When Cancer Type Stops Mattering: The High-Stakes Bet on Tumour-Agnostic Immunotherapy

For more than a century, cancer has been classified by the organ in which it arises, and that classification has dictated how tumours are studied, staged and treated. A landmark review published in eClinicalMedicine argues that this anatomical framework, while still dominant, is being steadily eroded by a more radical idea: that the molecular signature of a tumour, not the tissue harbouring it, should determine therapy. The watershed came in 2017, when the US Food and Drug Administration approved pembrolizumab for patients with microsatellite instability-high (MSI-H) or mismatch repair-deficient (dMMR) cancers regardless of where the tumour originated. That decision created the first truly tissue-agnostic oncology indication and opened the door to a growing list of pan-cancer biomarkers, including NTRK and RET fusions, BRAF V600E mutations, high tumour mutational burden (TMB-H) and HER2 overexpression. Yet the authors, led by Jia Liu and including Juliana Beal, Niamh Coleman, James Lynam and Vivek Subbiah, caution that the promise of treating cancers by their biology rather than their location rests on a far shakier evidence base than the headline approvals suggest.

The biological logic behind MSI-H/dMMR immunotherapy is the strongest in the field. When the mismatch repair machinery fails, cells accumulate insertion and deletion mutations that generate frameshift neoantigens, provoking a characteristic immune state: dense CD8-positive T-cell infiltration, heightened interferon signalling and adaptive upregulation of PD-1. Crucially, this immune phenotype is conserved across many tumour types, which is why pembrolizumab produced durable responses in roughly one-third of patients across more than 25 malignancies in the KEYNOTE-158 trial, with a median duration of response of 63.2 months. The paradigm extended beyond a single drug when the FDA granted accelerated approval to dostarlimab for previously treated recurrent or advanced dMMR solid tumours in 2021, supported by the GARNET study, in which the objective response rate reached 44 percent with responses lasting a median of 27.7 months. This, the authors argue, is what a genuine tumour-agnostic biomarker looks like: a shared, immune-sensitive biological state rather than a shared mutation.

Tumour mutational burden tells the opposite story. The premise was seductive: more mutations should yield more neoantigens and greater sensitivity to checkpoint blockade. But the pan-cancer experience exposed a fundamental flaw in that reasoning. A threshold of 10 mutations per megabase conflates biologically disparate cancers with different mutational processes, neoantigen quality, antigen-presentation machinery and immune microenvironments. TMB quantifies mutation burden while ignoring immunogenicity, T-cell infiltration and immune exclusion, all of which determine clinical response. In KEYNOTE-158, 102 patients with TMB-high tumours across ten histologies achieved an objective response rate of 29 percent, which was enough for the FDA’s 2020 accelerated approval of pembrolizumab for TMB-high solid tumours without satisfactory alternatives. Real-world datasets, however, show substantial heterogeneity by tumour type, with predictive value remaining strong mainly in cancers already characterised by immune infiltration. The immune ecosystem of a pancreatic adenocarcinoma differs fundamentally from that of a melanoma, regardless of whether a shared biomarker is present.

The review proposes that immunotherapy biomarkers be understood as layered rather than binary, spanning four domains. Tumour-intrinsic markers include MSI-H/dMMR, TMB, POLE and POLD1 mutations, viral antigens, DNA damage repair defects, mutational signatures and HLA loss. Immune microenvironment markers include PD-L1 expression, tumour-infiltrating lymphocytes, CD8 T-cell localisation, interferon signalling, myeloid infiltration and tertiary lymphoid structures. Host and systemic markers encompass immune fitness, microbiome composition, corticosteroid exposure and autoimmune risk, while dynamic markers include circulating tumour DNA kinetics, early radiographic patterns and immune-related toxicity. This layered model explains why single biomarkers perform inconsistently in tumour-agnostic practice. A high TMB result, for example, may be far less informative when it reflects treatment-induced hypermutation, such as temozolomide-associated hypermutation in recurrent glioma, rather than a clonal, endogenous immune-sensitive state.

The validation gap between candidate biomarkers is stark. PD-L1 protein expression has established tumour-specific assays and clinical utility, but its predictive value depends on the assay clone, scoring method, threshold and tumour lineage, and no PD-L1-expression-defined pan-cancer indication exists. PD-L1 gene amplification is rarer still, found in roughly 0.7 percent of tumours across more than 100 histologies in one pan-cancer series, and retrospective data suggest greater checkpoint inhibitor activity in amplified tumours, but no validated selection standard has emerged. Pathogenic POLE and POLD1 proofreading defects produce ultramutated, immunogenic phenotypes and correlate with checkpoint inhibitor efficacy, yet no regulatory indication rests on them alone, and the evidence is strongest in colorectal and endometrial cancers. Viral association, T-cell-inflamed gene signatures and tertiary lymphoid structures all carry biological plausibility and supportive evidence, but none constitutes a validated tumour-agnostic biomarker suitable for standalone prescribing.

