Reverse vaccinology was employed for the identification of surface-exposed proteins, including βb-OMPs and OM lipoproteins in the genome of L. interrogans serovar Copenhageni strain Fiocruz L1-130 (LIC). The bioinformatics workflow and total numbers of proteins identified by each bioinformatics algorithm (predictor) are shown (Figure 1). The predictors for subcellular localization (Cello, PSORTb, and Gneg-mPLoc) found 523 proteins (by 1 or more predictor) located in the OM of L. interrogans, 24 were identified by all three predictors. A total of 1,196 proteins were predicted to contain a signal peptide (SP) for translocation across the cytoplasmic membrane to, e.g., the OM. Of these, 72 proteins were identified by all three SP predictors (SignalCF, SignalP, and PrediSI). One or more of the transmembrane α-helix (TMH) predictors (Phobius, TMHMM, HHTOP, and MEMSAT) determined that 3,302 proteins did not contain a TMH, this was reduced to 2,929 proteins by all 4 predictors. The transmembrane β-barrel (βb) structure predictors (Bomp, HHomp, TMBETADISC-RBF, and MCMBB) identified 1,085 βb-OMPs, 20 of which were confirmed by all 4 predictors. Finally, 230 proteins were identified as lipoproteins by at least one of the predictors, 108 by both (LipoP and SpLip). A complete list of the proteins identified by each individual predictor is provided (Table S1 in Supplementary Material).

A total of 165 βb-OMPs were identified by at least 1 predictor, while only 1 protein (LIC10714) was identified by all predictors. In this study, a βb-OMP was defined as a protein predicted to contain a βb structure, an SP, and <2 TMHs. As an SP can be identified as a TMH, a protein containing a single, N-terminal, TMH was classified as a non-TMH protein. An OM lipoprotein was defined as a protein predicted to be located in the OM and that contained a lipobox. A total of 54 OM lipoproteins were identified by at least one predictor of each feature; however, no proteins were identified at the intersect between all the predictors for lipoproteins and cell localization. A list of the 165 βb-OMPs and 54 OM lipoproteins identified, the gene products as annotated in the LIC genome and the result for each predictor is provided (Table S2 in Supplementary Material).

Due to the particularities of each predictor, low agreement between the different bioinformatics tools was expected when identifying the same feature. To reduce the impact of a prediction when non-weighted (naïve) voting resulted in ambiguities (e.g., two negative and one positive result by three different predictors), we used an iterative weighted voting system. An in-house Python script was written to integrate the results and to identify those proteins with a high level of confidence in the consensus prediction among those selected by at least one predictor. The lists of 165 βb-OMPs and 54 OM lipoproteins were used as the script input and, after thousand iterations, each predictor received a final prediction weight (Table S3 in Supplementary Material). A final voting score >0.5 (0–1 scale) for each feature of interest (e.g., >0.5 for OM localization and >0.5 for lipoprotein prediction) was indicative of a high level of confidence in the prediction, resulting in the selection of 18 βb-OMPs and 9 OM lipoproteins (Table 1).

Eleven of the 18 βb-OMPs identified in this study were originally annotated as OMPs in the LIC genome, one was annotated as a cytoplasmic membrane protein, and six were annotated as hypothetical proteins (Table 1). Seven of the eight OM lipoproteins were annotated as hypothetical proteins or putative lipoproteins in the original genome annotation. The exception was LIC10024 that was annotated as an adenylate/guanylate cyclase (AGC). The KEGG database was used to identify any corresponding orthologs and their annotations in the genomes of the other pathogenic Leptospira spp. Of note, the genome annotations of LIC10496, LIC11458, LIC11506, and LIC12575 differed substantially to those of their orthologs (Table 1).

A functional annotation was performed using InterProScan for the 18 βb-OMPs and 8 OM lipoproteins (Table 1). InterProScan identified OMP-related domains in 14 of the βb-OMPs. However, it failed to identify functional domains in four of the predicted βb-OMPs (LIC11211, LIC11458, LIC13477, and LIC20087) and five out of eight OM lipoproteins (LIC10647, LIC10713, LIC11755, LIC12048, and LIC13411). Of the remaining OM lipoproteins, two (LIC11003 and LIC20172) contained domains with an unknown function and LIC10024 was confirmed as an AGC protein.

