Introduction
In 2017, the U.S. Food and Drug Administration (FDA) granted approval to the programmed death receptor-1 (PD-1) blocking antibody pembrolizumab for patients with unresectable or metastatic, microsatellite instability-high (MSI-H) or mismatch repair-deficient (dMMR) solid tumors that have progressed following prior treatment and who have no alternative treatment options as well as for patients with MSI-H/dMMR metastatic colorectal cancer (mCRC) after failure of fluoropyrimidine, oxaliplatin, and irinotecan. This approval was recognized as a breakthrough given the fact that this was FDA’s first tumor site-agnostic approval.
Several uncontrolled studies including patients with MSI tumors originating from different sites (albeit mainly from the colon and rectum) reported a high rate of in part heavily pretreated patients deriving a clinical benefit of treatment with pembrolizumab or nivolumab ± ipilimumab in terms of long-lasting disease stabilization and tumor remissions (Le et al.
2017; Overman et al.
2017,
2018). More recently, results from 45 patients with MSI-H/dMMR mCRC treated 1st line with nivolumab plus low-dose ipilimumab were reported (Lenz et al.
2019). After a median follow-up of 13.8 months, response rate and disease control rate were 60% and 84%, respectively, and 12-month progression-free survival and overall survival rates were 77% and 83%, respectively.
Moreover, the use of checkpoint inhibitors administered only twice during the waiting period for curative surgery in early stage colon cancer led to complete pathological remissions in four of seven patients with MSI-H colon cancer, while none of eight patients with a microsatellite stable tumor exhibited signs of major pathohistological remission (Chalabi et al.
2018).
MSI-H tumors are characterized by an increased tumor mutation burden (TMB), exhibit increased expression of neoantigens and display higher numbers of tumor-infiltrating lymphocytes (Willis et al.
2020). Susceptibility to treatment with checkpoint inhibitors is consequently increased (Nebot-Bral et al.
2019). However, the presence of MSI-H/dMMR features does not guarantee benefit from checkpoint inhibitor treatment and deeper understanding of the landscape of tumor mutations and their potential predictive potential is mandatory, especially when checkpoint inhibitors are being used in earlier treatment lines with potential alternative treatment options. For instance, loss-of-function mutations in genes relevant for antigen presentation or immune response such as
JAK1,
JAK2,
B2M and
STK11 have been identified as mediators of resistance to PD-1 inhibition despite overall high TMB (Shin et al.
2017; Skoulidis et al.
2018; Zaretsky et al.
2016).
Herein, we report on six patients with MSI-H/dMMR metastatic gastrointestinal (GI) cancers undergoing treatment with checkpoint inhibitors and along with their tumor mutational profile.
Discussion
Checkpoint inhibitors have shown excellent efficacy in patients with MSI-H/dMMR GI cancers. For instance, in KEYNOTE-164, two cohorts of patients with pretreated mCRC received 3-weekly pembrolizumab (Le et al.
2020). Response rates were 33% in patients pretreated with ≥ 2 (cohort A) or ≥ 1 (cohort B) rounds of pretreatment, respectively. While median survival was 31.4 months in cohort A, it had not been reached in cohort B. Similarly, in CheckMate 142, patients with pretreated MSI-H/dMMR mCRC receiving monotherapy with nivolumab after ≥ 1 line of pretreatment exhibited a response rate of 31% and a 1-year survival rate of about 70% (Overman et al.
2017). However, despite these excellent response rates and long-term disease control, it has to be noticed that the rate of primary progression ranged between 26 and 46% in both studies. Preliminary data of 1st line patients receiving nivolumab in a cohort of the CheckMate 142 study reported 16% patients exhibiting primary progression (Lenz et al.
2019). This data underlines the need for additional biomarkers beyond MSI-H/dMMR to minimize patients exposed to checkpoint inhibitors instead of potentially active chemotherapy regimens, especially for those patients scheduled to receive checkpoint inhibitors in earlier lines of treatment.
Recently, Schrock and coworker (Schrock et al.
2019) suggested that TMB might serve as an additional biomarker in mCRC. In a cohort of 22 patients with mCRC treated with checkpoint inhibitors a cut-point of 37.4 mutations per Mb (range 37–41 mutations per Mb) for TMB was reported to distinguish between responders and non-responders. While all 13 patients with TMB values above this threshold exhibited long term benefit, 6 out of 9 patients with lower values showed primary progression. Their threshold is considerably higher than the 11.7 mutations per Mb threshold published by Fabrizio et al. (Fabrizio et al.
2018) in a previous study to identify MSI-high CRC samples. The stricter threshold suggested by Schrock et al. (Schrock et al.
2019) indeed leads to a more stringent identification of responders, however, at the expense of missing some potential responders since there is an overlap in mutation ranges between responders and non-responders. In other words, though recent evidence has shown that higher TMB scores are generally associated with improved response to immune checkpoint blockade across a wide variety of cancer types (Samstein et al.
