Are You a TMBeliever? Mutations and Atezolizumab Response in Solid Tumors

The tenets of tumor biomarker development include analytic validity, clinical validity, and clinical utility. Tumor mutational burden (TMB) quantification was developed as a surrogate for immunogenic tumor neoantigens, under a roulette-like assumption that with more mutations comes an increased likelihood of immune-mediated tumor recognition and clearance when treating with immune checkpoint inhibitors. Based upon this premise, pembrolizumab, a PD-1 inhibitor, recently gained pan-tumor FDA approval in patients with TMB-high (TMB-H), that is, ≥10 mutations per megabase (mut/Mb) tumors, based upon data from the KEYNOTE-158 trial (1). While this pan-tumor approval is practice-changing, significant critiques were that KEYNOTE-158 excluded breast, colon, and other common cancers and failed to account for cancer-specific considerations. Subsequent publications challenge the specificity of 10 mut/Mb as a clinically meaningful pan-cancer cutoff, particularly after excluding known immunoresponsive mismatch repair–deficient (dMMR) and pol-hypermutated tumors (2). The challenges of harmonizing TMB analytic validity across assays were recently expertly reviewed in Cancer Discovery, and we direct readers here for a detailed discussion (3). The KEYNOTE-158 trial and subsequent analyses have left many in the field debating optimal TMB cutoffs and whether tissue-specific TMB-H definitions are needed to optimize patient selection.

In this issue, Friedman and colleagues build upon prior TMB literature by presenting results from the TMB-H cohort of the phase IIa MyPathway multibasket study (NCT02091141; ref. 4). Patients with solid tumors with locally defined TMB ≥10 mut/Mb by any chemiluminescent immunoassay were enrolled and treated with atezolizumab, a PD-L1 inhibitor, every 3 weeks, with a primary endpoint of objective response rate (ORR) in patients with centrally determined TMB ≥16 mut/Mb by the FoundationOne CDx assay. Of the 121 patients enrolled, 90 patients with 19 tumor types were evaluable for efficacy and underwent Foundation CDx testing. In total, there were 42 evaluable patients with TMB ≥16 mut/Mb, and 48 with TMB ≥10 and <16 mut/Mb. Patients with TMB ≥16 mut/Mb achieved an investigator-assessed ORR of 38.1% versus only 2.1% in those with a TMB ≥10 mut/Mb and <16 mut/Mb—both clinically and statistically significant. Centrally confirmed response rate is not yet available. Patients with higher TMB also achieved a superior median progression-free survival (mPFS) of 5.7 versus 1.8 months and a median overall survival (mOS) of 19.8 versus 11.4 months. These findings are consistent with prior publications reporting that a higher TMB score cutoff enriches for immune checkpoint blockade responders.

Beyond the response data, there are several relevant observations that can be gleaned from this work. To provide some overall context for the trial population, the authors use a large clinicogenomic solid tumor database (n = 73,693) and report that 8.4% of samples harbor a TMB of 16 mut/Mb or higher, although that number decreases to 7.1% after excluding microsatellite instability–high (MSI-H) samples. Notably, within the reported trial population, 17 patients were reclassified as TMB <10 mut/Mb upon central retesting, which highlights the need for further TMB harmonization between assays, as only 13 of 25 central versus local TMB categorizations were concordant when using the same sample, which declined to 3 of 12 when using different tissue samples. This is of interest, as a large recent series suggested that in general there was little temporal intrapatient TMB variability when using the same assay to test multiple time points (5). Of the evaluable patients, 11 of 40 patients with TMB ≥16 mut/Mb had MSI-H tumors versus only 1 of 46 patients with TMB ≥10 mut/Mb and <16 mut/Mb, and patients with MSI-H and TMB-H tumors achieved superior response and survival compared with those with TMB-H/microsatellite- stable (MSS) tumors. Of note, the only patient with an objective response and TMB <16 mut/Mb was MSI unknown. After excluding patients with MSI-H tumors, those with TMB/TMB-H ≥16 mut/Mb tumors achieved improved survival compared to those with lower TMB, suggesting additional mechanisms at play. The authors went on to demonstrate a 40% ORR in patients with TMB ≥16 mut/Mb who harbored POLE/POLD1 exonuclease domain mutations, highlighting yet another actionable mechanism driving a high TMB. Interestingly, while MSI-H or POLE/POLD1 mutations account for 9 of 16 evaluable responders, underlying hypermutating mechanisms were not identified in the remaining 7 responders. Within the limitations of the sample size, the authors did explore additional incremental TMB cutoffs ranging from 10 to 40 mut/Mb and noted further improvement in ORR, which reflects some further enrichment for MSI-H and POLE/POLD1-hypermutated tumors.

As discussed by the authors, the results from this trial question both optimal cutoff point selection and interassay variability, and their findings are strengthened by central testing using the approved Foundation One CDx companion diagnostic (6). However, the selection of a TMB of ≥16 mut/Mb is based on a retrospective analysis built largely on data from pooled studies heavily enriched for patients with urothelial and non–small cell lung cancer treated with atezolizumab (7). In this article, the ORR was 26.7% when using TMB ≥12 mut/Mb to define TMB-H versus 29.7% when TMB of ≥16 mut/Mb was used (7). Although it is tempting to compare the TMB-H definition used here to the ≥10 mut/Mb TMB-H definition used in KEYNOTE-158, there are several caveats that should be considered. Friedman and colleagues selected patients based upon TMB alone, resulting in inclusion of a larger number of tumor types (n = 20) versus KEYNOTE-158, which was restricted to only 10 tumor types, including several with already known sensitivity to PD-1/PD-L1 inhibitors. While antibody selection differs between trials, we would not expect differences between PD-1 and PD-L1 inhibitors to significantly impact observed response. Unfortunately, with the current immunotherapy landscape, it is unlikely we will see large, randomized trials comparing agents and/or TMB cutoffs that would be required to definitively resolve these lingering questions.

Regardless of the cutoff chosen, effective immune checkpoint blockade biomarkers must reflect neoantigenicity, as TMB attempts to do, but also T-cell activation, which it does not intrinsically assess. It is increasingly clear that nonsynonymous alteration quantification neither captures that frameshift indels are more immunogenic than nonsynonymous single-nucleotide variants nor differentiates intrinsic (MMR deficiency, pol mutations) versus extrinsic environmental (tobacco, UV) hypermutated etiologies (8). This partly explains the modest response rates seen even in patients with TMB ≥16 mut/Mb. When integrated with biomarkers of T-cell activation, such as IFNγ gene expression profiling, PD-L1 expression, or CD8⁺ T-cell infiltration, the predictive ability of TMB improves, though this approach has tumor type–specific implications and is limited by clinical implementability (9, 10).

In summary, Friedman and colleagues present well-conducted prospective support for the clinical activity of atezolizumab in a pan-solid tumor cohort with centrally defined TMB-H (≥16 mut/Mb) using a validated clinical assay. Their findings lend support to the idea that the current tumor-agnostic FDA approval for TMB ≥10 mut/Mb needs ongoing reevaluation and confirmation. In this work, response and PFS in the TMB ≥10 and <16 mut/Mb population mirror what one may expect with best supportive care alone in some tumors. This is a reminder that TMB is important but is unlikely to ever be a one-size-fits-all biomarker and that there is more work needed to make us all TMBelievers.