholerae and function as an SI defense factor. Our screen identified the suite of in vivo V. cholerae cell-surface proteins and also revealed a number of host-derived bacterial-binding proteins (HBBPs) that were not previously known to interact with bacteria. We found that HBBPs not only associate with V. cholerae, but also with specific subsets of murine gut commensal bacteria, suggesting that pathogen-targeting HBBPs may also facilitate intestinal bacterial homeostasis.

Results

CT drives the host proteomic response to V. cholerae.

The chemistry of the diarrheal fluid that accumulates in the cecum of infant rabbits infected with V. cholerae resembles that of human choleric diarrhea¹¹, but its proteomic composition and the pathogen factors that contribute to the diarrheal fluid proteome are not well defined. Infant rabbits were orogastrically inoculated with wild-type (WT) V. cholerae (an isolate from the Haiti 2010 outbreak²⁰), a derivative of the WT strain containing a deletion of ctxAB (V. cholerae Δctx) or purified CT (50 μg), to assess the contribution of CT in stimulating the proteomic response to V. cholerae intestinal colonization (Extended Data Fig. 1a). Consistent with previous reports^21,22, the concentration of WT and V. cholerae Δctx in the diarrheal fluid were similar, but there was a much greater abundance of diarrheal fluid in the animals that were inoculated with WT versus Δctx V. cholerae (Extended Data Fig. 1a-c). There was at least as much cecal fluid recovered from animals inoculated with CT alone as animals inoculated with WT V. cholerae. The small amount of cecal fluid obtained from negative control animals infected with buffer only (mock) was sufficient for proteomic analysis.

High-resolution TMT-MS¹ was then used to quantitatively analyze the proteomic composition of diarrheal fluid isolated from the four groups of rabbits. We identified 664 rabbit proteins with at least two peptides present in the fluid samples (Fig. 1a, Supplementary Data Set 1). Most of the proteins were predicted to be extracellular (Extended Data Fig. 1d), consistent with the idea that V. cholerae does not disrupt the integrity of the intestinal epithelial barrier^7,8 or stimulate the release of cytoplasmic proteins into the intestinal lumen. The protein composition of fluid from animals infected with V. cholerae Δctx was very similar to that of mock-infected animals, suggesting that in the absence of CT, V. cholerae intestinal colonization does little to alter the host intestinal luminal proteome (Fig. 1a,b). Fluid proteomes from animals infected with WT V. cholerae or treated with CT alone were also very similar (Fig. 1a-c), with relative fold changes in individual protein abundance strongly correlated (r = 0.93), (Fig. 1c and Extended Data Fig. 1e). These observations strongly indicate that CT drives not only accumulation of diarrheal fluid but also stimulates the secretion/release of hundreds of host proteins that are found in the cholera-like diarrheal fluid of infant rabbits.

Fig. 1 ∣. The diarrheal fluid proteome in infant rabbits is largely stimulated by CT.

a,b, Hierarchical clustering (a) and principal component (PC) analysis (b) of proteomes identified by TMT-based mass spectrometry. In a, heat map is sorted by log₂ fold differences between animals infected with WT V. cholerae or mock infected. c, Scatterplot showing relative fold changes in the abundance of proteins isolated from rabbits inoculated with either WT V. cholerae (Vc) or purified CT, each relative to the proteomes of mock-infected animals. The red dot indicates SP-D. d, Comparison of the gene sets enrichment based on GO molecular function pathways for the proteomes of rabbits infected with V. cholerae, V. cholerae Δctx or CT-infected rabbit versus control animals. NES, normalized enrichment score. Gene set enrichment was performed using fast GSEA and P values were corrected for multiple testing using the Benjamini–Hochberg procedure (as implemented in fGSEA). Pathways were considered to be significantly enriched if the adjusted P value was less than 0.25.

Pathway enrichment analysis of the cecal fluid proteomes revealed several pathways specifically associated with WT V. cholerae infection and CT treatment (Fig. 1d and Supplementary Data Set 2). These included several Gene Ontology (GO) Biological Process terms linked to immune responses, including ‘immunoglobulin production’ and ‘defense response to bacterium’. WT V. cholerae infection and CT treatment also led to similar reductions in the relative abundances of proteins classified as regulators of proteolysis, raising the possibility that CT modifies host protease activity²³.

Host proteins identified on the pathogen cell surface.

The bulk proteomic analysis of diarrheal fluid reports on the host response to V. cholerae, but does not yield knowledge of the host proteins that directly interact with the pathogen cell surface during infection. To identify these factors, we adapted an unbiased approach that has been used to define how viral or parasitic infection shapes the eukaryotic host cell-surface proteome to instead investigate the V. cholerae cell surface^2,3,24,25. Total surface proteins associated with V. cholerae cells isolated from the cecal fluid of infected rabbits were first washed and then labeled with the cell-impermeable primary amine biotinylation reagent Sulfo-NHS-SS-Biotin (Fig. 2a). After labeling, bacterial outer-membrane fractions were isolated and biotinylated proteins affinity purified. Then, TMT-based mass spectrometry was used to identify and quantify the labeled proteins. The biotinylated fraction contained the known outer-membrane protein OmpU, but not the cytoplasmic RNA polymerase subunit RpoB, suggesting that the bacterial lysis protocol did not lead to substantial cytoplasmic contamination of the biotinylated fractions (Extended Data Fig. 2a-c). Critically, host (rabbit) proteins bound to the V. cholerae cell surface as well as bacterial surface proteins were labeled and detected using this protocol. The surface-labeled V. cholerae proteins encompassed 32% of the total predicted outer-membrane proteome (Fig. 2b). In contrast, only 2.7% of the total predicted V. cholerae cytosolic proteome was labeled, reinforcing the idea that there was minimal cytoplasmic contamination in the labeled samples. CT was one of the surface-labeled proteins, suggesting that some CT remains associated with the cell surface before its release. Notably, 46% of the proteins previously identified in V. cholerae outer-membrane vesicles released during infection¹⁸ were identified in the surface-labeled proteome (Extended Data Fig. 2d), consistent with the idea that outer-membrane vesicles contain a subset of surface-associated proteins.

Fig. 2 ∣. Identification of surface-exposed V. cholerae proteins and V. cholerae-bound host-derived proteins.

a, Schematic of our surface-biotinylation approach. b, Proportion of the V. cholerae proteins identified by surface biotinylation relative to total number of open reading frames encoded in the V. cholerae genome for each predicted localization (extracellular (E), outer membrane (OM), periplasmic (P), inner membrane (IM) or cytoplasmic (C)). c, Hierarchical clustering of the proteins identified by surface biotinylation and the proteins identified in diarrheal fluid in three animals. Host-derived and bacterial proteins were sorted separately by a log₂ fold change before clustering. The dark-green bar represents host-derived proteins and the light-green bar represents bacterial proteins. d, Heat map showing the ratio of abundance of HBBPs identified with surface biotinylation versus their abundance in the diarrheal fluid of the corresponding animal. Proteins mentioned in the text are highlighted in red.

To contextualize the host proteins found in the biotinylated fractions, we also performed TMT-MS analysis on cell-free sterile-filtered fractions of the diarrheal fluid. A total of 564 proteins with at least two detected peptides were identified, including 382 rabbit and 182 bacterial proteins, across all conditions (Supplementary Data Set 3). Hierarchical clustering revealed that the proteins identified in the three surface-labeled samples and the three diarrheal fluid samples clustered together and exhibited distinct proteomic profiles (Fig. 2c). A subset of roughly 50 V. cholera proteins that were more abundant in the fluid samples was identified (Fig. 2c, light-green bar). These proteins included CT, Xds (VC2621), a secreted nuclease involved in immune evasion²⁶ and PrtV, a metalloprotease implicated in V. cholerae virulence²⁷. Several V. cholerae cell-surface-associated virulence factors known to be upregulated in vivo¹³, such as toxin-coregulated pilus components, the outer-membrane heme receptor HutA²⁸ and the accessory colonization factor AcfA²⁹ were included in the surface-labeled samples, providing further biological validation of the data set.

