An immunoinformatic approach driven by experimental proteomics: in silico design of a subunit candidate vaccine targeting secretory proteins of Leishmania donovani amastigotes

Background Visceral leishmaniasis (VL) caused by dimorphic Leishmania species is a parasitic disease with high socioeconomic burden in endemic areas worldwide. Sustaining control of VL in terms of proper and prevailing immunity development is a global necessity amid unavailability of a prophylactic vaccine. Screening of experimental proteome of the human disease propagating form of Leishmania donovani (amastigote) can be more pragmatic for in silico mining of novel vaccine candidates. Methods By using an immunoinformatic approach, CD4+ and CD8+ T cell-specific epitopes from experimentally reported L. donovani proteins having secretory potential and increased abundance in amastigotes were screened. A chimera linked with a Toll-like receptor 4 (TLR4) peptide adjuvant was constructed and evaluated for physicochemical characteristics, binding interaction with TLR4 in simulated physiological condition and the trend of immune response following hypothetical immunization. Results Selected epitopes from physiologically important L. donovani proteins were found mostly conserved in L. infantum, covering theoretically more than 98% of the global population. The multi-epitope chimeric vaccine was predicted as stable, antigenic and non-allergenic. Structural analysis of vaccine-TLR4 receptor docked complex and its molecular dynamics simulation suggest sufficiently stable binding interface along with prospect of non-canonical receptor activation. Simulation dynamics of immune response following hypothetical immunization indicate active and memory B as well as CD4+ T cell generation potential, and likely chance of a more Th1 polarized response. Conclusions The methodological approach and results from this study could facilitate more informed screening and selection of candidate antigenic proteins for entry into vaccine production pipeline in future to control human VL.


