Biomarkers of Pembrolizumab Response

What are key biomarkers of response and resistance to pembrolizumab (e.g., PD-L1, MSI-H/dMMR)?

The key biomarkers of pembrolizumab response are MSI-H/dMMR status (most consistent pan-cancer predictor with 32-44% response rates), PD-L1 expression (predictive in melanoma and lung cancer but not colorectal cancer), and tumor mutational burden (predictive across tumor types except microsatellite-stable colorectal cancer), with highest response rates achieved when both high TMB and high inflammatory markers are present together.

Abstract

Three major biomarker categories predict pembrolizumab response across cancer types: MSI-H/dMMR status, PD-L1 expression, and tumor mutational burden (TMB). MSI-H/dMMR status demonstrated the most consistent pan-cancer predictive value, with objective response rates of 32-35% in previously treated patients across diverse tumor types and 43.8% in first-line MSI-H/dMMR colorectal cancer. PD-L1 expression strongly predicted response in melanoma and non-small-cell lung cancer, with response rates increasing from 8% in PD-L1-negative melanoma to 57% in highly PD-L1-positive tumors, but did not demonstrate predictive value in colorectal cancer. High TMB (≥10 mutations per megabase) identified responders across tumor types, with 29% response rate versus 6% in low TMB patients, though it was not valuable in microsatellite-stable colorectal cancer.

Combined biomarker analyses revealed that TMB and inflammatory biomarkers (T-cell inflamed gene expression profile or PD-L1) capture complementary features of tumor immunobiology and perform better together than individually. Patients with both high TMB and high inflammatory markers achieved the highest response rates (37-57%), compared to moderate rates (11-42%) with elevation of only one biomarker and minimal responses (0-9%) when both were low. The low correlation between TMB and inflammatory markers suggests they reflect distinct biological features—neoantigen load and active T cell engagement, respectively—both of which appear necessary for optimal pembrolizumab response. Treatment line influenced outcomes, as first-line pembrolizumab in MSI-H/dMMR colorectal cancer achieved superior response rates and progression-free survival compared to later-line therapy.

Methods

We analyzed 10 sources from an initial pool of 200, using 9 screening criteria. Each paper was reviewed for 6 key aspects that mattered most to the research question.

Records from Elicit search

• n = 200
• Papers screened using: Pembrolizumab Treatment, Biomarker Analysis, Clinical Outcomes, Cancer Population, Study Design, Clinical Study, Sample Size, Efficacy Focus, Publication Type
• n = 200 Papers screened out
• n = 190 Papers included for extraction

Data extraction

We asked a large language model to extract each data column below from each paper. We gave the model the extraction instructions shown below for each column.

Biomarkers Studied:

• Specific biomarker name (e.g., PD-L1, MSI-H/dMMR, TMB, T-cell inflamed GEP)
• Whether studied for response prediction, resistance prediction, or both
• Primary vs secondary/exploratory biomarker analysis
• Any novel or emerging biomarkers beyond established ones

Biomarker Methods:

• Assessment method (e.g., IHC, sequencing, gene expression profiling)
• Specific assay or platform used
• Threshold/cutoff values defining positive vs negative (e.g., ≥50% for PD-L1, ≥10 mutations/Mb for TMB)
• How thresholds were determined (training vs validation sets, predefined vs data-driven)
• Timing of biomarker assessment relative to pembrolizumab treatment

Study Population:

• Cancer type(s) and stage
• Prior treatment history (treatment-naive vs previously treated)
• Line of therapy for pembrolizumab (first-line, second-line, etc.)
• Sample size for biomarker analysis
• Any population restrictions that might affect biomarker generalizability

Biomarker Performance:

• Response rates in biomarker-positive vs biomarker-negative patients
• Odds ratios, hazard ratios, or other effect measures with confidence intervals
• Sensitivity, specificity, positive/negative predictive values if reported
• Statistical significance (p-values) for biomarker associations
• Performance across different cancer types if assessed

