Hypermutation, Mismatch Repair Deficiency and Defining Predictors of Response to Checkpoint Blockade

Abstract

Mutational burden is positively correlated with tumor neoantigen load and studies have demonstrated an association between high tumor mutational burden (TMB) and response to checkpoint blockade. On the basis of a Phase 2 study, the anti-PD-1 therapy, pembrolizumab, was given FDA approval for use in any solid tumor with a high TMB (i.e. >10 mutations/MB) as assessed by the FoundationOne companion diagnostic. This was an important step in expanding a potentially efficacious treatment option to patients who are likely to benefit and have limited other therapies available. Following this approval, there has been debate regarding the wide applicability of this approval and the most appropriate use of TMB as a predictive biomarker, with several studies questioning the predictive utility of TMB in this context. We discuss the scientific rationale and utility of using TMB as a tool to predict response to immunotherapy as well as address this biomarker’s limitations.

Introduction

The Food and Drug Administration (FDA) granted the first tissue agnostic approval for antineoplastic therapy to pembrolizumab for treatment of mismatch repair (MMR) deficient (MMRd) tumors in 2017, with its label subsequently expanding to include any solid tumor with a tumor mutational burden (TMB) ≥10 mutations/megabase in 2020. The development of TMB and MMRd as predictive biomarkers stems from the observation that mutational burden is associated with improved clinical outcomes to checkpoint inhibitors across a number of different malignancies.^1,2 MMR genes assist in maintenance of genomic fidelity during DNA replication, and disruption of the MMR machinery can lead to somatic hypermutation and microsatellite instability (MSI).³ Hypermutation results in more tumor associated neoantigens and in turn, more tumor-specific T-cells.⁴ After the observation that MMRd colorectal cancer had a better response to anti-PD-1 therapy than mismatch repair proficient colorectal cancer, prospective studies validated the use of MMRd as a predictive biomarker.^2,3 These observations also led to the hypothesis that high TMB, even in the absence of MMRd, would result in more tumor associated neoantigens and increased immune recognition, with several studies showing a correlation between high mutation burden and checkpoint inhibitor response.^1,4–6

TMB as a Predictive Biomarker

Prospective data supporting TMB as an important biomarker for deploying immunotherapy comes from the Keynote-158 study, which was a single-arm, phase II multi-cohort study of pembrolizumab for previously treated advanced solid tumors.⁷ This study enrolled 790 TMB evaluable patients across ten rare solid tumor types and demonstrated an objective response rate (ORR) of 29% in patients with TMB-high (TMB-H) disease, which was defined as ≥10 mutations/megabase, compared to only 6% without TMB-H disease. Importantly, only 14% of patients with TMB-H tumors were also found to be MSI-High. When these MSI-High cases were excluded from the efficacy analysis the response rate was similar, with 28% of patients with TMB-H/MSI-Low disease achieving an objective response. Keynote-158 included patients with anal, biliary, cervical, endometrial, neuroendocrine, salivary, small-cell lung, thyroid, and vulvar cancers as well as mesothelioma.⁷ Notably, it did not include patients with the most frequently diagnosed tumor types including breast, prostate, and colorectal cancers. The resultant broad FDA approval stemming from this trial sparked a robust discussion about the universal applicability of TMB as a predictive biomarker across varied tumor types.^8–10

There have been several recent publications reigniting this debate. Rousseau and colleagues reported in the New England Journal of Medicine a retrospective analysis of 1661 patients with a variety of tumor types which demonstrated that TMB ≥10 mutations/megabase was not clearly associated with outcomes to checkpoint blockade and that in most cases, only TMB-H tumors that also had evidence of MMRd or mutations in pol-d benefitted from checkpoint blockade.¹¹ This was true for gastric, esophageal, colorectal, urinary tract, and brain tumors. The authors did find an association between TMB-H tumors and survival, even in the absence of MMRd, in patients treated with checkpoint blockade for non-small cell lung cancer (NSCLC), melanoma, and head and neck carcinomas, which they concluded was likely related to those cancers’ strong association with environmental carcinogens. McGrail and colleagues also recently published in Annals of Oncology a report of a retrospective analysis of thousands of patient samples which found that tumors with TMB ≥ 10 mutations/megabase, as determined using the FoundationOne assay, were not associated with response to checkpoint inhibitors in cancer subtypes where there was no correlation between CD8 T-cell infiltration and neoantigen load, such as prostate cancer, breast cancer, head and neck cancer, glioma, and others.⁹

These findings seemingly contradict prior prospective studies and also run contrary to the central hypothesis that mutationally derived neoantigens are fundamental to the activity of immune checkpoint inhibitors. Neoantigens are novel proteins on the cell surface of cancer cells that could be specifically recognized by neoantigen-specific T cell receptors. MMRd tumors have high mutational burden, and in theory a high neoantigen load, due to an inability to correct DNA errors that occur during cellular replication. Translational studies have shown that MMRd/microsatellite instability is associated with immunogenic neoantigen burden, which in turn associates with response to checkpoint inhibitor therapy.^3,12,13 Significant evidence supports the use of MMRd or MSI to select patients for immune checkpoint inhibitor therapy, which most likely relates to the association between MMRd/MSI and high TMB. This appears to hold true even for patients with cancers that have been traditionally less response to immunotherapy, such as prostate cancer.^14,15 Therefore, it stands to reason that high TMB as a result of any cause should associate with neoantigen load and response to checkpoint inhibition. To accept MMRd/microsatellite status as useful biomarkers but not mutational burden seems contradictory.

