Personalized mRNA cancer vaccines

Personalized mRNA cancer vaccines are therapies in which a patient's tumour is sequenced alongside their healthy tissue, the mutations unique to that tumour are ranked by algorithm, and a synthetic messenger RNA encoding the highest-ranked of them is manufactured for that one person. The injected mRNA instructs the recipient's own cells to make and display those mutant protein fragments, on the expectation that T cells will then recognise and kill cells carrying them. The field calls the class individualized neoantigen therapy; in August 2026 one such product reported the first positive phase 3 result.

§How it works

A neoantigen is a protein fragment that exists only because a tumour mutated. It is absent from every healthy cell, which makes it a target the immune system can attack without attacking the patient. The difficulty is that neoantigens are almost entirely private to one tumour, so there is nothing to stock on a shelf.

The workflow answers that by manufacturing per patient. A biopsy and a blood sample are sequenced and compared, separating the tumour's mutations from the person's inherited variation. An algorithm then scores those mutations for the chance that a fragment carrying them will be displayed on the cell surface by that patient's own major histocompatibility complex molecules and recognised by a T cell. Moderna's product, intismeran autogene, encodes up to 34 of the top-ranked mutations in a single synthetic mRNA, made as a single-patient batch at the company's Massachusetts facilities and shipped to the trial site for intramuscular injection.¹ The antigens are translated inside the recipient's cells rather than injected as peptides, so they enter the normal antigen-processing pathway.

Every advanced programme pairs the mRNA with a checkpoint inhibitor: priming new T cells accomplishes little if the tumour microenvironment then switches them off, so the vaccine supplies specificity and the antibody removes the brake. The trials measure a combination.

The delivery problem is unlike the one facing Somatic gene therapy: nothing has to reach a particular organ or integrate anywhere, so the platform reuses the lipid-nanoparticle manufacturing built for infectious-disease vaccines and needs none of the viral vectors gene therapy depends on. The mRNA is transient and the durable product is immunological memory, which is why the intervention grades as only partly reversible: the dose clears, the expanded T-cell clones do not, and neither would an autoimmune reaction they caused.

§Development history

1. 2010
   First therapeutic cancer vaccine approved
   The FDA approves sipuleucel-T for metastatic castration-resistant prostate cancer. Its pivotal trial extended median survival by roughly four months and did not delay disease progression at all.
2. 2013–2016
   MAGE-A3 fails twice
   GlaxoSmithKline's shared-antigen immunotherapeutic misses its endpoints in a phase 3 melanoma trial and then in the largest adjuvant lung cancer trial ever run. Both are abandoned.
3. 2017
   Neoantigen vaccines reach patients
   Two phase 1 studies published in Nature, one using synthetic peptides and one using RNA, show that vaccines built from an individual's own tumour mutations induce neoantigen-specific T cells in people with melanoma.
4. 2019
   KEYNOTE-942 opens
   Moderna and Merck begin a randomised phase 2b of mRNA-4157 plus pembrolizumab against pembrolizumab alone in resected stage IIIB–IV melanoma.
5. 2023
   Breakthrough therapy designation
   In February the FDA grants the combination breakthrough therapy designation for adjuvant high-risk melanoma, on phase 2b data showing a 44% reduction in the risk of recurrence or death.
6. 2023
   Pancreatic cancer signal
   A phase 1 study of BioNTech's autogene cevumeran in 16 patients with resected pancreatic cancer reports vaccine-expanded T cells in half of them, and longer recurrence-free survival in those who responded.
7. 2024
   Phase 2b published
   Weber and colleagues report KEYNOTE-942 in The Lancet: 18-month recurrence-free survival of 79% with the combination against 62% with pembrolizumab alone, in 157 patients.
8. 2025
   A competitor misses
   Autogene cevumeran plus pembrolizumab fails to significantly improve progression-free survival against pembrolizumab alone in first-line advanced melanoma, despite inducing neoantigen-specific T cells in most patients tested.
9. 2026
   First positive phase 3
   Merck and Moderna announce on 19 August that INTerpath-001 met its recurrence-free survival primary endpoint and its distant metastasis-free survival secondary endpoint at a pre-planned interim analysis.

§The melanoma result

INTerpath-001 enrolled 1,137 people whose stage IIB–IV cutaneous melanoma had been completely resected and who had received no prior systemic therapy. They were randomised two to one — double-blind, with placebo injections in the control arm — to intismeran autogene at 1 mg every three weeks for up to nine doses plus pembrolizumab, or to pembrolizumab alone, for about a year. Merck and Moderna reported on 19 August 2026 that an independent data monitoring committee found statistically significant and clinically meaningful improvements in recurrence-free survival and in distant metastasis-free survival, with no new safety signals.¹ Moderna's shares more than doubled in early trading.² That is the whole of the disclosed result: no hazard ratio, event count or survival curve.

The published record behind it is the 157-patient phase 2b, KEYNOTE-942, which reported 18-month recurrence-free survival of 79% with the combination against 62% with pembrolizumab alone.³ At a median follow-up of 60.3 months the companies reported a 49% reduction in the risk of recurrence or death and a 59% reduction in the risk of distant metastasis or death. Fatigue, injection-site pain and chills were the common adverse events attributed to the therapy; immune-related events occurred in 45.2% of combination patients and 44% of those on pembrolizumab alone, so the mRNA did not appear to compound the checkpoint inhibitor's toxicity.⁴

What a topline announcement cannot settle

The August 2026 release states that two endpoints were met and supplies no numbers. Recurrence-free survival is not survival, and the companies say the trial continues in order to evaluate the latter. The nearest published comparison is unhelpful in a specific way: at five years the phase 2b's exploratory overall-survival hazard ratio was 0.471, but on fourteen events, with a confidence interval running from 0.165 to 1.345.⁴ That interval is consistent with a large benefit and with none.

