Design of a multi-epitope Staphylococcus aureus vaccine integrating a protein A variant, α-hemolysin, and fibronectin-binding protein A through reverse vaccinology Authors: Hannia Michelle Prieto-Nevárez, Alejandro Tarango-García, Arturo Gutiérrez-Guerrero, Aristóteles Álvarez-Cardona, Saul O. Lugo-Reyes Published: 2026 ABSTRACT Staphylococcus aureus has become a major public health concern because of its increasing antibiotic resistance. Given limited clinical success of previous S. aureus vaccine candidates, this study aimed to design and computationally evaluate a novel multi-epitope construct integrating sequences derived from fibronectin-binding protein A, α-hemolysin, and a detoxified protein A variant. Candidate epitopes were computationally predicted and evaluated to determine immunogenicity, toxicity, and cross-reactivity. The epitopes were then joined using linkers with the addition of an adjuvant and an enhancer sequence to improve immunological response. The final construct showed favorable physicochemical properties, predicted solubility and high structural quality after refinement with 97.115% of amino acids in favored regions of the Ramachandran plot. Docking analysis supported plausible binding of selected epitopes to their representative HLA class I and II molecules. The complete vaccine construct showed stable in silico interaction with the TLR4/MD-2 heterodimer and a persistent receptor-vaccine interface throughout the 100 ns of the molecular dynamics simulation. The immune simulation predicted a predominantly humoral and CD4+ T helper response, with increments of B-cells, IgM and IgG1 production, memory CD4+ T-cells, macrophages, IFN-γ, IL-2, and IL-12. Codon optimization and in silico cloning supported the recombinant expression of the vaccine construct in Escherichia coli. Together, these findings identify this multi-epitope construct as a promising candidate for further experimental evaluation against S. aureus infection.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-54286-6. Introduction Staphylococcus aureus is a Gram-positive bacterium responsible for a broad spectrum of acute and chronic human infections, ranging from skin and soft tissue infections to osteomyelitis, respiratory tract infections, infective endocarditis, surgical site infections, device-related infections, and bacteremia1,2. Importantly, S. aureus remains a major cause of mortality worldwide, with bacteremia having an estimated fatality rate of 15–30% with around 300,000 deaths per year3. In addition to its role as an invasive pathogen, it is also a common colonizer of healthy individuals, with approximately 30% of the population colonized, while an additional 30% are intermittently colonized, thereby creating an important reservoir for transmission, recurrent infection and invasive disease4. S. aureus has become a major health threat due to its growing antibiotic resistance. For example, methicillin-resistant S. aureus (MRSA) causes 10 to 40% of bloodstream infections5,6. In the European region alone, an estimated 133,000 deaths were attributable to antimicrobial resistance in 2019, with MRSA identified as the leading pathogen-drug combination for deaths attributable to antimicrobial resistance in 27 countries, while S. aureus accounted for around 83,325 deaths associated with antimicrobial resistance7. This antibiotic resistance arises through various bacterial mechanisms, such as transduction, conjugation, transformation, transposons, integrons, or even accumulated mutations, thereby complicating the treatment and increasing morbidity and mortality8,9. Worldwide, antibiotic resistant bacteria are estimated to cause 700,000 deaths per year, a figure projected to reach 1 million deaths annually by 205010. This emphasizes the urgent need to develop new interventions, from novel antibiotics to the use of prophylactic vaccines11. In this context, reverse vaccinology has become a promising approach for S. aureus vaccine development because it may help reduce the use of antibiotics and also limit the development of antimicrobial resistance12. In multi-epitope vaccine design, bioinformatics-driven screening enables the early selection of conserved, surface-exposed, immunogenic, and non-toxic targets, while also permitting the exclusion of sequences with unfavorable allergenic or structural features before experimental testing13. As a result, reverse vaccinology improves the efficiency, safety, and cost-effectiveness of vaccine development by reducing dependence on extensive wet-laboratory screening, minimizing unnecessary experimental burden, and supporting the rational design of vaccine candidates intended to reduce reliance on antibiotics and help prevent the emergence and spread of resistant infections14. This strategy is particularly relevant in the case of S. aureus, for which several vaccine candidates have shown promising immunogenicity yet failed to provide consistent clinical protection in humans. For example, the capsular polysaccharide-based vaccine StaphVAX showed early protection to end-stage renal disease patients receiving hemodialysis, but this protection was limited in duration, with efficacy evident for approximately 40 weeks after vaccination15. Likewise, preoperative administration of the V710 vaccine was later associated with increased mortality among vaccine recipients who developed postoperative S. aureus infection, suggesting low IL-2 concentrations as one possible contributing factor16. More recently, SA4Ag vaccine was protective in preclinical models, yet it did not reduce postoperative invasive S. aureus infections even though it induced functional immune responses to each antigen compared with placebo, suggesting that adjuvants may be needed to enhance immunogenicity and that animal models don’t necessarily correlate with human protection3,17. Together, these outcomes suggest that previous vaccine strategies encountered limitations, potentially related to antigenic scope, the nature of the induced immune responses, and the difficulty of translating preclinical findings into clinical protection. In this context, effective vaccination against S. aureus will likely require broader and more rationally selected antigenic combinations capable of engaging both humoral and cellular immune responses13. To address these limitations, we designed a multi-epitope vaccine using a reverse vaccinology approach that combines sequences from three clinically relevant virulence factors of S. aureus: the fibronectin-binding protein A (FnBPA), α-hemolysin (Hla), and a mutant form of protein A (SpA) that disrupts IgG-binding while preserving immunogenic epitopes. FnBPA contributes to adhesion, biofilm formation in MRSA strains, obstructing immune cell infiltration, and phagocytosis171819. Hla is a pore-forming toxin that damages the membranes of human platelets, leukocytes, and endothelial and epithelial cells, and also promotes blood coagulation by activating platelets20. Lastly, SpA is a virulence factor that helps the bacteria evade the immune system, acting as a B-cell superantigen that non-specifically activates the immune response by binding to the Fc region of immunoglobulins, thereby disabling their function21,22. This immune evasion presents a challenge in vaccine development, therefore, in this study, the SpA epitope is replaced with a variant, SpAKKE that contains Q9K, Q10K and S33E amino acid substitutions. This variant has been shown to effectively elicit neutralizing antibodies eliciting SpA-specific IgG responses in mice and mouse models, protecting against S. aureus colonization or invasive bloodstream infections22. By combining epitopes from three nonredundant virulence factors with complementary roles in adhesion, toxin mediated injury, and immune evasion, and by incorporating adjuvants and immune response enhancers, this construct was designed to address key limitations of previous S. aureus vaccine candidates and to promote broader humoral and cellular immune coverage. To our knowledge, this is the first reported multi-epitope S. aureus vaccine to integrate the SpAKKE variant together with Hla and FnBPA. Materials and methods Antigen selection and computational filtering pipeline Candidate S. aureus proteins were initially identified through a systematic literature-based prioritization strategy following the rationale of designing a multi-epitope vaccine from experimentally characterized virulence factors. Proteins were not selected on the basis of isolated literature mention alone, instead the selection relied on inclusion criteria based on their role in S. aureus pathogenesis, accessibility to host immune recognition, evidence of in vivo protective efficacy, suitability for epitope prediction, and representation of complementary virulence mechanisms. In parallel, exclusion criteria were applied to deprioritize antigens with redundancy in pathogenic coverage andexcessive design complexity. The amino acid sequences of Hla and FnBPA sequences were obtained from the National Center for Biotechnology Information (NCBI) database. Based on the work of Shi, M et al.22 Q9K, Q10K