In comparison to the V. cholerae proteins, which were mainly enriched on the surface relative to the fluid, the abundance of most detected rabbit proteins was greater in the fluid samples. However, we identified 36 rabbit proteins that were more abundant in the surface-labeled samples (Fig. 2d). We termed these surface-enriched proteins HBBPs. One of these putative HBBPs was intelectin, a lectin previously found to be physically associated with and targeted for cleavage by V. cholerae during infection¹⁹. Some of the HBBPs (8/36) have known or predicted roles in host defense and inflammation; however, most are thought to function in pathways that are not directly related to host defense and are not known to associate with bacterial cells.

SP-D is an intestinal mucosal defense factor.

One of the most abundant proteins in diarrheal fluid that we also identified as an HBBP (that is, pathogen surface-enriched) was surfactant protein D (SP-D or sftpd) (Fig. 1d and Fig. 2d). SP-D is a C-type lectin conserved across rabbits, mice and humans that has been linked to pulmonary innate immune defense and has been recently reported to function in intestinal homeostasis by impacting the composition of the gut microbiota^30,31. SP-D was around 60-fold enriched in the CT-only and WT V. cholerae infections compared to the mock-infected controls, suggesting its induction is CT-dependent (Fig. 1d, red dot). Immunoblotting confirmed that SP-D was present in diarrheal fluid from infant rabbits infected with WT V. cholerae (Fig. 3a). Consistent with the surface-enrichment biotinylation results, SP-D was also readily detected on washed cecal fluid-derived V. cholerae (Fig. 3b). Immunofluorescence microscopy was used to further investigate the localization of SP-D and V. cholerae in situ during infection. In these experiments, infant rabbits were inoculated with a fluorescently tagged WT V. cholerae strain (V. cholerae-GFP) and SI sections stained with an antibody to SP-D (Extended Data Fig. 3a,b). Both V. cholerae and SP-D localized to the region immediately above the epithelium (Extended Data Fig. 3a). At higher magnifications (Extended Data Fig. 3a insert), SP-D and V. cholerae-GFP appeared to localize within the same matrix. A similar staining pattern was reported for WGA-positive mucin aggregates in V. cholerae-infected infant rabbits¹¹, suggesting that SP-D might colocalize with mucin and V. cholerae during infection.

Fig. 3 ∣. SP-D associates with V. cholerae cells in diarrheal fluid and in vitro.

a, Detection of SP-D in filtered diarrheal fluid by immunoblotting. Diarrheal fluids were collected, filtered and trichloroacetic acid precipitated before western blotting using anti-SP-D antibody. Mock corresponds to V. cholerae cells grown in laboratory. b, Detection of SP-D associated with V. cholerae cells isolated from diarrheal fluid of three rabbits. c,d, Recombinant SP-D associates with V. cholerae cells grown in LB. Immunoblots with anti-SP-D antibody of (from left to right) the purified SP-D protein (c) or SP-D ΔC-ter (C-terminal) (d), flowthrough (unbound protein), washes and bound fraction, and with bacterial cells treated with buffer only (mock) are shown in the last lane. Western blot analyses were performed at least three times with consistent results. e. SP-D aggregates V. cholerae. Bacterial cells were incubated in PBS containing 5 mM CaCl₂ for 1 h in the presence of SP-D (10 μg ml⁻¹), SP-D ΔC-ter (10 μg ml⁻¹) or no added protein and analyzed by light microscopy. Scale bars are 10 μm. Results from an experiment representative of three independent experiments are shown. f, Average expression level (base mean) and log₂ fold change (FC) of transcript abundance in V. cholerae cells incubated with SP-D (10 μg ml⁻¹) or denatured SP-D for 1 h. Genes with significantly different (adjusted P value <0.05) transcript abundance are highlighted in red (Wald test, as implemented in DESeq2). The asterisk * represents the outer-membrane proteins. g, Competitive indices for intestinal colonization for the indicated strains. Suckling mice were inoculated with 1:1 mixture of each V. cholerae mutant and a lacZ-negative derivative of WT V. cholerae. Competitive indices represent the output ratio (mutant strain CFU/lacZ WT strain CFU) divided by the input ratio. Competitive indices were compared to the WT (lacZ⁺ versus lacZ⁻) using a one-way ANOVA (F6,42 = 6.392, P < 0.0001) followed by Sidak’s multiple comparisons test. Data are mean for four independent animals.

We adapted a previously described bacterial whole-cell ‘pull-down’-like assay¹⁹ to test whether purified SP-D could directly bind V. cholerae in the absence of other host factors. In these experiments, V. cholerae cells grown in vitro were incubated with human SP-D (0.125 μg) and then the flowthrough (unbound), washes and bound fraction (lysed cells) were analyzed for the presence of SP-D. We detected strong SP-D signals in the flowthrough and bound fractions, with minimal protein in the wash fractions, demonstrating that SP-D can directly interact with V. cholerae cells in the absence of an intermediary host factor (Fig. 3c). In contrast, a truncated SP-D lacking its C-terminal carbohydrate recognition domain (CRD, SP-D ΔC-ter)^32,33 did not bind V. cholerae, suggesting that the CRD is required for SP-D’s direct interaction with V. cholerae (Fig. 3d).

We next investigated the phenotypic consequences of SP-D binding to V. cholerae. This lectin has been reported to kill E. coli as well as agglutinate the fungal pathogen Pneumocystis carinii (jiroveci)³² and the bacterial pathogen Streptococcus pneumoniae³⁴. There were modest CRD-dependent reductions in V. cholerae viability when the pathogen was incubated with SP-D (Extended Data Fig. 3c). Moreover, incubation with full length, but not SP-D ΔC-ter, led to marked V. cholerae agglutination, further demonstrating that SP-D binding can alter V. cholerae physiology (Fig. 3e and wide field in Extended Data Fig. 3d). RNA-seq of V. cholerae incubated with SP-D versus denatured SP-D, revealed 331 significantly differentially expressed genes (DEGs) (Fig. 3f red dots and Supplementary Data Set 4), uncovering the potent effect that SP-D binding has on V. cholerae’s gene expression profile. The DEGs included several outer-membrane proteins, some of which contribute to intestinal colonization, including the fatty acid transporter fadL²¹, suggesting that modulation of pathogen gene expression by SP-D binding may impair V. cholerae colonization. We tested whether some of the most highly SP-D-upregulated DEGs affect V. cholerae intestinal colonization using competitions assays in infant mice. Colonization of the ΔbreR strain, but not the other tested mutants, was attenuated (Fig. 3g), demonstrating that an SP-D-induced V. cholerae gene product, the bile-responsive transcriptional regulator BreR, can affect infection. SP-D was cleared from the V. cholerae surface within 2 h, suggesting that V. cholerae actively antagonizes SP-D recognition (Extended Data Fig. 3e, black curve). Consistent with this idea, we found that SP-D remained bound to V. cholerae lacking four in vivo secreted proteases (Δquad)—IvaP, VesA, VesB and VCA0812—for at least 6 h (Extended Data Fig. 3e, right side, red curve). Thus, these four proteases, whose activities are induced in the intestine¹⁹, appear to curtail the capacity of SP-D to remain bound to the pathogen.

