What Are Immunotherapy Biomarkers?
Immunotherapy biomarkers—PD-L1, tumor mutational burden (TMB), MSI-H/dMMR, and gene signatures—help predict response to checkpoint inhibitors when measured with fit-for-purpose assays in the right tumor context. Performance varies by indication and platform. Motif surfaces published predictive immunotherapy evidence with PMIDs and population modifiers before trial enrichment criteria are locked.
TL;DR: Immunotherapy Biomarkers
- MSI-H/dMMR tumors: ORR 30.8% with pembrolizumab in KEYNOTE-158 cohort K (Marabelle et al., 2022)
- PD-L1 and TMB are used in practice but performance varies by tumor type and assay (Yarchoan et al., 2017; Samstein et al., 2019)
- Multi-feature models like LORIS stratify ICB response using routine clinical and genomic variables (Chang et al., 2024)
- Primary, adaptive, and acquired resistance mechanisms differ; serial biomarker assessment matters (Sharma et al., 2017)
- Literature on PD-L1, TMB, MSI, and gene signatures should be scoped per tumor type before trial enrichment
- Motif extracts predictive associations with PMIDs; treatment decisions stay with your clinical team
From the Motif team: Last reviewed June 2026. Checkpoint inhibitor trials lean on published predictive biomarkers, but assay clones, scoring systems, and cutoffs differ across trials. Motif extracts associations from literature with PMIDs and subgroup modifiers before your protocol locks enrichment criteria.
Checkpoint inhibitors help a subset of patients dramatically, but many others see no benefit. Response rates in unselected populations are often in the low double digits, which is why enrichment strategies matter. Before writing protocol criteria, teams need a tumor-type-specific map of which biomarkers were tested, with what assays, and in which populations.
Ribas and Wolchok (2018) reviewed clinical development of PD-1 pathway inhibitors and the evolving role of biomarkers in patient selection.1 Predictive claims require treatment interaction evidence, not prognosis alone (Buyse et al., 2011).2
Microsatellite Instability (MSI-H / dMMR)
MSI-H reflects defective DNA mismatch repair (dMMR), typically assessed by PCR of mononucleotide repeats or by NGS signatures, with IHC for MLH1, MSH2, MSH6, and PMS2 as a complementary screen.
Le et al. (2015) reported pembrolizumab in mismatch-repair-deficient colorectal cancer with immune-related ORR 40% (4 of 10 patients) in the dMMR cohort versus 0% in MMR-proficient disease.3 Le et al. (2017) expanded to 86 dMMR patients across 12 tumor types with ORR 53% (95% CI 41% to 65%).4 These studies established MSI-H as a predictive biomarker class for checkpoint blockade, not merely a prognostic marker in colorectal cancer.
Marabelle et al. (2022) updated KEYNOTE-158 cohort K (non-colorectal MSI-H/dMMR tumors): ORR 30.8% (95% CI 25.8% to 36.2%), median duration of response 47.5 months.5 Marabelle et al. (2019) earlier reported ORR 34.3% in 233 patients.6 Response rates vary by tumor type within the MSI-H umbrella; basket aggregates hide small-n subgroups.
MSI testing method (PCR panel vs NGS vs IHC reflex) and tumor type affect prevalence and enrollment feasibility. Lynch syndrome germline context, prior immunotherapy, and hypermutation from polymerase epsilon (POLE) mutations are modifier strata that appear in specialized literature and should not be collapsed into a single MSI label without reading methods.
PD-L1 Immunohistochemistry: Clones, Scoring, and Trial Context
PD-L1 is the most widely used immunotherapy biomarker in practice, but it is not a single test. Hirsch et al. (2017) coordinated the Blueprint PD-L1 IHC Comparability Project across 22C3 (pembrolizumab), 28-8 (nivolumab), SP263, and SP142 clones, showing that analytic concordance at clinically relevant thresholds is assay- and pathologist-dependent.7 You cannot assume a 22C3 tumor proportion score (TPS) from one trial maps to an SP142 result in another.