and S33E amino acid substitutions were manually introduced into the SpA wild-type sequence to avoid immune evasion of the wild-type protein. For computational validation of candidate antigens, we followed the standard pipeline reported in23(Fig. 1). Post-translational modification analysis was performed to predict and remove signal peptides using SignalP 5.024. After this, extracellular localization of the protein was supported by Transmembrane Helix Prediction analysis with TMHMM 2.025,26. Each sequence was then assessed in VaxiGen to evaluate immunogenicity of the selected sequences, with “Bacteria” as the target organism27–29. Blast and homology analysis was also performed to evaluate potential cross-reactivity with human proteins using Basic Local Alignment Search Tool (BLAST), in which the search set organism was Homo sapiens (PARIS) (taxid:9606) using a blastp (protein-protein BLAST) algorithm. These three antigen sequences were used for epitope prediction. Fig. 1Research workflow for the in silico design of a multi-epitope S. aureus vaccine. The pipeline includes (1) antigen selection and computational filtering, (2) CTL, HTL and B-cell epitope prediction and screening, (3) vaccine construct design, (4) structural and immunological validation and (5) codon optimization and in silico cloning. Research workflow for the in silico design of a multi-epitope S. aureus vaccine. The pipeline includes (1) antigen selection and computational filtering, (2) CTL, HTL and B-cell epitope prediction and screening, (3) vaccine construct design, (4) structural and immunological validation and (5) codon optimization and in silico cloning. CTL epitope prediction To identify Cytotoxic T-lymphocyte (CTL) epitopes, the amino acid sequences of FnBPA, Hla and mutant SpA (SpAKKE) were analyzed using the NetMHCpan 4.1 EL Tool from the Immune Epitope Database (IEDB)30. CTL epitope prediction was performed using the 27 Human allele reference panel (HLA-A*01:01, HLA-A*02:01, HLA-A*02:03, HLA-A*02:06, HLA-A*03:01, HLA-A*11:01, HLA-A*23:01, HLA-A*24:02, HLA-A*26:01, HLA-A*30:01, HLA-A*30:02, HLA-A*31:01, HLA-A*32:01, HLA-A*33:01, HLA-A*68:01, HLA-A*68:02, HLA-B*07:02, HLA-B*08:01, HLA-B*15:01, HLA-B*35:01, HLA-B*40:01, HLA-B*44:02, HLA-B*44:03, HLA-B*51:01, HLA-B*53:01, HLA-B*57:01, HLA-B*58:01), which is intended to represent frequent HLA class I molecules in the global population. HTL epitope prediction Helper T-lymphocyte (HTL) epitopes were obtained using NetMHCIIpan 4.131 considering the 27 human allele panel as well (HLA-DRB1*01:01, HLA-DRB1*03:01, HLA-DRB1*04:01, HLA-DRB1*04:05, HLA-DRB1*07:01, HLA-DRB1*08:02, HLA-DRB1*09:01, HLA-DRB1*11:01, HLA-DRB1*12:01, HLA-DRB1*13:02, HLA-DRB1*15:01, HLA-DRB3*01:01, HLA-DRB3*02:02, HLA-DRB4*01:01, HLA-DRB5*01:01, HLA-DQA1*05:01/DQB1*02:01, HLA-DQA1*05:01/DQB1*03:01, HLA-DQA1*03:01/DQB1*03:02, HLA-DQA1*04:01/DQB1*04:02, HLA-DQA1*01:01/DQB1*05:01, HLA-DQA1*01:02/DQB1*06:02, HLA-DPA1*02:01/DPB1*01:01, HLA-DPA1*01:03/DPB1*02:01, HLA-DPA1*01:03/DPB1*04:01, HLA-DPA1*03:01/DPB1*04:02, HLA-DPA1*02:01/DPB1*05:01, HLA-DPA1*02:01/DPB1*14:01). Promiscuous high-affinity binders covering multiple HLA class I and II alleles were prioritized to maximize theoretical population coverage. B-cell epitope prediction Antibody epitope prediction was performed using the Bepipred Linear Epitope Prediction 3.032. Subsequently, an additional validation step was incorporated by verifying whether the selected linear epitopes were located within regions predicted to be discontinuous B-cell epitopes in DiscoTope 2.033. Evaluation of epitope immunogenicity, toxicity and allergenicity The top-ranked epitopes from each prediction were evaluated using the VaxiGen27–29 to assess their immunogenicity. To predict if the peptide was toxic, ToxinPred server was then used with the Swiss-Prot based method and an SVM (support vector machine) threshold of 0.034. Finally, AllerTOP v2.0 was used to determine possible allergenicity35. Epitopes that demonstrated the highest scores on the IEDB Tools, had an immunogenic probability between 66 and 100%, were non-toxic, and were probable non-allergens were selected for the vaccine construction. Epitope conservancy analysis Epitope conservancy analysis was performed separately for each antigen using S. aureus strain sequences retrieved from the NCBI database. For each target protein, ten strains were selected based on the availability of full-length corresponding sequences at the time of analysis, therefore, the strain sets differed across FnBPA, Hla, and SpA because the analysis was based on protein specific sequence availability. The following S. aureus strains were used: COL, USA300_FPR3757, PJFE-503, KUH140087, SA10, M1K194, CIG1242, HD1410, 21,345, CGS00 for FnBPA; Z0117SA0001, JARBHU1062, BG573, Sa795_KZ, Sa827_KZ, SAB_GP_03, DE_2010_3, SaK4_KZ, 057416_SL168, and 057407_SL157 for Hla; and A8117, 21,236, 502 A, FDAARGOS_14, Lr12, t442, ICE 630, HP20814011, BP2, and PYP/19 for SpA. Epitope linear sequence conservancy analysis was performed with a sequence identity threshold of ≥ 100%. Population coverage To estimate the percentage of individuals predicted to respond to the selected epitopes based on HLA genotypic frequencies, Major Histocompatibility Complex (MHC) binding and T-cell restriction data, Immune Epitope Database (IEDB) population coverage tool was used36. The geographic area “World” was selected and combined MHC class I and II population coverage was calculated for the selected epitopes and their corresponding HLA alleles. Vaccine design Linkers were selected according to the structural and immunological requirements of the multi-epitope construct. The EAAAK linker was used to connect the adjuvant to the epitope region because its rigid nature promotes spatial separation between domains and preserves structural independence. AAY linker was placed between CTL epitopes to enhance proteasomal processing and MHC-I presentation, whereas GPGPG was used between HTL epitopes to reduce junctional immunogenicity and support MHC-II responses. KK was inserted between B-cell epitopes to promote epitope separation and minimize artificial junctional epitopes37–39. The RS09 sequence (APPHALS) was incorporated at the N-terminus of the construct as an adjuvant, whereas PADRE (pan HLA-DR-binding epitope;AKFVAAWTLKAAA) was included as an enhancer of the immune response to the designed epitopes40,41. The final construct was arranged as follows: RS09-EAAAK-PADRE-EAAAK-CTL epitopes-HTL epitopes-B-cell epitopes, using the corresponding linkers described previously. The complete vaccine construct’s immunogenic potential was reassessed using VaxiGen27–29. Structural modeling, refinement and physicochemical properties ExPASy from Protparam server was used to compute physicochemical parameters of the final vaccine construct42. The full length amino acid sequence was used to calculate Instability index, Grand average of hydropathicity (GRAVY), Theoretical Isoelectric point (pI), and molecular weight. Protein solubility confirmation was further assessed with Protein-Sol server with the same vaccine amino acid sequence as input43. NetSurfP 3.0 was used to predict the vaccine’s secondary structure, estimating the proportions and positions of α-helices and β-strands44. The tertiary structure of the vaccine construct was modeled using AlphaFold to obtain the PDB file of the final construct for further analysis45. The structural quality of the vaccine’s model was initially evaluated using RamPlot with particular attention to the residues distribution (favored, allowed and outlier residues) on the Ramachandran plot46. GalaxyRefine from the GalaxyWEB server47 was then used to refine the vaccine construct to improve structural quality, particularly the portion of residues in favored regions of the Ramachandran plot. The refined model was subsequently re-evaluated using the same structural quality criteria, and the model showing the best stereochemical quality was selected for subsequent analysis. Molecular dockings MHC I and II molecules were downloaded from the Protein Data Bank (PDB) removing co-crystallized peptides with UCSF Chimera48. The receptor structures were preprocessed using a Python custom script to remove hydrogen atoms, solvent molecules, non-essential heteroatoms, and alternate locations. Docking between the CTL and HTL predicted epitopes and their respective HLA was performed with HPEPDOCK as supportive structural analysis to evaluate whether the selected epitopes could adopt plausible binding conformations within the HLA groove49. For each epitope, representative HLA alleles were selected individually based on the strongest final binding prediction and availability of experimentally resolved structures. Protein-protein docking was performed against the human TLR4/MD-2 complex. This receptor was selected due to the incorporation of the RS09 sequence (APPHALS) at the N-terminus of the vaccine construct. Since RS09 is a TLR4 adjuvant, its incorporation provided a biological rationale for selecting TLR4 for structural analysis50. Docking was carried out using the ClusPro 2.0 