Personalised Cancer Vaccines: The Biomarkers That Decide Who Benefits

Personalised Cancer Vaccines: The Biomarkers That Decide Who Benefits

For two decades, cancer vaccines were the immunotherapy that never quite worked. That changed on 19 August 2026, when Merck and Moderna announced that the phase 3 INTerpath-001 trial of intismeran autogene plus pembrolizumab met its primary endpoint in resected high-risk melanoma. It is the first positive phase 3 readout for an individualised neoantigen therapy, and for any mRNA-based cancer therapy.

Nine days later, BioNTech terminated its phase 2 trial of autogene cevumeran in ctDNA-positive resected colorectal cancer after the data monitoring board flagged a numerical imbalance in overall survival. Same broad technology, opposite outcomes.

For translational, NGS and immune-monitoring teams, the gap between those two results is the interesting part. A personalized cancer vaccine is a biomarker-driven product from the first step: the drug is designed from a sequencing result, the patients most likely to be enrolled are defined by residual disease, and proof of mechanism depends on measuring T cell responses that did not exist before treatment. This article sets out where the field stands in October 2026 and which biomarker questions now decide who benefits.

Key takeaways

• INTerpath-001 (1,137 patients, stage IIB–IV resected melanoma) met recurrence-free survival and distant metastasis-free survival at a prespecified interim analysis. Hazard ratios have not yet been disclosed, and no regulatory filing had been announced at the time of writing.
• BioNTech and Genentech ended the phase 2 colorectal trial of autogene cevumeran given as monotherapy; the pancreatic phase 2 (IMcode003), which combines it with checkpoint inhibition and chemotherapy, continues.
• Every individualised neoantigen therapy rests on a sequencing pipeline: tumour and germline DNA, tumour RNA, HLA typing and computational epitope ranking. Each step is an assay with its own failure modes.
• Exploratory KEYNOTE-942 data suggest the vaccine benefit is not confined to high tumour mutational burden, and that ctDNA status is strongly prognostic.
• Immune monitoring (ELISpot, TCR sequencing, multimer and single-cell methods) is the only direct evidence that the vaccine did what it was designed to do, and the pancreatic data show it can separate responders from non-responders.

What is a personalised cancer vaccine?

A personalised (or individualised) neoantigen vaccine is made for one patient. Tumour tissue and a normal sample, usually blood, are sequenced; somatic mutations that create new peptide sequences are identified; algorithms rank which of those peptides are most likely to be presented on the patient’s own HLA molecules and recognised by T cells; and a vaccine encoding the top candidates is manufactured and given, typically alongside a checkpoint inhibitor.

Merck and Moderna describe intismeran autogene (also known as V940 or mRNA-4157) as a synthetic mRNA coding for up to 34 neoantigens, designed from a sample of the patient’s tumour. BioNTech’s autogene cevumeran (BNT122) encodes up to 20 neoantigens, split across two mRNA strands, in a lipoplex formulation given intravenously. Viral-vector approaches follow the same logic: Transgene and NEC’s TG4050 is a modified vaccinia Ankara (MVA) vector encoding up to 30 patient-specific predicted neoantigens.

Neoantigens are attractive because they are absent from normal tissue, so the T cells they prime are not constrained by central tolerance and should not attack healthy cells. The catch is that most predicted neoantigens never elicit a response, and the ones that do are unique to each patient. That is why the biomarker stack matters so much here.

Where the leading programmes stand in October 2026

Intismeran autogene (Merck and Moderna)

The randomised phase 2b KEYNOTE-942 trial, published in The Lancet, randomised 157 patients with resected stage IIIB–IV melanoma 2:1 to mRNA-4157 plus pembrolizumab or pembrolizumab alone. Five-year follow-up presented in 2026 showed a 49% reduction in the risk of recurrence or death (HR 0.51, 95% CI 0.294–0.887) and a 59% reduction in the risk of distant metastasis or death (HR 0.411).

INTerpath-001 (NCT05933577) is the confirmatory study: global, double-blind, 1,137 patients with completely resected stage IIB–IV cutaneous melanoma randomised 2:1. Intismeran was given at 1 mg every three weeks for up to nine doses, with pembrolizumab 400 mg every six weeks for up to nine cycles. At a prespecified interim analysis the combination produced statistically significant and clinically meaningful improvements in recurrence-free survival and distant metastasis-free survival, with no new safety signals. Overall survival is still being followed.

Three things were not in the announcement: the hazard ratios, the subgroup data and a filing. The companies said the data will be presented at an upcoming international medical meeting and that they “will engage with regulators on filing submissions”. As of 13 October 2026, we found no announcement of a submitted or accepted application.

The wider INTerpath programme now spans nine phase 2 and phase 3 trials, according to the companies’ programme backgrounder. These include phase 3 adjuvant lung studies (INTerpath-002 in resected stage II–IIIB NSCLC; INTerpath-009 in patients without a pathological complete response after neoadjuvant chemo-immunotherapy; INTerpath-014 in resected high-risk stage I NSCLC), phase 2 studies in renal cell carcinoma (INTerpath-004), muscle-invasive and non-muscle-invasive bladder cancer (INTerpath-005 and -011), first-line advanced melanoma (INTerpath-012) and metastatic squamous NSCLC (INTerpath-013).