Scoring metrics differ by label:
- TPS (tumor proportion score): percentage of viable tumor cells showing membrane staining; central to KEYNOTE programs.
- CPS (combined positive score): tumor cells, lymphocytes, and macrophages; used in gastric, cervical, and other pembrolizumab indications.
- IC (immune cell score): SP142-specific; labels differ from TPS even when the same slide is read.
Reck et al. (2016) showed first-line pembrolizumab improved PFS versus chemotherapy in NSCLC with PD-L1 TPS of at least 50% (HR 0.50; 95% CI 0.37 to 0.68).8 That result is binding to the 22C3 pharmDx assay and the pre-specified 50% cutoff in that population. It does not generalize to all PD-L1-positive definitions across histologies.
Herbst et al. (2014) reported pembrolizumab versus docetaxel in previously treated NSCLC with PD-L1 TPS of at least 1% (KEYNOTE-010), with greater benefit in TPS of at least 50%.9 Early immunotherapy biomarker work was often exploratory or stratified post hoc; pivotal enrichment claims require pre-specified SAP rules and independent validation cohorts (Simon, 2013).10
Literature mapping must record: antibody clone, scoring system, positivity threshold, who read slides (central vs local), and whether the biomarker was primary, secondary, or exploratory in the trial SAP.
Tumor Mutational Burden: Assay, Cutoff, and Histology Dependence
TMB summarizes nonsynonymous somatic mutations per megabase (or exome equivalent). Yarchoan et al. (2017) reviewed association with PD-1/PD-L1 inhibitor response across tumor types but emphasized heterogeneous definitions and cutoffs.11
Samstein et al. (2019) analyzed MSK-IMPACT sequencing in 1,662 ICI-treated patients. Patients in the top 20% TMB within each histology had better overall survival (HR 0.52 vs bottom 80%).12 The top-20% rule is histology-relative, not a fixed mutations-per-Mb number portable across cancer types.
Hellmann et al. (2018) reported CheckMate 227 in advanced NSCLC: in patients with high TMB (at least 10 mutations/Mb by FoundationOne CDx), first-line nivolumab plus ipilimumab improved PFS versus chemotherapy (HR 0.58).13 TMB here is tied to a specific panel, bioinformatics pipeline, and tissue specimen. Blood-based TMB (bTMB) assays use different error profiles and require separate validation before copying tissue cutoffs.
Negative and null biomarker lessons matter as much as positive trials. Not every high-TMB tumor responds; not every PD-L1-negative tumor fails ICI. Enrichment strategies that ignore antigen presentation, interferon-gamma pathway alterations, and myeloid suppression biology oversimplify resistance (Sharma et al., 2017).14
Read our blog on genomic biomarkers in cancer therapy for NGS panel context and TMB measurement on tumor sequencing.
Core idea: Immunotherapy biomarker performance is assay- and indication-specific. Pooling PMIDs without platform metadata produces misleading enrichment assumptions.
Gene Expression and Microenvironment Signatures
Mariathasan et al. (2018) showed that TGFβ signaling in the tumor microenvironment can limit T-cell infiltration and response to PD-L1 blockade in urothelial cancer.15 Immune gene expression profiles capture broader context than a single IHC readout, but validation is signature- and indication-specific.
Multi-gene signatures require locked algorithms and external validation before clinical deployment. Read our blog on machine learning in biomarker validation for model locking and TRIPOD reporting standards.
Primary, Adaptive, and Acquired Resistance
Sharma et al. (2017) distinguish primary resistance (no initial response), adaptive resistance (initial response then escape), and acquired resistance (relapse after durable benefit).14 Mechanisms include loss of antigen presentation (β2-microglobulin or HLA alterations), alternate immune checkpoints, oncogenic pathway signaling, and immunosuppressive myeloid infiltration.
Literature review for trial design should tag whether papers report baseline predictors, on-treatment pharmacodynamic markers, or resistance mechanisms at progression. A high PD-L1 score at screening does not rule out acquired resistance months later.