server51. The receptor structure was obtained from the Protein Data Bank as the crystal structure of the human TLR4-human MD-2-E. coli LPS Ra complex (PDB ID: 3FXI)41. Because this structure contains two TLR4/MD-2 units, only a single heterodimer composed of chain A (TLR4) and chain C (MD-2) was retained for docking. Prior to docking, solvent molecules and all HETATM entries were removed from the receptor file. The refined three-dimensional structure of the complete vaccine construct was used as the ligand. Docking was performed using the four standard ClusPro coefficient sets: Balanced, Electrostatic-favored, Hydrophobic-favored, and Van der Waals plus electrostatics-favored. For each docking mode, the top 10 returned models were visually inspected in PyMOL. Pose selection for the subsequent molecular dynamics (MD) simulation was based on three criteria. First, the vaccine construct had to contact the ligand-recognition face of the TLR4/MD-2 heterodimer, corresponding to the region occupied by LPS (lipopolysaccharide) in the 3FXI crystal structure. Secondly, the N-terminal RS09 sequence had to be positioned in proximity to this ligand-recognition region. Finally, among the poses fulfilling the first two criteria, the model belonging to the highest-ranked cluster was selected (ClusPro recommends ranking docking models by cluster size). The final selected docked complex was then used for the molecular dynamics simulation. Molecular dynamics simulation The molecular dynamics simulation was performed on the final selected docked complex using GROMACS 2026.152. Topology generation was performed with the CHARMM36 force field using the pdb2gmx module53. The processed complex was placed in a dodecahedral box with a minimum distance of 1.2 nm from the box edges and solvated with explicit TIP3P water. The system was then neutralized and adjusted to a salt concentration of 0.15 M using gmx genion. Energy minimization was performed using the steepest descent algorithm for up to 50,000 steps with an emtol value of 1000.0 kJ/mol/nm. The cutoff scheme was set to Verlet, long-range electrostatic interactions were treated with the particle mesh Ewald method, and periodic boundary conditions were applied in all directions. The Coulomb and van der Waals cutoffs were both set to 1.2 nm, and a force-switch modifier was applied to van der Waals interactions from 1.0 to 1.2 nm. After minimization, the system was equilibrated under NVT conditions for 500 ps and under NPT conditions for 1 ns, both using a 2 fs integration step and position restraints. Hydrogen-containing bonds were constrained with the LINCS algorithm. Temperature coupling was performed with the V-rescale thermostat at 310 K using separate coupling groups for Protein and Non-Protein. During NPT equilibration, pressure was maintained at 1 bar under isotropic conditions using the C-rescale barostat with a compressibility of 4.5 × 10−5 bar−1. The final simulation was then carried out for 100 ns with a 2 fs time step under NPT conditions using the V-rescale thermostat at 310 K and the Parrinello-Rahman barostat at 1 bar. For trajectory analysis, an index file was generated to define the TLR4 chain, the MD-2 chain, the vaccine construct, the receptor complex, and their corresponding backbone groups. The production trajectory was first corrected for periodic boundary conditions using the nojump option and was then clustered and centered on the complete complex. Backbone root mean square deviation (RMSD) was calculated for the complete complex, for the receptor backbone, and for the vaccine backbone after alignment on the receptor backbone. Additional self-fit RMSD analyses were also performed separately for the receptor and vaccine backbones. Root mean square fluctuation (RMSF) was calculated separately for the backbone residues of TLR4, MD-2, and the vaccine construct. The radius of gyration, minimum distance between receptor and vaccine, intermolecular contact number, intermolecular hydrogen bond number, center-of-mass distance, and solvent-accessible surface area (SASA) were calculated using gmx gyrate, gmx mindist, gmx hbond, gmx distance, and gmx sasa, respectively. Buried surface area was estimated from the receptor, vaccine, and complex SASA values using the formula (SASA_receptor + SASA_vaccine - SASA_complex)/2. Secondary structure changes in the vaccine construct along the trajectory were evaluated using gmx dssp54. Conformational clustering was performed with the GROMOS method using a cutoff of 0.25 nm. Additional interface analyses were performed using custom Python scripts with MDAnalysis on the cleaned centered trajectory and the processed complex structure55,56. Interfacial residue contact occupancy was calculated using a 4.5 Å heavy-atom distance cutoff between receptor and vaccine residues, and interfacial hydrogen bond persistence was estimated using a donor-acceptor distance cutoff of 3.5 Å and a donor-hydrogen-acceptor angle cutoff of 150°, with both metrics reported as the percentage of analyzed frames in which each residue or residue pair was present. Graph generation and figure assembly were performed using custom Python scripts with Matplotlib. Molecular interactions The representative structure corresponding to the 100 ns frame of the molecular dynamics simulation was extracted in PDB format and used for intermolecular interaction analysis. Protein-protein interaction profiling between the vaccine construct and the TLR4/MD-2 complex was performed using the Protein-Ligand Interaction Profiler (PLIP)57. PLIP was used to identify the non-covalent interactions present at the receptor-vaccine interface, including hydrogen bonds, hydrophobic contacts, salt bridges, and other relevant intermolecular interactions. The identified interactions were then used for structural interpretation of the binding interface. Immune simulation The humoral and cellular responses to the multi-epitope Staphylococcus aureus vaccine were simulated using the C-ImmSim immune simulator58–60. Simulations were run with default parameters unless otherwise specified. Human leukocyte antigen (HLA) restriction was set to common class I and II alleles (HLA-A02:01, HLA-A01:01, HLA-B07:02, HLA-B07:04, HLA-DRB101:01, and HLA-DRB104:01). The simulation volume was fixed at 10, and a random seed of 12,345 was used to allow reproducibility. The complete simulation consisted of 1,050-time steps (approximately equivalent to 350 days) and included four vaccine administrations. For every injection, 1,000 units of antigen and 100 units of adjuvant were delivered at time steps 10, 90, 170, and 525. Codon optimization Codon optimization for heterologous expression in Escherichia coli was performed using the NovoPro server was used to improve GC (guanine-cytosine) content, Shine Dalgarno (SD) sequence, Deoxyribonucleic acid (DNA) repeats, Messenger ribonucleic acid (mRNA) secondary structure, codon usage bias, and 5’ region for the host system E. coli61. Then the codon sequence was modified with restriction sites of HindIII and BamHI at the N-terminus and C-terminus to perform directional cloning into pET-30a (+) using the Snapgene tool. This vector was selected as it is a T7 promoter-based bacterial expression vector widely used for high-level recombinant protein expression in E. coli, thereby supporting the feasibility of downstream production and purification of the designed vaccine candidate62. Results Antigen selection To identify the most suitable antigenic components for the multi-epitope vaccine, candidate S. aureus proteins were evaluated based on their biological relevance in pathogenesis, accessibility to host immune recognition, predicted antigenicity, and similarity to human proteins. Based on this screening strategy, three antigens, Hla, FnBPA, and the detoxified mutant SpAKKE, were selected as final vaccine targets after literature-based prioritization. These proteins were selected because they represent complementary virulence mechanisms of S. aureus, including toxin-mediated injury, adhesion/invasion, and immune evasion. Hla and FnBPA sequences were retrieved from NCBI, and the SpAKKE sequence was generated by introducing the Q9K, Q10K, and S33E substitutions into wild-type SpA. SignalP analysis identified signal peptides in the native sequences, which were removed prior to downstream analyses. TMHMM confirmed the extracellular accessibility of the processed antigens. VaxiGen predicted all three proteins to be probable immunogens with a probability of 100%. BLAST analysis against human proteins showed a percent identity of 24.81–37.61% for FnBPA, while no significant similarity was found for Hla and SpA. These findings supported the selection of Hla, FnBPA, and SpAKKE for subsequent epitope prediction. CTL epitope prediction The CTL epitope YYYDRGYTL, derived from FnBPA, exhibited a strong predicted affinity value for HLA-A*24:02 and HLA-A*23:01 with a NetMHCpan_El percentile rank of 0.01 and a 0.06 score for HLA-B*08:01. According to VaxiGen, it had a 100% probability of being immunogenic, showing its potential to elicit a cellular immune response. In addition, ToxinPred