The strategy, termed Proactive Immune Training (PIT), draws its inspiration from an observation that has become increasingly clear over the course of the COVID-19 pandemic. People who experienced natural infection followed by vaccination—so-called hybrid immunity—tend to mount broader and more durable protection against reinfection and severe disease than those receiving either alone. The explanation lies in the evolutionary arms race that unfolds during a natural infection: as the virus mutates within the host, immune pressure selects for B cells capable of recognizing an expanding repertoire of antigenic shapes. This dynamic training drives somatic hypermutation and the eventual emergence of broadly neutralizing antibodies that can accommodate variant-level changes. The catch is obvious. Acquiring immunity through infection carries real risks of severe illness and death, and existing mRNA vaccines present only static antigens that cannot evolve inside the host, so they never recreate this dynamic antigenic landscape.

PIT was designed to reproduce that landscape synthetically, without the danger of infection. Using SARS-CoV-2 as a model, the research team, led by Xiangrong Song of Sichuan University, first turned to computational prediction. They applied the PyR0 pipeline, a hierarchical Bayesian regression framework trained on approximately 6.4 million SARS-CoV-2 genomes from GISAID, to estimate the fitness contribution of potential spike protein mutations. The model, which achieved an R-squared of 0.983 for fitness estimates in prior validation, ranks mutations by a Z-score that incorporates estimation uncertainty. From this global fitness landscape the researchers selected 30 high-probability mutation sites in the Spike protein, including well-characterized immune escape hotspots such as residues E484, K417, and N501, along with sites like L452, S477, N501, and P681 that have repeatedly defined variants of concern.

With the 30 sites in hand, the team performed saturation mutagenesis, systematically substituting every possible amino acid at each position to generate a DNA plasmid library of 600 distinct Spike antigens: 570 unique single-point mutants plus the wild-type sequence. Quality control was rigorous. Next-generation sequencing confirmed that residual wild-type had been reduced to 0.25 percent of the library, coverage of the intended mutants reached 98.4 percent, and a Gini coefficient of 0.17 indicated that the genetic diversity was evenly distributed rather than skewed toward a few dominant clones. Translational fidelity was verified by expressing ten representative mutants in HEK293T cells and quantifying Spike production by ELISA, which showed expression levels consistent with the wild-type protein. The plasmid pool was then transcribed in vitro into a matching mRNA library, ensuring that each training antigen retained its structural integrity while presenting deliberate variation at the key hotspots where immune escape is most likely to occur.

Delivering 600 different mRNA species simultaneously required an efficient and homogeneous delivery vehicle. The researchers formulated their library in lipid nanoparticles built around CMP1, an ionizable lipid developed in-house and protected under US Patent 11,839,657. Using microfluidic mixing, they produced particles averaging roughly 100 nanometers in diameter with a low polydispersity index of about 0.2. Nanoflow cytometry revealed an mRNA encapsulation efficiency of 90 percent and a remarkably low empty-particle ratio of 1.7 percent. In vitro, CMP1-based nanoparticles doubled the transfection efficiency achieved by the common reagent Lipofectamine 2000, and cryo-electron microscopy confirmed uniform, quasi-spherical particles with intact lipid bilayers. In mice, the platform showed preferential uptake by splenic dendritic cells and neutrophils, and confocal imaging demonstrated superior endosomal escape compared with the ALC-0315 lipid used in clinically approved mRNA vaccines, a critical property because mRNA trapped in lysosomes is degraded before it can be translated.

The immunization regimen itself was carefully staged. Mice first received a prime with a conventional single-antigen vaccine encoding the Delta variant Spike protein, establishing a foundational pool of memory B cells. Fourteen days later, a subset received a booster dose of the PIT vaccine containing the full 600-antigen m-library. The logic mirrors natural infection: the prime creates a high-quality substrate of memory, while the boost challenges that memory with a dense landscape of related but distinct variants, driving affinity maturation toward clones that can tolerate or recognize multiple mutational states. Control groups received two doses of the single-antigen vaccine, two doses of the PIT vaccine alone, or a hexavalent cocktail of six whole-Spike mRNAs spanning the wild-type, Alpha, Beta, Gamma, Delta, and Omicron BA.1 variants, providing a direct benchmark against a rational multivalent design.