For each protein sequence, the I-TASSER software predicted five 3D models and each model was ranked in order of quality by a C-score. The top-ranking models for 16 of the βb-OMPs contained a typical transmembrane βb structure (Figure 2). While the LIC11268 model that contained a transmembrane βb structure was ranked third, it was included for further analysis. The structural models of the eight OM lipoproteins are shown (Data Sheet 3 in Supplementary Material). The refined βb-OMP and OM lipoprotein models were analyzed by COFACTOR, a structure-based method for assigning biological function to protein molecules. A COFACTOR output includes the top 10 closest structures in PDB ranked by TM-score, Gene Ontology (GO) terms associated with the protein model, and the root-mean-square deviation of atomic position (RMSD) related to the best templates used for modeling. GO terms, including molecular function, biological process, and cellular location, associated with the protein 3D models were predicted based on the GOs assigned to the template structures and provided a functional insight into the selected leptospiral proteins. When more than one GO term per category was predicted for each protein, some were related to more distant templates (the last of the 10 closest structures), they were collapsed to the closest parent term on the AmiGO2 database and are shown (Figure 3) according to their frequency in the βb-OMPs and OM lipoproteins. The complete list of GO terms predicted for each protein structure is provided (Table S5 in Supplementary Material). As expected, most of the βb-OMP molecular function and biological process GO terms were related to transportation. This was supported by the cell location GO term, most of the βb-OMPs were classified as membrane or OM. The GO terms predicted for the OM lipoproteins were distributed among several different categories. Most of the GO terms for molecular function were for catalytic/binding activity and the predicted biological processes were related to metabolism. The predicted cell location GO terms were diverse (Figure 3; Table S5 in Supplementary Material).

Due to the importance of phagocytosis in the clearance of leptospires during an infection, the presence of MHC class II-binding epitopes in the 18 βb-OMPs and 8 OM lipoproteins was evaluated. NetMHCII was used to predict 15 amino acid long peptides that can bind, at different affinity levels, to MHC class II molecules encoded by several HLAs. Predictions were made for 14 HLA-DRB alleles and only strong binder (SB) epitopes (IC50 <50 nM) were considered for analysis. Each predicted immunogenic epitope had a 9mer core that was aligned, and a consensus sequence was determined. The location of each SB epitope in their respective structural models was identified (Figure 2). The correct orientation of the βb structural models in the OM was deduced using the corresponding PDB structure as a template. The total number of 9mers and 15mers predicted for each βb-OMP and OM lipoprotein, as well as the number of surface-related 9mers and 15mers of the βb-OMPs is provided (Table 4). LIC20087 and the OM lipoproteins were not analyzed for surface-related epitopes (SREs) as they did not contain βb structures; therefore, it was not possible to determine their localization in the OM. Each immunogenic epitope (9mer core or consensus) was evaluated for conservation among the orthologous proteins. A representative set of the most conserved immunogenic SREs for each βb-OMP is shown (Figure 4). The SREs were relatively well conserved in all the βb-OMPs. The alignments of the βb-OMP immunogenic epitopes to the corresponding regions in the orthologs from pathogenic Leptospira spp. are highlighted in the alignment files (Data Sheet S1 in Supplementary Material). A list of the βb-OMP and OM lipoprotein SB epitopes for all MHC-II alleles is provided (Tables S6 and S7 in Supplementary Material, respectively).

