Pan-cancer analysis of the effect of biopsy site on tumor mutational burden observations.

Background: Tumor mutational burden (TMB) has been proposed as a predictive biomarker of response to immunotherapy. Efforts to standardize TMB scores for use in the clinic and to identify the factors that could impact TMB scores are of high importance. However, the biopsy collection site has not been assessed as a factor that may influence TMB scores.
Methods: We examine a real-world cohort comprising 137,771 specimens across 47 tissues in 12 indications profiled by the FoundationOne assay (Foundation Medicine, Cambridge, MA) to assess the prevalence of biopsy sites for each indication and their TMB scores distribution.
Results: We observe a wide variety of biopsy sites from which specimens are sent for genomic testing and show that TMB scores differ in a cancer- and tissue-specific manner. For example, brain or adrenal gland specimens from NSCLC patients show higher TMB scores than local lung specimens (mean difference 3.31 mut/Mb; p < 0.01, 3.90 mut/Mb; p < 0.01, respectively), whereas bone specimens show no difference. Conclusions: Our data shed light on the biopsied tissue as a driver of TMB measurement variability in clinical practice.

Introduction

Immunomodulatory cancer drugs, such as anti-programmed death-1 (anti-PD-1) antibodies, have transformed the clinical oncology landscape, leading to significant clinical benefit across multiple cancer types. However, not all patients benefit, highlighting the need for predictive biomarkers to guide clinical decision-making. Tumor mutational burden (TMB), a proxy for tumor-specific neoantigens leading to recognition by cytotoxic T cells, has been proposed to stratify patients likely to respond to anti-PD-1 therapy. TMB score is defined as the number of somatic nonsynonymous mutations per megabase (mut/Mb), as assessed by next-generation sequencing of targeted genomic regions.

TMB score has been reported to be associated with known mutagenic processes such as deficient DNA mismatch repair, smoking, and ultraviolet light exposure[1] and in some cases, chemotherapeutic treatment[2]. Within a cancer type, TMB scores can vary widely, with melanoma patients showing scores ranging from 1 to 1000 mut/Mb[3]. Previous work has shown that there were differences in TMB between primary and metastatic tissues, with metastatic tumors having higher TMB scores[4].

High TMB has previously been associated with anti-PD-1 response in multiple cancer types, most notably in non-small cell lung cancer (NSCLC)[5-7] and melanoma[8], indications in which TMB values are higher than most cancers[3]. Those associations with long-term survival benefits have led to attempts at identifying TMB-high patients in multiple indications accompanied by efforts to identify cancer-specific thresholds[9]. Recently, TMB was shown to correlate with higher response rates in patients treated with anti-PD-1 across multiple cancer types, leading to an FDA approval for tissue TMB-high patients with solid tumors. This approval was based on a threshold of 10 mut/Mb, applicable to all solid cancer types[10].

In parallel, important efforts have been made to develop TMB as a standardized and cost-effective clinical assay leading to a better understanding of how to best quantify and interpret TMB as a biomarker[11]. These efforts, however, have been focused on the technical aspects of TMB measurements, such as panel size, sequencing depth, bioinformatics pipelines and variant filtering and have ignored the source of the specimen as a potential factor influencing TMB.

Profiling the genetic diversity in cancer patients has been an active field of study across multiple cancer types. Large consortium efforts, such as the TRACERx initiative, have shown that within a single tumor lesion, there is considerable genetic heterogeneity, which can affect patient outcomes[12,13]. Focused on primary versus metastatic diversity, a study in NSCLC patients showed that EGFR expression was lower in metastatic tissues[14]. In contrast, primary and metastatic specimens from breast cancer patients had a high degree of concordance between the immunohistochemistry staining levels of estrogen receptor, progesterone receptor, HER‑2, and Ki‑67[15,16]. More recent work identified differences in TMB measurements between specimens from primary and metastatic biopsies[4]. Moreover, in a recent study profiling TMB measurements in specimens from lung adenocarcinoma patients, the authors showed that in addition to metastasis-specific differences, there were site-specific differences, notably in brain and adrenal gland metastases[17]. Here we extend those previous efforts and profile the effect of the biopsy site on TMB measurements in 137,771 specimens, biopsied from 47 tissues across 12 cancer types in a real-world cohort with genomic sequencing of tissue from Foundation Medicine (Cambridge, MA). Our comprehensive study shows that the biopsy site is associated with TMB score, with cancer-specific patterns.

Methods

Cohort and sample selection

We selected patient specimens profiled on bait sets DX1, D2, T5a, or T7 of the Foundation Medicine FoundationOne CDx or FoundationOne assay, as they are performed on tumor material (as opposed to blood). In cases where a patient had multiple specimens available, we kept the most recently collected specimen. Specimens for the same patients collected on the same date were chosen arbitrarily. To reduce the impact of tumor purity on results, each specimen was hand reviewed by a pathologist to ensure it was suitable for sequencing and we kept only specimens with ≥30% tumor purity, resulting in a collection of high-quality samples with a median coverage exceeding 500×. We filtered mutations with an allele frequency exceeding 5%. To reduce modeling noise and multiple testing, in each cancer type independently, we discarded specimens from tissues that had <50 specimens in total. Biopsy sites were assigned as primary by manual curation based on prior knowledge of each tumor type. Ethical approval, including a waiver of informed consent and a HIPAA waiver of authorization, was received from the Western Institutional Review Board (Protocol No. 20152817). Consented data that can be released are included in the article and its supplementary files. Patients were not consented for the public release of underlying sequence data.

Statistical association with TMB

We used the TMB score as previously defined[18]. Given that TMB score is log-normal distributed, we log-transformed the TMB score variable (log(TMB + 1)). Each cancer type was treated independently. We used multivariate linear regression and estimated marginal means to compute the difference in TMB score associated with the biopsied site. We reported the difference between all tissues and a reference/primary site, which differed by cancer type.

Difference estimates and confidence intervals (CIs) were back-transformed on “TMB scale (mut/Mb)” using re-gridding provided in the R package emmeans[19,20]. In order to control for sample quality factors that may influence the TMB score, we included covariates that measured the median sample sequencing coverage, tumor purity as estimated by a pathologist, and tumor purity based on FoundationMedicine computational model[21].

Percentage of patients assessed as TMB-high

We grouped specimens into TMB-high and TMB-low categories based on a 10 mut/Mb cutoff and calculated the difference in percentage and odds ratio (OR) for all tissues compared to a reference/primary site, which differed by cancer type. We tested for association using chi-square test. If a cancer type–tissue pair had <10 TMB-high or TMB-low specimens, the comparison was not performed.

Controlling for the presence of metastasis

We used available clinical annotations from the Flatiron-Foundation Medicine Clinico-Genomic Database (CGDB; Flatiron Health, New York, NY) from breast, non-small cell lung, and melanoma cancers to stratify patients according to whether they had a metastasis reported prior to primary specimen collection. Each cancer type and target specimen type were modeled independently, as described above. Sample counts for each comparison can be found in Supplementary Data 5.

Tissue pair analysis

We aggregated all specimens by patients and filtered for patients who had at least 2 biopsies from different sites in a 90 days window. The most frequent tissue pair was primary lung and metastatic brain specimens in lung cancer patients, with nine patients.