The results of the prime-boost sequence were striking. The SA+PIT regimen elicited significantly higher serum IgG binding titers and pseudovirus neutralization titers across a panel of ten SARS-CoV-2 variants, including Alpha, Beta, Gamma, Delta, Omicron BA.2, BA.2.75, BA.5, and BF.7, compared with two doses of the single-antigen vaccine. Notably, BA.5 and BF.7 had not yet emerged when the PyR0 model was trained, so the breadth against these variants demonstrates genuine anticipatory capacity. When compared head-to-head with the hexavalent cocktail, the two strategies performed similarly against BA.2, which is antigenically close to the cocktail’s Omicron component, but the PIT-boosted group generated significantly higher binding and neutralizing titers against the more divergent BA.5 and BF.7 variants, with P values below 0.01. The authors interpret this as evidence that dense mutational coverage fills the antigenic gaps between discrete variants of concern, providing better protection against evolutionary drift within the predicted mutational space than a mixture of known strains.

Cellular immunity showed a parallel enhancement. Flow cytometry of splenocytes revealed that the SA+PIT regimen significantly increased the frequency of IFN-γ-producing CD4-positive and CD8-positive T cells, expanded the T follicular helper cell population that supports antibody maturation, and boosted IFN-γ production by B cells, while IL-2 responses remained comparable between vaccinated groups. Single-cell T cell receptor and B cell receptor sequencing, performed ten days after a rapid day-0/day-7 immunization schedule, showed markedly greater diversity in the BCR heavy chain and both kappa and lambda light chain repertoires, with a larger fraction of unique CDR3 sequences, the hypervariable regions that determine antigen specificity. The repertoires also displayed a transcriptomic shift toward the IgA subclass, hinting at broader isotype diversification, although mucosal IgA protein was not directly measured. Importantly, safety evaluations at a higher 50-microgram dose showed no significant changes in blood chemistry markers and no histopathological abnormalities in the heart, liver, spleen, lung, or kidney.

The authors are candid about the study’s limitations. The work remains a proof of concept based on surrogate endpoints such as ELISA binding and pseudovirus neutralization; no live-virus challenge experiments were performed, and protective efficacy must be confirmed in K18-hACE2 transgenic mice or non-human primates. Mechanistic cohorts were small, the BALB/c mouse model incompletely recapitulates human immunity, the study did not benchmark against clinically licensed mRNA vaccines such as BNT162b2 or mRNA-1273, and the manufacturing complexity of a 600-mutant library presents substantial hurdles for clinical-grade production. Germinal center dynamics, affinity maturation kinetics, and the precise epitopes underlying cross-reactivity also remain to be mapped. Yet the framework is inherently adaptable: the same combination of computational fitness prediction, saturation mutagenesis, and LNP delivery could, in principle, be extended to influenza or entirely novel pathogens, with artificial intelligence models predicting mutational landscapes directly from sequence data. The broader significance is conceptual. Vaccines have long been built to match the past; PIT suggests they could instead be built to anticipate the future, presenting the immune system with a cloud of evolutionary possibilities and letting it train against them before the virus ever gets the chance.

Subject of Research: A proactive mRNA vaccine strategy using a library of predicted SARS-CoV-2 Spike mutants to broaden antiviral immunity against future variants

Article Title: A proactive immune training strategy to generate anticipatory immunity against viral evasion

Article References: Qin, S., Teng, Y., Xin, J., Zhang, Y., Huang, L., Zhao, S., Jiao, X., Yin, X., Liu, S., Sun, J., Xu, W., Xie, Y., Liu, J., Kong, L., & Song, X. (2026). A proactive immune training strategy to generate anticipatory immunity against viral evasion. Journal of Advanced Research. https://doi.org/10.1016/j.jare.2026.08.059

Image Credits: AI Generated

DOI: 10.1016/j.jare.2026.08.059

Keywords: SARS-CoV-2, mRNA vaccine, Proactive Immune Training, viral immune evasion, broadly neutralizing antibodies, lipid nanoparticles, saturation mutagenesis, variant of concern, BCR repertoire, hybrid immunity, PyR0 model, vaccinology