Autogene cevumeran (BioNTech and Genentech)

The pancreatic programme produced one of the field’s most cited translational datasets. In the phase 1 study, patients received surgery, a single dose of atezolizumab, autogene cevumeran and modified FOLFIRINOX. The first report (Rojas et al., Nature, 2023) found vaccine-induced T cells in 8 of 16 vaccinated patients. The extended analysis (Sethna et al., Nature, 2025), at a median follow-up of 3.2 years, reported that median recurrence-free survival was not reached in responders versus 13.4 months in non-responders (HR 0.14). The randomised phase 2 IMcode003 trial (NCT05968326) is ongoing.

Colorectal cancer went the other way. On 28 August 2026 BioNTech terminated BNT122-01 (NCT04486378), which compared adjuvant autogene cevumeran monotherapy with watchful waiting in ctDNA-positive, resected high-risk stage II or stage III colorectal cancer. The futility boundary had been crossed in October 2025; the new DSMB recommendation followed a numerical imbalance in overall survival. No new safety signals were identified, and BioNTech says it will analyse the data to inform patient selection.

Other platforms

TG4050 is the most advanced viral-vector example. In the randomised phase 1 part of a phase 1/2 trial in resected HPV-negative head and neck squamous cell carcinoma, patients were assigned to vaccination immediately after adjuvant therapy or at recurrence. Transgene and NEC reported in September 2026 that all immediately vaccinated patients remained disease-free at three years, alongside the Nature Communications publication describing durable neoantigen-specific CD8+ T cell responses. The cohort is small and the randomised phase 2 part is still running.

The biomarker questions that decide who benefits

1. Neoantigen prediction: the sequencing and computation pipeline

Every individualised neoantigen therapy starts with an assay that is really a chain of assays:

• Whole-exome sequencing of tumour and matched normal to call somatic variants and exclude germline ones. Tumour purity and clonality estimates matter, because a low-purity biopsy can miss variants or misjudge which ones are present in every cancer cell.
• Tumour RNA sequencing to confirm the mutated gene is expressed. A mutation in a silent gene produces no peptide to present.
• HLA typing, ideally to high resolution for both class I and class II, since binding predictions are allele-specific. HLA loss of heterozygosity in the tumour can remove the very allele a vaccine epitope depends on.
• Binding and presentation prediction, combining peptide-HLA affinity, processing and, increasingly, mass spectrometry-trained presentation models, to rank candidates. Moderna describes using integrated algorithms; NEC brings AI-based prediction to TG4050.

None of these steps is standardised across companies, and the selection algorithm is effectively part of the product. For diagnostics and NGS providers, the practical issues are turnaround, sample quality from small surgical specimens, and analytical validation of a pipeline whose output is a drug design rather than a clinical report.

2. Tumour mutational burden and clonality

Intuitively, more mutations should mean more candidate neoantigens and a better chance of a hit. The data are less tidy. Exploratory KEYNOTE-942 analyses reported benefit from adding mRNA-4157 in both TMB-high and TMB-low patients, and the investigators presented TMB, PD-L1 and an interferon-gamma signature as markers associated with good outcomes rather than as selection criteria. Pancreatic cancer, a low-TMB tumour, produced vaccine responses in half of treated patients.

TMB is still a useful descriptor: it sets the size of the candidate pool and correlates with checkpoint inhibitor benefit in the comparator arm. Our article on tumour mutational burden from research to clinical reality covers the assay harmonisation issues. Clonality may matter more. Targeting truncal mutations, present in every tumour cell, should reduce escape, and the Sethna paper found that recurrent pancreatic tumours had been pruned of vaccine-targeted cancer clones, which is exactly the escape route clonal selection is meant to close.

3. ctDNA: selection, stratification and monitoring

Adjuvant vaccine trials are trials in patients with no visible disease, so ctDNA is the closest thing to a measurable target. It is being used in three ways:

• Selection. BNT122-01 enrolled only ctDNA-positive patients after colorectal surgery, a population with high relapse risk where a vaccine has a clear job to do. Its termination does not discredit ctDNA selection, but it shows that molecular residual disease may be too far advanced for a monotherapy vaccine to control.
• Stratification. In KEYNOTE-942, longitudinal ctDNA analyses presented in 2023 showed that patients who were ctDNA-positive at baseline, or turned positive, did worst, while within each ctDNA category the combination generally appeared better than pembrolizumab alone. ctDNA was strongly prognostic; whether it predicts vaccine benefit is not established.
• Monitoring. Tumour-informed ctDNA assays share a starting point with vaccine design, the patient’s exome, so the same sequencing run can serve both purposes.

For more on assay choice and endpoints, see our guide to ctDNA and MRD in immunotherapy trials.