LIC10496 was originally annotated as a hypothetical protein and LIC12575 was annotated as a cytoplasmic membrane protein. However, their orthologs were annotated as TolC-like OMPs in the other pathogenic Leptospira spp. Furthermore, functional annotation by InterProScan predicted that they contained an OM efflux protein domain (Table 1), a property of TolC proteins, and a component of Type I secretion systems. In diderm bacteria, TolC is an OM efflux protein that forms a trimeric channel composed of a 12-stranded βb that spans the OM (4 transmembrane β-sheets per monomer) and long periplasmic α-helices that span the periplasm. The channel is connected to the cytoplasm by a TMH inner membrane protein channel. TolC transporters are involved in the transport of a variety of substrates, from small molecules to protein secretion. TolC proteins from Edwardsiella tarda (45), Listeria monocytogenes (46), and Salmonella paratyphi (47) were reported to be protective vaccine antigens. The structural models of the leptospiral TolC proteins were highly similar to an E. coli TolC protein (48). Both were characterized as high-quality models, with high ModFOLD and QMEAN scores and the lowest number of disallowed residues on the Ramachandran plot (Table 3). Interestingly, the extracellular loops connecting the β-sheets in the βb were predicted to contain MHC-II epitopes. Both leptospiral TolC-like proteins were conserved in all Leptospira spp., except for L. wolffii that has no ortholog for LIC10496. Structure-based GO predictions for these proteins suggested a possible relation to copper ions, siderophores, or protein transport (Table S5 in Supplementary Material). A previous study linked LIC12575 to the leptospiral response when exposed to in vivo like conditions (49). These data support the idea that LIC10496 and LIC12575 are promising vaccine candidates.

A TonB-dependent receptor (TBDR) domain was identified in all these proteins and this was supported by the structural models that predicted TBDRs. All the leptospiral TBDR models were reliable in terms of quality, this was expected as there is a high degree of similarity among TBDRs even in different bacteria (50). TBDRs are a βb proteins comprised of 22 amphipathic β-strands, and a globular plug domain that is folded up inside the barrel. Indeed, this structural folding was observed for LIC10881* and LIC10896* that were originally partial CDSs, supporting our finding that LIC10882 and LIC10895 are not proteins but are instead part of the LIC10881 and LIC10896 CDS, respectively, as in seen in other Leptospira spp. Interestingly, LIC10881* was absent in saprophytic and intermediate Leptospira spp., suggesting a potential role in pathogenesis. Gram-negative bacteria contain variable numbers of TBDRs, the E. coli genome contains 7 TBDRs (50); while other bacteria can contain up to 65 TBDRs (51). The LIC genome contains 13 genes that are annotated as TBDRs (52). While orthologs of LIC10714 in other Leptospira spp. genomes were annotated as TBDRs, LIC10714 was annotated as an OM receptor protein. LIC10714 contained the characteristic domains and the model included a 22-stranded βb with high structural similarity to FhuA from E. coli (53), even though it was suggested to be a leptospiral FecA (54). An ortholog of LIC10714 in L. biflexa was knocked out, resulting in a mutant with impaired ability to use iron citrate, iron chloride, iron sulfate, and aerobactin as iron sources, suggesting a role in iron metabolism (55). LIC10714 was also found to bind fibronectin (renamed as MFn2) and it was suggested that the extracellular loops could play a role in the bacterial adhesion process (56). The structural model of LIC10881* was similar to TbpA, a transferrin transporter in pathogenic Neisseria spp. (57). The LIC10896* 3D model was similar to FpvA, a pyoverdine-Fe transporter from Pseudomonas aeruginosa (58). LIC10964 was predicted to be a hemin transporter encoded by phuR, this gene is absent from all saprophytic Leptospira spp. except for L. vanthielii (Table 2). Of note, phuR was upregulated in the infection-mimicking model of leptospiral growth within dialysis membrane chambers (DMCs) implanted in the rat peritoneal cavity (59). The LIC10964 model was highly similar to the N. meningitidis zinc transporter, ZnuD (60). The LIC12374 and LIC20151 models were structurally similar to the E. coli BtuB cobalamin (vitamin B12) transporter (61). It has been suggested that pathogenic Leptospira spp. are autotrophic for vitamin B12 (12).

The actual molecule transported by a TBDR is difficult to predict by bioinformatics and should be determined experimentally. TBDRs play an important role in pathogenicity, they are conserved among the pathogenic Leptospira spp., and a significant portion of its structure was predicted to be surface-exposed. MHC-II epitopes were predicted for the TBDRs and several were located on the surface-exposed region of the protein. While TBDRs have been explored as vaccine candidates in other bacterial diseases (62–64), they have not yet been evaluated as experimental vaccines against leptospirosis. TBDRs transport essential molecules and if this is blocked, e.g., by antibodies, it would be a potentially lethal event.