It is plausible that other factors may account for why TMB in these studies has not consistently been shown to predict response to immune checkpoint blockade. The findings reported by McGrail and colleagues may be influenced by their dataset. The Cancer Genome Atlas (TCGA) was used in order to delineate whether a cancer type had a positive association between mutation load and CD8+ T-cell infiltration. There may be some problems with this approach, for example, in many cancer types (e.g., prostate, renal cell, bladder, breast) TCGA analyzed primary tumors from mostly localized cases, which is unlikely to accurately reflect the mutational or immunologic landscape of metastatic tumors. There is also data that mutation burden correlates with CD8+ infiltrating lymphocytes in prostate cancer, which is contradictory to the authors’ classification of prostate cancer.¹⁶ Further, the clinical dataset used in this paper to explore the association between TMB and ICI response in prostate cancer, used a study of anti-CTLA4 monotherapy, which may not extrapolate to anti-PD1 response. The absolute numbers of patients analyzed for ICI response in any given cancer type is also low.

Defining the Optimal TMB Threshold

The results reported by Rousseau, et al may be due to lack of optimization of their process for estimating TMB or a consequence of the threshold used to define TMB-H as opposed to biologic differences in tumors with MMR and pol-d gene mutations.¹¹ For example, our institutional data shows that across 30 consecutive cases with microsatellite instability, the median TMB is 20 mutations/megabase, indicating that identifying cases by MMR gene status may simply select for higher TMB.¹⁷ While ≥10 mutations/megabase is the approved threshold accepted by the FDA to define hypermutated solid tumors as part of pembrolizumab’s label, the evidence to support this specific threshold is slim, and a higher mutational load may be more appropriate to use for selecting patients. Indeed, a linear correlation between average TMB in solid tumor types and response rate to anti-PD-1 or anti-PD-L1 monotherapy has been reported.^1,18 In addition, median TMB score was higher in responders than in non-responders in the Keynote-158 trial and in other retrospective analyses of diverse tumor types.^7,19 Many of the tumor types that routinely respond to checkpoint inhibitor therapy, such as melanoma and non-small cell lung cancer, have high rates of TMB without MMRd, indicating that using MMRd status alone to select patients for checkpoint inhibitor therapy would omit many patients likely to benefit.^20,21 Further, the validation dataset referenced by Rousseau and colleagues reported improved survival in ICI-treated cases with TMB in the top 10% vs. 10–20%.¹¹ Additionally, although the paper reported no benefit of checkpoint inhibitor therapy in breast cancer, a recent non-randomized study of pembrolizumab in metastatic breast cancer with high TMB found a 37% disease control rate and a 21% objective response rate.²² Only one patient in this trial was known to be MSI-H.

There may also be different TMB cutoffs that are appropriate for different tumor types. In a retrospective analysis of 1,578 patients with microsatellite stable tumors treated with immune checkpoint inhibitors, response rates were higher for patients in the highest decile of tumor mutational burden (≥18 mutations/megabase).¹⁰ Using tissue-specific cutoffs resulted in even higher response rates for TMB-H tumors in most cancer types studied.¹⁰ In contrast, while using the FDA approved threshold of 10 mutations/megabase was associated with higher response rates to anti-PD-1 or anti-PD-L1 therapy, this association was not consistent across all tumor types studied.¹⁰ Similarly, the mismatch repair proficient tumors that were found to have benefit from checkpoint inhibitors in the analysis by Rousseau and colleagues were those that had strong associations with environmental carcinogens.¹¹ TMB may also have a role as a prognostic biomarker for survival even in the absence of immunotherapy, and the mutational cutoff for this purpose may be distinct.²³

