The Science
How Personalised mRNA Cancer Vaccines Work — and What We Know
This page covers the mechanism, the human clinical evidence, the comparative oncology foundation, and the limitations — in full. It is written for anyone who wants the complete scientific picture.
How Personalised mRNA Cancer Vaccines Work
Cancer cells carry mutations that healthy cells do not. Some of those mutations produce altered proteins — called neoantigens — that appear on the surface of cancer cells. The immune system can, in theory, recognise these neoantigens as foreign and attack the cells that display them. In practice, tumours evolve ways to hide from the immune system, and the immune response often needs help getting started.
A personalised mRNA cancer vaccine provides that help. It works in four steps: sequence the tumour's DNA and identify the mutations; predict which mutations produce neoantigens the immune system is most likely to recognise; encode those neoantigens into an mRNA sequence packaged in lipid nanoparticles; and inject it.
The patient's own cells read the mRNA, produce the neoantigen proteins, and present them to the immune system. The immune system then mounts a targeted response against any cell displaying those neoantigens — including the tumour cells that carry the original mutations.
The mRNA itself is transient. It degrades within days and does not integrate into DNA. The lipid nanoparticle delivery system is the same technology used in billions of COVID-19 vaccine doses, with a well-characterised safety profile.
What the Human Clinical Data Shows
The strongest evidence comes from human clinical trials, which are more advanced and more rigorously controlled than anything yet done in animals. The results are promising but vary significantly depending on the cancer type, disease stage, and whether the vaccine is combined with other immunotherapies.
The Strongest Results: Post-Surgical Adjuvant Use
Personalised mRNA vaccines have shown their best results when given after a tumour has been surgically removed, to prevent recurrence. In this setting, the immune system only needs to eliminate small amounts of residual cancer cells rather than fight an established tumour.
Melanoma
Moderna/Merck — mRNA-4157/V940 + pembrolizumab
49% reduction in risk of recurrence or death at five-year follow-up compared to pembrolizumab alone in resected high-risk melanoma.
This is the most mature dataset and the basis for an ongoing Phase III trial. The first personalised mRNA cancer vaccine regulatory approval is expected based on this programme.
Pancreatic Cancer
BioNTech — autogene cevumeran
50% of patients mounted a measurable immune response. Responders showed dramatically better outcomes — at 3.2 years, only 2 of 8 responders had relapsed, compared to a median recurrence-free survival of 13.4 months for non-responders.
Pancreatic cancer has a 13% five-year survival rate. Any signal in this cancer type is significant. The correlation between immune response and clinical outcome is the strongest evidence that the vaccine mechanism works as intended.
Triple-Negative Breast Cancer
BioNTech — individualised neoantigen vaccine
11 of 14 patients disease-free after more than six years. T-cell responses induced by vaccination were still detectable years later.
Durable responses and persistent immune memory suggest that a single course of treatment can produce long-lasting protection against recurrence.
Renal Cell Carcinoma
Phase I study
100% recurrence-free survival at 40.2 months post-surgery in vaccinated patients.
Small cohort, but the signal is strong. Another indication where the adjuvant setting appears highly favourable.
Harder Territory: Advanced and Metastatic Disease
Results in advanced cancer — where the tumour is large, established, and actively suppressing the immune system — are less impressive. Early melanoma trials showed objective response rates in small cohorts, but these included concurrent immunotherapy, making it difficult to isolate the vaccine's contribution.
In metastatic solid tumours more broadly, durable responses have been rare. The immune system struggles to overcome a large, immunosuppressive tumour even when primed by a vaccine.
This matters for the veterinary application. Many pet owners will present animals with advanced, inoperable disease — exactly the setting where human data is weakest. The honest expectation should be that personalised mRNA vaccines are most likely to help animals with smaller, localised tumours (especially post-surgical) and least likely to help animals with widespread metastatic disease.
Why Companion Animals Are Valid Translational Models
The US National Cancer Institute's Comparative Oncology Program has spent decades validating dogs as translational models for human cancer. The scientific basis is strong: canine and human cancers share the same oncogenes (TP53, PIK3CA, BRAF), develop through similar molecular pathways, and respond to the same immunological mechanisms.
Dogs develop cancer spontaneously — unlike laboratory mouse models where tumours are artificially induced. This means canine tumours exhibit the same heterogeneity, immune evasion, and microenvironment complexity seen in human patients. A treatment that works in dogs with spontaneous cancer is more likely to translate to humans than one that works only in engineered mouse models.
Specific parallels are well-documented. Canine osteosarcoma mirrors human osteosarcoma in genomic landscape, metastatic pattern, and clinical behaviour. Canine melanoma shares key driver mutations with human melanoma. Canine mast cell tumours, while not directly homologous to a common human cancer, provide a model for studying neoantigen-driven immune responses in a spontaneous tumour setting.