Serial biopsy and liquid biopsy studies report emergent alterations under checkpoint blockade. Read our blog on liquid biopsy biomarkers for ctDNA monitoring in that setting.
Machine Learning Models for Response Prediction
Chang et al. (2024) developed LORIS (logistic regression-based immunotherapy-response score) from 2,881 ICB-treated patients across 18 solid tumor types using six routinely available clinical, pathologic, and genomic features. The model outperformed TMB and PD-L1 alone in their analysis.16 Prospective validation is still needed before treating such scores as clinical decision rules.
Algorithmic scores still require analytical validity on the inputs they use (sequencing quality, PD-L1 scoring lab, staging data completeness). Literature-derived feature lists should be pre-registered before fitting on institutional cohorts.
Combination Checkpoint Blockade and Biomarker Gaps
Anti-PD-1 plus anti-CTLA-4 combinations increase response rates in some settings but add toxicity. Biomarkers that predict benefit from monotherapy do not always predict benefit from combination without indication-specific trial evidence. Literature review should separate monotherapy PMIDs from combination regimen PMIDs before writing enrichment criteria.
Immune-related adverse events (irAEs) literature is growing but is not interchangeable with efficacy biomarker evidence. Protocol teams should not use irAE predictors as response enrichment markers without explicit validation PMIDs.
Scoping Evidence in Motif Before Trial Design
- Search PubMed, PMC, and Europe PMC for the tumor type and checkpoint inhibitor class
- Extract predictive associations with comparators and interaction p-values where reported
- Flag population modifiers (line of therapy, prior treatment, molecular subtype)
- Compare conflicting cohort results before locking enrichment criteria
- Export cited associations for protocol background or SAP sections
Common failure modes:
- Pooling PD-L1 thresholds across trials that used different assays
- Treating MSI results from one tumor type as automatically transferable without reading subgroup sizes
- Ignoring studies that report benefit only in a narrow modifier stratum
- Confusing literature evidence with a patient-selection algorithm Motif does not run
- Using prognostic TMB association as predictive enrichment without treatment interaction
Motif surfaces stratification evidence from literature with PMIDs. See cited literature review for scoping workflows.
Related Articles
- Genomic biomarkers in cancer therapy: MSI, TMB, and targeted therapy context
- Personalized medicine biomarker analysis: precision oncology landscape
- FDA biomarker validation: companion diagnostic and qualification paths
Frequently Asked Questions
What biomarkers predict immunotherapy response?
Published predictors include MSI-H/dMMR status (Marabelle et al., 2022), PD-L1 expression (Ribas & Wolchok, 2018), tumor mutational burden (Yarchoan et al., 2017; Samstein et al., 2019), tumor microenvironment signatures (Mariathasan et al., 2018), and multi-feature models such as LORIS (Chang et al., 2024). Performance varies by tumor type and assay.
Is MSI-H a tissue-agnostic immunotherapy biomarker?
MSI-H/dMMR tumors have FDA-recognized indications for pembrolizumab in defined advanced settings. KEYNOTE-158 cohort K reported ORR 30.8% in non-colorectal MSI-H/dMMR cancers (Marabelle et al., 2022). Prevalence and testing methods differ by tumor type.
Why do PD-L1 results differ between trials?
Antibody clones (22C3, 28-8, SP263, SP142), scoring systems (TPS vs CPS vs IC), positivity thresholds, and central vs local pathology differ across trials. The Blueprint project showed incomplete analytic concordance between assays at clinical cutoffs (Hirsch et al., 2017). A TPS of at least 50% on 22C3 from KEYNOTE-024 is not interchangeable with an SP142 read from another study (Reck et al., 2016).
Is there a universal TMB cutoff for immunotherapy?
No. Samstein et al. (2019) used top-20% TMB within each histology; Hellmann et al. (2018) used at least 10 mutations/Mb by FoundationOne CDx in CheckMate 227 NSCLC. Cutoffs are panel-, specimen-, and indication-specific. Applying one threshold across histologies without validation PMIDs is a common protocol error.