4. Immune monitoring of vaccine-induced T cells

This is the pharmacodynamic core of any neoantigen programme, and the pancreatic study shows why. Responders were defined by ex vivo IFN-gamma ELISpot, then followed by TCR beta-chain sequencing to trace individual vaccine-expanded clones (CloneTrack) and single-cell RNA and TCR sequencing to phenotype them (PhenoTrack). That combination showed the clones were absent before vaccination, adopted a cytotoxic, tissue-resident memory-like state, and in most cases persisted about three years later.

The operational point is that sample logistics drive assay quality. Viable PBMC collection, cryopreservation and shipping, often from community sites, determine whether a functional readout is possible at all. Tissue at recurrence is valuable for showing antigen loss or HLA loss, but is rarely collected systematically.

5. Manufacturing timelines and sample logistics

An individualised product cannot be made until sequencing and design are complete, which pushes the first dose weeks after surgery. Trial designs have to build that window in: the pancreatic study, for example, gave a single dose of atezolizumab and then the vaccine between surgery and chemotherapy. Each step in the chain (tissue fixation, nucleic acid quality, sequencing turnaround, design, manufacture, release) adds time and a point of failure. If filings follow and these products reach routine care, the speed and reliability of tumour sample handling at the treating centre become part of whether a patient receives the drug at all.

What this means for biomarker, NGS and CRO teams

• The design assay becomes a regulated component. Regulators will want to see how variant calling, HLA typing and epitope ranking are validated and controlled, even though there is no single analyte.
• ctDNA is moving into the adjuvant mainstream. The open question after the colorectal result is whether ctDNA positivity should select patients in or out of vaccine combinations, and at what level.
• Immune monitoring needs to be standardised and scalable. Multimer approaches do not scale to individualised epitopes; TCR sequencing with functional deconvolution increasingly does.
• Turnaround is a clinical variable. Every week of sequencing, design and manufacture delays the first dose in an adjuvant setting where relapse risk is front-loaded.

These are the same issues that dominate the genomics and IO meeting circuit each spring, and they cut across both of ImmunoMark’s streams.

Discuss personalised cancer vaccines at ImmunoMark 2027

The nearest place to take these questions is ImmunoMark Summit San Diego, our immuno-oncology and biomarker conference on 10–11 February 2027. On Day 1, Stream 1: Next-Gen Immuno-Oncology includes the session “Personalised Cancer Vaccines and Neoantigen Discovery for Next-Generation Immunotherapies”, while Stream 2 runs in parallel on biomarkers and companion diagnostics. The launch edition brings together 300+ senior attendees, with keynotes, panels, case studies, pre-arranged 1:1 meetings and a Day 1 drinks reception.

The conversation continues at the immuno-oncology conference in Boston in June 2027, where Stream 1 will again cover next-generation IO modalities for an audience of 300+.

Register for ImmunoMark San Diego 2027

Register for ImmunoMark Boston 2027

If your company wants to meet this audience, whether you provide NGS, neoantigen prediction, ctDNA or immune-monitoring services, see the San Diego sponsorship opportunities and the Boston sponsorship opportunities, or read more about sponsoring a cancer immunology conference with ImmunoMark.

FAQs

What is a personalized cancer vaccine?

A personalised cancer vaccine, also called an individualised neoantigen therapy, is designed for one patient. Tumour and normal samples are sequenced, mutations that create new peptides are identified, algorithms rank which are most likely to be presented on the patient’s HLA molecules, and a vaccine encoding the top candidates (up to 34 for intismeran autogene) is manufactured and given, usually with a checkpoint inhibitor.

Has an mRNA cancer vaccine succeeded in a phase 3 trial?

Yes. On 19 August 2026 Merck and Moderna announced that INTerpath-001, testing intismeran autogene plus pembrolizumab against pembrolizumab alone in 1,137 patients with resected stage IIB–IV melanoma, met its primary endpoint of recurrence-free survival and a key secondary endpoint of distant metastasis-free survival at an interim analysis. Hazard ratios had not been released and no regulatory filing had been announced as of October 2026.

Does tumour mutational burden predict who benefits from a neoantigen vaccine?

Not clearly. Exploratory KEYNOTE-942 analyses reported benefit in both TMB-high and TMB-low melanoma, and autogene cevumeran induced T cell responses in half of treated patients with pancreatic cancer, a low-TMB tumour. TMB sets the size of the candidate pool, but clonality, expression, HLA presentation and the patient’s immune fitness also shape response.

Why did BioNTech stop the autogene cevumeran colorectal trial?

In August 2026 BioNTech terminated the phase 2 trial in ctDNA-positive resected colorectal cancer after the data safety monitoring board identified a numerical imbalance in overall survival and judged that continuing was unlikely to change the efficacy outcome. The vaccine was given as monotherapy against watchful waiting. No new safety signals were found, and the pancreatic phase 2 trial continues.

How are vaccine-induced T cell responses measured?

Common methods are IFN-gamma ELISpot against each vaccine peptide, intracellular cytokine staining by flow cytometry, peptide-MHC multimer staining, bulk TCR sequencing to track expanded clones over time, and single-cell RNA and TCR sequencing to phenotype them. The pancreatic autogene cevumeran study combined ELISpot with TCR clone tracking to separate responders from non-responders.