Introduction

Acinetobacter baumannii, an ESKAPE pathogen, has gained the attention of medical fraternity worldwide due to its nosocomial infection in hospital setup mainly ICUs and emergence of multi-drug resistance mechanism in it1,2,3,4,5,6. A. baumannii have developed MDR, XDR, and PDR strain7. Bacterial pathogen interacts with the host and has also developed several strategies to evade the host immune system. Therefore, it is high time to develop suitable therapeutics or vaccine against the A. baumannii.

Previous studies reveal that outer membrane proteins OmpA, biofilm-associated protein, poly-N-acetyl-β-(1–6)-glucosamine8, trimeric autotransporter protein, K1 capsular polysaccharide, outer membrane vesicles (OMV) and formalin-inactivated whole cells could serve as vaccine candidates and provide partial immunity against lethal doses in various mouse models9. Recently, it is shown that subtractive genomics and reverse vaccinology as a powerful tool to identify drug target and vaccine candidates. Although both these approaches have been used separately to design the novel drugs and vaccines against Gram-negative bacteria10,11. Currently, vast information about genomes and proteomes of A. baumannii strains are available and promising vaccine candidates, or novel proteins can be identified using the computational tools12. Subtractive genomics approach subtracts pathogen genes that are required the survival of pathogen but not present in the host13. This is important to find druggable protein, which may be considered for the therapeutics development. The selected druggable proteins may be used for the development of the chimeric-subunit vaccine or multi-subunit vaccine that appears as a very promising and effective treatment option to control the diseases caused by this pathogen14. Once shortlisted, these candidates can be cloned and over-expressed in E. coli and purified by affinity chromatography. Their immunogenicity can be validated in-vivo in suitable animal models.

In addition to essential proteins, virulence factors and resistant determinants also mediates bacterial attachments that may contribute to the pathogenicity of the bacterium15. Cytoplasmic proteins are usually considered for small molecule drug development while membrane or secreted proteins are considered for vaccine development16. Therefore, the present study aims to identify the druggable essential and virulence proteins from the different strains of A. baumannii. The identified proteins will be used for selection of promiscuous immunogenic non-allergenic epitopes. The selected epitopes will be used in the development of chimeric subunit vaccine. The designed chimeric-subunit vaccine can be developed to cure the infection caused by the A. baumannii.

Methods

Data collection of proteome

In the present study, list of all available strains of A. baumannii have been downloaded from UNIPORT server. Acinetobacter species have more than 32 genospecies, which include major four genospecies like A. baumannii, A. calcoaceticus, A. pittil, and A. nosocomialis. A. baumannii was filtered out from the list and used for further analysis. Bacterial proteome redundancy is a barrier to the effective use of the dataset for multiple reasons, removing redundant sequences is desirable to avoid highly repetitive search results for queries that closely match with an over-represented sequence. Hence, all the searched strains from the UNIPORT proteome were separated according to their redundancy and non-redundancy. All the selected 52 proteomes (including reference proteome) were downloaded from UNIPORT database, and 51 proteomes were subjected to BLASTp against reference strain (SDF). The obtained shared proteins were used for further analysis. The proteins having sequence length less than 100 amino acids were also considered17.

Data collection of the genome and phylogenetic analysis

Genome data of selected 52 strains of A. baumannii were obtained from NCBI. INSDC (International Nucleotide Sequence Database Collaboration) numbers were used for the complete genome where as WGS number used for the draft sequence. The whole genome DNA sequence was searched for rRNA sequences using RNAmmer18. One 16S rRNA gene was randomly sampled per strain because there are only small sequence differences among 16S rRNA genes within the same genome and the same species. Phylogeny tree was constructed using MEGA6 where 16S rRNA gene sequences from the genome of the all A. baumannii strains were used19. The alignment program ClustalW was used for multiple sequence alignment of the sequences. From the alignment, a distance neighborhood joining tree was constructed, using 1000 bootstraps to find the best fitting distance tree20.

CD-HIT analysis

Subtractive analysis (Fig. 1) of proteins were performed using CD-HIT to identify the duplicate proteins by clustering techniques. Sequence identity cut-off was kept at 0.6 (60% identity) as sequence having identity >60% similar/related structure and functions21. Global sequence identity algorithm was selected for the alignment of the amino acids. The bandwidth of 20 amino acids and default parameters for alignment coverage were selected.

Figure 1
figure 1

Illustration of predefined comparative and subtractive proteomics systemic workflow.

Screening of essential proteins

The database of essential genes (DEG) (http://tubic.tju.edu.cn/deg/) includes essential protein-coding genes determined by genome-wide gene essentiality analysis. DEG consists of experimentally identified 22,343 essential protein-coding genes and proteins, 646 non-coding RNAs, promoters, regulatory sequences, and replication origins from 31 prokaryotes and 10 eukaryotes22. The queried proteins having homologous hit in DEG are likely to be essential. BLASTp search was performed for the proteome of A. baumannii against DEG bacterial proteins with cut-off parameters of 1e−04 E-value, bit score of 100, BLOSUM62 matrix and gapped alignment mode were selected to screen out the essential proteins.

Analysis of virulence factors (VF’s)

Virulence factors help bacteria to modulate or degrade host defense mechanism with the help of adhesion, colonization, and invasion resulted cause the disease. VFDB, a database consists of four categories of VFs namely offensive, defensive, non-specific and virulence-associated regulated proteins from 25 pathogenic bacteria were used in the present study23. The proteome of A. baumannii was subjected to BLASTp search against the database of protein sequences from VFDB core dataset (R1) with default hit with cut-off bit score >100, and E-value was 0.0001.

Analysis of resistance proteins

ARG-ANNOT (Antibiotic Resistance Gene-ANNOTation) is a bioinformatics tool that detects existing and putative new antibiotic resistance (AR) genes or proteins in bacterial genomes or proteome. ARG-ANNOT contains 1689 antibiotic resistant protein sequences from various classes including aminoglycosides, beta-lactamases, fosfomycin, fluoroquinolones, glycopeptides, phenicols, rifampicin, sulfonamides, tetracyclines, and trimethoprim. A local BLAST program was run for the proteome of A. baumannii in Bio-edit software against antibiotic resistant sequences in ARG-ANNOT with cut-off E-value of 1e−0424.