Although both proteins were originally annotated as hypothetical proteins, InterProScan found an alginate export domain in LIC11268 (Table 1). However, the structural model of LIC11268 suggested it was a βb-OMP with high structural similarity to the neisserial TbpA transferrin transporter, a known TBDR (57). LIC13477 was identified as an alginate transporter based on structural annotation (Table 3) and was structurally similar to the OMP AlgE from P. aeruginosa (65). AlgE contains 18-stranded β-sheets in the transmembrane βb and is different to other TBDRs. The predicted models of both LIC11268 and LIC13477 contained an 18-stranded βb. Although an alternative, low quality, model of LIC11268 contained 20-stranded β-sheets (data not shown). Alginate is a polysaccharide in biofilms (66) and both L. biflexa and L. interrogans contain a complete set of genes for alginate biosynthesis (52, 67). Nevertheless, alginate exporters have not yet been identified in L. interrogans. The in silico modeling of LIC11268 and LIC13477 suggested they may play a role in the transport of alginate to the extracellular milieu, but this will need to be confirmed experimentally. Both proteins were predicted to contain a βb structure, potentially spanning the leptospiral OM. Immunogenic epitopes were identified in the surface-exposed regions and both proteins were conserved among pathogenic Leptospira spp., suggesting they are promising vaccine candidates.

Both proteins were originally annotated as OMPs, while in the current study, they were found to contain the bacterial surface antigen D15/Oma87 domains (Table 1). Of note, the orthologs of LIC11623 and LIC12254 in other Leptospira spp. were annotated as Oma87-like proteins. These BamA-like proteins are responsible for the assembly of βb-OMPs in Gram-negative bacteria. BamA is the OM component of the βb assembly machinery (Bam) complex and consists of a transmembrane βb and five polypeptide translocation-associated (POTRA) domains that extend into the periplasm. Both the βb and the five POTRA domains were identified in the LIC11623 model, and the closest PDB structure was BamA from Neisseria gonorrhoeae (68). Interestingly, although the LIC12254 model was similar to BamA, three of the POTRA domains (P1, P2, and P3) were missing. The closest PDB structure was HdBamAΔ3 from H. ducreyi, it too lacks three POTRA domains (68). Immunization with recombinant D15/Oma87 induced protective immune responses against Haemophilus influenza (69) and Pasteurella multocida (70). In addition, LIC12254 was only found in two of the saprophytic Leptospira spp., suggesting a possible role in pathogenesis. Although LIC11623 was described as a BamA-like protein in L. interrogans (52, 54), neither of these proteins have been evaluated as a vaccine candidate.

LIC11458 was identified as LptD/OstA, a member of the OM LPS export porin (LPS-ep) family (Tables 1 and 3). In diderm bacteria, LptD is responsible for LPS assembly in the OM (71). The structural model of LIC11458 was similar to the Shigella flexneri LptD, containing a transmembrane βb and a periplasmic β-jellyroll domain (72). The predicted transmembrane architecture of this protein forms a large 26-stranded βb with immunogenic surface-exposed epitopes. LPS is an essential virulence factor in pathogenic Leptospira spp. and, in contrast to the highly variable LPS molecule, LIC11458 was conserved among Leptospira spp. An immune response directed against LIC11458, as well as stimulating opsonizing leptospires, could potentially impair LPS assembly, ultimately killing the bacteria.

