M. fermentans is a clinically important opportunistic mycoplasma, a group of bacteria that lack a peptidoglycan cell wall. It has been detected in human genitourinary tissues and associated with respiratory and urogenital infections, chronic inflammatory conditions including rheumatoid arthritis, and neurological disease. The source article describes the organism as capable of subverting immune recognition and following a prolonged or chronic course of infection. Its prevalence remains uncertain. Macrolides are used as first-line drugs, although resistant strains have been reported, and no specific vaccine is currently available. To identify vaccine targets without first cultivating the pathogen, the researchers applied reverse vaccinology and immunoinformatics to the M64 strain genome.

The analysis began with the amino acid sequences of six stress-associated proteins: DnaK, DnaJ, trigger factor, a putative heat-shock ATP-dependent protease, an ATPase subunit and thioredoxin reductase. The proteins were evaluated for predicted antigenicity, solubility, toxicity, molecular weight, instability, signal peptides and transmembrane regions. All six met the initial screening criteria, which included antigenicity scores above 0.4, solubility scores above 0.45 and no predicted toxicity. DnaK and DnaJ are components of the HSP70 chaperone system, which assists protein folding and helps bacteria respond to thermal and other stresses. Thioredoxin reductase is part of a redox-regulatory system involved in maintaining cellular redox balance and responding to oxidative stress. The targeted proteins may therefore be relevant to M. fermentans survival in host immune cells, where organisms encounter changes in temperature, pH, nutrient availability and other conditions.

Computational epitope screening reduced the six proteins to 15 peptide candidates. ABCpred, VaxiJen 2.0, AllerTOP 2.0 and ToxinPred were used to identify linear B-cell epitopes with strong predicted recognition and no predicted allergenicity or toxicity. Six peptides were retained, with ABCpred scores ranging from 0.85 to 0.91. For cytotoxic T-cell epitopes, the researchers used the IEDB MHC I binding server and NetMHCpan 4.1, selecting peptides with median binding percentile ranks below 0.5 and positive immunogenicity scores. This process yielded four candidates. A parallel MHC II workflow, including an interferon-gamma prediction tool, produced five helper T-cell epitopes. Molecular docking examined the predicted interactions between peptide structures and selected HLA molecules in the peptide-binding groove, while the IEDB population-coverage tool assessed the HLA alleles represented across 16 regions.

Predicted population coverage was generally above 98%, but one regional estimate was substantially lower: South Africa was estimated at 49.73%. The contrast indicates that broad coverage cannot be assumed to be uniform across populations. The immune simulation itself used a restricted set of HLA parameters—HLA-A*30:01, HLA-A*11:01, HLA-DRB1*04:05, HLA-DRB1*15:01, HLA-B*08:01 and HLA-B*07:02—and therefore did not represent every human HLA background. The coverage results are computational estimates based on selected alleles, not measurements of immune protection, safety or efficacy in any population. The low-confidence global structure of the artificial construct also meant that its docking and molecular-dynamics results had to be interpreted as model-dependent analyses rather than direct observations of molecular behavior.

The final construct placed beta-defensin at the N-terminus behind a rigid EAAAK linker, followed by the B-cell epitopes connected with KK linkers, the cytotoxic T-cell epitopes joined with AAY linkers and the helper T-cell epitopes separated with GPGPG spacers. A dendritic-cell-targeting signal peptide was added at the C-terminus to support antigen presentation. The linkers were selected to provide spacing and conformational flexibility while separating the functional components. Sequence-based tools classified the resulting 318-amino-acid protein as a probable immunogen with a predicted probability of 100% and as non-allergenic, although those outputs remain predictions. Its predicted molecular weight was 34.6 kDa, its Protein-Sol solubility score was 0.729, its instability index was 17.99, its aliphatic index was 61.07 and its GRAVY value was −0.8971. These numerical features describe expected sequence-level properties rather than demonstrating expression, stability in cells or performance in a host.