Selection of non-homologous proteins to host

BLASTp search of the comprised list from the above three independent searches were performed against non-redundant protein sequence (nr) database of the host Homo sapiens (taxid: 9606). The comparison of proteins with human host protein finds the non-hit proteins lists that denote non-human homologous proteins of the pathogens. This will help to design the pathogen specific therapeutics drugs25.

Pathogen-specific pathways

KEGG (Kyoto encyclopedia of gene and genome) is a pathway database26 that is to find metabolic pathways of non-homologous proteins of A. baumannii. Similarly, host metabolic pathways were also enlisted along with their K number. The enlisted host and pathogen metabolic pathways were manually compared to identify the pathways that present only in the pathogen but not in the human host. The list of proteins that plays a role in unique pathways was enlisted. The proteins were also separated according to their role only in pathogen-specific unique pathways and common pathways present in both pathogen and host.

Druggability analysis

Druggability analysis of the short-listed proteins searched against all drug targets present in the DrugBank database27. The drug targets with bit score >100 and E value < 0.005 were considered as a potential drug target. Biological location of druggable proteins was classified based on the consensus location predicted using various online servers like PSORTb28 and CELLO v2.5 (http://cello.life.nctu.edu.tw). The low molecular weights (100–110 kDa) target proteins increase the accessibility value of the targets proteins.

Prediction of antigenic protein

From subtractive proteomics approach, outer membrane proteins were selected that contained the potential for vaccine development. These protein targets were used for the bioinformatics study (Fig. 2) to identify the potent epitopes that can enhance the immune response. Antigenic property of all the selected proteins was determined using VaxiJen web server29. A threshold of 0.5 was considered as the potent antigenicity. Identified potent epitope containing proteins were used for afterward analysis.

Figure 2
figure 2

Illustration of reverse vaccine devolvement workflow.

Protein-protein interaction network analysis

Protein interaction carried to find out the most potential metabolic functional associations among all identified proteins through protein interaction database STRING (http://string-db.org). The STRING database aims to provide a critical assessment and integration of protein-protein interactions, including direct as well as indirect associations.

T cell MHC Class I epitope prediction

Predictions of MHC class I epitopes were made by the NetCTL server30 for the selected proteins. In this server, epitopes were chosen on the basis of the high overall combinatorial score for each peptide’s intrinsic potential. In integrating with several predictions methods such as proteasomal cleavage, TAP transporter associated efficiency with antigen processing, and MHC-I affinity predictor, scores of all three methods were merged together and achieved combined score for all the predicted epitopes. Prediction threshold value 0.75 was set for epitope identification.

At the last stage, 40 T-cell epitopes were subjected to MHC-I binding prediction, using the immune epitope database analysis resource (IEDB AR)31. MHC molecules represent the antigenic peptides which are recognized by the T-cells. IEDB recommended as a default prediction method and uses the consensus method consisting of ANN32, SMM33, and CombLib34 and NetMHCpan35. Identified T-cell epitopes with HLA alleles were selected by IC50 values and percentile rank. Lower the percentile rank the higher the interaction shown between the peptide and MHC molecules. IC50 values divided in three different categories: high binding affinity, IC50 value < 50 nM; intermediate binding affinity, IC50 value < 500 nM; and low binding affinity, IC50 value < 5000 nM36.

Class I immunogenicity prediction

Epitope/MHC complex should have the ability to evoke an immune response. Hence we have used MHC I immunogenicity prediction tool using IEDB server37. Default parameters were selected to perform the immunogenicity prediction. The epitopes which, immunogenicity prediction gained the positive value were selected for further analysis. (http://tools.immuneepitope.org/immunogenicity/).

Antigenic, Conservancy and Toxicity Analysis

All the promiscuous epitopes obtained from immunogenicity tool were analysed using VaxiJen version 2.0 server for their antigenic properties at a threshold value of 0.5. This server depends on Auto Cross Covariance (ACC) transformation, and alignment-independent predicted antigenic epitopes by their physicochemical behaviour. To assess the epitopes conservancy level within genotype sequences, IEDB conservancy analysis38 was used. The sequence identity parameters were set to default. This analysis aims to calculate the degree of conservancy of epitopes within a given protein sequence39. ToxinPred (http://www.imtech.res.in/raghava/toxinpred/index.html) online server tool analysed physicochemical properties of epitopes to predict the toxicity level. It confirms that the specific immune responses which are induced in the host cell will target only the bacteria rather than host tissue40. The parameters were set to default.

T cell MHC Class II epitope prediction

T-cell epitopes binding to MHC Class II molecules were predicted using IEDB-AR server. The T-cell epitopes were computed using the consensus method41,42. The consensus prediction approach uses a combination of both stabilization matrix alignment method and average relative binding matrix method.

Cluster analysis of the MHC restricted Alleles

MHCcluster v2.0 server43 identified the cluster of MHC restricted allele with appropriate peptides to further strengthen our prediction. This is the additional crosscheck of the predicted MHC restricted allele analysis from the IEDB analysis resources. The output from this server is a graphical tree and static heat map for describing the peptides and HLAs functional relationship.