analysis yielded an SVM score of −0.65, classifying the peptide as non-toxic. AllerTop analysis classified the epitope as a probable non-allergen. KPQSKKSEL epitope from FnBPA had a 0.01 NetMHCpan_El percentile rank and was predicted to bind to HLA-B*07:02 and HLA-B*08:01. ToxinPred analysis revealed an SVM score of −0.89 with a non-toxic prediction, VaxiGen showed a 66% probability of the peptide being immunogenic and was classified as probable non-allergen by AllerTop. The predicted CTL epitope from Hla was GANKSGLAW with a NetMHCpan_El percentile rank of 0.06 and predicted binding to the HLA-B*58:01 allele; VaxiGen showed a 66% probability of being immunogenic, AllerTop classified it as a probable non-allergen, and an SVM score of −0.65 from ToxinPred predicted it as non-toxic. For the SpAKKE the selected epitopes were DEAKKNAFY and DPSQSANVL. DEAKKNAFY showed a 0.07 NetMHCpan_El percentile rank with affinity for HLA-B*44:03, 0.07 to HLA-B*44:02, 1.1 to HLA-A*26:01, 1.3 to HLA-A*01:01, and 1.9 to HLA-B*35:01. VaxiGen showed a 66% probability of being immunogenic, it was classified as a probable non-allergen by AllerTop analysis, and an SVM score of −0.73 predicted it as non-toxic according to the ToxinPred analysis. DPSQSANVL epitope had a NetMHCpan_El percentile rank of 0.22 with predicted affinity to HLA-B*35:01, 0.29 to HLA-B*51:01, 0.36 to HLA-B*53:01, 0.51 to HLA-B*07:02, and 0.66 to HLA-B*08:01. VaxiGen showed a 100% probability of being immunogenic, it was a probable non-allergen according to the AllerTop analysis and a score of −0.89 in ToxinPred supported it as non-toxic. HTL epitope prediction The HTL epitope RIFEYLGNNEDIAKS derived from FnBPA had a NetMHCpan_EL percentile of 0.02 with a predicted affinity to HLA-DRB1*04:05 and HLA-DRB1*04:01, a 66% probability of being immunogenic (VaxiGen), classified as probable non-allergen by AllerTop, and a −1.19 score in ToxinPred interpreted as non-toxic. The SSGFSPDFATVITMD epitope from Hla had a NetMHCpan_EL percentile of 0.11 and predicted binding to HLA-DRB3*01:01 and 2.0 to HLA-DRB3*02:02, it had a 66% probability of being immunogenic according to VaxiGen, classified as probable non-allergen by AllerTop, and had a −0.89 score in ToxinPred (non-toxic). LDPNKASSLLSSGFS had a NetMHCpan_EL percentile of 0.08 to HLA-DQA1*01:02/DQB1*06:02, VaxiGen predicted a 100% probability of being immunogenic, it was classified as probable non-allergen by AllerTop, and had a −0.70 score in ToxinPred (predicted as non-toxic). FYQVLNMPNLNADQR had a NetMHCpan_EL percentile of 0.56 to HLA-DRB1*04:01, 1.8 to HLA-DRB1*15:01, 3.5 to HLA-DRB1*04:05, 4.8 to HLA-DRB4*01:01, 6.4 to HLA-DRB1*01:01,7.9 to HLA-DRB3*02:02, 8.1 to HLA-DRB1*08:02, 8.6 to HLA-DRB1*12:01, 8.8 to HLA-DRB1*11:01, 11 to HLA-DQA1*01:01/DQB1*05:01, 14 to HLA-DRB5*01:01, and 16 to HLA-DRB1*13:02. VaxiGen showed a 100% probability of being immunogenic, AllerTop classified it as probable non-allergen, and ToxinPred calculated an SVM score of −1.04 classifying it as non-toxic. B-cell epitopes The predicted B-cell epitope for FnBPA was SQDNSGDQ, showing a 100% predicted probability of being an immunogen in VaxiGen, an SVM score of −0.92 in ToxinPred classified it as non-toxic, and AllerTop classified it as a probable non-allergen. The predicted B-cell epitope for Hla was NEVAQISDYYPRNSIDTK with a 66% probability of being immunogenic according to the VaxiGen analysis, a −0.73 SVM score (non-toxic), and was predicted to be non-allergenic by AllerTop. The predicted epitope for SpAKKE was QNQHDEAKKNA, with VaxiGen showing a 100% probability of being immunogenic, AllerTop classified it as a probable non-allergen, and ToxinPred calculated an SVM score of −0.64, classifying it as non-toxic. Epitope conservancy analysis To evaluate whether the selected epitopes were conserved among different S. aureus strains, epitope conservancy analysis was performed using ten strains for each antigen with a sequence identity threshold of 100%. The results showed variable levels of conservation among the predicted epitopes. For FnBPA-derived epitopes, SQDNSGDQ showed the highest conservancy, being fully conserved in 90.00% (9/10) of the analyzed strains. In contrast, YYYDRGYTL was conserved in 30.00% (3/10) of strains, whereas KPQSKKSEL and RIFEYLGNNEDIAKS showed lower conservancy values of 20.00% (2/10) each. Among the Hla-derived epitopes, GANKSGLAW, SSGFSPDFATVITMD, and LDPNKASSLLSSGFS were fully conserved in 100.00% (10/10) of the analyzed strains, indicating strong sequence preservation across isolates. In contrast, NVTGDDTGKIGGLIGANVSIG showed a lower conservancy of 60.00% (6/10). For the SpAKKE-derived epitopes, QKLNDSQ, DPSQSANVL, and FYQVLNMPNLNADQR were conserved in 100.00% of the evaluated strains, whereas DEAKKNAFY showed no conservancy under the applied threshold (0.00%). Overall, the conservancy analysis identified several highly preserved epitopes, particularly within Hla and SpAKKE, supporting their suitability for inclusion in the multi-epitope vaccine construct. Population coverage analysis The population coverage calculation globally yielded 88.81% projected population coverage, an average of 2.34 of epitope hits/HLA combinations recognized by the population, and a PC90 value of 0.89, which is the minimum number of epitope hits/HLA combinations by 90% of the population (Fig. 2). Fig. 2Population coverage of the selected epitopes in the Global population, predicted with the IEDB Population Coverage Tool. The x-axis shows the number of epitope hits/HLA combinations recognized per individual; blue bars (left y-axis) represent the percentage of individuals with each number of epitopes–HLA hits, and the blue line (right y-axis) indicates the cumulative percent of population coverage for the combined class I and class II epitopes. Population coverage of the selected epitopes in the Global population, predicted with the IEDB Population Coverage Tool. The x-axis shows the number of epitope hits/HLA combinations recognized per individual; blue bars (left y-axis) represent the percentage of individuals with each number of epitopes–HLA hits, and the blue line (right y-axis) indicates the cumulative percent of population coverage for the combined class I and class II epitopes. Vaccine design The resulting amino acid sequence was the following: , resulting in 210 amino acids. The complete vaccine construct had a 100% probability of being an immunogen. Secondary structure prediction and physicochemical properties The Expasy Instability index score was 33.22, classifying the protein as stable. The molecular weight for the 210 amino acids was 21752.08, and its theoretical pI was 8.58. The GRAVY score was − 0.513, which means that the construct is hydrophilic, soluble, and antigenically accessible. This solubility was confirmed by the Protein-Sol server, which gave a predicted scaled solubility of 0.612. According to this server, the secondary structure consists of 32.9% of α-helices, 0.50% β-strands, and 66.7% coils. Approximately 92.4% of residues were surface-exposed (RSA (Relative Solvent Accessibility) ≥ 0.25), and linker regions (GPGPG motifs) showed high coil propensity (> 0.7), suggesting flexibility that may facilitate proper folding and epitope presentation. Tertiary structure and refinement The initial tertiary structure obtained from AlphaFold was analyzed using RamPlot, in which the majority of amino acids were forming parallel beta sheets and right-handed α-helical conformations. To reduce the Ramachandran outliers, the GalaxyWeb was used to refine the structure. As a result, Ramachandran favored amino acids reached 97.1%, the poor rotamers were reduced to zero, and the clash score was 0.0. RamPlot was then used to confirm the resulting tertiary structure, with 97.115% of favored amino acids, 2.885% allowed amino acids, and 0% of disallowed amino acids (Fig. 3). Fig. 3Three-dimensional Ramachandran plot of the refined vaccine model. The distribution of backbone dihedral angles (φ and ψ) is shown as a three-dimensional frequency map, where vertical bars represent residue density in different conformational regions. The highest concentration of residues was observed in the favored regions corresponding mainly to right-handed α-helical and β-sheet conformations, supporting the acceptable stereochemical quality of the refined structure. Three-dimensional Ramachandran plot of the refined vaccine model. The distribution of backbone dihedral angles (φ and ψ) is shown as a three-dimensional frequency map, where vertical bars represent residue density in different conformational regions. The highest concentration of residues was observed in the favored regions corresponding mainly to right-handed α-helical and β-sheet conformations, supporting the acceptable stereochemical quality of the refined structure. Molecular docking Docking demonstrated all selected CTL epitopes formed energetically favorable complexes with their respective HLA. The epitopes YYYDRGYTL, KPQSKKSEL, GANKSGLAW, and DEAKKNAFY yielded docking scores of −167.897, −270.684, − 210.064, and − 187.943, respectively, indicating strong