To directly assess whether SP-D impacts V. cholerae intestinal colonization sftpd^+/+/sftpd^+/− (SP-D⁺) and sftpd^−/− (SP-D⁻) infant mouse littermates from heterozygous breeders (sftpd^+/−) were orally inoculated with WT V. cholerae and bacterial burdens in the proximal and distal SI were determined 18 h after infection. There were significantly higher V. cholerae burdens in proximal, but not distal, SI samples from SP-D⁻ compared to SP-D⁺ mice (Fig. 4a), suggesting that SP-D contributes to subintestinal region-specific host defense. Consistent with our proteomic data showing that CT triggered SP-D production (Fig. 1d), sftpd expression in the intestine was also elevated in infant mice infected with WT V. cholerae but not V. cholerae Δctx (Extended Data Fig. 4a). Furthermore, in the absence of CT, a protective role for SP-D was not observed and the burden of V. cholerae Δctx in the proximal SI was not greater in SP-D⁻ versus SP-D⁺ mice (Extended Data Fig. 4b).

Fig. 4 ∣. SP-D is an intestinal mucosal defense factor.

a, V. cholerae SI colonization in littermate SP-D⁺ and SP-D⁻ mice. Bacterial burdens recovered from proximal and distal SI 18 h after V. cholerae inoculation. Data are mean for 14 and six independent SP-D⁺ and SP-D⁻ animals, respectively. Note SP-D⁺ include both heterozygotes (sftpd^+/−) and homozygous (sftpd^+/+) animals. Statistical significance was determined using a Mann–Whitney U-test. b, Volcano plot showing the log₂ fold change in gene expression and the statistical significance of the differential expression in the proximal SI of WT mice compared to sftpd^−/− mice infected with V. cholerae. Upregulated genes in WT mice are shown in blue, and upregulated genes in sftpd^−/− mice are shown in red. c, Volcano plot showing the log₂ fold change (FC) in gene expression and the statistical significance of the differential expression in the distal SI of WT mice compared to sftpd^−/− mice infected with V. cholerae. d, Gene sets enrichment pathways for the transcriptome of V. cholerae-infected sftpd^−/− mice versus WT-infected animals.

RNA-seq of proximal and distal SI samples from V. cholerae-infected SP-D⁻ and SP-D⁺ mice was also carried out to assess how production of this lectin modifies intestinal responses to infection. In the proximal SI, there were 173 DEGs in SP-D⁺ versus SP-D⁻ mice (Fig. 4b, Extended Data 4c and Supplementary Data Set 4); in stark contrast, no DEGs were observed in the distal SI sample (Fig. 4c and Extended Data Fig. 4c), consistent with the finding that SP-D-mediated protection is limited to the proximal SI. Upregulated genes in the proximal SI of sftpd^−/− mice were significantly enriched for peptidase activities (Fig. 4d), suggesting that one consequence of SP-D absence is an increase in host protease expression.

(lactoperoxidase (LPO), annexin A1 (AnxA1) and zinc-alpha-2-glycoprotein (ZAG) directly bind V. cholerae.

In contrast to SP-D, three additional HBBPs chosen for further study, LPO, AnxA1 and ZAG (or AZGP1), were not known to bind to bacteria. Each of these proteins has been reported to be in the extracellular space and LPO and AnxA1 have been implicated in innate defense^35-37. LPO generates the antimicrobial molecule hypothiocyanite from H₂O₂, and is present on mucosal surfaces including the intestinal epithelium^37,38. AnxA1 is generally considered to be a host cell death marker³⁹, and V. cholerae proteases have been found to modulate its abundance in the intestines of infected rabbits¹⁹. ZAG is a serum/body fluid protein has been associated with diverse nonimmune functions, and exhibits structural similarity to the MHC-I heavy chain^40,41.

We first probed V. cholerae isolated directly from the diarrheal fluid of infected rabbits for the presence of LPO, AnxA1 and ZAG. Each protein was readily detected on the bacterial cells collected from infected animals, suggesting that LPO, AnxA1 and ZAG associate with V. cholerae cells during infection (Fig. 5a-c). Next, we tested whether these proteins directly interact with V. cholerae grown in the laboratory using the binding assay described above (Fig. 5d-f). For all three proteins, a band corresponding to the molecular weight of the respective purified protein was detected in the elution fraction, though the amount of ZAG bound was not as great as the other two proteins. Little or no protein was observed in any of the wash fractions, suggesting that each protein can interact with V. cholerae in the absence of additional host factors (Fig. 5d-f). Apparent cleavage of AnxA1 was detected in both assays, consistent with the previous report that it can be targeted by V. cholerae proteases (Fig. 5b,e, see ref. ¹⁹).

Fig. 5 ∣. LPO, AnxA1 and ZAG interact with V. cholerae.

a–c, Detection of LPO, AnxA1 and ZAG associated with V. cholerae cells in diarrheal fluid. V. cholerae cells collected from diarrheal fluid of infected rabbits were washed twice and lysed. Proteins were separated by 10% acrylamide SDS–PAGE and immunoblots for LPO (a), AnxA1 (b) and ZAG (c) were performed. An asterisk * represents AnxA1 cleavage. d–f, Detection of LPO, AnxA1 and ZAG associated with V. cholerae cells grown in the laboratory. Immunoblot detection of recombinant LPO (amino acids 244–346) (d), AnxA1 (e) and ZAG (f) incubated with V. cholerae cells cultured in LB. From left to right, the purified proteins, flowthrough (unbound protein), washes and bound fraction were analyzed alongside with bacterial cells treated with buffer only (mock). An asterisk * represents AnxA1 cleavage. Western blot analyses were performed at least three times with consistent results. g–h, ZAG and LPO glycan binding to glycan microarrays. Binding of recombinant human ZAG (5 and 50 μg ml⁻¹) (g) and LPO (5 and 50 μg ml⁻¹) (h) to microbial glycan arrays. Data are shown as mean ± s.d. (n = 4 technical replicates) (glycan array data organized by genus are in Extended Data Fig. 5 and the full data set with coefficient of variation in Supplementary Data Set 5). RFU, relative fluorescence units.

We reasoned that the capacity of LPO, AnxA1 and ZAG to bind microbes was not likely to be restricted to V. cholerae and hypothesized that these HBBPs may bind to conserved microbial cell-surface structures such as glycans or phospholipids. To test whether these three proteins bind to microbial glycans, we used previously developed glycan microarrays that contain >300 purified bacterial polysaccharides isolated from a broad range of diverse microbes (but not V. cholerae)⁴². Both ZAG and LPO exhibited dose-dependent binding signals (>500 relative fluorescence units) to different polysaccharides, whereas AnxA1 did not (Fig. 5g-h, Extended Data Fig. 5a-c and Supplementary Data Set 5). LPO bound to polysaccharides from different microbes, including Acetobacter methanolieus, and Klebsiella (Fig. 5g, top five hits). The four different Klebsiella glycans are all enriched in galactose residues and all five consist of predominantly linear rather than branched chains. ZAG bound to Salmonella and Shigella boydii lipopolysaccharide and the capsular polysaccharide from S. pneumoniae 34 (Fig. 5h, top five hits). The ZAG-binding glycans did not exhibit an obvious common feature, suggesting a broader range of recognition. Together, these data indicate that although ZAG and LPO are not canonical lectins, they can directly bind to structurally diverse microbial glycans. Since AnxA1 belongs to the annexin superfamily of calcium-dependent phospholipid-binding proteins, it probably binds noncarbohydrates such as phosphatidylserine, a lipid constituent of the bacterial membrane^36,39. Consistent with this idea, purified AnxA1, which did not interact with any of the glycans on the microarray, strongly bound known annexin phospholipid ligands, including phosphatidylethanolamine, phosphatidic acid and phosphatidylserine (Extended Data Fig. 5d), all of which are found in the V. cholerae membrane⁴³.

HBBPs interact with gut commensal bacteria.