B-cell epitope prediction

Prediction of linear B-cell epitopes for proteins was achieved by using online server BCPREDS, i.e., B-cell Epitope Prediction Server (http://ailab.ist.psu.edu/bcpred/predict.html)44 and FBCPred45 server. BCPred based on five different kernel methods having fivefold cross-validation by SVM. FBCPred utilizes subsequent kernel for the prediction of linear length B-cell epitopes. The cut-off score of BCPreds is >0.8 for prediction of linear B-cell epitopes11. The goal of B-cell epitope prediction was used to determine the antigen recognized by B lymphocytes and initiate humoral immunity. IEDB B-cell epitope prediction server predicted the linear epitopes using biochemical properties such as amino acid composition, hydrophobicity, hydrophilicity, surface accessibility, and secondary structure. This server is consists of BepiPred linear epitope prediction46, Karplus-Schulz flexibility prediction47, Chou-Fasman beta-turn prediction48, Kolaskar Tongaonkar antigenicity49, Emini surface accessibility prediction50, and Parker hydrophilicity prediction51. ElliPro server52 of IEDB identify the linear and conformational epitopes of B-cell. This server predicted epitopes with a score, defined as a Protrusion Index (PI) value averaged over epitope residues. In the method, the protein’s 3D shape is approximated by some ellipsoids. Residues with larger scores are associated with greater solvent accessibility. Discontinuous epitopes are also defined based on PI values.

Construction of model vaccine

To construct the novel vaccine with low toxicity, allergenicity, and highly immunogenicity, we have analysed the different combination of sequence constructs. During this vaccine construction, firstly, sequence 62–106, 359–419 of B0VMD0 (uniport ID), and epitope sequence 688–787, 79–139, 366–382 of B0VUZ6 protein were joined with the help of amino acid linkers. Secondly, epitopes 162–204, 231–336 of B0VMD0 protein, and epitope 174–248, 636–659 of B0VUZ6 proteins were added with the help of linkers. For enhancing the immunogenicity of these two sequence constructs were added with the four different adjuvants L7/L12 ribosomal protein, beta-defensin, HBHA protein (M. tuberculosis, accession no. AGV15514.1), and HBHA conserved sequence53 respectively. PADRE peptide sequences were also incorporated along with the adjuvants. PADRE peptide induced CD4+ T-cells that improve efficacy and potency of peptide vaccine54. Adjuvant HBHA and L7/L12 ribosomal protein are agonists to the TLR4/MD2 complex while beta-defensin adjuvant is agonist to TLR1, TLR2, and TLR4. The TLR’s interaction polarizes CTL responses which have a robust immuno-stimulatory effect53. HEYGAEALERAG and GGGS linkers were conjugated with HTL, CTL and B epitopes, whereas adjuvants sequences were linked with the help of EAAAK linkers at both N- and C-terminus. Yang et al. proved that ‘GGGS linkers superior to ‘AAY’ as a ‘linkers’ of epitope-based vaccines55. The fused vaccine constructs were used for further analysis.

Allergenicity, Antigenicity and solubility evaluation

Allergenicity was predicted using AlgPred program56. AlgPred uses all these parameters (IgE epitope + ARPs BLAST + MAST + SVM) combined to predict the allergenicity of peptide with the accuracy of 85%. A threshold score of prediction has been considered to be −0.40. Prediction score less than threshold value point towards the non-allergic behaviour of the peptide. For identification of vaccine antigenicity, ANTIGENpro program57 (accuracy of 76%) and VaxiJen v2.0 program were used. SOLpro program58, predicts the propensity of vaccine to be soluble upon over-expression in E. coli. Two-stage SVM architecture was employed based on multiple representations of the primary amino acid sequence. The final result was summarizing the predictions with overall accuracy of 74% at corresponding probability (≥0.5).

Secondary structure prediction

The PSIPRED v3.3 program was used59 for the secondary structure analysis (http://bioinf.cs.ucl.ac.uk/psipred/). Psi-Blast was performed to identify and select the related sequences showing significant homology to the vaccine protein. An average Q-3 score of PSIPRED 3.2 server is 81.6%.

Prediction of various physicochemical properties

Vaccine sequences were functionally characterized using Expasy ProtParam server (http://expasy.org/cgi-bin/protpraram). This server is widely used for determining the physicochemical characteristics like the number of amino acids, molecular weight, PI values, hydropathicity GRAVY values, instability index, aliphatic index, and estimated half-life of the protein of a generated protein model. It can compute various physicochemical properties on the basis of pK values of different amino acids. Instability index of protein predicts whether it is stable or unstable. Instability index value is <40 for stable protein and >40 for unstable protein. The volume occupied by the aliphatic side chains is known as the aliphatic index of protein. Grand average of hydropathicity was calculated by the sum of hydropathicity obtained for all of the amino acid residues divided by the total number of amino acid residues present in the protein.

Molecular Docking and Molecular Dynamics Simulation

With the help of the Phre2 online tool, we have made models of all the four vaccine constructs. PDB ID of all HLA alleles were downloaded from protein data bank, RCSB. Molecular docking of the final four vaccine constructs were performed with six different HLA alleles i.e. 1A6A(HLA-DR B1*03:01), 3C5J(HLA-DR B3*02:02), 1H15(HLA-DR B5*01:01), 2FSE(HLA-DR B1*01:01), 2Q6W(HLA-DR B3*01:01), and 2SEB(HLA-DRB1*04:01) were performed using PatchDock to show HLA-peptide interactions. For more, refinement and re-scoring of rigid body molecular docking score, FireDock (Fast Interaction Refinement in Molecular Docking) server was used. It gives best 10 solutions for final refinement. The refined models were based on the binding score and global binding energy. Similarly, docking of vaccine construct (V4) with TLR 4/MD2 complex (PDB ID 2Z65) was performed by PatchDock server. The generated top 10 models were refined and re-scored by FireDock server. The refined candidates were ranked by their respective binding energy. Molecular dynamics simulation of V4-TLR4 complex was performed using Gromacs v5.1.2 as published method60.

Codon optimization of the vaccine construct and In-silico cloning

Java Codon Adaptation Tool (JCAT) was performed to adapt the codon usage of vaccine to E. coli host strain61. Vaccine amino acid sequence was back-translated to DNA, and subsequently adapted for codon usage to E. coli. The adaptation was based on Codon Adaptation Index values (CAI), that were computed by employing an algorithm. The rho-independent transcription terminators, prokaryotic ribosome binding sites and cleavage sites of some restriction enzymes were avoided. Moreover, to clone the adapted gene sequence of final vaccine construct in E. coli pET28a vector, using Snapgene tool to ensure the vaccine construct expression.