Background
Leishmania spp. are obligate intracellular pathogens of phagocytic host cells. Two species, Leishmania donovani and L. infantum cause visceral leishmaniasis (VL), a neglected tropical disease and second only to malaria in parasitic cause of death. With a chance of case fatality of 100% in an inadequate treatment scenario, over 90% of VL cases occur in relatively poor communities of Bangladesh, India, Sudan, South Sudan, Ethiopia and Brazil [1]. The VL elimination program known as Kala-azar Elimination Programme (KEP) has contributed to a remarkable decline in the incidence of VL over recent years in the Indian subcontinent and now it is approaching the maintenance phase of VL elimination [2]. However, sustained elimination cannot be possible without proper and prevailing immunity development in the endemic population against Leishmania parasites in the post-elimination era due to the chance of reservoir mediated re-emergence of the disease [3]. A vaccination strategy can induce longterm protection with proper immunity in order to prevent development of disease in the most economical way, regardless of its mode of implementation.
In recent years, enormous progress has been made in the design of vaccines against leishmaniasis using liveattenuated or killed parasites, cellular extracts, and individual and/or recombinant antigens of parasites. The first-generation vaccine, which includes live-attenuated, killed and fractionated parasites, is the only class of human prophylactic VL vaccine that entered phase III clinical trials so far. However, this vaccine failed to achieve satisfactory results [4]. The second-generation vaccines are produced from recombinant Leishmania antigens (single peptides/polypeptides). Among several approaches, LEISH-F3, a multicomponent vaccine formulated with GLA-SE adjuvant showed promising results in phase I as a robust immune response inducer in healthy people [5]. Earlier, LEISH-F1 in combination with MPL-SE adjuvant also showed strong antigen-specific immune response in healthy people living in a L. donovani endemic area [6]. More recently, a third-generation DNA vaccine approach that employed simian adenovirus expressing a novel synthetic gene encoding Leishmania antigens, hence termed as ChAd63-KH, has shown potentiality to be a safe and immunogenic therapeutic vaccine for human VL and post kala-azar dermal leishmaniasis (PKDL) in a phase I trial [7]. Despite the ongoing progresses in vaccine development, the priority objective has not yet been achieved, i.e. the development of safe, effective, durable and low-cost prophylactic vaccine for human visceral leishmaniasis [8].
Besides producing memory lymphocytes towards a long-term immunity pathway, an ideal vaccine against Leishmania will stimulate parasite-specific cellular immunity that include a strong Th1 response to eliminate infections. In this regard, the use of epitopes or epitopecontaining peptides is advantageous since epitopes can be evaluated for immuno-recognition and epitope-specific response. Since epitopes/peptides themselves remain poorly immunogenic, the approaches that have been gaining interest are based on the development of peptidebased formulations in combination with potent adjuvant components (peptide, lipids, virus particles, nanoparticles etc.) [9]. However, mapping of epitopes in immunogenic proteins remains crucial in peptide vaccine development. In addition to in vitro methods of epitope mapping such as phage display library, immunodominance and peptide competition assays, immunoinformatic mapping can be a powerful approach to facilitate screening of desired epitopes in immunogenic proteins [9]. Recent findings of leishmaniasis vaccine research also suggest that in silico predicted MHC class I and class II restricted epitopecontaining peptides derived from Leishmania antigens alone, as a cocktail, as a chimeric peptide or in combination with adjuvant can be substantially immunogenic in vitro and/or in vivo [10][11][12][13]. Thus, the application of immunoinformatics-based pipeline can facilitate largescale screening of peptide epitopes from Leishmania proteome for rational design of potent vaccines.
While derivation of potentially immunogenic peptides can be performed by analyzing (in vitro and/or in silico) either the whole parasite proteome, proteins known to elicit immunological outcome, or the known peptide libraries [9], two essential criteria have been suggested for consideration to initially select potential vaccine antigens for leishmaniasis: (i) known antigen that is expressed in the disease-causing mammalian stage of the parasite; and (ii) selected adjuvants that elicit a cellular, Th1-biased immune response for the immunizations in humans [14]. The human stage-associated Leishmania proteins that facilitate intracellular survival and infective process of the parasite thus constitute attractive targets for anti-Leishmania vaccine design. In order to adapt in mammalian host, the promastigote stage of Leishmania undergoes morphological and metabolic changes when transformed into amastigote stage upon entry and invasion. This is accompanied by a cascade of programmed changes in mRNA abundance, translation rate, and/or protein processing. However, interpretation of Leishmania transcriptome is likely controversial on whether relative changes in mRNA abundance is substantial [15,16], constitutive or negligible [17][18][19]. Moreover, stagespecific upregulation for some transcripts [20] does not necessarily reflect in altered functional protein profile because of post-transcriptional [21] and post-translational regulation [22] evident for Leishmania species. While mRNA abundance may not be a perfect indicator of protein levels in eukaryotes [19,23,24], the relationship between mRNA and protein abundance is suggested to be dynamically changing as L. donovani adapts to amastigote condition, with correlation in changes for only a small proportion [22]. Moreover, the changes in protein level can also vary between clinical and cultured amastigotes [25]. In this scenario, experimentally evaluated proteome analysis can better highlight the key changes, which have important implications for diagnostics, drug target identification and vaccine design.
Reverse vaccinology [26] has been becoming increasingly popular in supported vaccine design by the combined use of genomics, transcriptomics, proteomics and immunoinformatics. Here, we propose an approach to design a subunit vaccine based exclusively on mass spectrometry (MS)-driven comparative proteomic information associated with amastigotes, since genome/transcriptome information can be deviant as Leishmania adapts to amastigote condition. Invasion and survival of L. donovani in the mammalian host largely involves the export of virulence factors and immune-modulatory components into the host cytosol. However, constant exposure of these secretory proteins to host immune system can lead to immunological tolerance and strong parasite-specific humoral response, which can be problematic for vaccine design. Hence, we limited our focus on rational screening of immunogenic T cell-specific epitopes in such secretory proteins, which have substantial coverage of endemic population as well. We then combined the epitopes into a single recombinant protein molecule. We explored into the physicochemical properties and receptor binding interaction of the multi-epitope vaccine, followed by molecular dynamics simulation of the vaccine-receptor complex and simulation of immune response. Overall, we showed that the experimental proteome data-driven immunoinformatic approach can facilitate informed screening of potential subunit vaccine candidates from truly produced human stage-associated parasitic proteins of pathological/physiological importance. The immunogenic potential evaluated in silico can also provide rationality for experimental validation of the modeled subunit vaccine.

Methods
All the computational tasks using online and offline tools in this study were carried out with the concurrent version of the tools between September 2018 and December 2018. Graphpad Prism v.7 software was used for descriptive data calculation, comparison of means and to reproduce graphs using software generated numeric values when applicable. A two-tailed P-value of < 0.05 was considered as significant. The methodological flowchart of the study is given in Fig. 1. The details of the methodological steps are given in Additional file 1: Text S1.

Antigen selection
Literature reports on the proteome profile of L. donovani were screened in the PubMed (NCBI) database to index articles published between year 2000 and 2018, and reports on MS-driven comparative abundance of proteins in promastigotes and amastigotes were selected to generate a local database of proteins. Sequences of identical or closely similar L. donovani proteins, which had more abundance in amastigotes, were retrieved. An MS-derived secretome profile of L. donovani was also retrieved from the published literature [27]. Secretory proteins in the local database were screened by both cross-matching the secretome and proteome data using ViroBLAST [28], as well as by evaluating combined prediction of SignalP [29], SecretomeP [30] and TMHMM [31]. Antigenicity of the potential secretory proteins was estimated using ANTIGENpro [32] and VaxiJen [33] programs.