A previous study found that SP-D bound to roughly 2% of fecal bacteria, prompting us to hypothesize that validated HBBPs from our screen might also bind to commensal organisms within the gut microbiota³⁰. We developed a modified ‘IgA-Seq’-like method⁴⁴ to isolate and identify microbes that are bound by LPO, AnxA1 or ZAG in the intestine (Fig. 6a). SYBR Green-positive (viable) and HBBP-coated microorganisms isolated from the feces of specific-pathogen-free (SPF) mice were detected by flow cytometry using biotinylated anti-HBBP antibodies and Cy7-conjugated streptavidin (Fig. 6a and Extended Data Fig. 6a). This protocol revealed that LPO, AnxA1 or ZAG bound 1–10% of the SPF murine fecal microbiota. LPO coated a higher fraction of microbes (6.5% ± 3) than AnxA1 (3% ± 2) and ZAG (2% ± 3) (Fig. 6b,c). Thus, these three HBBPs, like IgA and SP-D, interact with gut symbionts.

Fig. 6 ∣. HBBPs interact with gut commensal bacteria.

a, Schematic of the workflow for detection of HBBP bound to fecal microbiota. b, Flow cytometry of microbiota stained with Streptavidin-PE-Cy7 only, or antibodies to ZAG, LPO or AnxA1. c, Quantification of flow cytometry data from b. Data are mean for 10, 11 and 14 independent experiments for LPO, AnxA1 and ZAG HBBPs-binding experiments, respectively. d, Relative abundance of order or family-specific OTUs after 16S rRNA-seq of sorted cells from a. The bound (positive) and unbound (negative) fraction of the microbiota is shown. Each bar represents the average from four individual mice.

We next used fluorescence-activated cell sorting to sort the HBBP-bound (and hence fluorescently tagged) and unbound bacterial fractions in each sample, and then carried out 16S ribosomal RNA-seq analysis to classify the populations (Fig. 6d). The positive (HBBP-coated) populations for each HBBP formed distinct clusters whereas the negative (unbound) populations all clustered together, suggesting that LPO, AnxA1 and ZAG bind to distinct microbial taxa (Extended Data Fig. 6b,c). Furthermore, operational taxonomic unit (OTU) distributions at the order and family level between coated and uncoated populations differed and coated bacteria were enriched for different OTUs for all three of these proteins (Fig. 6d). In particular, (1) LPO-coated bacteria were abundant in Lactobacillaceae, Turicibacteraceae and Coriobacteriaceae; (2) AnxA1-coated bacteria were highly abundant in Lactobacillaceae and (3) ZAG-coated bacteria were abundant in Lachnospiraceae, Ruminococcaceae and Turicibacteraceae. Higher-resolution analysis at the genus level revealed HBBP-specific enrichment of additional taxa (Extended Data Fig. 6d).

Discussion

Here leveraging the infant rabbit model of cholera enabled analysis of the choleric diarrheal fluid proteome as well as the specific proteomic landscape of the V. cholerae cell-surface–host interface during infection. We discovered that CT, V. cholerae’s signature virulence factor, is almost solely responsible for the pathogen’s impact on the host’s secretion or release of proteins during infection. We charted the protein landscape of both bacterial and host proteins present at the in vivo pathogen surface and identified a class of 36 putative HBBPs. Targeted investigation of SP-D, one of the most abundant proteins in diarrheal fluid and an HBBP, revealed the association of this lectin with the V. cholerae surface and its impedance of V. cholerae colonization in the proximal SI, demonstrating HBBPs can be critical to determining infection outcomes. We found that additional HBBPs (LPO, AnxA1 and ZAG) bound to distinct bacteria in the gut microbiota, suggesting that HBBPs may modulate the composition and function of host-associated microbial communities.

CT is thought to facilitate V. cholerae dissemination and transmission by inducing diarrhea. Our data show that in addition to this activity, CT can both stimulate and impede the secretion/release of host proteins into the intestinal lumen, some of which (for example, SP-D) may control the course of primary V. cholerae infection. Several proteins from protease inhibitor families were less abundant in both WT infection and after CT administration compared to V. cholerae Δctx infection (Fig. 1e). Thus, CT may increase the abundance of intestinal proteases, modifying proteolytic outcomes and thus the proteomic composition of choleric diarrhea. As another function of CT is goblet cell exocytosis and mucin release⁴⁵, some CT-induced host proteins may be goblet cell granule residents. Alongside recent studies that suggest that CT affects the nutrient composition of the intestinal milieu in infected animals²¹, our results collectively reveal an unappreciated and noncanonical role of CT in the modulation of host responses to V. cholerae.

We identified the C-type lectin SP-D as a new CT-induced intestinal mucosal defense factor, impeding V. cholerae colonization of the proximal SI. SP-D binds to l-glycero-d-mannoheptose (Hep), a constituent of the partially conserved lipopolysaccharide inner core of many Gram-negative bacteria, including V. cholerae³³. Along with effects on V. cholerae physiology, SP-D protected against V. cholerae intestinal colonization in infant mice, providing a new role for this lectin that has been linked to pulmonary defense against infection^34,46,47. We found that SP-D production influences gene expression on both sides of the host–pathogen interaction. In addition to host genes linked to innate immunity such as gasdermins, SP-D binding modified expression of V. cholerae genes that themselves influence intestinal colonization such as fadL and breR, offering molecular hypotheses for the increased sensitivity of the SP-D⁻ mice. Protection afforded by SP-D against V. cholerae colonization appeared to be restricted to the proximal SI. We previously observed a similar localized phenotype for V. cholerae colonization in infant mice lacking d-amino acid oxidase⁴⁸, and propose that SP-D and d-amino acid oxidase are representative of a class of region-specific SI mucosal defense factors. While proximal SI-specific expression of SP-D could explain our findings, additional mechanisms underlying regional defense could be involved.

The consequences of CT-dependent host factors on V. cholerae intestinal infection may seem paradoxical. We suggest that V. cholerae’s stimulation of SP-D release, which inhibits pathogen colonization, may represent a modest trade-off compared to the massive net gain that CT provides V. cholerae with respect to transmission. Thus, even if CT triggers host defenses such as SP-D, the pathogen still benefits overall from the toxin’s presence. Finally, our data indicate that V. cholerae’s suite of in vivo-produced proteases can degrade bound SP-D, suggesting that pathogens may be able to actively and perhaps specifically antagonize HBBP recognition and function. Investigation of the broader dependence of HBBPs on CT production will reveal the extent of this phenomenon.

Our surface-labeling approach revealed the in vivo V. cholerae surface proteome as well as host proteins bound to the pathogen’s surface. Among the most abundant bacterial surface proteins during infection were TcpA and a methyl-accepting chemotaxis protein (VCA0176), two known V. cholerae antigens⁴⁹, suggesting that our approach can uncover useful antigenic targets. We identified a class of host proteins (HBBPs) that were more abundant on the V. cholerae surface than in diarrheal fluid, including intelectin, a known V. cholerae-targeting HBBP whose function remains elusive^19,50. Only around 25% of the identified HBBPs from our screen have previously been linked to host defense/inflammation. Most of the other HBBPs were classified as enzymes or linked to metabolism, raising the possibility that these factors might impact in vivo V. cholerae physiology by modifying the surface-proximal milieu.

Three additional validated V. cholerae-binding HBBPs—AnxA1, LPO and ZAG—also bound 1–10% of fecal microbiota, a similar range as reported for IgA and SP-D^30,44. ZAG and LPO bound to specific and distinct microbial glycans and AnxA1 to specific phospholipids (Extended Data Fig. 5). Each of these three proteins bound to distinct microbial taxa in vivo, suggesting that these and other HBBPs play a role in host microbial surveillance and could modify the composition and/or function of the intestinal microbiome. These interactions may not necessarily be antagonistic and could have important consequences on host physiology. For example, mice deficient in SP-D have distinct gut microbiota and immune profiles³⁰. Ultimately, defining the proteomic composition of the microbe–host interface will deepen our understanding of interkingdom interactions that underlie homeostasis and disease, and offer new targets for therapeutic interventions.