LIC11086 was predicted to contain a MetA-pathway domain for phenol degradation. The structural model of LIC11086 was highly similar to the crystal structure of Pput2725, a protein from Pseudomonas putida F1, a microorganism that can biodegrade hydrocarbons in the environment (73). Both structures contained a 12-stranded barrel with an N-terminal segment preceding the first β-strand that blocks the barrel. Proteins with this domain are predicted to transport hydrophobic molecules through the membrane, usually trichlorophenol and some are relatively well characterized (74, 75). LIC11211 was originally annotated as a hypothetical protein. The structural model of LIC11211 resembled TodX, an aromatic hydrocarbon transporter, also from P. putida (76). TodX is a 14-stranded βb, with an N-terminal flexible hatch domain. Both Pput2725 and TodX are members of the FadL family, a transporter of hydrophobic molecules in E. coli (77). P. putida and other biodegradation bacteria, such as Ralstonia pickettii, have intracellular pathways for the degradation of toxic hydrocarbons that enter the cells via FadL, with structural changes in the hatch domain (76). The role of hydrophobic molecule transporters in Leptospira spp. is unknown. LIC11506 was predicted to contain an OMP domain that is restricted to the Leptospiraceae (Leptospira and Leptonema). The structural model of LIC11506 was similar to E. coli OmpG, a βb with 14 antiparallel β-strands involved in the transport of carbohydrates into the cell (78). All three predicted leptospiral βb-OMPs have orthologs in all pathogenic strains; however, the LIC11506 and LIC11211 orthologs were absent in two species of intermediate pathogenicity (Table 2).

Both proteins are encoded on chromosome II, LIC20019 was annotated as a hypothetical protein while LIC20087 was annotated as an OMP. However, functional annotation showed that while LIC20019 contained a putative OMP porin 6 domain, exclusive to Leptospira spp., no such domain was identified in LIC20087. Furthermore, LIC20019 was modeled as a perfect βb, while LIC20087 was the only protein among the βb-OMPs identified in the present study that did not contain a βb. The LIC20087 model displayed structural similarity to the Type II pilus protein PitB from Streptococcus pneumoniae (79). The presence of pili or their function in Leptospira spp. is unknown; however, it was predicted to be surface-exposed and represents a potential vaccine candidate. As it is not expected for pili to be intrinsic to the OM, LIC20087 was not evaluated for the presence of surface-exposed immunogenic epitopes. Whether LIC20087 was misidentified as a βb-OMP due to the presence of several β-sheets in the pilus subunit protein or due to a poor threading template model, will need to be further investigated. This protein was shown to be immunogenic during infection and a possible candidate for early leptospirosis diagnosis (80). LIC20019 was, however, modeled as an βb-OMP, with high structural similarity to the plasminogen activator Pla from Yersinia pestis (81). Pla is an OM protease (omptin) whose inactivation drastically reduces Y. pestis virulence (82). Omptins are widely distributed among Enterobacteriaceae and have several functions (83). Some leptospiral proteins have been reported to bind plasminogen, however, just a few were investigated for cellular localization. Besides an apparent redundancy in the extracellular component binding proteins in Leptospira spp., LIC20019 was predicted to be exposed on the bacterial surface and to be highly immunogenic, a potentially strong vaccine candidate.

Leptospiral lipoproteins that were previously shown to be protective in the hamster model of leptospirosis, such as LigA (LIC10465), LigB (LIC10464), and LemA (LIC11058) were not selected among the list of nine predicted OM lipoproteins in this work. While these proteins were predicted by both LipoP and SpLip as lipoproteins (Table S1 in Supplementary Material), they were not consistently predicted to be located in the OM by the localization predictors. LemA was included in the list of the 54 OM lipoproteins (Table S2 in Supplementary Material) as it was identified by Cello as an OM protein. These observations are indicative that the list of 54 OM lipoproteins and 165 βb-OMPs included potential vaccine candidates that should not necessarily be excluded from future studies.