Epitope screening
For screening of cytotoxic T-lymphocyte (CTL, 9-mer) epitopes and helper T-lymphocyte (HTL, 15-mer) epitopes, both affinity and allele coverage were considered. Initially, CTL and HTL epitopes were screened for above cut-off prediction scores in NetCTL [34] and lower percentile rank in IEDB (Immune Epitope Database) recommended MHC-II consensus module (http://tools .iedb.org/tcell /), respectively. Epitopes that are superior in human leukocyte antigen (HLA) cross-allele coverage were preliminarily selected. Then, both CTL and HTL epitope sets were filtered through specific (MHC-I and MHC-II modules of IEDB) and common (VaxiJen) immunogenicity prediction tools. CTL epitopes were further filtered in TAPpred [35] for more accurate prediction of TAP transporter binder. As per requirement of a proper anti-Leishmania immune response, T cell epitopes containing B cell recognition region (predicted by Bepipred [36]) were excluded, and all IL-10 inducing epitopes (predicted by IL-10Pred [37]) were removed. BLAST search against the non-redundant human protein database was carried out to rule out self-epitopes, while search against the RefSeq [38] protein database of Leishmania was performed to find out the conservancy of the epitopes in other Leishmania species.
In order to calculate the theoretical population coverage (TPC) (http://tools .iedb.org/popul ation /) of each epitope, binding affinity to HLA allele-specific MHC molecules was set at percentile rank cut-off of 1.0 for CTL epitopes (IEDB recommended), and IC 50 cut-off of 100 nM for HTL epitopes (10-times lower than the IEDB recommended value). Our target was to reach more than 90% population coverage by both CTL and HTL epitope sets in each of three most endemic areas of VL (India, Sudan and Brazil) with maximum number of alleles per epitope, while keeping the subunit length at minimum. In accordance, CTL epitopes with minimum TPC of 40% and at least eight HLA alleles were selected for vaccine construction. On the other hand, most of the screened HTL epitopes had more than 80% coverage in our observation, and therefore HTL epitopes having greater than 90% coverage were selected.

Chimeric vaccine construction and evaluation
The vaccine construct was arranged by joining the CTL and HTL epitopes with linkers [39,40], and preceded by a synthetic peptide adjuvant [41]. The selection of chimeric arrangement was based on antigenicity (ANTIGENpro and VaxiJen) and allergenicity (AlgPred [42] and Aller-TOP [43]) scores, retaining of cleavage, TAP transporterand MHC-binding propensity of target epitopes, and generation of none/least number of non-specific and/ or IL-10 inducing epitopes due to recombination. The capability to induce IFN-γ and IL-10 by the chimera was predicted by scanning in IFNepitope [44] and IL-10Pred module, respectively. Simulation of immune response based exclusively on the chimeric construct was performed in C-ImmSim [45] server, whereas two previously reported candidate Leishmania vaccine peptides [46,47] were used to evaluate whether C-ImmSim prediction corroborates to the dynamicity of antigenic constructs. For structural analysis, the tertiary structure of the construct was produced in I-TASSER [48] modeling server followed by refinement using YASARA [49] force-field and GalaxyRefine [50] web tools. Stability of vaccine construct, a prerequisite for antigen processing, was assessed using physicochemical features predicted by both sequence-based (ProtParam [51]) and structural (SCooP [52] and CamSol [53]) analysis tools. Furthermore, chimeric vaccine-specific linear (Bepipred and BCPREDS [54]) and conformational (Ellipro [55]) B cell epitopes were predicted. The structural model was used to dock to Toll-like receptor 4 by using ClusPro [38] docking server and the binding interactions were analyzed. Molecular dynamics (MD) simulation was performed by using Desmond v5.3 (Schrödinger, LLC, New York, USA) software to check the conformational stability of vaccine-receptor docked complex.

In silico cloning
To validate the immunoinformatic findings, evaluation of immuno-reactivity through serological analysis is the preliminary step and this requires expression of the candidate vaccine. For this purpose, codon optimization was carried out by using JCAT [56]. A cloning model was then developed for this construct by using SnapGene (GSL Biotech, California, USA) tool and by inserting the optimized coding sequence into a plasmid vector.