Methodologic Challenges in TMB Assessment

Accurately calling TMB-H tumors from targeted exome panels can be challenging. An ideal approach is to use whole exome sequencing (WES) to calculate mutational burden, with matched germline testing to eliminate normal germline variations. Routinely using WES in clinical care, however, is impractical due to high costs, difficulty in attaining adequate tissue samples, and large amounts of data to process. Panel-based sequencing is a more cost-effective alternative, as it encompasses a more limited set of recurrently-mutated genes in cancer. There are several inherent limitations in the use of panel-based sequencing to calculate TMB, which can influence both cross-assay validity as well as variability between cancer types. Due to the fact that they focus on frequently mutated genes, panel-based sequencing can often overestimate TMB and requires bias correction to control for this fact. The size of the genome included in the panel, read depth, and intra-tumoral heterogeneity can also all influence the calculation of TMB. Precision of TMB calculation is correlated with panel size, with imprecision increasing significantly in panel sizes <1 Mega base pairs (Mbp).²⁴ Further variability is introduced when assessing TMB through sequencing cell-free circulating tumor DNA (ctDNA). Although typically more convenient than obtaining a tissue sample, there has not been an established consensus that TMB estimates derived from sequencing plasma-based ctDNA are concordant with tissue assessments. Overall, the predictive utility of TMB assessed through plasma-based studies remains less well defined.^25,26 Future clinical studies are needed to validate the use of plasma-based TMB assays as predictive biomarkers for immune checkpoint blockade.

The FoundationOne CDx assay, which is the FDA approved companion diagnostic for selecting patients for pembrolizumab therapy based on TMB, determines TMB by counting synonymous and nonsynonymous variants present with at least 5% allele frequency on a limited gene panel of 0.8 Mb and it does not use a paired normal sample to account for germline mutations.²⁷ In one comparison of WES to the Foundation One assay, it was found that the FoundationOne CDx led to overestimation of TMB in 25 out of 31 cancer types analyzed.⁹ Overestimation of TMB by targeted sequencing panels is exacerbated in cancers with lower mutation burden because driver mutations make up a larger proportion of total mutation number.⁹

Additionally, not all mutations are equally likely to generate neoantigens. In a cohort of patients with NSCLC, nonsynonymous mutation burden correlated better with objective response rate and progression-free survival with immune checkpoint blockade than total exonic mutation burden.⁶ These limitations may partially explain why an association between high TMB and immunotherapy response is not always reproduced. It is also possible that there are biologic differences between mutations that arise in the context of MMRd compared to other causes of hypermutation. Mismatch repair machinery is responsible for repairing single nucleotide mismatches as well as correcting small insertions and deletions. When this machinery is deficient, tumors accumulate a large number of single-nucleotide variants and frameshift mutations.²⁸ Frameshift mutations are more likely to generate neoantigens, and it has been demonstrated in colorectal cancer that tumors with large numbers of frameshift mutations have higher numbers of CD8+ T-cell infiltrates.²⁹ Other drivers of hypermutation, such as chemotherapy, carcinogen exposure, or other oncogenic driver mutations have other genomic signatures which could be less immunogenic.³⁰

Future Directions

The predictive ability of TMB may be further enhanced by the use of complementary biomarkers. Somatic mutations give rise to tumor neoantigens only in the minority of cases and there are other factors that affect the ability of T-cells to recognize and respond to neoantigens. Clonality of neoantigens may also play a role, as tumor neoantigens expressed by only a minority of tumor cells may not be effectively targeted. MgGranahan, et al. found that in a small cohort of patients with NSCLC and melanoma, clinical efficacy of checkpoint blockade was related to higher neoantigen clonality.⁴ Using a combination of TMB with other markers of immune response may be more predictive than TMB alone. In a meta-analysis of clinical trials using TMB, gene expression profiling (GEP), PD-L1 immunohistochemistry (IHC), or multiplex IHC to predict response to anti-PD-1/PD-L1 therapy, multimodal strategies, such as TMB combined with GEP, performed better than TMB alone.³¹

While the current FDA approval is broad, discarding TMB as a predictive biomarker for immunotherapy response would be unnecessarily restrictive for many patients. Central to the development of any biomarker is the ability to consistently assess the biomarkers status. To that end, future studies assessing TMB should employ validated approaches for correcting bias that arises when estimating TMB from panel-based sequencing assays. We expect that the use of TMB will be further refined and optimized with time, but that it will remain a useful predictive tool for selecting patients most likely to benefit from immunotherapy.

Statement of Translational Relevance.

Immune checkpoint blockade has revolutionized cancer treatment in the last decade. Not all patients respond to treatment, however, and toxicities can be severe and potentially life-threatening. This has highlighted the importance of optimal biomarker discovery for identifying patients most likely to benefit from immune checkpoint blockade. Multiple studies have shown that tumor mutational burden (TMB) associates with response to checkpoint blockade; however, questions remain regarding how to best deploy this biomarker in clinical practice given that not all tumors with high TMB appear to benefit. In addition, TMB estimates are prone to bias when assessing from panel-based sequencing assays and require error correction in order to accurately determine mutational load. Regardless, TMB remains an important biomarker and future studies should focus on identifying the optimal TMB threshold to predict response to immunotherapy and bioinformatic approaches for estimating TMB should be standardized.