Clinical Outcomes:

• Objective response rates (ORR) by biomarker positive vs negative groups
• Progression-free survival (PFS) medians and confidence intervals by biomarker status
• Overall survival (OS) outcomes by biomarker status
• Duration of response by biomarker groups
• Complete vs partial response rates if differentiated by biomarker

Combination Analysis:

• Specific biomarker combinations tested (e.g., TMB + GEP, PD-L1 + MSI)
• Response rates and survival outcomes for combination biomarker groups
• Whether combinations performed better than individual biomarkers
• Correlation between different biomarkers
• Any proposed multi-biomarker algorithms or scoring systems

Results

Characteristics of Included Studies

The included studies evaluated pembrolizumab across diverse cancer types and disease settings. Three major categories of predictive biomarkers emerged: MSI-H/dMMR status, PD-L1 expression, and TMB, with one study also evaluating a T-cell inflamed gene expression profile. Four studies focused specifically on MSI-H/dMMR colorectal cancer, while others examined broader tumor type cohorts. Sample sizes ranged from 63 to 805 patients evaluable for biomarker analysis. Most studies enrolled previously treated patients, with one notable exception being KEYNOTE-177, which evaluated pembrolizumab as first-line therapy.

Biomarker Assessment Methods

MSI-H/dMMR status was assessed using two complementary approaches: immunohistochemistry to detect loss of mismatch repair protein expression (MLH1, MSH2, MSH6, PMS2) and PCR-based analysis of microsatellite loci. TMB assessment methods varied between studies, with one using whole-exome sequencing with a threshold of ≥123 mutations per exome and another employing the FoundationOne CDx assay with a threshold of ≥10 mutations per megabase. The TMB threshold in the Cristescu study was determined using the Youden Index, a data-driven approach. For PD-L1, immunohistochemistry using the 22C3 antibody was standard, but positivity thresholds varied from ≥1% to ≥50% tumor cells. The 50% cutoff was determined using receiver-operating-characteristic curves from a training cohort.

Predictive Performance of Individual Biomarkers

MSI-H/dMMR Status

MSI-H/dMMR status demonstrated consistent predictive value across multiple cancer types. Response rates in MSI-H/dMMR cancers ranged from 32-35% in previously treated patients, with notably higher response (43.8%) in the first-line setting. Duration of response was generally not reached across studies, with more than three quarters of responders maintaining responses for 24 months or longer in the noncolorectal cohort. In the first-line MSI-H/dMMR CRC study, 83% of responders had ongoing responses at 24 months compared to only 35% with chemotherapy.

PD-L1 Expression

PD-L1 expression demonstrated a strong association with pembrolizumab response in melanoma and NSCLC. In melanoma, objective response rates increased progressively from 8% in PD-L1-negative tumors (MEL score 0) to 57% in tumors with MEL score 4. Both progression-free survival (HR 0.76, 95% CI 0.71-0.82) and overall survival (HR 0.76, 95% CI 0.69-0.83) were significantly longer in patients with higher PD-L1 expression (p<0.001). In NSCLC, patients with PD-L1 expression in at least 50% of tumor cells had a response rate of 45.2% compared to 19.4% overall, with median PFS of 6.3 months versus 3.7 months overall and median OS not reached versus 12.0 months. However, patients with PD-L1-negative tumors could still achieve durable responses in both melanoma and NSCLC.

Tumor Mutational Burden

The KEYNOTE-158 biomarker analysis prospectively evaluated TMB across 10 tumor types. Patients with high tissue TMB (≥10 mutations per megabase) had an objective response rate of 29% (95% CI, 21-39%) compared to 6% (95% CI, 5-8%) in non-TMB-high patients. This represented a nearly 5-fold difference in response rates based on TMB status. Of the 105 patients with TMB-high tumors, 16 (15%) experienced grade 3-5 treatment-related adverse events.