Data availability

All data generated or analysed during this study are included in this published article and its Supplementary Information files.

Ethical approval

The present study does not involved human or animal samples.

Results

Subtractive proteomics approach shortlisted A. baumannii strains

We have downloaded the list of 1753 different strains of Acinetobacter genospecies from the UNIPORT. A total of 1578 strains of Acinetobacter baumannii were filtered from the lists and used for further analysis. All the 1578 strains of A. baumannii were manually separated according to their redundancy and non-redundancy. In this manual comparison, we have made 29 different groups of non-redundant strains that contain the list of redundant strains. Twenty-three strains of A. baumannii did not show any redundancy with the other Acinetobacter strains. Therefore a total number of 52 (29 + 23) strains were selected for the analysis. Proteome of A. baumannii SDF strain was considered as reference proteome that contains most representative and best annotated set of proteins of all A. baumannii strains. The proteomes of all 52 A. baumannii strains were retrieved from UNIPORT database. With the help of BLASTp, we have found the similar protein present in the all 51 proteomes of A. baumannii with respect to reference proteome SDF.

Phylogenic analysis of selected 52 A. baumannii strains showed their inter-relation

INSDC numbers or WGS numbers of all 52 strains of A. baumannii were downloaded from NCBI (Supplementary Table ST-1). The results of RNAmmer analysis yielded no rRNA sequences for five genomes (A. baumannii 118362, 855125, 940793, ABBL059 and ABBL059). These five genomes would have some unknown base stretches that prevent RNAmmer from identifying rRNA sequences20. Genome sequence lack 16S rRNA sequences that might be a result of sequence assembly. Since ribosomal RNA sequences often are repeated sequences, the assembly process might not be able to conclusively place the rRNA in the DNA and might discard the sequences all-together62. Distance-based phylogenic analysis measured dissimilarity observe between pairs of sequence alignment. Length of branch represented an amount genetic change of 0.05. The sum of branch length (SBL) value 0.437 was observed (Supplementary Figure SF1) in the analysis of distance based phylogenic tree.

Identification of essential proteins of A. baumannii

From the DEG result, 811 proteins of A. baumannii were found to be essential, and rest of them were known as non-essential proteins. These 811 essential proteins are necessary for the survival of A. baumannii. Blocking these bacterial proteins will cause the death of the micro-organism, which will make the protein targets more important in drug discovery.

Identification of virulence factors of A. baumannii

Analysis of virulence factors explored the significance of bacteria in various diseases necessitates the search for novel VFs. The VFDB result showed that 339 proteins were found to be associated with virulence of A. baumannii. These proteins may also be considered as a very important target to inhibit the pathogenesis of A. baumannii.

Identification of resistant determinant in A. baumannii

Protein involved in the resistance and drug efflux proteins could be act as potential therapeutic targets to inhibit the resistant strain of A. baumannii. We have obtained 10 antibiotic resistant proteins with cut-off E-value of 1e−04. These ten proteins are involved in the degradation of antibiotics and efflux of the antibiotics.

Identification of non-human homologous proteins in A. baumannii

The manual comparative analysis of the proteins showed that 35 proteins contained the both essential and virulence property. The search of the non-redundant proteins of essential, virulence and resistance datasets is necessary to evaluate homology of these protein with human, to find non-human homologous proteins in A. baumannii that further enhances the specificity of the designed drug or vaccine. We have found 174 VF’s protein, 347 essential protein and 5 antibiotic resistance protein that are non-human homologous protein. Remaining 165 VF’s, 459 essential and 3 resistance proteins showed similarity with human proteins, therefore, filter out from further use. Detection of non-human homologous proteins to find out the proteins were likely to lead to drug development exclusively interact with proteins of A. baumannii.

Involvement of selected A. baumannii proteins in the unique and common pathway

We have done the manual comparison between the enlisted metabolic pathways of A. baumannii and pathways of human (present in KEGG Database). The result showed that 33 unique pathways (Supplementary Table 2) are present in the A. baumannii pathogen and rest 104 pathways were common in both bacteria and host. These 33 unique pathways were found to possess 126 proteins. Out of 126 proteins, 34 proteins play a role in unique A. baumannii specific pathways while 92 proteins have a role both in unique and common pathways of pathogen and human. Therefore, 34 proteins have been selected for further analysis. The 358 proteins are not present in any metabolic pathway are considered as pathway independent protein analysis.

Identification of putative drug targets for therapeutics development

Pathway dependent protein target

Druggability analysis showed that only two proteins of unique pathway were druggable in nature that include penicillin-binding protein 1B protein (that have role in peptidoglycan biosynthesis) and channel-tunnel spanning the outer membrane and periplasm segregation of daughter chromosomes protein (that have role in bacterial secretion system, two-component system, beta-lactam resistance and CAMP resistance pathway). The detailed drug name, drugbank ID, E-value and bit score value of the target proteins are mentioned in Table 1. These proteins uniquely play a role in different pathways of A. baumannii thus can be targeted for vaccine or drug discovery.

Table 1 Novel druggable targets involved in different metabolic dependent or independent pathways.

Pathway independent protein target

The 358 metabolic pathways independent proteins of A. baumannii screened for druggability test. The result showed the 11 druggable target that includes DNA gyrase subunit A, bifunctional purine biosynthesis protein PurH, ferredoxin-NADP reductase, malonyl coA-acyl carrier protein transacylase, dihydropteroate synthase, pyridoxine 5′-phosphate synthase, putative UDP-glucose 6-dehydrogenase (Ugd), putative acetyl-coA synthetase/AMP-(Fatty) acid ligase, and 3-oxoacyl-(acyl-carrier-protein) reductase proteins were present in the cytoplasm whereas putative D-ala-D-ala-carboxypeptidase, penicillin-binding protein, putative oxidoreductase, and short chain dehydrogenase/reductase family present at the membrane choose as druggable targets on the basis of E-value and the bit score value. The detailed drug name, drugbank ID, the E value and bit score value mentioned in Table 1.