Methods

Ethics statement.

Animal experiments were conducted according to protocols approved by the Brigham and Women’s Hospital Committee on Animals (Institutional Animal Care and Use Committee protocol number 2016N000334 and Animal Welfare Assurance of Compliance number A4752-01) and in accordance with recommendations in the National Institute of Health’s Guide for the Care and Use of Laboratory Animals and the Animal Welfare Act of the United States Department of Agriculture.

Bacterial strains and growth condition.

V. cholerae strain H1, a clinical isolate from 2010 and its Δctx derivative (Supplementary Data Set 6)²⁰ were cultured in Luria-Bertani (LB) medium or on LB agar plates at 37 °C unless otherwise stated, supplemented with streptomycin at a concentration of 200 μg ml⁻¹. V. cholerae cells carrying the pUA-GFP plasmid, which contains a green fluorescent protein (GFP) gene under strong constitutive promoter) was used for immunostaining of infected infant rabbit SI and cultured overnight at 30 °C in LB supplemented with streptomycin (200 μg ml⁻¹) and kanamycin (50 μg ml⁻¹).

Infant rabbit infection studies.

For inocula preparation, overnight bacterial cultures were diluted 1:100 in 50 ml of LB and cultured with aeration at 37 °C until an optical density (OD₆₀₀) of 0.5–0.9 was achieved. roughly 2 × 10¹⁰ CFU were pelleted by centrifugation at 5,000g for 5 min, the supernatant was removed and cell pellets were resuspended in 10 ml of 2.5% sodium bicarbonate solution (2.5 g in 100 ml water, pH 9.0) to a final cell density of around 2 × 10⁹ CFU ml⁻¹. Serial dilutions of the inoculum were plated to enumerate the inoculum dose. Infant rabbit infections were performed as previously described¹¹. Briefly, 2-day-old litters of mixed gender New Zealand White rabbit were cohoused with a lactating dam (Charles River) for the duration of the experiment. Each infant rabbit was orogastrically inoculated with 500 μl of the inoculum, using a size 4 French catheter. Following inoculation, the infant rabbits were monitored at least twice a day for signs of illness and euthanized around 16–18 h postinfection. For purified CT experiments, 50 μg CT (Sigma, C8052) was used per rabbit (500 μl of a 100 μg ml⁻¹ solution in sodium bicarbonate). Animals infected with CT were euthanized 3–6 h postinoculation.

Mice colonization assay.

C57BL/6 Sftpd^−/− mice were purchased from Jackson laboratory and were bred at the Harvard Institutes of Medicine animal facility. Littermates that were the offspring of heterozygous Sftpd^+/− breeders were used in this study. Infant mice were genotyped postmortem at the end of the colonization assay. Intestinal colonization in infant mice was conducted as described⁵¹. Briefly, bacterial cells were grown overnight at 30 °C and then diluted 1:1,000 in LB. Infant mice were orogastrically inoculated with 50 μl (roughly 10⁵ CFU) and then killed after roughly 18 h. SIs were equally divided into proximal and distal segments. Dilutions of SIs homogenates were plated on LB agar plates supplemented with 200 μg ml⁻¹ streptomycin to enumerate CFU. Statistical significance was determined using a Mann–Whitney U-test. Infant mice were genotyped postmortem at the end of the colonization assay using tail chips and PCR according to Jackson laboratory protocol using primers 24516 (TGT TGA TGC ATG TTA TGT GAT GA), 24517 (CCT AGG GAA GGC TAG GGA GT) and oIMR2088 (AGA CTG CCT TGG GAA AAG CG).

Immunofluorescence microscopy.

Immunofluorescence images were analyzed from six rabbits infected with V. cholerae-GFP; two or three sections of the SI per rabbit were examined. Briefly, tissue samples used for immunofluorescence were fixed in 4% PFA for 2 h, and subsequently stored in 30% sucrose before embedding in a 1:2.5 mixture of OCT (Tissue-Tek) and stored at −80 °C, as previously described⁵². Frozen sections were then cut at a thickness of 10–15 μm using a cryotome (catalog no. CM1860UV; Leica). Sections were first blocked with 5% bovine serum albumin (BSA) in PBS for 1 h and then stained overnight at 4 °C with a primary anti-SP-D antibody (1:500, R&D Systems, AF1920), diluted in PBS with 0.5% BSA and 0.5% Triton X-100, anti-GFP labeled with Alexa 488 (1/1,000, SAB4600051). After washing three times with 1× PBS containing 0.5% Triton X-100, sections were incubated with Alexa Fluor 647 phalloidin (1/1,000; Invitrogen) and anti-Goat Alexa Fluor 568 (1/1,000, ThermoFisher, A-11055) for 1 h at room temperature, washed and stained for 5 min with 4′,6-diamidino-2-phenylindole at 2 μg ml⁻¹ for 10 min and covered with ProLong Diamond mounting medium. Following staining, slides were imaged using a Nikon Ti Eclipse equipped with a metal-oxide-semiconductor (sCMOS) camera (Andor Zyla) for wide-field microscopy.

Preparation of diarrheal fluid for MS analysis and immunoblotting.

Diarrheal fluids were filtered through sterile polyester membranes with a pore size of 0.22 μm before precipitation with trichloroacetic acid 15%, 45 min on ice. Precipitated proteins were washed once in acetone and resuspended in 1× blue loading buffer (NEB, B7703S).

For immunoblotting, bacterial pellets or precipitated proteins were resuspend in blue loading buffer (NEB, B7703S), boiled at 95 °C for 10 min and loaded on 10% gels (BioRad) for electrophoresis. Proteins were transferred from the gel to nitrocellulose membranes and immunoblotted. Antibodies for western blot assays were used at the following concentrations: anti-SP-D (1:2,000, R&D Systems, AF1920), anti-LPO (1:2,000, LSBio, LS-C25068), anti-AnxA1 (1:500, ThermoFisher, 71-3400), anti-ZAG (1:2,000, ThermoFisher, H00000563-B01P), anti-RNA Polymerase (1:2,000, Biolegend, 663903) and anti-OmpU (1:500, homemade, gift from the Mekalanos laboratory). The membranes were developed with SuperSignal West Femto maximum-sensitivity substrate (ThermoFisher) and visualized with a ChemiDoc Scientific imaging system (BioRad).

Peptide labeling with tandem nass tags and MS.

For the diarrheal fluid proteome identification experiment (Fig. 1), two replicates of V. cholerae and of V. cholerae Δctx and three replicates of CT and of mock-infected fluid were used. For the identification of the surface-exposed V. cholerae proteome (Fig. 2), three replicates of surface-biotinylated proteins and three replicates of the total fluid proteome were used. Samples were submitted in 1× blue loading buffer (NEB, B7703S) to the ThermoFisher Center for Multiplexed Proteomics at Harvard Medical School (Boston, MA, USA) for isobaric TMT-based quantitative proteomics. Briefly, after adjusting proteins to equal concentrations, 40 μl of each sample was loaded on 10% Bis/Tris gels and run at 120 V for 10 min in MES buffer. Gels were then stained with Coomassie for 2 h and destained overnight with water at room temperature. Gel bands were then cut out and destained overnight at 4 °C. Gel bands were reduced with dithiothreitol and alkylated with iodoacetamide and in-gel digests were performed overnight with trypsin. Peptides were eluded, dried, cleaned by stage tip and labeled with TMT reagents. Labeling reactions were combined, cleaned and dried down. Peptides were resuspended in 5% acetonitrile, 5% formic acid and one-third of the sample was shot on an Orbitrap Fusion Mass spectrometer. Peptides were detected (MS1) and quantified (MS3) in the Orbitrap Fusion Mass spectrometer. Peptides were sequenced (MS2) in the ion trap. MS2 spectra were searched using the SEQUEST algorithm against a Uniprot composite database derived from the combined V. cholerae and Oryctolagus cuniculus (rabbit) proteomes containing its reversed complement and known contaminants. Peptide spectral matches were filtered to a 1% false discovery rate using the target-decoy strategy combined with linear discriminant analysis. Proteins were quantified only from peptides with a summed signal to noise threshold of ≥200 and MS2 isolation specificity of 0.5. Mass spectrometry data files have been deposited to the ProteomeXchange Consortium via PRIDE⁵³ with the data set identifiers PXD027076 and 10.6019/PXD027076.