Can a PD-L1-negative patient still respond to checkpoint inhibitors?
Yes. PD-L1 is an enrichment marker with imperfect sensitivity and specificity. Sharma et al. (2017) review resistance and response biology beyond a single IHC readout. Literature review for trial design should include studies reporting responses in biomarker-negative strata, not only enrichment successes.
What is the difference between primary and acquired immunotherapy resistance?
Primary resistance is no initial response; acquired resistance is relapse after durable benefit. Mechanisms include antigen presentation loss, alternate checkpoints, and pathway escape (Sharma et al., 2017). Baseline predictors do not eliminate acquired resistance risk.
Can machine learning replace PD-L1 and TMB for patient selection?
Models like LORIS combine clinical and genomic features and outperformed single markers in retrospective analysis (Chang et al., 2024), but prospective validation and locked deployment standards are still required before clinical decision use.
How should teams review immunotherapy biomarker literature before trial enrichment?
Map PMIDs by tumor type, assay platform, line of therapy, and interaction statistics. Motif extracts cited predictive associations with population modifiers for protocol and SAP background sections.
References
- Ribas, A., & Wolchok, J.D. (2018). Cancer immunotherapy using checkpoint blockade. Science, 359(6382), 1350-1355. PMID: 29567705
- Buyse, M., et al. (2011). Biomarkers and surrogate end points. Nature Reviews Clinical Oncology, 7(6), 309-317. PMID: 20368571
- Le, D.T., et al. (2015). PD-1 blockade in tumors with mismatch-repair deficiency. New England Journal of Medicine, 372(26), 2509-2520. PMID: 26028255
- Le, D.T., et al. (2017). Mismatch-repair deficiency predicts response to PD-1 blockade. Science, 357(6349), 409-413. PMID: 28596308
- Marabelle, A., et al. (2022). Pembrolizumab in MSI-H cancers: KEYNOTE-158 update. Annals of Oncology, 33(10), 1032-1042. PMID: 35680043
- Marabelle, A., et al. (2019). Pembrolizumab in noncolorectal MSI-H/dMMR cancer. Journal of Clinical Oncology, 38(1), 1-10. PMID: 31682550
- Hirsch, F.R., et al. (2017). PD-L1 IHC assays for lung cancer: Blueprint Comparison Project. Journal of Thoracic Oncology, 12(2), 208-222. PMID: 28376109
- Reck, M., et al. (2016). Pembrolizumab versus chemotherapy for PD-L1-positive NSCLC. New England Journal of Medicine, 375(19), 1823-1833. PMID: 26712084
- Herbst, R.S., et al. (2014). Pembrolizumab versus docetaxel for PD-L1-positive NSCLC (KEYNOTE-010). Lancet, 387(10027), 1540-1550. PMID: 25061874
- Simon, R.M. (2013). Genomic biomarkers in predictive medicine. EMBO Molecular Medicine, 5(6), 813-818. PMID: 23818349
- Yarchoan, M., et al. (2017). Tumor mutational burden and response to PD-1 inhibition. New England Journal of Medicine, 377(25), 2500-2501. PMID: 29262275
- Samstein, R.M., et al. (2019). Tumor mutational load predicts survival after immunotherapy. Nat Genet, 51(2), 202-206. PMID: 30643254
- Hellmann, M.D., et al. (2018). Nivolumab plus ipilimumab in lung cancer with a high tumor mutational burden. New England Journal of Medicine, 378(22), 2093-2104. PMID: 29658845
- Sharma, P., et al. (2017). Primary, adaptive, and acquired resistance to cancer immunotherapy. Cell, 168(4), 707-723. PMID: 28187290
- Mariathasan, S., et al. (2018). TGFβ attenuates tumour response to PD-L1 blockade. Nature, 554(7693), 544-548. PMID: 29443960
- Chang, T.G., et al. (2024). LORIS predicts outcomes with immune checkpoint blockade. Nature Cancer, 5(6), 943-957. PMID: 38831056