Identification of subcellular localization and prioritization of selected target proteins

Although non-homologous proteins of the pathogen represent potential targets for therapeutic and vaccine candidates, prioritization and more filtration of the identified proteins could help to minimize the time, labor and resources for developing the therapeutic agent and optimize the success of getting the best drug and/or vaccine against the pathogen. Hence, additional parameters that determine the suitability of a drug and vaccine target were used to characterize the identified target proteins. The different characterizations of proteins like, sub-cellular localization by CELLO and PSORTb, molecular weights by Swiss-Prot database, and TMHMM which identify trans-membrane proteins. Prediction of sub-cellular localization of druggable targets using various online search tools showed that in the pathway dependent proteins both proteins presented in the outer membrane. In the pathway independent proteins, 9 proteins presented in the cytoplasm and two proteins were presented on the membrane. Cytoplasmic proteins can be considered for small molecule drug development while membrane or secreted proteins for vaccine development. On the other hand, the number of trans-membrane helices (Table 2) was set to be less or not more than 1. Proteins that were identified to have more than one trans-membrane helices were excluded due to the difficulties in cloning, expression and purification of proteins exhibiting multiple trans-membrane spanning regions.

Table 2 Identification of durggable targets characteristics and cellular localization.

Selection of antigenic membrane proteins

After the prioritization of target proteins, we found the four proteins (B0VUZ6, penicillin-binding protein 1B; B0VMD0, channel-tunnel spanning the outer membrane and periplasm segregation of daughter chromosomes; B0VTR5, putative D-ala-D-ala carboxypeptidase penicillin-binding protein; and B0VT18, putative oxidoreductase, short-chain dehydrogenase/reductase family) were present on the membrane hence considered for vaccine candidate molecules. VaxiJen web server showed that B0VUZ6 and B0VMD0 were identified as the most potent antigenic protein having a maximum total prediction score of 0.5541 and 0.5789 respectively (Supplementary Table ST-3). B0VMD0 is a protein involved in virulence while B0VUZ6 is an essential protein of A. baumannii. These two proteins with diverse role, and may be better situated for the development of vaccine candidates as they inhibit virulence as well as essential for the normal survival of A. baumannii. Hence, these two proteins are used for development of chimeric subunit vaccine against A. baumannii.

Intra-species interaction of selected membrane proteins with other proteins

Penicillin binding protein 1A family protein (Fig. 3B) show interaction with the mrcB (murein polymerase), Dac F(serine-type D-Ala-D-Ala carboxypeptidase), F911_00507 (penicillin binding protein, transpeptidase domain protein), mrdA (penicillin binding protein 2), murG (UDP-diphospho-muramoylpentapeptide-beta-N-acetylgluco-saminyltransferase), mtgA(mono-functional biosynthetic peptidoglycan transglycosylase), HMPREF0010_01990 (serine type D-Ala-D-Ala carboxypeptidase), murC (UDP-N-acetylmuramate-alanine ligase), F911_01113 (uncharacterized proteins), and HMPREF0010_02514 (Type IV pilus assembly protein PilM). Out of these, murG, murC, and mtgA play role in cell wall formation and mrcB, plays a role not only in cell wall formation but also in the synthesis of cross-linked peptidoglycan from the lipid intermediates. Similarly, channel-tunnel spanning the outer membrane and periplasm segregation of daughter chromosomes protein (Fig. 3A) exhibited interactions with macB (macrolide export ATP binding/permease protein), F911_01966 (putative acriflavine resistance protein A), acrA (acriflavine resistance protein A), bpeE (multidrug efflux pump BpeE), F911_01943 (uncharacterized protein), F911_01944 (GH3 auxin-responsive promoter), F911_01946 (uncharacterized protein), F911_01941 (efflux pump membrane protein), F911_01014 (putative ATP synthase F0,A subunit) and bpeB (inner membrane multidrug efflux protein BpeB). These protein are very crucial in the survival of the A. baumannii. Therefore, targeting these two proteins (B0VUZ6 and B0VMD0) also influence the interacting proteins.

Figure 3
figure 3

Protein-protein interaction diagram of (A) Channel-tunnel spanning the outer membrane and periplasm segregation of daughter chromosomes (B) penicillin binding protein predicted by STRING tool.

Selection of potent T-cell epitopes in proteins shortlisted for chimeric vaccine design

On the basis of the high combinatorial score, best T-cell epitopes were predicted by NetCTL server using protein sequence of these two selected proteins. The software identified 17 epitopes in B0VMD0 proteins while 27 epitopes in B0VUZ6 (Supplementary Table ST-4). Peptides with a higher score represent higher processing capabilities. T-cell epitopes of both proteins were subjected to MHC-I binding prediction, using the IEDB server. The epitopes that elicited higher affinity (IC <200 nM) and percentile rank (≤0.2) were subjected to afterward analysis (Supplementary Table ST-5). These epitopes were further filtered on the basis of their interaction with class I MHC molecules. On the basis of MHC I molecules interaction, 10 out of 17 epitopes of B0VMD0 antigenic protein and 12 out of 27 epitopes of B0VUZ6 were selected for further analysis.

Selection of Immunogenicity in selected T-cell epitopes

Besides the epitope predictions, the binding affinity between peptide/MHC complex and TCR was further analysed. A high immunogenicity score is deemed to have high ability to stimulate naive T cells and induce cellular immunity. The epitopes selected above (10 from B0VMD0 and 12 from B0VUZ6) using the above methods were subjected to IEDB immunogenicity prediction. The epitopes’ immunogenicity scores ranged from 0.24661 to −0.04637. In this analysis, 6 out of 10 epitopes of B0VMD0 protein and 8 out of 12 epitopes of B0VUZ6 contain the positive value score. These 14 (6 + 8) epitopes were selected for the further analysis (Supplementary Table ST-5).