Gene set enrichment analysis.

The G:Profiler version e104_eg51_p15_3922dba (http://biit.cs.ut.ee/gprofiler/) webtool was used for finding enriched GO cellular component terms in the rabbit intestinal proteome. A score above 1.8 for negative log of adjusted P values was considered significant. Gene set enrichment was performed as previously described⁵⁴ using fast GSEA (fGSEA) in R (v.1.8.0) and RStudio (v.1.3.1093)⁵⁵ with modifications. Only genes with annotation were considered. The normalized mean proportion for each protein was divided by the value of that protein in the uninfected data set and log₂ transformed to create a fold change. These log₂ fold change values were use as the ‘rank’ for fGSEA.

Hierarchical clustering.

The signal to noise of each protein was first normalized by calculating the proportion of the total signal represented by that protein in each sample. Clustering was then performed on these values in R using heatmap.2 with the default Pearson correlation method.

In vitro protein-V. cholerae-binding assay.

Binding assays were carried out as previously described¹⁹. Briefly, bacteria were grown to OD of roughly 0.4 in LB and then centrifuged (5,000g, 5 min at room temperature). Bacterial pellets were washed twice in 25 ml HEPES-buffered saline (140 mM NaCl, 1.5 mM Na₂HPO₄, 50 mM HEPES, pH 7.5) supplemented with 5 mM CaCl₂. Bacterial cells were then incubated with 0.25 μg of purified human SP-D (R&D Systems, 1920-SP-050), ZAG (R&D Systems, 4764-ZA-050), AnxA1 (R&D Systems, 3770-AN-050) or LPO (MyBiosource, MBS954610) for 30 min at room temperature and washed twice with an equal volume of buffer. Bacterial pellets were then resuspended in 1× blue loading buffer (NEB, B7703S). Unbound input and the two washes were treated with 4× blue loading buffer (NEB, B7703S) and incubated at 95 °C for 10 min before SDS–PAGE and immunoblot analysis. All binding experiments were repeated at least three times with consistent results.

Bacterial aggregation assay.

V. cholerae cells were grown to an OD of roughly 0.4 in LB, centrifuged (5,000g, 5 min at room temperature) and then resuspended and washed in PBS supplemented with 5 mM CaCl₂. Bacterial suspensions were incubated with human SP-D at a concentration of 10 μg ml⁻¹ (R&D Systems, 1920-SP-050) or human SP-D ΔC-ter at a concentration of 10 μg ml⁻¹ (EzBiolab) for 1 h at room temperature without agitation and observed by light microscopy for agglutination using a Nikon Ti Eclipse equipped with a metal-oxide-semiconductor (sCMOS) camera (Andor Zyla). Figures were made using Fiji software (v.2.1.0/1.53c).

Pathogen surface-enrichment proteomics.

Cecal fluid was harvested 16–18 h after inoculation of infant rabbits. The fluid was then filtered through a 5 μM filter to remove particulate matter and eukaryotic cells. Bacteria were isolated from diarrheal fluid by centrifugation (5000g, 5 min at room temperature). Bacterial pellets were washed twice and resuspended in PBS supplemented with 1 mM CaCl₂, 0.5 mM MgCl₂ and 1.5 mM d-biotin at room temperature. Cell-surface biotinylation was performed as described⁵⁶ with modifications. Sulfo-NHS-LC-biotin (ThermoFisher, 21335) was added to a final concentration of 200 μM for 20 min at room temperature. The reaction was stopped by addition of 2 volumes of buffer (80 mM Tris pH 7, 100 mM NaCl, 30 mM KCl, 1 mM CaCl₂ and 0.5 mM MgCl₂). After washing the bacterial cells three times with the same buffer they were resuspended in 50 mM Tris pH 7, 50 mM NaCl, 10 mM MgCl₂, DNase (0.1 mg ml⁻¹), lysozyme (0.1 mg ml⁻¹) and Complete protease inhibitor mixture (Roche). Cells were broken using an Emulsiflex-C3 (Avestin) and the crude membrane fraction was isolated by ultracentrifugation at 45,000g for 45 min. Membrane-containing fractions were washed twice in 50 mM Tris pH 7, 150 mM KCl, 10 mM EDTA and Complete protease inhibitor mixture (Roche). Membranes were then solubilized overnight at 4 °C in presence of 0.5% n-dodecyl β-d-maltoside (Sigma, D4641). Lysates were used for coimmunoprecipitation using Dynabeads M-280 Streptavidin (ThermoFisher, 11205D) overnight at 4 °C. Magnetic beads were washed three times with 1 ml of Tris pH 7, 100 mM NaCl and 0.2% Tween 20 and resuspended in 50 μl of 1× blue loading buffer (NEB, B7703S) and heated for 10 min at 96 °C.

HBBP binding to glycan arrays.

Human ZAG (R&D Systems, 4764-ZA-050), human LPO (MyBiosource, MBS954610) and human AnxA1 (R&D Systems, 3770-AN-050) were provided to the Protein-Glycan Interaction Resource at the National Center for Functional Glycomics (Beth Israel Deaconess Hospital, Boston) for hybridization to the microbial glycan microarray. The microbial glycan microarrays were prepared as previously described⁴². The printed array includes polysaccharides derived from 313 different bacteria printed at 500 μg ml⁻¹, in replicates of six. To interrogate the arrays, ZAG, LPO and AnxA1 were diluted to 5 and 50 μg ml⁻¹ in binding buffer (20 mM Tris-HCl, pH 7.4, 150 mM NaCl, 2 mM CaCl₂, 2 mM magnesium chloride (MgCl₂), 1% BSA and 0.05% Tween 20) and applied directly to the array surface for 1 h. After incubation, the array was washed by soaking with binding buffer four times. ZAG was detected with anti-ZAG (2 μg ml⁻¹, AZGP1 antibody, H00000563-B01P MaxPab) and antimouse IgG-Alexa-488 (5 μg ml⁻¹) diluted in binding buffer, applied directly to the array surface and allowed to incubate for 1 h. Similarly, LPO was detected with anti-LPO (2 μg ml⁻¹, ThermoFisher PA5-18917) and anti-goat IgG-Alexa-488 (5 μg ml⁻¹) diluted in binding buffer, and then applied directly to the array surface for 1 h. Similarly, AnxA1 was detected with Anti-AnxA1 (2 μg ml⁻¹, Sigma, AMAB90558) and antimouse IgG-Alexa-488 (5 μg ml⁻¹). The arrays were washed in binding buffer (four times), binding buffer without BSA and Tween 20 (four times) followed by deionized water (four times) and scanned. The high and low fluorescence values from the six replicates were eliminated and the remaining four values were averaged. Data were plotted with Excel (Microsoft) as average relative fluorescence units versus print identification number. The top five HBBP-glycans interactions for ZAG and LPO showed in Fig. 5g,h were defined as threefold over background and exhibiting a dose-responsive binding; AnxA1 binding did not meet these criteria.

