Computational pipeline yields three mRNA vaccine designs against Rift Valley fever virus

Rift Valley fever virus, a mosquito-borne zoonotic pathogen that has repeatedly devastated livestock herds and killed people across Africa and the Arabian Peninsula, still has no fully approved vaccine for human use. A new study published in the journal Discover Immunity reports the design of three multi-epitope mRNA vaccine candidates against the virus, built entirely through computational immunoinformatics. The authors, led by Elijah Kolawole Oladipo, stress that every result in the work is predictive and that the constructs remain untested in any laboratory or animal system.

The research team began with 1,476 Rift Valley fever virus protein sequences retrieved from the National Center for Biotechnology Information database as of 4 June 2025. After screening for antigenicity and immunogenicity, only 130 sequences met the selection criteria. Multiple sequence alignment using the MUSCLE algorithm in MEGA 12 narrowed this set to 23 proteins with sufficiently conserved regions, all of which retained their predicted antigenic and immunogenic character. Conserved regions were prioritised because epitopes shared across viral strains offer a route to broader protection.

From these 23 conserved proteins, the researchers predicted three classes of epitopes using paired, independent servers. Cytotoxic T-lymphocyte epitopes were identified with NetMHCpan 4.1 and NetCTL v1.2, helper T-lymphocyte epitopes with NetMHCIIpan 4.0 and the IEDB MHC-II binding tool, and linear B-cell epitopes with ABCPred and Bepipred. Only consensus epitopes found by both tools in each pair, and which passed filters for antigenicity, non-allergenicity and non-toxicity, were carried forward. Helper T-cell epitopes were additionally screened for their predicted ability to induce interferon-gamma, interleukin-6 and interleukin-10.

To reduce the risk of autoimmune cross-reactivity, the selected epitopes and the final vaccine constructs were compared against the human reference proteome using BLASTp. The analysis found no significant similarity between the vaccine sequences and human proteins, which the authors interpret as a reduced likelihood of off-target immune responses. Each of the three constructs incorporated 8 linear B-cell epitopes, 8 cytotoxic T-cell epitopes and 5 helper T-cell epitopes, joined by a defined linker scheme: EAAAK linkers connecting the adjuvant to helper T-cell epitopes, GPGPG linkers within the helper T-cell segment, KK linkers between helper T-cell and B-cell segments, and AAY linkers for B-cell to cytotoxic T-cell junctions and within the cytotoxic segment.

The mRNA architecture included a 5-prime cap, a 5-prime untranslated region, a Kozak sequence, a tissue plasminogen activator signal peptide, the Human Beta Defensin 1 adjuvant, an MHC-I trafficking domain, a 3-prime untranslated region and a poly-A tail. Antigenicity scores from VaxiJen 2.0 were 0.5075, 0.5161 and 0.5467 for Constructs 1, 2 and 3 respectively, all above the server’s default threshold of 0.4. AllerTop 2.0 and ToxinPred 2 assessments classified all three constructs as non-allergenic and non-toxic.

Population coverage analysis with the IEDB tool, based on HLA allele frequency distributions, predicted that the selected cytotoxic and helper T-cell epitopes would be recognised across 95.94 percent of the global population for Construct 1, 94.71 percent for Construct 2 and 90.52 percent for Construct 3. Regional estimates for Construct 1 ranged from 98.54 percent in Europe and 94.92 percent in North America down to 72.04 percent in Oceania. The authors caution that these figures reflect predicted HLA binding rather than demonstrated immunogenicity, and that lower coverage in Oceania and South America underscores the influence of HLA diversity.

Physicochemical profiling with ExPASy ProtParam placed the molecular weights of the three constructs between roughly 138.7 and 139.5 kilodaltons, with theoretical isoelectric points of 9.05 to 9.24, instability indices below the threshold of 40, aliphatic indices of 75.40 to 77.55 and negative GRAVY scores indicating hydrophilic character. Protein-Sol predicted all three constructs to be soluble, with scores of 0.372, 0.350 and 0.364. Tertiary structures predicted with AlphaFold 3 and refined with GalaxyRefine showed 95.9, 94.4 and 93.1 percent of residues in favoured Ramachandran regions for Constructs 1, 2 and 3 respectively, which the authors describe as stereochemically sound models.

Molecular docking with the ClusPro server against Toll-like receptors TLR-2 and TLR-3, chosen because these innate immune receptors are involved in viral nucleic acid sensing, produced strong predicted binding interactions for all three constructs with both receptors. Normal mode analysis using the iMODS server yielded eigenvalues ranging from 3.33 x 10-8 to 1.67 x 10-7 across the vaccine-receptor complexes, which the authors interpret as indicative of rigid structural behaviour. However, they explicitly note that normal mode analysis cannot confirm long-term stability under physiological conditions and recommend extended all-atom molecular dynamics simulations with explicit solvent, including root mean square deviation, fluctuation, radius of gyration and binding free-energy calculations, in future work.

Immune simulation with the C-ImmSim server predicted that a single dose of each construct would activate dendritic cells, macrophages and epithelial cells, followed by proliferation of CD4-positive helper T cells, CD8-positive cytotoxic T lymphocytes and natural killer cells within the first few days, along with B-cell activation, plasma cell differentiation and antibody production across multiple isotypes. Codon optimisation for expression in Escherichia coli K-12 using the Java Codon Adaptation Tool produced codon adaptation index values of 0.9, 1.0 and 1.0, with GC contents of 55.2, 51.1 and 50.9 percent, and the sequences were virtually cloned into a pET-28a(+) vector.

Throughout the discussion, the authors are careful to frame the findings as computational prioritisation only. They acknowledge that epitope-prediction platforms carry training-dataset biases and incomplete modelling of antigen processing and T-cell recognition, that docking cannot confirm receptor activation or downstream signalling, and that C-ImmSim cannot reproduce mRNA delivery, antigen expression, host HLA diversity or protective efficacy in vivo. The conclusion is unambiguous: the three constructs show favourable predicted properties, but rigorous in vitro and in vivo validation is required to confirm antigen expression, immune activation, safety, immunogenicity and protective efficacy before any of them can be considered a viable vaccine candidate against Rift Valley fever virus.

Subject of Research: In silico design of multi-epitope mRNA vaccine candidates against Rift Valley fever virus

Article Title: In silico design of a multi-epitope mRNA vaccine candidate against rift valley fever virus using integrative immunoinformatics approaches

Article References: Oladipo, E. K., Ogunniran, J. A., Sokari, B. W., Akosile, A. O., Gbadegesin, H. O., Babarinde, J. M., Samson, O. O., Majolagbe, O. N., Olatunde, S. K., & Oloke, J. K. (2026). In silico design of a multi-epitope mRNA vaccine candidate against rift valley fever virus using integrative immunoinformatics approaches. Discover Immunity, 3(1), Article 16. https://doi.org/10.1007/s44368-026-00035-w

Image Credits: AI Generated

DOI: 10.1007/s44368-026-00035-w

Keywords: Rift Valley fever virus, mRNA vaccine, immunoinformatics, reverse vaccinology, epitope prediction, molecular docking, immune simulation, codon optimization, computational vaccine design, zoonotic disease, vaccine candidates, in silico