Screening of secretory amastigote proteins from experimental studies
Following literature screening, three out of 28 PubMed indexed experimental studies on L. donovani proteome were considered for the study. Two of the studies compared proteomic abundance of promastigotes with that of amastigotes as Leishmania adapts to the changes in conditions resembling the host [57,58], while the other one compared splenic amastigotes to axenic amastigotes [25]. A total of 118 out of 134 proteins, which had a relative increase of at least 1.5-fold or were reported exclusively in the protein profile of amastigotes and/or splenic amastigotes, were found to have an identity of 90% or above for absolute query coverage with L. donovani proteins of similar functional annotations. After cross-matching of this group of proteins to 151 L. donovani secretory proteins revealed experimentally [27], 16 proteins were found to have an identity percentage and query coverage of 96.75 ± 1.1% and 99.44 ± 1.13%, respectively. Based on the presence of classical or non-classical secretion signal sequences along with minimum (no more than one) transmembrane helices, one common and an additional 17 secretory proteins were included to the pool. Among 33 amastigote-associated potential secretory proteins, 26 were selected based on their antigenicity probability scores of ≥ 0.5 as predicted by both ANTIGENpro and VaxiJen (Table 1, Additional file 2: Data S1).

Multi-epitope subunit L. donovani vaccine: construction and properties
A total of 79 CTL 9-mer epitopes were initially screened in NetCTL. Among them, only nine epitopes from six proteins were predicted to be non-self, highly immunogenic and high-to-moderate TAP-transporter binder non-B cell epitopes. These epitopes covered theoretically, an average of 66.46 ± 7.88% and a cumulative of 98.57% of the world population. Similarly, HTL 15-mer epitopes were screened to ensure both affinity and coverage. Fourteen selected HTL epitopes from eight proteins were finally predicted to be non-self, highly immunogenic non-B cell epitopes, with a mean theoretical coverage of 96.62 ± 1.35% and a cumulative of 99.52% of the world population. All the CTL and HTL epitopes except for H2-10 and H2-13 were conserved (100% identical) in L. infantum, whereas, less conservancy was found in representative proteins of L. major (13/23) and L. mexicana (10/23). Properties of individual CTL and HTL epitopes are given in Tables 2 and 3, respectively.
The construct of 397 amino acid residues comprised of the 9 CTL and 14 HTL epitopes, with AAY and GPGPG linkers added in the intra-epitopic positions of CTL and HTL epitopes, respectively. It preceded in N-terminal by TLR4 peptide adjuvant, APPHALS, linked by EAAK linker to the vaccine. The selected rearranged model had the antigenicity score of 0.8 calculated by ANTIGENpro, and 0.74 (bacteria model) and 0.65 (parasite model) by VaxiJen. Furthermore, the construct was found to be non-allergenic for human use. When re-analyzed by the screening tools, all the original CTL and HTL epitopes were found consistent with the pre-screening immunogenicity, cleavage and TAP binding properties in the rearranged model. On the other hand, the arrangement of the construct resulted in generation of only three regions (15mer overlapping) of IL-10 inducing epitopes and three non-specific CTL epitopes (9-mer) comparable to the potency of target epitopes (Additional file 3: Figure S1).
infantum) [ heparin binding [128,129] Immunogenic in vivo and immunoreactive in sera of human (L.
major) [130] 26 E9BED5 Cysteine peptidase A (CPA) Predictive (classical) Increased 0.670/0.939 [25] Host-parasite interaction [131] Th1 pathway stimulatory and cross-protective in vivo (L. infantum) [10]  No prediction was consistent with the C-ImmSim simulated immune response in terms of high IFN-γ production after hypothetical immunization (three doses) in a population characterized by a combination of frequent and VL susceptible HLA alleles [59][60][61]. Since the hypothetical cytokine levels in simulated immune response represent only the outcome of algorithmically set dynamic cellular interactions for a defined time period after antigen priming [37], it was important to evaluate whether the simulation module can respond dynamically to different constructs [45]. Therefore, we simulated immune responses of two additional peptide vaccine candidates: peptide 1 (L. infantum derived fusion peptide [46]) and peptide 2 (L. donovani GP63 derived peptide [47]), which were experimentally found to exhibit varying cytokine response in comparison to soluble Leishmania antigen (SLA). Substantial difference was observed in terms of immunosuppressive IL-10 and TGF-β induction capacity between peptide 1 and peptide 2; however, determination of statistical significance was not possible in the simulation module. Nevertheless, the outcome can be considered consistent with the general trend of in vitro immune response (compared to SLA), with peptide 2 being more prominent IL-10 inducer compared to SLA as reported in [47]. In terms of cytokine induction potential, simulation outcome of our designed construct conformed more closely to that of peptide 1, which did not induce IL-10 level higher than that by SLA in vitro [46] (Fig. 2).