Rare cancers expose the paradigm’s sharpest dilemma. Many are systematically excluded from pivotal trials yet demonstrate immunotherapy sensitivity without validated predictive biomarkers, as seen in Merkel cell carcinoma, alveolar soft-part sarcoma, thymic carcinoma and selected neuroendocrine subtypes. The SWOG S1609 DART trial, a phase 2 basket study of ipilimumab plus nivolumab across 53 rare cancer cohorts, enrolled 798 patients and confirmed responses in 24 of 53 cohorts, with an overall response rate of 13 percent and durable benefit in angiosarcoma, neuroendocrine tumours and rare gynaecological malignancies. The Australian MoST-CIRCUIT trial reported an objective response rate of 54 percent in advanced ovarian and endometrial clear cell cancers, most of which were mismatch repair-proficient and low TMB, and 63 percent in non-colorectal dMMR/MSI-H cancers across 17 tumour types. These results reveal immune-sensitive cohorts while leaving clinicians without biomarkers capable of identifying individual responders.

This is what the authors call the clinical grey zone: a patient with a rare cancer, exhausted standard options, emerging evidence of immunotherapy activity, no validated biomarker, no reimbursed molecular test and no feasible trial. A strictly biomarker-defined access model may be too restrictive, but unrestricted prescribing risks unwarranted toxicity, false hope and low-value care. The authors propose a pragmatic framework weighing four linked domains: strength of clinical evidence, biological and biomarker evidence within tumour-lineage context, competing treatment options, and patient-specific benefit-risk. Treatment can be recommended when convincing evidence or a validated immune-sensitive state exists and no superior option is available; considered with multidisciplinary review when uncertainty remains; and avoided outside a trial when evidence is absent, a superior therapy exists, or toxicity risk outweighs plausible benefit. The framework is deliberately not a numerical score, because strong evidence in one domain may override uncertainty in another.

Australia has become the world’s most consequential test case. In 2026, the Pharmaceutical Benefits Scheme consolidated and expanded public reimbursement of ipilimumab plus nivolumab and pembrolizumab for eligible patients with advanced or metastatic cancers considered immunotherapy-sensitive, including rare and uncommon cancers lacking conventional trial data. The listing, described as a world first and the product of partnership between government, industry, clinicians and patient advocates, includes shared costs between government and sponsor and a reassessment after three years. Critically, the reimbursement change did not create a corresponding tumour-agnostic regulatory indication, and the listing does not require a positive biomarker. Because broad pan-cancer next-generation sequencing is not universally funded in Australia, the country may be the first to fund pan-cancer immunotherapy without simultaneously funding the biomarker testing infrastructure needed to deploy it, transferring substantial selection responsibility to treating clinicians.

The economics of immunotherapy are also shifting. Annual treatment costs for PD-1 and PD-L1 inhibitors in the United States commonly exceed US$100,000 per patient, and multiple cost-effectiveness analyses have questioned the sustainability of broad population-level use at originator prices. Competition, biosimilar development and alternative procurement models may progressively reduce price as the sole determinant of access. As financial barriers ease, the question of appropriate use intensifies: severe immune-related adverse events occur in 15 to 20 percent of patients receiving checkpoint inhibitors, with fatal outcomes in 0.3 to 1.2 percent of cases. Reimbursement policy, the authors argue, cannot be considered independently of patient selection, and structured decision frameworks become the mechanism for translating genomic and immunologic data into responsible clinical guidance.

The lessons extend beyond checkpoint inhibitors to emerging pan-tumour platforms, including antibody-drug conjugates, bispecific antibodies and cellular therapies. Shared target expression among tumour types does not necessarily define a conserved therapeutic vulnerability: for antibody-drug conjugates, target density, internalisation kinetics, spatial heterogeneity, payload sensitivity and tumour lineage all influence efficacy despite expression of the same antigen. The review’s conclusion is measured but pointed. Tumour-agnostic immunotherapy has not made tumour type obsolete; it has shown that in selected settings biology can be more clinically informative than anatomy. Whether Australia’s experiment succeeds will depend on four prerequisites: funded molecular testing, expert decision support, prospective registries capturing outcomes and toxicity, and mechanisms for reassessment. Without those safeguards, the authors warn, the model risks both waste and harm, funding the drug before funding the pathway that decides who should receive it.

Subject of Research: Tumour-agnostic immunotherapy biomarkers and implementation of pan-cancer checkpoint inhibitor reimbursement

Article Title: Tumour-agnostic immunotherapy: aligning biological rationale, biomarker precision, and real-world implementation

Article References: Liu, J., Beal, J., Coleman, N., Lynam, J., & Subbiah, V. (2026). Tumour-agnostic immunotherapy: aligning biological rationale, biomarker precision, and real-world implementation. eClinicalMedicine, 100, Article 104211. https://doi.org/10.1016/j.eclinm.2026.104211

Image Credits: AI Generated

DOI: 10.1016/j.eclinm.2026.104211

Keywords: tumour-agnostic immunotherapy, MSI-H/dMMR, tumour mutational burden, checkpoint inhibitors, pembrolizumab, biomarkers, rare cancers, basket trials, Australia PBS reimbursement, precision oncology, tumour microenvironment, antibody-drug conjugates