Toxicity, conservancy and antigenicity prediction of selected 14 epitopes

To confirm the specific immune responses induced by epitopes, ToxinPred was used to confirm epitopes non-toxicity. Epitopes toxicity analysis score (Table 3) confirmed that the selected 14 epitopes were non-toxic. Conservancy status of selected epitopes of B0VMD0 and B0VUZ6 were determined using IEDB conservancy analysis. Epitope which showed more than 50% conservancy status were selected for further analysis. The result showed that three epitopes of B0VMD0 like DVSEANAQY (70%), SSSFALDLV(60%) and HVLNVAEAY(60%) are conserved in 10 other genotype sequences. Similarly, in the protein B0VUZ6, all the 8 epitopes showed conservancy score more than 50% such as MALLIELHY (96.49%), ISTEDRNFY (96.49%), LSAIESGRY (90.35%), STFTNNLRK (85.56%), LLDRYGLNV (92.98%), SSGTGRAAY (85.96), GVESTIPAY (70.18) and YVHTRGFDY (60.53%) and found to be conserved in 114 other genotype sequences. Potent epitope antigenicity of selected 11 epitopes (3 of B0VMD0 and 8 of B0VUZ6) were analysed using VaxiJen web server. Result showed that, a total of seven epitopes (2 of B0VMD0 such as DVSEANAQY, SSSFALDLV, and 5 of B0VUZ6 such as MALLIELHY, LLDRYGLNV, SSGTGRAAY, ISTEDRNFY and YVHTRGFDY) contained the potent antigenicity (Table 3). These 7 epitopes were selected for further study.

Table 3 Predicted MHC class I epitopes Toxicity, Antigenicity and conservancy analysis.

Selection of MHC-II epitopes in proteins shortlisted for chimeric vaccine design and their conservancy analysis

In addition to MHC-I epitope predication, both the selected proteins (B0VMD0 and B0VUZ6) were further subjected to MHC-II binding prediction, using IEDB server. The epitopes that elicited higher affinity (IC <200 nM) and low percentile rank were subjected to afterward analysis (Table 4). Conservancy analysis showed that two epitopes of B0VMD0 showed the conservancy more than 50% such as LEQLNMMNAKLKEGL (60%) and MVDVLLAQRNAFSAK (70%) in 10 other genotype sequences. Similarly, six epitopes of B0VUZ6 showed the conservancy more than 50% which include FERGIGFFALIFSIL (84.21%), ALSIYLIRLDNIIRE (92.11%), GRAAYNSLSPALKLA (64.04%), LSTFTNNLRKFGVES (85.96%), FTGFNRALDAKRQVG (78.01%), STEDRNFYHHHGISI (85.09%). These 8 epitopes (2 of B0VMD0 and 6 of B0VUZ6) were selected for further analysis.

Table 4 Predicted of MHC class II epitopes by IEDB server and conservancy analysis.

MHC restriction and cluster analysis of selected epitopes

Further, MHC class I and MHC class II restricted allele were further analysed on the basis of the IC50value. All the predicted epitopes were assessed for the MHC interaction analysis independently for 7 MHC-1 epitope and 8 MHC-II epitopes. Furthermore, the interacted alleles were reassessed by cluster analysis and results are shown as a heat map of MHC-1 (Fig. 4A) and MHC-II (4B), and dynamic tree (Supplementary Figure SF-2). Epitopes shown in Fig. 4 are clustered on the basis of interaction with different HLA alleles and red colour indicating strong interactions while yellow colour shows weaker interactions with appropriate annotation63.

Figure 4
figure 4

Cluster analysis of the HLA alleles for both MHC molecules through heat map representation. (A) Representing the cluster of the MHC-I. (B) Representing the cluster of MHC-II molecules. Epitopes are clustered on the basis of interaction with HLA and shown as red colour indicating strong interaction with appropriate annotation. Yellow zone indicates the weaker interaction. Here, all the available alleles are shown.

Selection of B-cell epitopes in proteins shortlisted for chimeric vaccine design

In addition to MHC-I and MHC-II epitopes (cellular immunity), these two proteins were also used to predict the B-cell epitope (humoral immunity). Humoral immunity is also required for bacterial elimination. To predict B-cell epitopes, firstly location of linear B-cell epitopes was identified using BCPREDS, FBCpred, and IEDB (Supplementary Tables ST6 and ST7). Resulted epitopes were further shortlisted on the basis of BepiPred linear epitope prediction, Karplus-Schulz flexibility prediction, Chou-Fasman beta-turn prediction, Kolaskar Tongaonkar antigenicity, Emini surface accessibility prediction, and Parker hydrophilicity prediction. The result of predicated B-cell epitopes of B0VMD0 and B0VUZ6 are shown in Figs 5 and 6 respectively. Interaction of shortlisted 10 B-cell epitopes (4 of B0VMD0 and 6 of B0VUZ6) with the HLA-II allele’s protein was performed using IEDB and result is shown in Supplementary Table ST-8. In the result, highest binding HLA alleles were selected on the basis of percentile rank (<0.2) and IC50 value (<200 nM).

Figure 5
figure 5

Protein ID B0VMD0 B cell epitope (A) Bepipred Linear Epitope, (B) Chou & Fasman Beta-Turn Prediction, (C) Emini Surface Accessibility Prediction, (D) Karplus & Schulz Flexibility Prediction, (E) Kolaskar & Tongaonkar Antigenicity, (F) Parker Hydrophilicity Prediction.

Figure 6
figure 6

Protein ID B0VUZ6 B cell epitope (A) Bepipred Linear Epitope, (B) Chou & Fasman Beta-Turn Prediction, (C) Emini Surface Accessibility Prediction, (D) Karplus & Schulz Flexibility Prediction, (E) Kolaskar & Tongaonkar Antigenicity, (F) Parker Hydrophilicity Prediction.