2019), some patients still benefit from immune checkpoint blockade despite rather low mutation rates. This is evident in the cohort of Schrock et al. (3 patients) and in our cohort (1 patient; P6 with durable complete response to pembrolizumab). Another issue that needs to be considered when discussing TMB as a clinical biomarker in MSI-H patients, is the high variability and inconsistent reporting of (current) TMB assessment methods across different studies, which can create confusion for oncologists and may impact critical treatment decisions.
Originally, TMB was determined by whole exome sequencing (WES) and usually calculated as the number of non-synonymous mutations per exome or Mb reflecting the mutation load in all protein coding regions of the genome. Due to the increased interest in TMB for prediction of response to immune checkpoint inhibition and because WES is not yet routinely used in clinic, recent efforts have begun to validate TMB estimation based on targeted NGS panels, which are already implemented in routine molecular diagnostics for oncogenic mutation detection (Stenzinger et al.
2020). Currently, a minimum panel size of 0.8–1 Mb is widely accepted for TMB estimation for clinical purposes (Allgauer et al.
2018; Buttner et al.
2019), although in silico simulations based on TCGA exome data suggest a panel size of 1.5–3 Mb for an optimized cost–benefit ratio (Buchhalter et al.
2019). Overall, accuracy and precision of TMB estimation tend to increase with panel size (Garofalo et al.
2016) while below 0.5 Mb variance rises drastically, in particular for samples with low TMB (Chalmers et al.
2017). However, not only panel size matters but also panel composition as certain differences in TMB estimations have been observed between panels depending on their genomic composition (Xu et al.
2019). With the increased use of gene panels for TMB estimation, TMB definitions started to diverge from the original WES-based definition and other mutation types such as nonsense mutations, synonymous mutations and small indels were included, however, inconsistently across different laboratories and studies (Chan et al.
2019). Due to the enrichment of cancer relevant genes, it is suggested to remove oncogenic driver events by filtering against databases such as COSMIC (Bamford et al.
2004; Chalmers et al.
2017). As illustrated by our data, the mutation types included and removal of potentially oncogenic somatic mutations by exclusion of COSMIC-listed variants has a considerable effect on TMB estimation. However, COSMIC does not only contain oncogenic/cancer-relevant mutations but somatic mutations in general, a fact that may lead to an overcorrection of TMB estimated based on cancer gene panels. Furthermore, COSMIC is an evolving database, i.e., the number of cataloged mutations will increase over time, questioning its value for correcting panel-based TMB estimates for clinical purposes.
Two other important sources contributing to variability among TMB scores are the method of germline filtering and the minimum VAF threshold. Our data show that if no matched normal sample is sequenced along with the tumor sample, TMB estimation is more robust if filtered against at least a panel of normal samples (PON) that was processed in the same way as the case samples. The PON will not only help augment population frequency databases such as GnomAD, which are highly filtered and curated, but also help remove systematic and assay specific sequencing artifacts, which can be widespread even with matched normal samples. This is even more important when FFPE tissue is used as FFPE material is more artifact prone than fresh/frozen samples. Based on our observations, the commonly used VAF cutpoint of 5% (F1Dx, Oncomine Tumor Mutation Load Assay) may be too low and may thus increase the risk of including false positives, in particular when using FFPE tissue and only computational germline filtering. As filtering against a local PON could at least partially resolve this issue, this should be considered as practical option for routine molecular diagnostics.
Given the plethora of factors influencing exact TMB values (Stenzinger et al.
2019), exact TMB values are only comparable within individual studies using the exact same preanalytical workflow, sequencing methodology and bioinformatics pipeline. However, some confounding factors such as different levels of tumor cell purity, which significantly influence the number of called mutations even if the exact same workflow is followed, will remain in routine practice and are hard to control. For instance, harmonizing tumor purity would require disintegrating the tissue and enriching for the tumor cells by immunofluorescent markers. However, despite all the discrepancies and uncertainties regarding absolute TMB values, recent data from the QuIP study indicate a reasonable agreement of assignment to TMB categories between different laboratories and panels (Stenzinger et al.
2020). Apart from that, it is evident that the majority of responders with metastatic GI cancers greatly exceeds the currently suggested thresholds of 12 or 37 mutations per Mb, respectively, relatively independent from the specific gene panel used, mutation types included, and specific thresholds applied. The small group of patients with response to immune checkpoint inhibition and only low or borderline TMB, on the other hand, may need a biomarker other than TMB for identification. A potential option could be mutation signatures, which are a reflection of the underlying mutational processes, and other factors such as cell type composition/tumor-infiltrating lymphocytes (Loupakis et al.
2020). Ultimately, multi-omics testing may be the most reliable way to identify responders and non-responders to immune checkpoint inhibition. If TMB is pursued as a clinical biomarker for immunotherapy, consistent standards for TMB estimation and reporting are needed to minimize variability, to ensure reliable and reproducible identification of responders, and to allow comparison across studies (Fancello et al.
2019). To make informed clinical decisions, clinicians/oncologists need to be aware that different methods for TMB testing and reporting exist.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.