predicted binding. Similarly, all HTL epitopes displayed favorable interactions with docking scores of −246.242 for RIFEYLGNNEDIAKS, −251.744 for SSGFSPDFATVITMD, −203.897 for LDPNKASSLLSSGFS and − 342.257 for FYQVLNMPNLNADQR (Table 1). In each case, the highest-ranked pose was visually inspected in UCSF Chimera, and all peptides resided within the binding groove of the HLA proteins. Table 1Immunoinformatic screening and HLA-binding profile of the selected CTL, HTL and B-cell epitopes included in the multi-epitope vaccine construct and complete vaccine docking against TLR4/MD2.AntigenEpitopeEpitope typeRepresentative HLA alleleHLA percentile rankAntigenicityAllergenicityToxicity (SVM score)Conservancy (%)Docking scoreFnBPASQDNSGDQB-cellNANAImmunogen with 100% prob.Probable Non-allergen−0.9290.00%NAYYYDRGYTLCTLHLA-A*24:020.01Immunogen with 100% prob.Probable Non-allergen−0.6530.00%−167.897KPQSKKSELCTLHLA-B*07:020.01Immunogen with 66% prob.Probable Non-allergen−0.8920.00%−270.684RIFEYLGNNEDIAKSHTLHLA-DRB1*04:010.02Immunogen with 66% prob.Probable Non-allergen−1.1920.00%−246.242HlaNVTGDDTGKIGGLIGANVSIGB-cellNANAImmunogen with 66% prob.Probable Non-allergen−0.7360.00%NAGANKSGLAWCTLHLA-B*58:010.06Immunogen with 66% prob.Probable Non-allergen−0.65100.00%−210.064SSGFSPDFATVITMDHTLHLA-DRB3*01:010.11Immunogen with 66% prob.Probable Non-allergen−0.79100.00%−251.744LDPNKASSLLSSGFSHTLHLA-DQA1*010.08Immunogen with 66% prob.Probable Non-allergen−0.70100.00%−203.897SpAKKEQKLNDSQB-cellNANAImmunogen with 100% prob.Probable Non-allergen−0.64100.00%NADEAKKNAFYCTLHLA-B*44:030.07Immunogen with 66% prob.Probable Non-allergen−0.730.00%−187.943DPSQSANVLCTLHLA-B*35:010.22Immunogen with 100% prob.Probable Non-allergen−0.89100.00%−150.843FYQVLNMPNLNADQRHTLHLA-DRB1*04:010.56Immunogen with 100% prob.Probable Non-allergen−1.04100.00%−342.257Complete vaccine constructNATLR4/MD-2NAImmunogen with 100% prob.Probable Non-allergenNANA−919.5 Table 1Immunoinformatic screening and HLA-binding profile of the selected CTL, HTL and B-cell epitopes included in the multi-epitope vaccine construct and complete vaccine docking against TLR4/MD2.AntigenEpitopeEpitope typeRepresentative HLA alleleHLA percentile rankAntigenicityAllergenicityToxicity (SVM score)Conservancy (%)Docking scoreFnBPASQDNSGDQB-cellNANAImmunogen with 100% prob.Probable Non-allergen−0.9290.00%NAYYYDRGYTLCTLHLA-A*24:020.01Immunogen with 100% prob.Probable Non-allergen−0.6530.00%−167.897KPQSKKSELCTLHLA-B*07:020.01Immunogen with 66% prob.Probable Non-allergen−0.8920.00%−270.684RIFEYLGNNEDIAKSHTLHLA-DRB1*04:010.02Immunogen with 66% prob.Probable Non-allergen−1.1920.00%−246.242HlaNVTGDDTGKIGGLIGANVSIGB-cellNANAImmunogen with 66% prob.Probable Non-allergen−0.7360.00%NAGANKSGLAWCTLHLA-B*58:010.06Immunogen with 66% prob.Probable Non-allergen−0.65100.00%−210.064SSGFSPDFATVITMDHTLHLA-DRB3*01:010.11Immunogen with 66% prob.Probable Non-allergen−0.79100.00%−251.744LDPNKASSLLSSGFSHTLHLA-DQA1*010.08Immunogen with 66% prob.Probable Non-allergen−0.70100.00%−203.897SpAKKEQKLNDSQB-cellNANAImmunogen with 100% prob.Probable Non-allergen−0.64100.00%NADEAKKNAFYCTLHLA-B*44:030.07Immunogen with 66% prob.Probable Non-allergen−0.730.00%−187.943DPSQSANVLCTLHLA-B*35:010.22Immunogen with 100% prob.Probable Non-allergen−0.89100.00%−150.843FYQVLNMPNLNADQRHTLHLA-DRB1*04:010.56Immunogen with 100% prob.Probable Non-allergen−1.04100.00%−342.257Complete vaccine constructNATLR4/MD-2NAImmunogen with 100% prob.Probable Non-allergenNANA−919.5 Immunoinformatic screening and HLA-binding profile of the selected CTL, HTL and B-cell epitopes included in the multi-epitope vaccine construct and complete vaccine docking against TLR4/MD2. Docking of the vaccine construct against the single TLR4/MD-2 heterodimer in ClusPro 2.0 produced multiple candidate poses. These were visually inspected in PyMOL, and models that did not satisfy the first two selection criteria were excluded. From the remaining poses, the complex with the highest-ranked cluster was chosen. The selected docked complex originated from the Hydrophobic-favored mode, comprising 85 members, with a center score of −919.5 and a lowest-energy score of −1358.7. The docked complex was chosen for the MD simulation (Fig. 4). Fig. 4Surface representation of the selected docking pose of the multi-epitope vaccine construct with a single human TLR4/MD-2 heterodimer. The complex was generated using ClusPro 2.0 and visual representation was generated with ChimeraX. The vaccine construct is shown in red, TLR4 chain A in blue, and MD-2 chain C in green. Surface representation of the selected docking pose of the multi-epitope vaccine construct with a single human TLR4/MD-2 heterodimer. The complex was generated using ClusPro 2.0 and visual representation was generated with ChimeraX. The vaccine construct is shown in red, TLR4 chain A in blue, and MD-2 chain C in green. Immune simulation CD8+ cytotoxic T-lymphocyte responses were limited. After the first injection there was an increase of non-memory CD8+ T-cells, the rest of the graph showed only small random fluctuations over time, with no peaks coinciding with the vaccine doses, and memory CD8+ T-cells remained absent. Regarding the CD8+ T-cell population in different states, resting CD8+ T-cells decreased after the first dose from about 1,100 cells/mm³ to 200 cells/mm³ between days 50 and 100, then gradually returned to baseline by day 350. Active CD8+ T-cells increased after the first dose from zero to approximately 900 cells/mm³ within the first 50–100 days and then steadily declined back toward zero. Duplicating and anergic CD8+ T-cells stayed at zero throughout. Natural killer cells also showed no significant response to vaccination (Supplementary Fig. S1). Dendritic cell populations showed only a brief, early response to the multi-epitope S. aureus vaccine. Immediately after the first vaccination, there was a short-lived decrease in resting and active dendritic cells, accompanied by a transient rise in internalized, presenting-1 and presenting-2 dendritic cells, consistent with antigen uptake and presentation. These changes quickly returned to baseline, and subsequent vaccine doses did not produce obvious additional shifts in any dendritic cell subset, except for very slight increases in presenting-2 cells (Supplementary Fig. S1). Epithelial cell populations revealed that only the active epithelial population was detectable, without discernible peaks or shifts associated with any of the four vaccine doses (Supplementary Fig. S1). Total macrophage numbers stayed fairly consistent throughout the 350 days. Internalized and presenting-2 macrophages displayed brief spikes after each vaccine dose (largest after the first dose, smaller thereafter) that rapidly returned to zero. Active macrophages increased after each injection and remained increased for weeks, only to eventually fall to baseline numbers (Supplementary Fig. S1). Total B-cells rose from a baseline of approximately 450 cells/mm³ to progressively higher peaks after each dose. Non-memory B-cells declined after the first dose and remained lower than baseline, with only transient rises after subsequent doses, whereas memory B-cells were absent at baseline and increased with each injection, peaking after the third dose and demonstrating a recall response at the fourth dose. Isotype B-cells showed that IgG1 B-cells were not detectable in the beginning, then appeared after the first dose and became exponentially higher with the second and third vaccinations, displaying a recall response after the fourth dose. In contrast, IgG2 B-cells were only minimally affected by vaccination and remained at very low levels throughout the simulation. IgM B-cells were the predominant isotype across all time points and showed modest increases above their own baseline with each vaccine dose. B-cells divided by their functional state showed that each vaccine dose triggered short-lived increases of duplicating, presenting-2, and anergic cells, after which they quickly returned to baseline. Active B-cells decreased sharply after each dose, followed by an increase; these increases were larger with each dose, reaching their highest point at the third dose, after which they returned to their baseline. The fourth dose demonstrated a recall response, as active B-cells increased from baseline to a level higher than the first and second doses. Plasma B-cell simulations demonstrated that each vaccination induced short-lived spikes of plasma B-cells that quickly returned to baseline. After the first dose, IgM and IgG1 plasma cells rose to similar levels, while IgG2 remained almost unchanged. Subsequent doses produced progressively stronger responses (Supplementary Fig. S2). CD4+ T helper cell responses were characterized by the progressive increase of the memory subset. Memory CD4+ T-cells were absent at baseline but appeared after the first dose and increased further with the second and third doses. Between the third and fourth doses, memory cells declined slightly but remained above their starting level. The fourth immunization demonstrated a recall response in memory CD4+ T-cells, after which