Comparison of all the predicted epitopes to select the final epitopes

For the construction of final chimeric subunit vaccine sequence, predicted B-cell epitopes were treated as a template for MHC-I, and MHC-II epitopes and those epitopes whose sequences were found to be overlapping in B-cell epitopes were shortlisted and selected for final vaccine constructs (Supplementary Tables 9 and 10).

Construction of final vaccine with different adjuvants, linkers, and PADRE sequence

An adjuvant is an ingredient of a vaccine that helps create a stronger immune response in the patient’s body. Different adjuvants were chosen in the vaccine constructs to analyze the effects of the adjuvants on the allergenicity and antigenicity of the constructs. Along with adjuvants sequence, PADRE sequence was also added to overcome the problems caused by polymorphism of HLA-DR molecules in the worldwide population. It showed the 100-fold more potency than the universal T helper epitopes64, and it was explained that PADRE containing vaccine construct exhibited better CTL responses and protection than the vaccine without the T helper epitope64. A total of eight vaccine constructs were prepared and further analyzed. All eight vaccine constructs were added with the respective adjuvants with the help of EAAAK linker. MHC-I, MHC-II, B-epitopes and PADRE sequence were joined together by HEYGAEALERAG and GGGS linkers. All the epitopes were joined together with these two linkers because they are reported not to alter the conformation of designed vaccine construct55. Details of the vaccine constructs have been mentioned in Table 5.

Table 5 Allergenicity prediction of all vaccine construct using Algpred server.

Allergenicity, antigenicity and solubility prediction of different vaccine constructs

AlgPred server was used to predict the non-allergic behaviour of vaccine constructs. Allergenicity value of all eight vaccine constructs were shown (Table 5). Predicted vaccine constructs score indicates that four constructs (first vaccine construct with HBHA adjuvant (V1), with HBHA conserved sequence (V2), with beta-defensin (V3) as well as second vaccine construct with beta-defensin (V4) were found to be non-allergic in nature. Antigenicity of shortlisted four non-allergenic vaccine constructs (V1, V2, V3, V4) were further predicted using ANTIGENpro and followed by VaxiJen 2.0. (Supplementary Table ST-11). The predicted antigenicity values were more than 0.90 in ANTIGENpro and 0.75 in VaxiJen 2.0 that shows a good antigenic nature of these four vaccine constructs. SOLpro server predicts the protein solubility and corresponding probability should be ≥0.5. All four vaccine constructs showed the solubility more than 0.7 (Supplementary Table 11), which exhibited that vaccine construct will be highly soluble during its heterologous expression in the E. coli.

Physicochemical analysis of vaccine constructs

Various physicochemical properties such as molecular weight, number of amino acid, PI value, aliphatic index, hydropathicity index and instability index, of the all four constructs were calculated from ProtParam server (Supplementary Table 12). The molecular weights of constructs were found between 41–59 kDa that shows its good antigenic properties. GRAVY (a hydropathic index) was found to be less than −0.5 (a negative value), which indicates hydrophilic nature of constructs. Aliphatic and stability value also showed that all the four vaccine constructs have good characteristics to initiate an immunogenic reaction.

Structure prediction of selected four vaccine constructs

Secondary structures of final four vaccine constructs were identified using PSIPRED. The predicted structure of all four vaccine constructs have shown to have 55–60% alpha helix, 10–15% extended strand, 7–12% beta turn, and 20–30% coil structure (Fig. 7). The models of all four vaccine constructs were generated and the Ramachandran plot was used to validate these models. Modeled tertiary structure of vaccine 4 and Ramachandran plot have been shown in Fig. 8.

Figure 7
figure 7

Secondary structure prediction of vaccine constructs (V1 to V4) using PESIPRED server.

Figure 8
figure 8

Tertiary Structure prediction and validation of vaccine construct V4. (A) Tertiary structure of model construct V4. (B) Ramachandran plot of the modelled V4 showing 92.0 residues in the allowed region.

Molecular docking

Docking of vaccine constructs (V1 to V4) with HLA allele’s protein

Linear T-cell epitopes binds to HLA molecules and activate the adaptive immunity against pathogen. The immune response elicited against the epitopes may be genetically restricted or epitopes may be recognized by one individual but not by other65. Hence, a vaccine construct against A. baumannii should have a potential to induce immune response against the number of epitopes, which can be recognized in the context of different HLA allele’s proteins. Promiscuous vaccine can bind to different HLA allelic proportion of human populations. We have docked all four vaccine constructs with the six different HLA allele’s protein. Vaccine construct 4 (V4) have the lowest global binding energy value with different HLA alleles i.e. 2FSE (HLA-DR B1*01:01); −45.02, 2SEB (HLA-DR B1*04:01); −32.17, 1H1S (HLA-DR B5*01:01); −12.92, 1A6A (HLA-DR B1*03:01); −0.90, 3C5J (HLA-DR B3*02:02); −11.22, and 2Q6W (HLA-DR B3*01:01) −8.74 as shown in Table 6. On the basis of the different criteria, we have analysed all four different constructs and finalized the one suitable, and best vaccine construct i.e. V4 that can control A. baumannii infection.

Table 6 Docking score of different vaccine construct (V1 to V4) with the different HLA alleles.

Docking, and Molecular Dynamics Simulation of vaccine construct (V4) and TLR 4

TLR 4 is agonist to beta-defensin. The interaction between the TLR4 and adjuvant enhance the immune response. Therefore, we have performed docking between vaccine construct (V4) with TLR 4. The result showed that docking score of 17970 (−31.11 binding energy) that showed good interaction between vaccine construct V4 and TLR-4/MD2 complex (Fig. 9). The molecular dynamics simulation confirms that V4-TLR-4/MD2 complex get stabilised after 2 ns and remain stable (Fig. 10).

Figure 9
figure 9

Docked complex of vaccine construct V4 with human TLR4-MD2 complex. The vaccine construct docked within the TLR-4 receptor.

Figure 10
figure 10

Molecular dynamics simulation of V4- TLR4-MD2 complex. The result shows the RMSD obtained for the complex which showed that complex is stable after 2 ns at 0.8 nm.