Of the eight OM lipoproteins selected, LIC10024 was the only lipoprotein originally annotated as an AGC. LIC11003 was annotated as LipL71 and the remaining six proteins were annotated as conserved hypothetical proteins or putative lipoproteins. In the current study, LIC11003 and LIC20172 were identified as peptidoglycan-binding proteins LysM (also known as LruA) and LruC, respectively. LruA and LruC were previously described as leptospiral recurrent uveitis-associated proteins A, B, and C (84, 85). LruC was experimentally demonstrated to be an OM lipoprotein but was not exposed on the bacterial surface, it was located in the inner leaflet of the OM (84). In contrast, LruA was exposed on the leptospiral surface, with a possible role in the modulation of interactions with human apolipoprotein A-I (ApoA-I), contributing to leptospiral virulence (86). Furthermore, LruA was shown to be essential for L. interrogans virulence in the hamster model (86). None of the Lru proteins have been evaluated as vaccine candidates and in order to do so, it will be necessary to exclude the regions that are responsible for leptospiral-related uveitis. LIC10024 was originally annotated as a membrane bound AGC, with an undefined cellular location. Another leptospiral protein (LA4008/LIC13201) with an AGC domain was shown to have host cell cAMP-elevating activity (87). The role of AGC proteins in leptospiral pathogenicity and the location of LIC10024 in the cell remain to be determined. Following in vitro analysis, LIC13411 was identified as a leptospiral adhesin that binds to VE-cadherin, an endothelial cell receptor (88). LIC13411 was also demonstrated to be present in the OM, supporting the findings of the current study. Unlike the βb-OMPs, the structural models of the OM lipoproteins could not be used to predict function or localization. The best matches to PDB structures included three eukaryotic proteins and a viral protein (Table 3). Most of the PDB structures identified were not bacterial surface-related proteins. This could be due to the absence of known conserved domains and low similarity to known protein structures. However, all eight lipoproteins were predicted to be OM lipoproteins and were highly conserved among Leptospira spp., including several immunogenic epitopes, suggesting they could be potential vaccine candidates.

The genome-derived proteome (chromosomes I and II) of L. interrogans serogroup Icterohaemorrhagiae serovar Copenhageni strain Fiocruz L1-130 was downloaded from the Leptospira Genome Project website (http://aeg.lbi.ic.unicamp.br/world/lic) in FASTA format, corresponding to GenBank accession numbers AE016823.1 and AE016824.1, respectively. High-quality draft, improved high-quality draft, or complete genomes for 20 additional Leptospira spp. (Table S4 in Supplementary Material) were obtained from GenBank.

The amino acid sequences corresponding to all 3,773 CDS in both LIC chromosomes were used as the input sequences for 16 bioinformatics programs for the identification of βb integral transmembrane proteins and OM-associated lipoproteins. Subcellular localization was predicted using PSORTb v. 3.0.2 (89), CELLO v. 2.5 (90), and GNeg-mPLoc v. 2.0 (91). The presence of a SP was predicted using SignalP v. 4.1 (92), Signal-CF (93), and PrediSi (94). TMHs were predicted using MEMSAT3 (95), TMHMM v. 2.0 (96), HMMTOP v. 2.0 (97), and Phobius (98). Proteins with more than one TMH (cut-off <2 TMH) were excluded from further analysis. Proteins with a single TMH, especially when located at the N-terminus, were included as a predicted SP can be confused with a TMH. The βb-OMPs were predicted using HHomp (99), BOMP (100), MCMBB (101), and TMBETADISC-RBF (102). Leptospiral lipoproteins were predicted using LipoP v. 1.0 (38) and SpLip (37). Each program was used with the default settings and when available the option for Gram-negative bacteria was selected. Sequence data and results retrieval for each predictor was automated using Python scripts whenever possible.

Low agreement is expected between different bioinformatics tools when predicting OMPs (103). To reduce the impact of a prediction when naïve (unweighted) voting would result in ambiguities (e.g., two negative results and two positive results), we used an iterative weighted voting system. A Python script was used to integrate the results, increasing confidence in the consensus prediction of a protein feature identified by more than one predictor. The Consensus by Voting with Iterative Re-weighting based on Agreement (CoVIRA) algorithm, script, and several examples are available on a GitHub repository (https://github.com/biopro/covira) and the algorithm followed the logic: PtASj=∑i=1NPd−1(Vi×PdWi), where PtAS_j, Protein Agreement Score—the value per protein by a given predictor. j=1,2,…, NPt

NPt—the total number of proteins evaluated.

V_i, Vote, always 1 or 0—one (1) when both predictors were in agreement, zero (0) for no agreement.

PdW_i, predictor weight—the calculated weight of each predictor in the total number of votes per protein, varied by iteration.

NPd—total number of predictors used per protein feature.

In the first iteration, PdW was PdWi=1NPd.