Structural modelling produced a refined AlphaFold model but also exposed substantial uncertainty. PSIPRED and SOPMA assigned 33.33% of the sequence to alpha helices, 17.30% to extended beta strands, 7.23% to beta turns and 42.14% to random coils. The model was refined with GalaxyRefine and assessed with ProCHECK, ProSA-web and ERRAT. In the refined structure, 95.2% of residues occupied the most favoured Ramachandran regions and 4.8% occupied additional allowed regions, with none in generously allowed or disallowed areas. The ERRAT quality factor increased from 97.64 to 100, and the ProSA-web Z-score shifted from −3.66 to −4.08. By contrast, pLDDT values were high only near the N-terminal region and generally low elsewhere, while predicted aligned error was elevated for many nonadjacent residue pairs. The study’s own analysis suggests that the artificial architecture and flexible linkers may contribute to this pattern, although that explanation was not established causally. Refinement improved local stereochemical and atomic-contact assessments but did not verify the accuracy of the global fold.

ElliPro identified four exposed linear B-cell regions in the refined model, with scores of 0.783, 0.761, 0.674 and 0.672. The coding sequence was codon-optimised for Escherichia coli K12, yielding a 959-nucleotide sequence with 49.2% GC content and a codon adaptation index of 1.0. In silico insertion into the pET-28a (+) vector produced a 6,288-bp construct, showing that an expression design could be created computationally; actual protein production was not reported. RNAfold predicted stable mRNA secondary structures, with minimum free energies of −289.90 kcal/mol for the optimal structure and −306.95 kcal/mol for the centroid structure. A BLASTp search against the human reference proteome found no significant similarity among the 15 selected pathogen-derived epitopes. The principal match was to human beta-defensin 103 precursor, reflecting the deliberately included adjuvant rather than the selected M. fermentans epitopes. Sequence analysis alone could not exclude immunological consequences associated with that human-derived component.

The immune simulations predicted coordinated humoral, cellular and innate responses, but they did not observe an immune system. In the C-ImmSim model, IgM, IgG1 and IgG2 rose during simulated secondary and tertiary responses as antigen levels declined, while simulated B-cell populations remained elevated. T-cell, natural killer cell, macrophage, dendritic-cell and epithelial-cell responses also increased under the model’s assumptions. Molecular docking placed the construct in a predicted complex with TLR2, a receptor selected because of its role in immune responses involving M. fermentans lipoproteins; the reported total MM-GBSA binding energy was −44.28 kcal/mol. Three independent 100-nanosecond molecular-dynamics simulations showed a common early adjustment followed by bounded fluctuations. Radius of gyration decreased from approximately 3.7 nanometres to roughly 3.3–3.5 nanometres, but the replicates differed in root-mean-square deviation, solvent-accessible surface area, hydrogen-bond counts and free-energy landscapes. Normal-mode analysis further characterized predicted flexibility. Together, these analyses are consistent with a computationally plausible candidate, not an established biological effect.

The results identify a testable candidate rather than a demonstrated vaccine. All evidence came from sequence analysis, epitope prediction, structural modelling, docking, immune simulation and molecular dynamics, with no reported in vitro expression, biochemical characterization, immunization study, animal challenge or human trial. Prediction algorithms can disagree with biological reality, and a protein assembled from epitopes and functional modules does not necessarily adopt a single stable native conformation. The authors also note that protein-focused reverse-vaccinology pipelines may omit potentially immunogenic carbohydrate-based or glycolipid-based determinants. Experimental expression and verification, followed by evaluation of immunogenicity, safety and protection in appropriate models, are required before the construct can be said to prevent M. fermentans infection, reduce disease or provide durable immunity.

Subject of Research: Computational design of a multi-epitope vaccine against Mycoplasma fermentans stress-associated proteins

Article Title: Rational design of a multi-epitope vaccine targeting the stress-associated proteins for Mycoplasma fermentans via immunoinformatics

Article References: He, J., Mi, J., Tang, H., Zhang, Z., & Chen, Z. (2026). Rational design of a multi-epitope vaccine targeting the stress-associated proteins for Mycoplasma fermentans via immunoinformatics. Frontiers in Bioinformatics, 6, Article 1994081. https://doi.org/10.3389/fbinf.2026.1994081

Image Credits: AI Generated

DOI: 10.3389/fbinf.2026.1994081

Keywords: Mycoplasma fermentans, multi-epitope vaccine, reverse vaccinology, immunoinformatics, heat shock proteins, thioredoxin reductase, epitope prediction, immune simulation, molecular dynamics, vaccine design, Rational, design