§Why the field's history argues for caution

Therapeutic cancer vaccines have a long record of inducing immune responses and not changing outcomes. Sipuleucel-T, approved in 2010 and still the reference case, extended median survival by roughly four months in a 512-patient trial while having no measurable effect on time to progression.⁵ The MAGE-A3 immunotherapeutic, built on a single antigen shared across many tumours, failed in a phase 3 melanoma trial of roughly 1,300 patients and again in an adjuvant lung cancer trial that screened close to 14,000 surgical patients.⁶

The standard explanation is that shared antigens are also present, at low levels, in normal tissue, so the T cells capable of attacking them have largely been deleted or suppressed. Neoantigens escape that argument by construction, which is why the two 2017 phase 1 studies that first vaccinated melanoma patients against their own tumour mutations were read as a change of strategy rather than another attempt at the same one.⁷ Whether the prediction step is good enough is the weakest link. When a multi-team consortium compared neoantigen pipelines against T-cell recognition in the patients' own samples, the pipelines disagreed sharply about which peptides to rank first, and only a minority of top-ranked predictions were recognised.⁸ Immunogenicity prediction is a place where machine learning applied to biology is better than chance and short of reliable, and where neither structural design methods nor the simulation ambitions of Virtual cell models have closed the gap.

§Making one dose per person

The manufacturing model is the part with no precedent at scale. Each dose is a distinct product, made after the patient is enrolled, from that patient's sequencing data. Nothing can be stockpiled, a delay in sequencing is a delay in treatment, and quality control has to be repeated for a batch of one. Approved autologous cell therapies show that per-patient biologics can be commercialised, and also what that costs: they carry some of the highest list prices in medicine and are given at a handful of qualified centres.

No price has been announced for intismeran autogene and no regulator has approved it, so its cost is unknown. The eligible population is already defined: Merck and Moderna cite more than 330,000 melanoma diagnoses worldwide in 2022, and pembrolizumab is approved in the United States as adjuvant therapy after complete resection of stage IIB, IIC or III disease.⁴ Adding a bespoke biologic on top of an existing standard of care makes who can obtain a therapy a question about health systems rather than biology, and the sequencing infrastructure it assumes does not exist in most countries.

§Beyond melanoma

Moderna and Merck listed nine phase 2 and phase 3 trials of intismeran autogene as of mid-2026, across melanoma, non-small cell lung cancer, bladder cancer and renal cell carcinoma. None had reported.⁴

The competing programme is BioNTech and Genentech's autogene cevumeran, an RNA-lipoplex therapy encoding up to 20 neoantigens per patient and given intravenously. Its phase 1 in resected pancreatic cancer expanded neoantigen-specific T cells in eight of sixteen patients, and those eight went longer without recurrence — a comparison of responders against non-responders inside one arm, not against a control.⁹ In first-line advanced melanoma the same therapy did not significantly improve progression-free survival against pembrolizumab alone, at a hazard ratio of 0.78 and a p value of 0.31, despite inducing neoantigen-specific T cells in about five of every six patients tested.¹⁰ That is the sharpest available reminder that an immune response is not an outcome.

Does melanoma generalise?

Melanoma carries one of the highest mutation burdens of any common cancer — more candidate neoantigens per tumour — and already responds to checkpoint blockade better than most. Proponents argue the mechanism is antigen-agnostic and should transfer wherever there are enough mutations to choose from. Sceptics answer that it is being tested first where the immune system has the most to work with, and that the tumours responsible for most cancer deaths are the low-mutation ones where it has the least. The two positions agree about melanoma and disagree about lung and kidney cancer, which is why those readouts matter more than the one just delivered.

§Outlook

The result changes what is claimable about mRNA as a therapeutic platform more than what is known about cancer, and it rests on an interim analysis whose numbers nobody outside the sponsors and the monitoring committee has seen. Judging it needs the full dataset, an overall-survival readout, and the trials in tumours with fewer mutations.

For the arguments this wiki tracks, the significance is narrower than the coverage suggests. An adjuvant therapy that delays recurrence in resected melanoma addresses one disease in one setting, the pattern the geroscience argument treats as insufficient: eliminating a single cause of death shifts population life expectancy far less than the effort implies, because the same people remain at risk of everything else. The more general contribution may be methodological — a therapy specified by a patient's own sequence data and built in weeks is the production logic that patient-derived models and individualised simulation assume and have not delivered. Whether it survives contact with an ageing immune system is untested: the trials enrol patients fit enough for surgery and checkpoint blockade, and the narrowing of the T-cell repertoire with age is exactly the constraint a therapy that depends on priming new clones would be expected to hit. Nothing about the current readiness of the approach supports treating it as a general answer to late-life cancer mortality.

§See also

• Cancer
• Lipid nanoparticles
• Somatic gene therapy
• Casgevy
• Immunosenescence
• AI drug discovery
• Geroscience hypothesis
• Access and inequality

§References