From the second to the 1,000 iteration, the PdW for the next iteration was PdWi=PtASi¯|max{PtAS1¯,…,PtASNPd¯}−min{PtAS1¯,…,PtASNPd¯}|,

where PtASi¯—the arithmetic mean per protein per predictor: PtASi¯=∑PtASjNPt.

A predictor with low accuracy was expected to generate a low agreement score and as accuracy improved, so would the agreement score. Finally, the last vote, based on the final weight of each predictor, was performed to identify those βb-OMPs and OM lipoproteins with a high confidence prediction. A final vote value >0.5 (0–1 scale) for each feature of interest (SP, OM localization, etc.) was considered as a confident prediction, and these proteins were selected for further analysis. The voting algorithm was validated using a set of proteins with and without an SP. The presence of an SP was evaluated using SignalP, PreDiSi, and SignalCF. The results were analyzed by naïve voting, each predictor had an equal weight. Based on the same results, CoVIRA was executed to calculate a final score and prediction for each protein. The results were analyzed by a receiver operating characteristic (ROC) curve, and the CoVIRA prediction was improved compared to the naïve voting (see Data Sheet S2 in Supplementary Material).

Orthologs to the selected proteins were identified in the leptospiral genome sequences (Table S1 in Supplementary Material) using the reciprocal best hit (RBH) method based on protein BLAST (BLASTp) searches. Protein sequences with >70% of similarity and >40% coverage that were also the best reciprocal hit were considered orthologs. A multiple sequence alignment was performed among orthologs using the online MUSCLE tool (MUltiple Sequence Comparison by Log-Expectation) (104). The βb-OMPs and OM lipoproteins with orthologs in other pathogenic Leptospira spp. were screened against the human, bovine, canine, equine, ovine, and swine genomes using the online Blastp server (taxIDs: Homo sapiens 9606, Bos taurus 9913, Canis familiaris 9615, Equus caballus 9796, Ovis aries 9940, and Sus scrofa domesticus 9825). Leptospiral proteins with >40% similarity to any host proteins were excluded from the final target list of potential vaccine candidates.

The 3D structures of conserved βb-OMPs and OM lipoproteins were generated by protein threading. Prior to modeling, the SP sequence was manually removed from the final FASTA amino acid sequence. βb-spanning OM proteins and OM lipoprotein structures were predicted using the I-TASSER server (41, 42). I-TASSER predicted up to five alternative models and the model with the higher C-Score was refined using ModRefiner (105). Quality assessment of models was performed using ModFold4 (106), Qmean6 (107), and Procheck (108). ModFold and Qmean returned a score (0–1) that inferred the overall quality of the structure. ModFold also assigned a P-value and a degree of confidence (poor, low, medium, high, and cert) to the model. Procheck was used to analyze the stereochemistry of the refined model by evaluating the Ramachandran plot of each protein structure (109). The 3D structures were visualized using UCSF Chimera (110) and PyMol (111).

Functional annotation was performed using the InterProScan tool (112, 113). In addition, the UniProt Knowledgebase was screened using the locus tag (LIC number) to identify functional information and to access the annotation in other Leptospira spp. by KEGG database. In addition to the primary amino acid sequence analysis, a 3D structure-based functional annotation was performed using COFACTOR (114), to identify the PDB structures with the closest structure to the target protein model, and assigned GO terms for the protein model based on these PDB structures.

The presence of MHC-II linear epitopes in the amino acid sequence of the selected proteins was predicted using NetMHCII v. 2.2 (115, 116). Fourteen HLA-DRB alleles were used in the prediction of immunogenic epitopes for each protein, including the most frequent alleles in human populations. The location of SB immunogenic epitopes (threshold values of IC50 <50 nM) were determined in the 3D structures of the OMPs, those most likely located within the surface-exposed portion of the OMPs were selected. The orientation of the OMPs models in the OM was determined by an interpretation of the orientation of the closest PDB structure from the COFACTOR analysis. The predicted MHC-II epitopes were aligned to the corresponding region in the ortholog proteins from other pathogenic Leptospira spp. (Table 2) by multiple sequence alignment generated by MUSCLE, and sequence logos were generated for each epitope using WebLogo (117). Images of the structural models, highlighting the SREs were generated by PyMol.