they gradually decreased and stabilized at approximately 600 cells/mm³ until the simulation’s end (Supplementary Fig. S2). T helper subsets showed a strongly Th1-skewed response to the vaccine. The first three doses demonstrated an increasingly stronger Th1 response, with the third dose peaking slightly above 10,000 cells/mm³. After this third peak, the Th1 population declined before the fourth vaccination, which again boosted Th1 cells close to 9,000 cells/mm³, followed again by a gradual decrease (Supplementary Fig. S2). Immunoglobulin levels showed progressive increases with each vaccine dose. After the first dose, IgM and IgG1 antibody titers rose to similar levels, while IgG2 remained very low throughout the whole simulation. The second dose elicited a higher response from both isotypes, with IgM higher than IgG1. The third dose produced the strongest humoral response, with marked increases in both IgM and IgG1. Before the fourth dose, all antibody titers declined. The fourth vaccination reproduced the pattern seen with the third dose, although it is to be noted that the baseline for all antibodies increased, with IgG1 stabilizing at 2,000 by the end of the simulation (Supplementary Fig. S3). Overall, cytokine and interleukin predictions suggested robust, repeated bursts of Interferon Gamma (IFN-γ), IL-2, and IL-12 production with diminishing TGF-β and IL-10 responses over time, compatible with a sustained Th1 immune profile (Supplementary Fig. S4). Molecular dynamics simulations During the 100 ns simulation, the complex backbone RMSD increased rapidly during the first few nanoseconds and then remained mainly within a range of around 1.0–1.5 nm, with a slight upward drift over the remainder of the trajectory, reaching a final value of 1.517 nm and a last-20-ns average of 1.482 nm (Fig. 5). The receptor backbone RMSD remained lower throughout the simulation, with a mean of 0.539 nm and a final value of 0.602 nm (Supplementary Fig. S5). The vaccine construct showed a self-fit backbone RMSD mean of 1.337 nm and a final value of 1.453 nm (Supplementary Fig. S5), whereas the vaccine backbone RMSD calculated after alignment on the receptor backbone was higher and showed an early increase followed by continued fluctuations with a slight gradual upward drift, with a mean of 3.031 nm, a final value of 3.766 nm, and a last-20-ns average of 3.682 nm (Fig. 5). The receptor-vaccine center-of-mass distance increased early relative to the starting structure and then remained within a relatively narrow range for the rest of the simulation, with a mean of 4.868 nm and a final value of 5.008 nm (Fig. 5). The radius of gyration remained within a narrow interval over the trajectory, with a mean of 3.530 nm and a last-20-ns average of 3.534 nm (Supplementary Fig. S6). The buried interface area fluctuated during the simulation, with a mean of 15.987 nm² and a last-20-ns average of 16.820 nm² (Fig. 5). The number of intermolecular hydrogen bonds also varied over time, with a mean of 6.352 hydrogen bonds and a last-20-ns average of 7.075 (Fig. 5). Receptor SASA averaged 341.041 nm², vaccine SASA 151.097 nm², and total complex SASA 460.162 nm². The vaccine and complex SASA were higher at the beginning of the trajectory and lower over the final part of the simulation (Supplementary Fig. S6). Backbone RMSF differed across the three components of the system. Chain A of TLR4 showed broader residue-level fluctuations than chain C of MD-2, whereas the vaccine construct showed the highest fluctuations overall. Average backbone RMSF values were 0.196 nm for chain A, 0.103 nm for chain C, and 0.592 nm for the vaccine construct (Supplementary Fig. S7). Analysis of the vaccine’s secondary structure demonstrated it retained helix as its dominant single secondary-structure class throughout the trajectory, with a mean helix content of 40.1% and a mean beta-sheet content of 1.2% across 210 residues, while the remainder corresponded mainly to turn/bend and coil/other states (Fig. 5). Fig. 5Molecular dynamics simulation of the TLR4/MD-2 heterodimer in complex with the multi-epitope vaccine construct over a 100 ns simulation. (A) Complex backbone RMSD. (B) Vaccine backbone RMSD after alignment on the receptor backbone. (C) Receptor-vaccine center-of-mass distance. (D) Buried surface area at the receptor-vaccine interface. (E) Number of intermolecular hydrogen bonds between the receptor and the vaccine construct. (F) Secondary-structure composition of the vaccine construct over the trajectory, shown as the percentage of residues assigned to helix, beta-sheet, turn/bend, and coil/other states. Molecular dynamics simulation of the TLR4/MD-2 heterodimer in complex with the multi-epitope vaccine construct over a 100 ns simulation. (A) Complex backbone RMSD. (B) Vaccine backbone RMSD after alignment on the receptor backbone. (C) Receptor-vaccine center-of-mass distance. (D) Buried surface area at the receptor-vaccine interface. (E) Number of intermolecular hydrogen bonds between the receptor and the vaccine construct. (F) Secondary-structure composition of the vaccine construct over the trajectory, shown as the percentage of residues assigned to helix, beta-sheet, turn/bend, and coil/other states. Interface occupancy analysis identified a recurrent set of receptor residues at the contact surface, including Leu87, Met85, Phe126, Val82, Ile124, Asn158, Phe121, Ser127, Lys125, Lys128, Asn86, Pro157, Gln156, Asn83, and Tyr131, most of which were present at the interface in roughly 85% to 100% of analyzed frames (Supplementary Fig. S6). Persistent interfacial hydrogen-bond pairs were also observed, with the highest occupancies reaching approximately 40%; the highest-ranked pairs included ALA1→GLU321, LYS341→GLU8, SER171→ASN158, LYS179→ASN158, and TRP19→MET85 (Supplementary Fig. S6). Taken together, these analyses indicate that the complex underwent an early rearrangement relative to the starting docked pose while maintaining a persistent receptor-vaccine interface throughout the 100 ns simulation. Molecular interactions PLIP analysis of the representative structure extracted at 100 ns identified 23 non-covalent receptor-vaccine interactions, consisting of 7 hydrogen bonds, 14 hydrophobic interactions, and 2 salt bridges. Hydrophobic interactions were the most frequent interaction class and involved vaccine residues Ala13, Lys14, Phe15, Val16, Trp19, Ala23, Ile190, and Ile194, contacting receptor residues Phe121, Ile124, Phe126, Leu87, Ile80, Val82, Asn83, and Pro157. Hydrogen bonds were detected between Ala23-Asn83, Gly126-Lys125, Gln153-Asn158, Gly170-Lys128, Ser171-Asn158, Ala5-Arg322, and Ser7-Tyr296. In addition, two salt bridges were identified, formed by Asp117-Lys130 and Asp125-Lys125 (Fig. 6). Fig. 6PLIP-based interaction analysis of the representative 100 ns MD simulation frame of the multi-epitope vaccine construct in complex with the TLR4/MD-2 heterodimer. The top section summarizes the non-covalent intermolecular interactions identified by PLIP, grouped into hydrogen bonds, hydrophobic interactions, and salt bridges, with the corresponding interacting residue pairs listed for each category. In the bottom section, the structural view of the receptor-vaccine interface highlights the interacting residues, where molecular structures in pink correspond to the vaccine construct and molecular structures in yellow and light grey correspond to TLR4/MD-2. The dotted lines represent the molecular interactions, with blue representing hydrogen bonds, red representing hydrophobic interactions, and green representing salt bridges. PLIP-based interaction analysis of the representative 100 ns MD simulation frame of the multi-epitope vaccine construct in complex with the TLR4/MD-2 heterodimer. The top section summarizes the non-covalent intermolecular interactions identified by PLIP, grouped into hydrogen bonds, hydrophobic interactions, and salt bridges, with the corresponding interacting residue pairs listed for each category. In the bottom section, the structural view of the receptor-vaccine interface highlights the interacting residues, where molecular structures in pink correspond to the vaccine construct and molecular structures in yellow and light grey correspond to TLR4/MD-2. The dotted lines represent the molecular interactions, with blue representing hydrogen bonds, red representing hydrophobic interactions, and green representing salt bridges. Codon optimization The codon-optimized sequence showed a Codon Adaptation Index (CAI) of 0.81 and a GC content of 57.30%. The optimized nucleotide sequence was provided in the supplementary material. For in silico cloning, HindIII and BamHI restriction sites were incorporated, and the sequence was inserted into the pET-30a(+) vector (Fig. 7). Fig. 7The optimized codon sequence containing the HindIII and BamHI restriction sites was cloned into the pET-30a(+) expression vector in the MCS (multiple cloning site) region. The optimized codon sequence containing the HindIII and BamHI restriction sites was cloned into the pET-30a(+) expression vector in the MCS (multiple cloning site) region. Discussion