Phospholipid-binding assay.

Membrane strips containing spots of equal amounts (100 pmol) of different phospholipids (PIP-strips; Echelon Biosciences) were used as recommended by the manufacturer. Briefly, strips were blocked with 3% BSA and incubated with AnxA1 (5 μg ml⁻¹) protein overnight at 4 °C. Bound AnxA1 was detected by anti-AnxA1 (2 μg ml⁻¹, Sigma, AMAB90558) and visualized by secondary antibody coupled to peroxidase followed by chemiluminescence detection. Protein band intensities were measured using Fiji and normalized to the blank spot (no lipid spotted). Experiments were performed in duplicate with consistent results.

Analysis of HBBP-microbiota binding.

Fecal pellets from SPF C57BL/6 mice were collected and directly resuspended in phosphate-buffered saline (PBS) (100 mg of feces in 100 μl) supplemented with 1% BSA and 1 mM CaCl₂, and filtered with a 40-μm cell strainer to remove particulate matter. Bacterial suspensions were centrifuged (5,000g, 5 min) and washed twice in the same buffer. Then 20 μl of bacterial suspension was incubated with 2 μg of biotinylated anti-LPO (LSBio, LS-C684314), anti-ZAG (R&D Systems, BAF4764) or anti-AnxA1 (LSBio, LS-C317217) for 30 min on ice. After washing three times and resuspending in PBS, then supplementing with 1% BSA and 1 mM CaCl₂, bacteria were incubated with 1 μg of Streptavidin-PE-Cy7 (ThermoFisher, SA1012) for 15 min on ice. After washing, bacterial genomic DNA was stained with 1/10,000 dilution of SYBR Green followed by two washes. Bacterial suspensions were then analyzed by flow cytometry (Sony, SH800) and HBBP-positive or HBBP-negative population were sorted.

16S rRNA gene sequencing and analysis.

The PureLink Microbiome DNA Purification Kit (ThermoFisher, A29790) was used according to the manufacturer’s protocol to extract the DNA from the sorted microbiota. The 16S rRNA amplification were done as previously described⁴⁸. Briefly, the V1–V2 region of 16S rRNA was PCR amplified (12.5 ng purified DNA per reaction; Phusion polymerase, New England Biolab) for 25 cycles (95 °C for 30 s, 50 °C for 30 s and 72 °C for 30 s) (primer pair 27Fmod/338R, ref. ⁴⁸). PCR products were then purified (MinElute, QIAGEN) and resuspended in 25 μl of 10 mM Tris-HCl pH 8.5. The V1–V2 PCR products were indexed with the Nextera XT Index kit (Illumina) by PCR (2.5 μl PCR product, Nextera XT Index primers, Phusion polymerase) for eight cycles (95 °C for 30 s, 55 °C for 30 s and 72 °C for 30 s). The 16S rRNA amplicons with indices were purified (MinElute, QIAGEN), resuspended in 25 μl of 10 mM Tris-HCl pH 8.5, quantified with a Qubit v.2.0 Fluorometer (Life Technologies), pooled at a concentration of 4 nM, denatured, diluted to a final concentration of 4 pM and sequenced using the MiSeq Reagent Kit v.3 (600-cycle, paired-end, Illumina) on a MiSeq sequencer (Illumina). Sequencing reads were demultiplexed using MiSeq Reporter v.2.0 and further processed using QIIME2 (ref. ⁵⁷). Briefly, paired-end reads (FASTQ files) were merged with FastqJoin and quality filtered with a Q score cutoff of 20. Merged sequencing reads were denoised using DADA2 (ref. ⁵⁸). Taxonomic classification was generated using a pretrained naïve Bayes classifier based on the bacterial 16S rRNA Greengenes reference database and QIIME2 v.qiime2-2020.8 (https://qiime2.org). Differential relative abundance of individual taxa at the genus level was determined using the ALDEx2 plugin in QIIME2 (ref. ⁵⁹).

Real-time quantitative PCR (qPCR).

RNA was extracted using the RNeasy Plus Mini Kit (Qiagen, 74134) and complementary DNA was generated from 2 μg RNA using a High-Capacity cDNA Reverse Transcription kit (ThermoFisher, 4368814). Quantitative real-time PCR was performed using a Step One Plus Real-Time PCR machine using Fast SYBR Green Master Mix kit (Life Technologies) and probes for SFTPD (Supplementary Table 6) and GAPDH (Supplementary Table 6). Undiluted cDNA was used in the qPCR reactions. Expression levels were calculated using the ΔΔC_T method normalized to GAPDH.

Bacterial permeability.

Bacterial membrane integrity was assessed by differential staining with the permeant fluorescent probe SYTO 9 and the impermeant fluorescent probe Propidium Iodide (Live/Dead kit BacLight Termo Fisher L7007). Fluorescence was measured using a Nikon Ti Eclipse equipped with a metal-oxide-semiconductor (sCMOS) camera (Andor Zyla) for wide-field microscopy.

Bacterial RNA-seq.

Total RNA was extracted with TRIzol (Life Technologies) according to the manufacturer’s instructions. Library preparation and messenger RNA-seq was performed by the Microbial Genome Sequencing Center (Pittsburgh PA). Sequencing reads and genome resources were uploaded to the Galaxy web platform, which was used to process and map reads. Trimmed reads were mapped to the KW3 V. cholerae reference genome (National Center for Biotechnology Information, NCBI) using RNA STAR (Galaxy Tool v.2.6.0b-1). featureCounts (Galaxy Tool v.1.6.4+ galaxy1) was used to build a count matrix from mapped reads. The count matrix was analyzed using DESeq2 (v.1.22.2) in R to compare the abundance of mapped transcripts between different sample groups and identify DEGs. Parametric dispersion was used, and the required shrinkage of effect size was performed using the apeglm package in R. DEGs with adjusted P values of less than 0.05 were considered to be differentially expressed. Transcriptomic data files have been deposited to the NCBI Gene Expression Omnibus (GEO) repository⁶⁰ with the identifier GSE179530.

Host mRNA-seq.

RNA extraction and mRNA-seq library preparation and sequencing were performed as previously described⁵⁴. Briefly, RNA was extracted from Trizol using the RNeasy mini kit from Qiagen with some modifications. Trizol samples were incubated at 65 °C until just thawed (5–10 min), then 200 μl of chloroform was added to each 1-ml sample tube. The tubes were inverted ten times for mixing and incubated at room temperature for 3 min. The samples were spun at 12,000g for 15 min at 4 °C, to separate the aqueous and organic layers. The clear aqueous phase was removed, an equal volume of 70% ethanol was added, and the sample was mixed by inversion ten times before incubating at room temperature for 5 min. The entire volume was transferred to an RNeasy spin column and spun for 1 min. The samples were washed with prewash buffer twice, with wash buffer once, spun empty to remove residual buffer twice and eluted with 50 μl RNase free water. Total RNA was assessed for quality and integrity (RINe) using a High Sensitivity RNA ScreenTape (Agilent) at the HMS Biopolymers Facility. RNA of high quality (RINe >8) was prepared for mRNA-seq using the KAPA mRNA HyperPrep kit (Roche). Libraries were quantified using a High Sensitivity D1000 ScreenTape (Agilent) and High Sensitivity Qubit. Libraries were sequenced on a NextSeq 550 run with paired-end 150 bp reads. Sequencing reads and genome resources (GRCm38) were processed on the Harvard Medical School O2 cluster. First, reads were trimmed using Trim Galore with automatic adapter sequence detection. Then, trimmed reads were mapped to the mouse reference genome and annotation using RNA STAR. Feature Counts was used to build a count matrix from mapped reads using the Ensembl annotation as a guide. Count matrices were exported from O2 and imported into R (v.1.4.1717). The non-normalized count matrix was also analyzed using DESeq2 (v.1.30.1) to compare the abundance of transcripts between different inoculum types to identify DEGs. Parametric dispersion was used, and shrinkage of effect size was performed using the package apeglm. Genes with an adjusted P value of less than 0.05 were considered to be differentially expressed. Normalized read counts were generated using a regularized log (rlog) transformation and used to perform principal component analysis of each sample. Raw reads and count matrices have been deposited into the GEO repository⁶⁰ with the identifier GSE179530.