Tertiary structure of the chimeric protein and cloning model
Since there was no significant template hit, the choice of 3D model among outputs generated by I-TASSER was based on: (i) cluster size of model replicas; (ii) frequency of model in simulation trajectory; and (iii) C-score. The selected model has the highest C-score of − 1.56 which is close to the I-TASSER recommended score (− 1.5) for accuracy, and has the highest frequency in the top cluster by size. After further refinement of the protein topology, the PROCHECK [62] server returned a G-score of − 0.04, which indicates that the backbone and side chain of the model correspond to high-probability stereochemical conformations.  In terms of chimera-specific B cell response, Bepipred predicted six B cell epitopes of 8-12 residues in length above the threshold score, while BCPREDS predicted 11 non-overlapping and linear 20-mer B cell epitopes with specificity scores > 0.99. Residues in those linear epitopes accounted for 41% residues of the 08 non-overlapping conformational epitopes (Fig. 3, Additional file 5: Table S1).
This sequence was used to generate in silico cloning model for E. coli (K12) expression. After optimization of the codon, the codon adaptation index (CAI) value of  For insertion into the E. coli pET28a(+) expression vector, two restriction sites for XhoI and NdeI enzymes were added in the 3ʹ-and 5ʹ-end, respectively, of the vaccine coding strand enclosed by 6-histidine residues at both ends (Additional file 6: Figure S3).

Molecular docking of vaccine in TLR4
Molecular docking of the vaccine construct with TLR4 in ClusPro 2.0 docking server generated 30 models ranked by cluster size of the representative pose. The selected docked complex had the largest cluster size (ClusPro recommended) with second-lowest binding energy score (− 1282.3) among the top ten models. The chimeric construct seemed to occupy partially into the lateral concave surface, but not the convex surface, with strong hydrophobic interactions mostly with the beta-sheet adjacent residues at the C-terminal domain of TLR4 ectodomain (ECD) and also with its adapter protein, MD2, with support of several hydrogen bonds, thus establishing ligand mediated cross-link between TLR4 and MD2 (Fig. 4).

Molecular dynamics (MD) simulation of vaccine-TLR4 complex
Molecular dynamics simulation of the docked complex was performed by using OPLS_2005 force field. Using the Simulation Quality Analysis tool of the Desmond software, the mean potential energy for the complex was obtained as − 6.4e5 kilocal/mol (Additional file 7: Figure  S4). The radius of gyration (Rg) obtained for the docked complex showed that the mean distance in rotating complex from the center of mass is 4.31 nanometers (SD: 0.2 nanometers) about which the model becomes consistent after 4 ns (Fig. 5a). The number of intermolecular hydrogen bond (H-bond) between the side chains of vaccine protein and TLR4 initially fluctuated probably due to solvent effect before matching the trend of Rg in reaching steadiness after 4 ns. This suggests the role of H-bonds in the overall compactness of the complex (Fig. 5b). The trends of Rg and H-bond plots indicate that 6-8 strong H-bonds were persistent over simulation period between vaccine and TLR4, and this might be crucial for stable binding.
The root mean square deviation (RMSD) of the vaccine-TLR4 complex for backbone atoms over the simulation period was 4.0 Å (SD: 0.49 Å), while it was 3.2 Å (SD:  (Fig. 5d). Although the trends toward reaching convergence were very similar, higher RMSD value of the complex than the vaccine-unbound TLR4 indicates structural mobility in the complex due to vaccine interaction and this is likely attributable to the higher RMSD of vaccine protein along the MD simulation time. In congruence, rearrangement of several bonds between the vaccine and TLR4 was observed between pre-simulation and post-simulation models, while the total number of non-covalent bonds increased from 41 in pre-simulation model to 64 in post-simulation model (not shown). Visualization of the interacting residues also indicates that, in comparison to unbound (and also pre-simulation) structure, the post-simulation bonding rearrangement is coupled with increased number of H-bond at the C-terminal domains between TLR4 and TLR4* (second TLR4 ECD) (Additional file 8: Figure S5). This implies likely chance of positive interactions between the TLR4 monomers in physiological condition following vaccine interaction. Overall, the conformation of vaccine-bound receptor supports structural flexibility, which might be in favor of biological response of the receptor.