S. aureus remains a difficult target for vaccine development. Despite extensive preclinical and clinical investigations, no vaccine for human use has yet been released to the public63,64. This long-standing difficulty reflects more than antigen choice alone, since S. aureus possesses multiple immune-evasion mechanisms and prior host exposure can generate non-protective immune imprinting that interferes with the development of effective vaccine-induced immunity65,66. Several recent computational vaccine studies against S. aureus have focused on narrower antigenic spaces, including hemolysin-centered constructs or broader reverse-vaccinology pipelines driven primarily by antigen prediction67,68. In contrast, the present study was designed around three mechanistically distinct virulence functions, toxin-mediated injury, adhesion/invasion, and immune evasion, represented by Hla, FnBPA, and a detoxified SpA variant, respectively, thereby incorporating immune evasion and complementary immune mechanisms as a central design principle rather than as a secondary consideration18–21. Recent experimental work supports this direction, as next-generation S. aureus vaccine strategies have increasingly emphasized direct targeting of immune-evasion determinants such as SpA and LukAB instead of relying on a single antigenic class69,68. The present findings support our vaccine construct as a promising in silico candidate grounded in a biologically complementary design rationale, although its protective efficacy remains to be established experimentally. Regarding immunological in vivo evidence of the selected antigens, a 2024 study that analyzed samples from children and then tested vaccination in neonatal mice demonstrated that anti-Hla responses are relatively limited early in life, whereas immunization with a structurally modified Hla antigen produced an approximately 100-fold increase in toxin-neutralizing antibodies, enhanced the T follicular helper response during infection, reduced skin disease after challenge, and also conferred protection to offspring after maternal immunization69. The human relevance of this antigen is also supported by a study that followed 235 children with community-onset S. aureus skin infection for 12 months and found that higher anti-Hla antibody levels were associated with a lower risk of recurrence, supporting anti-Hla humoral immunity as a correlate of protection70. FnBPA’s vaccine relevance has also been supported in vivo. Previous studies have shown that immunization with FnBPA protected rats against experimental endocarditis and reduced S. aureus infection of vascular patches in mice71,72. Other works with FnBPA in mice have further suggested that protection is not explained by antibodies alone. A 2017 study reported induction of IL-17 A-producing cells after FnBPA vaccination and loss of protection in IL-17 A-deficient animals73. A 2018 study also showed that neutralization of IL-17 A, but not IFN-γ, reversed vaccine-mediated protection in both sepsis and skin infection models74. This is biologically relevant because IL-17 A is an important mediator of anti-staphylococcal barrier immunity, promoting neutrophil influx and antimicrobial peptide production, and studies of nasal colonization have shown that efficient clearance of S. aureus at mucosal surfaces depends on IL-17 A-driven responses75,76. SpAKKE was included to address immune evasion, since SpA binds the Fc region of immunoglobulins and also crosslinks VH3-type B-cell receptors, thereby impairing opsonophagocytosis and distorting adaptive B-cell responses21. This rationale is strengthened by preclinical studies of detoxified SpA variants. For example, the 2021 study that identified the detoxified SpA antigens used in this study demonstrated these variants elicited SpA-neutralizing antibodies and protected animals from colonization and bloodstream infections22. These findings further strengthen the antigen selection strategy and biological rationale of the study. In relation to epitope selection in our vaccine construct, the overall results were favorable. The final CTL, HTL, and B-cell epitopes demonstrated acceptable predicted immunogenicity-related profiles with non-toxic and probable non-allergen classifications, and the combined class I and II panel yielded a projected world population coverage of 88.81%. The main result requiring a more cautious interpretation was epitope conservancy. Hla contributed the most consistent results, with multiple selected epitopes fully conserved across the analyzed strains, supporting its role as one of the strongest components of the construct. In contrast, FnBPA-derived epitopes showed lower conservancy, suggesting greater strain-level variability despite otherwise acceptable results. SpAKKE showed a mixed pattern, with some epitopes fully conserved, whereas DEAKKNAFY showed 0.0% conservancy under the applied 100% identity threshold. Because this epitope was derived from the engineered mutant SpAKKE sequence rather than native SpA, the absence of exact matches is partially understandable in that context, and its retention in the construct was derived from the fact that its other selection results remained favorable enough to justify inclusion despite the lack of conservancy. Taken together, these findings support the selected epitopes as a reasonable vaccine design, while also showing that uniform conservancy was not achieved across all epitopes. The immune simulation suggested a response profile dominated mainly by humoral and CD4+ T helper mediated immunity, since repeated immunizations were associated with progressive increases in total B-cells, memory B-cells, IgM and IgG1 production, plasma B-cell responses, memory CD4+ T cells, macrophage activation, and repeated bursts of IFN-γ, IL-2, and IL-12, whereas CD8+ responses remained limited. This pattern is relevant in the context of S. aureus vaccine development because protection against this pathogen is increasingly understood to require more than antibody production alone, and previous work has emphasized that functional humoral responses should be accompanied by appropriate CD4+ T-cell polarization14. In this regard, the Th1 profile observed here may be considered favorable, since previous experimental work has shown that memory Th1 cells and IFN-γ can contribute to protection against invasive S. aureus infection77. Recent vaccine studies are also consistent with this interpretation, as a vaccine targeting the immune-evasion determinants SpA and LukAB reduced bacterial burden and dissemination while using a Th1 adjuvant strategy intended to restore functional opsonophagocytic antibody responses68. However, the present simulation did not demonstrate a Th17 response and consequently did not show an IL-17-associated profile, which is an important limitation in the context of anti-staphylococcal immunity. This point is relevant because IL-17 has been shown to be crucial for protection at barrier and mucosal sites, particularly in skin and soft tissue infection and in nasal colonization, where it promotes neutrophil recruitment and antimicrobial defense78. Both the favorable and unfavorable findings of the immune simulation should be interpreted with caution, since they represent supportive in silico analysis only, and experimental studies will be required to determine the actual immunological response elicited by the vaccine. A limitation of the immune simulation was the weak CD8+ T-cell response. Under the specific C-ImmSim conditions used, non-memory CD8+ T-cells increased only after the first injection, whereas memory CD8+ T-cells remained absent, so the simulation did not support the expected vaccine’s CTL component. However, we do not believe that this finding invalidates the rationale for including CTL epitopes in the construct. In a multi-epitope vaccine, class I-restricted epitopes are incorporated to broaden the theoretical capacity for MHC-I presentation, but epitope prediction and peptide-HLA docking mainly indicate presentation potential and do not by themselves guarantee a measurable CD8+ response in a given in silico setting79,80. This point is particularly relevant here because CTL epitope selection was performed across a broad HLA class I reference panel, whereas the immune simulation was conducted under a narrower, user-defined HLA setting and therefore reflected only a limited part of the class I presentation landscape considered during epitope selection81. In addition, robust CD8+ priming to subunit antigens depends strongly on dendritic-cell cross-presentation and adjuvant context, which are not fully captured by immune simulation82. We therefore interpret the limited CD8+ signal mainly as a limitation of the simulated formulation and context rather than as a complete rejection of the CTL design rationale. Peptide-HLA docking in the present study should be interpreted only as supportive structural analysis, since it was used to assess whether the selected CTL and HTL epitopes could adopt plausible groove-binding conformations in representative HLA molecules, not to demonstrate intracellular