This is not a new or speculative claim. The NCI has funded comparative oncology trials for over fifteen years. Multiple drugs now used in human oncology were tested in dogs first. The innovation here is applying personalised mRNA vaccines — a technology proven in human trials — through the same comparative framework.
The Comparative Oncology Dataset
Every treated animal generates a matched dataset: tumour genomic profile, the somatic mutations identified, the neoantigens selected for the vaccine, the vaccine construct design, and the measured tumour response. This is a closed loop — prediction linked to outcome — and it is exactly what the field of neoantigen prediction needs to improve.
Current neoantigen prediction algorithms are imperfect. In one recent human trial, a median of only 2 out of 20 predicted neoantigens actually triggered T-cell responses. The algorithms predict binding affinity reasonably well, but binding does not equal immunogenicity. The gap between prediction and actual immune response is the central unsolved problem in personalised cancer vaccines.
A dataset of hundreds of canine cases — each linking predicted neoantigens to measured tumour response — addresses this problem directly. It reveals which prediction features actually correlate with clinical outcomes, allowing the prediction model to improve with every case treated. This is the Foundation Medicine insight applied to personalised mRNA oncology: the data asset compounds, and the first mover's advantage grows with every case.
For human pharmaceutical companies — BioNTech, Moderna, Merck — this dataset is directly relevant. They are running their own personalised cancer vaccine trials in humans and seeking real-world evidence to de-risk their programmes. A comparative oncology dataset showing which neoantigen predictions produced tumour responses in dogs, with matched genomic and outcome data, is evidence they cannot generate on their own timeline.
What Gets Better With Every Case
Neoantigen prediction
As the outcome dataset grows — linking predicted neoantigens to measured tumour responses — the prediction model improves. Which binding affinity thresholds actually matter? Which expression levels correlate with immune response? Which clonality patterns predict durable responses? Every treated animal provides answers.
Checkpoint inhibitor combinations
Gilvetmab — a canine anti-PD-1 monoclonal antibody — has received a conditional USDA licence. As veterinary checkpoint inhibitors become available and protocols are refined, combination therapy is expected to significantly improve response rates, mirroring the human experience where vaccine plus checkpoint inhibitor consistently outperforms either alone.
Patient selection
With data from dozens and eventually hundreds of treated animals, the ability to predict which patients will benefit improves. Tumour mutation burden thresholds, cancer type suitability, DLA allele confidence — all become more precise with accumulated evidence. Better patient selection means fewer animals undergo treatment unlikely to help them.
Second-line vaccines
When a tumour does not respond or partially responds, the genomic data can be re-analysed. New mutations that have emerged, or alternative neoantigen selections based on updated prediction models, can be used to design a second vaccine targeting different epitopes. This iterative approach is already being pursued in human trials.
Have a Question About the Science?
We welcome questions from veterinary professionals, researchers, and dog owners. Get in touch.
What We Do Not Yet Know
Intellectual honesty requires stating the limitations clearly. These are the gaps in current knowledge that affect this work.
Neoantigen prediction accuracy
Current algorithms predict MHC binding affinity reasonably well but struggle to predict immunogenicity — whether a neoantigen that binds will actually trigger a useful immune response. Human trial data suggests that most predicted neoantigens do not elicit T-cell responses. Improving this prediction is the core scientific challenge and the core purpose of the outcome dataset.
Canine DLA allele databases
The canine MHC (DLA) allele database is significantly less comprehensive than the human HLA database. Some breeds have well-characterised DLA alleles; others — particularly mixed breeds — may have alleles with limited representation in the binding prediction training data. This introduces uncertainty in neoantigen prediction quality that varies by patient.
Tumour heterogeneity
A single biopsy captures only part of a tumour. Different regions of the same tumour may carry different mutations, and metastatic sites may have diverged from the primary tumour. A vaccine designed from one biopsy may target neoantigens absent from other parts of the tumour. This is a fundamental challenge shared with human personalised cancer vaccines.
No veterinary-specific efficacy data yet
All response rate estimates on this site are derived from human clinical trials. We do not yet have efficacy data from treated companion animals. The first treated cohort will generate this data. Until then, human data and the comparative oncology literature are the best available evidence — but they are not veterinary-specific evidence.
Advanced disease
The human evidence is weakest in the setting most veterinary patients will present with: advanced, inoperable cancer with established tumour burden. Response rates in this setting are expected to be lower than in the adjuvant (post-surgical) setting. Combination with checkpoint inhibitors may improve outcomes but adds complexity and cost.
Questions About the Science?
We welcome questions from veterinary professionals, researchers, and dog owners. If you want to discuss the scientific basis in more detail, or if you have expertise to contribute, please get in touch.