Extended Data

Extended Data Fig. 1 ∣. Diarrheal fluid volume and composition.

(a) Schematic of the experimental protocol for identification of the proteome in diarrheal fluid isolated from rabbits inoculated with V. cholerae, V. cholerae Δctx, purified cholera toxin (CT) or buffer alone (Mock). (b) Bacterial burdens recovered from diarrheal fluid harvested from rabbits infected with V. cholerae and V. cholerae Δctx. Data are mean for 3 independent animals per group. (c) Diarrheal fluid volumes collected from rabbits infected with V. cholerae, V. cholerae Δctx, purified cholera toxin (CT) and buffer (Mock). Data are mean for 3 independent animals for V. cholerae, V. cholerae Δctx and mock, and 4 independent animals for CT-infected animals. (d) Predicted localization of rabbit proteins identified in diarrheal fluid. Bioinformatic analysis was performed using the G:Profiler (http://biit.cs.ut.ee/gprofiler/) webtool. (e) Scatterplot of relative fold changes in protein abundances isolated from rabbit infected with V. cholerae Δctx (Delta) compared to wild-type V. cholerae (Vc), each relative to the proteomes of mock infected animals. The red dot indicates SP-D.

Extended Data Fig. 2 ∣. Validation of the surface biotinylation assay.

(a) Controls validating the surface-biotinylation screen. (a) Proteins isolated following surface-biotinylation protocol with (+) or without (−) the biotinylation step were separated by 10% acrylamide SDS–PAGE and silver-stained. (b) Presence of cytoplasmic RNA polymerase β and (c) outer-membrane OmpU were assessed by immunostaining with anti-RNApol and anti-OmpU antibodies, respectively. T: total V. cholerae lysate. Western blot analyses were performed at least three times with consistent results. (d) Venn diagram showing the comparison of V. cholerae proteins identified with our surface-biotinylation screen and V. cholerae outer membrane vesicles (OMV’s) proteomes from (34); 181 and 110 are the total number of proteins from each group.

Extended Data Fig. 3 ∣. SP-D associates with V. cholerae cells.

(a,b) Immunofluorescence micrographs of rabbit small intestines inoculated with V. cholerae-GFP. Bacterial cells were detected by GFP fluorescence, SP-D was detected with a goat anti-SP-D antibody followed by anti-goat antibody coupled to Alexa fluor 468. Phalloidin (for actin labeling) is stained with an antibody coupled to Alexa fluor 647 and DAPI (for DNA labeling) is shown is blue. (b) Only anti-goat antibody coupled to Alexa fluor 468 was used to assess non-specific staining of the second antibody. Scale bar is 100 μm. (c) V. cholerae cells were incubated in PBS containing 5 mM CaCl2 for 1 hour at 37 °C in the presence of SP-D (10 μg/ml), SP-D ΔC-ter (10 μg/ml) or a denatured SP-D and then incubated with fluorescent nucleic acid stains SYTO 9 and Propidium Iodide to assess viability. Data are mean ± s.d for three biological replicates. (d) Wide field of micrographs shown in Fig. 3e are represented by the dotted white square. Additional fields are also shown in rows #2 and #3. Scale bar is 10 μm. (e) Degradation of SP-D over time upon incubation with V. cholerae C6706 or the protease deficient strain V. cholerae C6706 Δquad. Graph shows SP-D protein band intensity normalized to RNApolβ over time. Western blot analyses were performed at least three times with consistent results.

Extended Data Fig. 4 ∣. Transcriptomic analysis.

(a) mRNA levels of the SFTPD gene measured by qRT-PCR and normalized to GAPDH mRNA levels in arbritrary unit (A.U) in mice infected with V. cholerae, V. cholerae Δctx or mock. Data are shown as mean ± s.d. (n = 4 technical replicates). (b) V. cholerae Δctx small intestinal colonization in littermate sftpd−/− and sftpd +/+ mice. Bacterial burdens recovered from proximal and distal small intestine 18 hrs after V. cholerae Δctx inoculation. Data are mean for 12 and 7 independent SP-D+ and SP-D− animals respectively. Note S-PD + include both heterozygotes (sftpd +/−) and homozygous (sftpd +/+) animals. Statistical significance was determined using a Mann-Whitney U t test. (c) Principal Component Analysis (PCA) plot of RNA-seq data from four biological replicates of WT or sftpd−/− mice proximal small intestine infected with V. cholerae. (d) Principal Component Analysis (PCA) plot of RNA-seq data from four biological replicates of WT or sftpd−/− mice distal small intestine infected with V. cholerae. (e) Heat map of rlog-transformed read counts for 4 animal replicates (WT or sftpd−/− infected with V. cholerae) for top 30 and bottom 30 genes by rank.

Extended Data Fig. 5 ∣. LPO, AnxA1 and ZAG binding to microbial glycans.

(a-c) Results of ZAG (a), LPO (b) and AnxA1 (c) binding to Microbial Glycan Microarray organized by genus and species (red is 5 and blue is 50 μg ml⁻¹). Data are presented as the mean ± s.d. (n = 4 of a technical replicate for each immobilized glycan). Note: scales on Y axes are different. The complete datasets are available in Supplementary Data Set 4. (d) AnxA1-lipid interaction assessed by protein-lipid overlay assays. Graph shows quantification of the Annexin A1 protein band intensity normalized to the intensity of a blank spot band. Lysophosphatidic acid (LPA), Lysophosphocholine (LPC), Phosphatidylinositol (PtdIns), Phosphatidylinositol(3)-phosphate (PtdIns(3)P), Phosphatidylinositol (4)-phosphate (PtdIns(4)P), Phosphatidylinositol (5)-phosphate (PtdIns(5)P), Phosphatidylethanolamine (PE), Phosphatidylcholine (PC), Sphingosine-1-Phosphate (S1P), Phosphatidylinositol(3,4)-bisphosphate (PtdIns(3,4)P2), Phosphatidylinositol (3,5)-bisphosphate (PtdIns(3,5) P2), Phosphatidylinositol(4,5)-bisphosphate (PtdIns(4,5)P2), Phosphatidylinositol (3,4,5)-trisphosphate (PtdIns(3,4,5)P3), Phosphatidic acid (PA), Phosphatidylserine (PS).

Extended Data Fig. 6 ∣. Microbiota-bound 16S analysis.

(a) Representative gating strategy illustrating bacterial populations coated with HBBP. (b) Principle component analyzes based on the Bray Curtis β-diversity metric showing that samples of AnxA1, LPO and ZAG positive populations each form separate clusters whereas all the HBBP negative populations cluster together. (c) Alpha rarefaction plot. Shown are the number of different observed features as a function of the number of sequences analyzed and generated with QIIME2. (d) Relative abundance differences between bound and unbound fractions of the gut symbionts, taxa with significant p values are shown as red circles (two-sided Welch’s t statistical as implemented in aldex2). clr: center log-ratio; f: family; g: genus.

Data availability

All data, reagents and strains presented in this study are reported in the paper and associated Supplementary Information. Proteomic datasets were deposited to the ProteomeXchange Consortium via PRIDE (Data set identifier PXD027076 and 10.6019/PXD027076). RNA-seq datasets were deposited to the NCBI GEO repository (data set identifier GSE179530). Source data are provided with this paper.