Immune simulation to predict secondary response
Hypothetical administration of three doses of vaccine construct four weeks apart with 1000 unit/dose was performed to simulate the immune response generated by immunization. In silico immune simulation RMSD of the backbone atoms fitted to complex (green) and ligand-free receptor (black), with respect to initial structure (c), and RMSF for side-chain atoms fitted to complex (green) and ligand-free receptor (black) with respect to initial structure of TLR4 ECD (627 aa; divided into three domains of leucine rich repeats or LRR) (d). Receptor positions interacting with vaccine (final frame) are represented with red circles plots hinted at antigenic recognition and subsequent response in terms of antibody production, and active as well as memory B cell and T cell generation in the population with a VL susceptible HLA profile after hypothetical immunization. The primary response to the proposed chimera can be characterized by a marked increase in chimera-specific IgM and IgG production. After subsequent doses, a corresponding decrease in antigen concentration indicates gradual increase in memory B cell production with persistence. Furthermore, expansion of CD4+ T lymphocytes with memory development following the initial dose was observed. CD8+ T lymphocytes response was also high for the susceptible population reaching at its peak after the second dose. Repeated exposure of 12 doses, on the other hand, did not seem to cause clonal expansion of any epitope-specific T cells-as indicated by the Simpson's index D, which is inversely related to diversity (Additional file 9: Figure S6).