In-silico cloning of chimeric vaccine construct (V4) for its heterologous expression

Reverse translation and codon optimization were performed from Codon Usage Wrangler, in order to evaluate the cloning and expression of multi-epitope vaccine within the expression vector. Reverse translation generates cDNA sequence, which was further analyzed by codon optimization analysis. Codon optimization evaluates the sequence on the basis of GC content of the cDNA sequence. The optimum range was found to be between 30–70% for all the four constructs. This is followed by CAI calculation. The high value of CAI indicates high expression of the gene. Our constructs have the CAI value in the range of 0.95–1.0 (Supplementary Table ST-11). Final vaccine construct (V4) was cloned into pET28a vector for its heterologous cloning and expression in E. coli (Fig. 11).

Figure 11
figure 11

In-silico restriction cloning of gene sequence of final vaccine construct V4 into pET28a expression vector showing V4 sequence red colour surrounded between BglII (401) and AscI (1543).

Discussion

Acinetobacter baumannii infection has emerged as a severe problem, reasons for a large number of deaths in worldwide. Despite this, there is no permanent cure and prevention for the multi drug resistant A. baumannii. Some approaches have been tried which includes the screening of herbal compounds66,67,68, nanomaterial based approach69,70 and in-silico approach71,72 to find an alternative to the current drug used against A. baumannii. Similarly, subtractive genomics of different strains of A. baumannii has also been tried to identify the druggable targets for A. baumannii10. Recently, proteome based approaches are used to develop subunit vaccine73.

In the present study, we have used comparative proteomes of 52 different strains of A. baumannii to identify their shared and unique features such as essential proteins, resistant determinants, and virulent proteins. Identified membrane proteins with the role in virulence and essential survival of A. baumannii, were selected for development of vaccine construction. The immune system reacts to some of foreign molecules with the help of T and B cells and generates efficient immune response against pathogens. Interestingly, this immune response has been replicated with synthetic peptides or epitopes against various infectious organisms, including ESKAPE pathogen like A. baumannii. Promiscuous epitopes were designed against MHC-class-I, II alleles and B cells, using protein sequence of selected two membrane proteins i.e. channel-tunnel spanning the outer membrane periplasm segregation of daughter chromosomes proteins and penicillin binding protein 1B. Only those peptides, having 100% binding affinity towards all the HLA alleles and IC50 values < 200 nM were considered as promiscuous peptides. Predicted peptides were analysed for allergenicity, antigenicity, and solubility, physiochemical properties, etc. Final vaccine constructs were designed with the help of different adjuvants and amino acids linkers. Furthermore, docking analysis was done to explore the binding affinity of promiscuous epitopes with the different HLA molecules i.e. DRB1*0101, DRB3*0202, DRB5*0101, DRB3*0101, DRB1*0401, and DRB1*0301. To analyse the effect of adjuvant and enhance the immune response against vaccine construct, we have also checked interaction between vaccine construct V4 with the TL4/MD2 complex. In the present study, promiscuous peptides predicted, and analysis of promising epitopes by using in-silico tools are unique peptides to best of our knowledge. Chimeric subunit vaccine development against the A. baumannii may use to generate the immune response against all the finalized epitopes (epitopes of chimeric peptides of finalised proteins) and potential to bind with more than one HLA alleles. Therefore, the present study will help to develop a suitable therapeutics against A. baumannii and may help to reduce the mortality and morbidity caused by its infection. In addition to the therapeutic use of the vaccine, people in intensive care units or in hospitals setup can be vaccinated before admission in order to protect them from this nosocomial pathogen. In the vaccine construct, we have added adaptor beta-defensin, Pan-DR epitopes, linker along with multi-epitope sequences that may enhance the significant A. baumannii specific immune responses. It is reported that PADRE sequence will reduce the polymorphism of HLA DR molecules in the population54. In the murine model, vaccine showed the very high CTL responses than the vaccines without Pan-DR sequence64. Murine beta-defensin act as ligands for TLR4 to promote the maturation of dendrite cells74,75. In addition, with G-rich linkers in vaccine enhanced the immunogenicity of the multi-subunit vaccine constructs55. Therefore, we have taken all important factors that can induce the immunogenicity of our vaccine constructs.

Conclusion and future prospects

In the present study, we have attempted two approaches subtractive proteomics and reverse vaccinology approach to finding out the suitable non-human homologous targets for the development of vaccine construct. Selected vaccine targets proteins were used to develop the chimeric subunit vaccine. This study started with the retrieval of all A. baumannii strains and shortlisted them according to their redundancy and non-redundancy. BLAST of non-redundant strains was performed with respect to reference proteome and identified the shared proteins. All proteins were shortlisted on the basis of their essentiality, virulence factors, resistance gene, non-human homologous proteins and druggability analysis. After shortlisted, thirteen proteins were screened as suitable druggable targets. Out of which four proteins were present on the membrane and two proteins were identified as suitable vaccine antigenic targets. This step was followed by the prediction of immunogenic B-cell and T-cell epitope to generate the humoral and cell-mediated immunity, respectively. Predicted epitopes were merged using suitable linkers and adjuvant to enhance the immunogenicity and effective separation of epitopes within the human body. Allergenicity, conservancy, toxicity, antigenicity and secondary structure analysis were also confirmed followed by the physiochemical properties evaluation. Molecular docking was also performed to check the binding affinity and stability of HLA molecules and TLR-4 with respect to vaccine complex. At last, in-silico cloning was performed to ensure the stability and effective expression of vaccine construct (V4). Furthermore, the proposed vaccine needs to be experimentally validated to ensure its use to control A. baumannii infections by effective immunological memory. Our predicted in-silico results were based on diligent analysis of sequence and various immune databases. In-silico studies save both time and costs for researchers and can guide the experimental work, with higher probabilities of finding the desired solutions and with fewer trial and error repeats of assays. The proposed vaccine candidates need a validation in animal models.