Combined Biomarker Analyses

The pan-tumor genomic analysis by Cristescu et al. systematically evaluated TMB and T-cell inflamed GEP as joint predictors. These biomarkers exhibited only modest correlation with each other, suggesting they capture distinct features of neoantigenicity and T cell activation respectively.

Patients with both high TMB and high GEP achieved the highest response rates (37-57%), representing a population with the highest likelihood of benefit. Those with high levels of only one biomarker had moderate response rates (11-42%), while patients with low levels of both biomarkers had minimal responses (0-9%). Longer progression-free survival was observed in patients with higher levels of both TMB and GEP. When TMB was jointly assessed with PD-L1 expression, comparable findings emerged. The correlation between TMB and GEP was low, while TMB showed moderate correlation with PD-L1, and GEP demonstrated significant correlation with PD-L1.

The KEYNOTE-028 study similarly found that correlations between TMB and inflammatory markers (GEP or PD-L1) were low. Higher response rates and longer PFS were demonstrated with higher levels of T-cell inflamed GEP, PD-L1 expression, and/or TMB. Response patterns indicated that patients with tumors having high levels of both TMB and inflammatory markers represented a population with the highest likelihood of response.

Synthesis: Context-Dependent Biomarker Performance

While all three major biomarker categories (MSI-H/dMMR, PD-L1, TMB) predicted pembrolizumab response, their relative utility varied by cancer type and clinical context. MSI-H/dMMR status emerged as the most consistent pan-cancer predictor, with response rates of 32-35% across diverse tumor types in the previously treated setting and 43.8% in first-line MSI-H/dMMR CRC. This consistency likely reflects the fundamental biology whereby MSI-H/dMMR tumors harbor hundreds to thousands of somatic mutations encoding potential neoantigens, creating highly immunogenic tumors with upregulation of immune checkpoint proteins.

PD-L1 expression showed strong predictive utility in melanoma and NSCLC, but was not demonstrated to be a valuable biomarker in colorectal cancer. This cancer type-specific performance may reflect differences in tumor microenvironment biology. In MSI-H CRC, the presence of tumor-infiltrating lymphocytes and T cell activation appears sufficient for response regardless of PD-L1 levels, whereas in other cancers, PD-L1 expression may be more critical for identifying tumors with active immune engagement. Alternatively, in MSS (microsatellite stable) CRC, low levels of tumor-infiltrating lymphocytes and high levels of suppressive cells like regulatory T cells and macrophages limit immunotherapy response, suggesting that broader immune contexture rather than PD-L1 alone determines outcomes.

TMB demonstrated predictive value across tumor types, with high TMB patients achieving 29% response rates versus 6% in low TMB patients. However, TMB was not identified as valuable in MSS CRC, potentially because other resistance mechanisms dominate in this context, including activation of the WNT/β-catenin pathway, which promotes immunosuppression, and presence of liver metastases enriched with TGF-β.

The combination biomarker analyses reveal an important pattern: TMB and inflammatory biomarkers (T-cell inflamed GEP or PD-L1) capture complementary features of tumor immunobiology. Their low correlation suggests TMB reflects neoantigen load while inflammatory markers reflect active T cell engagement. Patients with high levels of both achieve response rates of 37-57%, substantially higher than those with elevated levels of only one biomarker (11-42%). This suggests that both adequate neoantigen presentation (captured by TMB) and pre-existing T cell inflammation (captured by GEP/PD-L1) are required for optimal pembrolizumab response.

Treatment line also influences biomarker-stratified outcomes. In MSI-H/dMMR CRC, first-line pembrolizumab achieved 43.8% response rate with median PFS of 16.5 months, compared to 32-35% response rates and median PFS of 2.3-4.1 months in previously treated patients. This difference may reflect better performance status, lower tumor burden, and less immunosuppression from prior chemotherapy in treatment-naive populations. Notably, 83% of first-line responders maintained ongoing responses at 24 months compared to only 35% of chemotherapy-treated patients, underscoring the durability advantage of pembrolizumab in biomarker-selected populations.