Discussion
Proteomics-driven identification of potential vaccine candidates can be a sound approach for selecting promising antigens, which are elicited against environmental stimuli analogous to host response upon pathogen invasion and are physiologically relevant for pathogens within the host [64]. Availability of pathogen proteome information upon infection of the host can provide opportunities for in silico mining of novel vaccine candidates, and this approach has been utilized for in silico design of an epitope-based vaccine against Theileria parasites of ruminants [65]. For a dimorphic human parasite like Leishmania, it is important to target human stage-associated antigenic proteins that are physiologically important for parasites to infect and establish in a new host. In recent years, several studies utilized immunoinformatic approaches of epitope screening in designing epitope-based vaccines. Khatoon et al. [66], Singh et al. [67] and Vakili et al. [68] have previously reported the theoretical potential of in silico designed vaccines for visceral leishmaniasis. Notably, in a recent study by Vakili et al. [69], the group further evaluated successfully the immunogenic potential of the multiepitope vaccine, derived in part from known antigens, by administering the chimeric construct in experimental mice. This suggests that the in silico designed vaccines with epitopes derived from appropriate protein targets have the potential to progress toward advanced phases of vaccine development for visceral leishmaniasis. While the in silico studies by Khatoon et al. [66] and Singh et al. [67] largely utilized available genomic databases of L. donovani to select vaccine targets, Dikhit et al. [11,70] performed thorough investigations involving in silico, in vitro and in vivo analysis to screen and validate immunogenic epitopes obtained from proteins that are increasingly expressed at the infective parasite stage. Such highly expressed proteins are likely important for physiological and/or infective process of the parasite and thus can be more effective vaccine targets. In this study, we took an approach to select such amastigote proteins in terms of contrasting abundance or specificity (abundant up to the level of detection) from comparative proteome profiles of L. donovani promastigotes and amastigotes. Based on the propensity of those proteins for secretion in vitro and/or having secretory signal sequence, we further combined immunoinformatic tools to identify candidate antigens that have secretory potential. A comparison of the methodological and outcome features among several studies that have employed in silico design and evaluation of epitope-based candidate vaccines against visceral leishmaniasis to date is summarized in Table 4. Overall, our reported vaccine construct was found to be comparable to the earlier exclusively-in silico reports in terms of antigenicity, population coverage and receptor interaction. However, experimental studies remain crucial to validate the immunogenic potential of the designed vaccine.
Analyzing amastigote secretome through intra-macrophagic studies is considered difficult, while significant difference in secretome between amastigotes and promastigotes is unlikely due to relatively low stage-specific differences in gene expression [27]. However, due to the dynamicity in the relationship between mRNA and protein abundance as L. donovani adapts to the amastigote condition, comparative levels of abundance of these secretory proteins can be a more reliable indicator. Hence, our screening approach is relevant within the context. Perhaps, the most studied amastigote-specific vaccine candidate in L. donovani happens to be a cellular stress countering abundant surface antigen, A2, which has shown to confer whole or epitope-specific efficacy in multiple immunization models [8,71]. The vaccine construct reported in the present study comprised of immunogenic T cell-specific epitopes (as predicted immunoinformatically) from 13 amastigote-associated proteins. Five of them are known to associate with virulence in the mammalian host (fructose-1,6-bisphosphatase, putative protein disulfide isomerase, putative lipophosphoglycan biosynthetic protein, leishmanolysin and cysteine protease), while others have putative roles in countering the host-induced stress response (thioredoxin-like protein, glutathione peroxidase, stress-inducible protein STI1 homolog), host-microbicidal activity regulation (proteasome endopeptidase) and protein synthesis (elongation factor 2). Three proteins were uncharacterized according to the proteomic studies. Protein domain and homology (to proteins of other Leishmania   [68] species) suggest that two of these proteins may potentially play a role in drug resistance phenotype (E9BUW4) and protection from intracellular stress (E9BDB8), while the specific function of alpha/beta hydrolase domaincontaining protein (E9BQ40) in amastigotes has not yet been deciphered. On the other hand, lack of reports on experimental evaluation of immunogenicity of several Leishmania proteins, which have been included in our set of antigenic proteins is apparent. Among the 13 proteins of current interest, only six (elongation factor 2, proteasome endopeptidase complex, putative protein disulfide isomerase, leishmanolysin, cysteine protease and putative lipophosphoglycan biosynthetic protein) or their species homologs are known to have proven immunoreactive properties (Table 1). Nevertheless, the increased abundance of the unexplored proteins suggests their likely role of pathological/physiological significance in host invasion and/or survival. The antigenicity scores further corroborate to the potentiality of these proteins as antigenic. Immunological evaluation of these amastigote stage-associated proteins may unravel novel Leishmania antigens in future.
In the context of functional roles of selected proteins, our designed vaccine has the potential to benefit the host by generating appropriate immune response both in the early and progressive phase of systemic infection. Furthermore, almost all of the epitopes were found in corresponding proteins of L. infantum, indicating potential cross-protection against this visceralizing species. Most of the VL cases are reported from the endemic zones of the Indian subcontinent, East Africa and South America. Thus, in designing an epitope-based subunit vaccine, it is important to estimate the fractions of population in the target endemic zones based on HLA genotypic frequencies. The immunogenic non-self CTL epitopes in the vaccine modeled here is estimated to cover 96.8%, 91.7% and 93.9% of the allelic populations of Brazil, India and Sudan, respectively, with experimentally evaluated truly binding affinity [72], while for HTL epitopes, it is almost 100% for each of these populations. The vaccine construct has antigenic properties while it was not found to be an allergen. The structure was found thermodynamically stable and surface-soluble, while the core is hydrophobic, a favorable feature for antigen processing. Vaccine-specific, but not parasite protein-specific humoral response was predicted, and this can be used as a biomarker of vaccine efficacy [46,73] without eliciting a parasite-specific B cell response. Moreover, the construct structure showed a good binding affinity in previously reported binding cavity of TLR4 [74][75][76][77].