processing, peptide loading, or in vivo T-cell priming83. The same applies to the TLR4/MD-2 docking. Selection of TLR4/MD-2 was biologically reasonable because the construct includes the N-terminal RS09 sequence, which has been reported as a synthetic TLR4-stimulating peptide capable of inducing NF-kB activation and inflammatory signaling41. However, docking of the complete recombinant construct to a single TLR4/MD-2 heterodimer should still be regarded as a structurally motivated exploration of a plausible receptor interface rather than as evidence of receptor activation. Within that framework, the 100 ns molecular dynamics simulation provides additional support because the trajectory was more consistent with early relaxation of the docked pose than with interface failure. In protein-protein MD, RMSD alone is not sufficient to evaluate complex behavior, since preservation of the interface is better judged together with contact persistence and interface area84. In our case, the modest upward drift of the complex RMSD was not accompanied by signs of dissociation, as the receptor backbone remained comparatively stable, the vaccine self-fit RMSD remained lower than the receptor-aligned vaccine RMSD, the radius of gyration stayed narrow, the receptor-vaccine center-of-mass distance stabilized after its initial increase, buried interface area and intermolecular hydrogen bonds persisted, and interface analyses showed recurrent contact residues and persistent interfacial hydrogen-bond pairs. Together, these findings are more consistent with relative repositioning and conformational accommodation of the flexible construct against the receptor than with wholesale unfolding or progressive loss of the interface84,85. Thus, the MD results strengthen the structural plausibility of the selected pose after relaxation from the initial docking state, but they still do not demonstrate signaling competence, innate immune activation, or biological binding in vivo. Comparison with recent in silico S. aureus multi-epitope vaccine studies shows that favorable computational profiles have also been reported in constructs built around more narrow antigenic scopes. A 2025 LukED-based design centered only on LukD and LukE selected only epitopes with > 90% conservancy, and reported a stable interaction with TLR1/2 by docking and molecular dynamics, but it did not include an immune simulation and achieved a global population coverage of 65.15%. In that study, TLR1/2 selection was linked to the hBD3 adjuvant, and the docking pose that advanced for further analysis was described as the output with the best interaction pattern and the lowest weighted score86. Another 2025 study focused on Clumping factor A (ClfA) selected 61 conserved epitopes, reporting over 50% global population coverage together with an immune simulation showing increased immunoglobulins, memory B-cell and CD4+ T-cell populations, macrophages, and cytokines such as IFN-γ, IL-10, IL-12, and IL-23. However, upon inspection of the immune simulation graphs in the study, the vaccine did not demonstrate evidence of in silico memory CD8+ generation. In addition, although that study compared TLR2, TLR3, and TLR4 and stated that final docked models were selected on the basis of HADDOCK score and cluster size, receptor prioritization for downstream interpretation still relied mainly on docking-derived affinity and interface metrics, with TLR4 emphasized because of its more favorable binding free energy and interaction profile87. Finally, a vaccine design based solely on acetolactate synthase reported 97.55% global population coverage, stronger docking to TLR4 than TLR2, and an immune simulation characterized by increasing total and memory B-cell populations, sustained immunoglobulin responses, elevated IFN-γ, and a strong CD4+ T-cell response, although it did not report an epitope conservancy analysis and did not present a dedicated CD8+ simulation panel. In that study, the TLR4 complex was selected for further analysis primarily because it showed the more favorable binding affinity and interface-interaction profile relative to TLR288. Taken together, these studies show that recent S. aureus multi-epitope designs based on single antigens have often produced favorable docking, variable population-coverage percentages, and variable immune-simulation results, although conservancy analysis, memory CD8+ T-cell numbers in the immune simulations, and the biological rationale used to prioritize receptor complexes for molecular dynamics have varied across studies. In particular, none of these studies clearly described predefined structural criteria for choosing a specific docked pose for molecular dynamics in the way applied in the present work, and in most cases receptor selection depended mainly on docking-derived affinity or interface-related metrics rather than on a more explicit biological rationale. The present study resembled these reports in showing acceptable epitope screening profiles, favorable structural refinement, and supportive receptor interaction analyses. In this context, the projected world population coverage of 88.81% may be interpreted as being on the higher end relative to these recent reports, exceeding the LukED and ClfA constructs while remaining below the acetolactate synthase construct. On the other hand, the immune profile of the present study should not be interpreted as objectively less favorable than these studies, since direct comparison of CD8+ memory remains weak or unmentioned across the three studies. The present study applied a more structured biological rationale for innate-receptor and docking-pose selection, since TLR4/MD-2 was selected because of the inclusion of the RS09 sequence and the docked pose for molecular dynamics was chosen according to predefined structural criteria. Overall, this comparison suggests that the main distinction of the present construct lies not in an objectively superior in silico performance in all categories, but in the fact that it integrates three mechanistically distinct virulence functions within a single design while applying a more explicitly biologically guided framework for receptor selection and structural interpretation. The limitations of this study include the following. First, vaccine design and evaluation were performed only in silico, so experimental validation is still needed to confirm expression, purification, antigenicity, structural integrity, and biological function. Second, the CTL component remains insufficiently validated, since the immune simulation showed only a limited CD8+ response after the first dose and no memory CD8+ T-cell formation. In addition, proteasomal cleavage and TAP transport predictions were not performed. Third, cytokines were not predicted at the individual epitope level and were inferred only from the C-ImmSim output. Fourth, epitope conservancy was uneven, with several FnBPA-derived epitopes showing low conservancy and one retained SpAKKE-derived CTL epitope showing 0% conservancy under the applied 100% identity threshold. Finally, peptide-HLA docking and TLR4/MD-2 docking, even when supported by molecular dynamics, remain supportive structural analyses only and do not demonstrate intracellular processing, receptor activation, or biological signaling. Future studies should therefore include experimental validation, processing-level CTL analysis, receptor activation assays, and in vivo evaluation of immunogenicity, protection, and immune memory. Conclusion An in silico multi-epitope vaccine incorporating FnBPA, Hla and a mutated SpA against S. aureus was successfully designed, demonstrating promising in silico results such as high stability, solubility, Ramachandran favored scores, and high population coverage worldwide. Furthermore, molecular docking supported plausible binding conformations of the selected epitopes to their representative HLA molecules and suggested a stable interaction between the vaccine construct and the TLR4/MD-2 heterodimer. The immune simulation showed a B-cell response, an IgG1 response after the first immunization and IgM throughout the immune simulation. It also indicated CD4+ T-cells and a Th1 response. Molecular dynamics simulation supported the persistence of the receptor-vaccine interface throughout 100 ns with hydrophobic interactions predominating among the identified non-covalent interactions, together with hydrogen bonds and salt bridges. In silico cloning reinforced the feasibility of inserting the construct into the pET-30a (+) expression vector for potential expression in E. coli. These findings identify this construct as a promising candidate for future in vitro and in vivo studies. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1 Supplementary Material 2 Supplementary Material 2 Supplementary Material 3 Supplementary Material 3 Supplementary Material 4 Supplementary Material 4 Supplementary Material 5 Supplementary Material 5 Supplementary Material 6 Supplementary Material 6 Supplementary Material 7 Supplementary Material 7 Supplementary Material 8 Supplementary Material 8 Supplementary Material 9 Supplementary Material 9