The structural interface between TLR4 and the peptide adjuvant (APPHALS) used here has been extensively studied before. The position occupied by the adjuvant peptide in the TLR4-MD2 complex has been suggested to be varying depending on its position in the vaccine model and the canonical activation of the receptor is thought to be mechanized by insertion of peptide adjuvant in MD2 [78]. Since we used already activating but hypo-responsive TLR4-MD2 crystal structure removed of LPS for docking [79], it was not possible to speculate about the agonistic behavior of the bound vaccine. Nevertheless, our docking model is suggestive of non-MD2 (non-canonical) binding of adjuvant linked peptide, in which the vaccine intrinsic segment may have more affinity than the peptide adjuvant for binding to TLR4. The binding interface along with the molecular dynamics (MD) simulation of the docked complex in the solvent system hint at a sufficiently stable cross-link of TLR4 and MD2 with no major bond rearrangement between TLR4 and MD2, and between TLR4* and MD2 heterodimer formations. Although, the simulation time was short, this is reasonable as none of the vaccine residues interacted at crucial MD2-binding sites [74]. On the other hand, H-bond was found to increase between TLR4 ECDs (where vaccine is bound to one TLR4 ECD) in the vaccine-bound form compared to the unbound TLR4, which suggests potential event of positive interactions and movement between the ECDs. Additionally, reduction in electrostatic surface potential at the vaccine-bound TLR4 interface was observed after docking, which was consistent in post-simulation structural interface. Simultaneously, it was observed that a homo-dimer destabilizing His458-His458* repulsion [76] at pre-dock TLR4 was nullified and superseded post-dock by a solvent stable pihydrophobic interaction. It is thus possible that a change in the interpolated charge difference between pre-dock and post-dock TLR4 interface could have contributed to the bonding rearrangement between TLR4 ECDs. Notably, this rearrangement also involved participation of other critical histidine (His431, His555) residues at the TLR4-TLR4* interface [80] unlike the unbound structure (Additional file 8: Figure S5). Overall, these events are congruent with non-canonical TLR4 activation model mediated by microbial peptides, metals and cationic lipid nano-carriers, which are suggested to not confer canonical interaction with other monomers but to induce bond rearrangement among receptor monomers upon interaction [74][75][76][77]. Although the exact mechanism remains to be elucidated, our observations suggest that the vaccine construct may possess a characteristic peptide feature of a non-canonical TLR4 ligand [81,82], which may facilitate TLR4-TLR4* dimerization for downstream activation of immune cells. The trends of backbone RMSD, Rg and H-bond of the vaccine-bound complex over the simulation period complied with structural flexibility rather than rigidity of the complex. The RMSF values of the complex side-chain indicate that the higher fluctuations in TLR4 were of those residues, which are vaccineunbound and located in the solvent exposed loop mostly at or around glycines [83]. Increased residual fluctuation at LRR10-12 and around Gly397 may also be attributed to the mutations introduced at the position 299 and 399 in TLR4 structure (4G8A), as reported in [79], which was used to dock the vaccine protein. Nevertheless, it is unlikely that vaccine interaction would induce dissociation in structural interface of natural TLR4-MD2 since none of the highly fluctuating TLR4 residues had any direct interaction with the vaccine or MD2.
Simulation outcome of hypothetical immunization in VL susceptible HLA alleles (hypothetical heterozygous combination) was consistent with the predicted immunogenicity of the vaccine. Furthermore, we showed that the simulation outcome can be dynamic for different constructs when we used the same criteria in the simulation program and the same HLA profile to test two known vaccine candidates for VL. Importantly, for these peptides, IL-10 production was reported previously as either prominent (peptide-2) or lessened (peptide-1) in comparison to SLA in vitro. It is not expected that simulation results will reflect experimental outcomes; however, we observed a general trend of difference in immunosuppressive cytokine (e.g. IL-10) induction potential between the two peptides from the simulation outcome, with peptide-2 having more potent IL-10 induction capacity. Although statistical significance could not be inferred from the simulation plots, the difference seems consistent with the experimental result. Understandably, the predicted epitopes (not shown) in the simulation program did not comply mostly with our target set of epitopes due to the difference in the epitope prediction algorithm [45]. However, when compared to the simulation outcome of the known peptides, the general trend was comparable to both peptides for IFN-γ induction, while TGF-β and IL-10 were predicted to be considerably less pronounced than that by peptide-2. Besides IL-10, TGF-β has potent immunosuppressive properties, enhances disease progression and may prevent cure and protective immunity development against leishmaniasis [84,85]. Thus, the simulation prediction of higher propensity of the construct to induce a more Th1-polarized response rather than Th2 is consistent with our desired immunogenicity.
Despite the difference in the epitope set, simulation dynamics over time can be extrapolated for the estimated set of epitopes of our construct since it is also comprised of diverse T cell epitopes and vaccine-specific B cell immunogenic regions as predicted by several immunoinformatic tools. It has been proposed previously that the simulation dynamics can be consistent with a realistic immunization process in terms of primary and secondary immune responses [45]. Likewise, clearance of antigen, production of antibody, development and persistence of memory B cells as well as CD4+ T cells over several months were assumed in the simulation outcome. For primary activation and maintenance of CD8+ T cells, CD4+ T cells (both Th1 and Th2 type) [86] are believed to be required [87,88], where cytokines such as IFN-γ, IL-2 and IL-4 could be involved [89][90][91][92]. The simulation outcome suggests chance for expression of high levels of IFN-γ and IL-2, which may potentiate CD8+ T cell expansion. On the other hand, it is unlikely that the vaccine would trigger clonal expansion of epitope-specific T cells since we combined potent epitopes from several amastigote-associated proteins of comparable affinity, and it was consistent with the simulation dynamics for repeated exposure of 12 doses, as indicated by Simpsonʼs index (D). Rather, high level of IL-2 production can be expected for diverse epitope-mediated immune response functional over long time in vaccine-mediated immunity.
Experimental validation is utmost to prove this computational work. Next phases of the reverse vaccinology approach would ideally involve assessing the recombinant immunogenic protein expressed in the E. coli (strain K12) system as proposed here, in vitro stimulation of peripheral blood mononuclear cells from active VL patients as well as healthy endemic people for cytokine production, and evaluation in challenge models. While a multi-epitope vaccine molecule generated by using a reverse vaccinology approach can induce specific responses in in vivo and in vitro assays, a single recombinant molecule can also reduce the cost of production [93,94]. The in silico designed vaccine reported here confers substantial immunogenic potential to be considered for in vitro experimental evaluation in the next phase of the study.

Conclusions
Screening and design of large-scale subunit/peptide vaccine candidates can be facilitated by a reverse vaccinology approach prior to experimental validation. This modelling study took a systematic approach to apply a series of immunoinformatic tools to extract T cell-specific epitopes from MS-driven human stage-associated L. donovani proteins with secretory potential, and design a subunit vaccine with a broad population coverage. Development of such a prophylactic vaccine for VL may complement therapeutic strategies against active infections as well. Overall, collective approaches of in silico, in vitro and in vivo investigations are utmost to develop a universal subunit vaccine against human VL.