Biological and technical factors in the assessment of blood-based tumor mutational burden (bTMB) in patients with NSCLC

Abstract

Background: Patients treated with immunotherapy are at risk of considerable adverse events, and the ongoing struggle is to accurately identify the subset of patients who will benefit. Tumor mutational burden (TMB) has emerged as a promising predictive biomarker but requires tumor tissue which is not always available. Blood-based TMB (bTMB) may provide a minimally invasive assessment of mutational load. However, because of the required sequencing depth, bTMB analysis is costly and prone to false negative results. This study attempted to design a minimally sized bTMB panel, examined a counting-based method for bTMB in patients with stage I to IV non-small cell lung cancer (NSCLC) and evaluated both technical factors such as bTMB and tissue-based TMB (tTMB) cut-off, as well as sample-related factors such as cell-free DNA input mass which influence the correlation between bTMB and tTMB.

Methods: Tissue, plasma, and whole blood samples collected as part of the LEMA trial (NCT02894853) were used in this study. Samples of 185 treatment naïve patients with stage I to IV NSCLC were sequenced at the Roche Sequencing Solutions with a custom panel designed for TMB, using reagents and workflows derived from the AVENIO Tumor Tissue and circulating tumor DNA Analysis Kits.

Results: A TMB panel of 1.1 Mb demonstrated highly accurate TMB high calls with a positive predictive value of 95% when using a tTMB cut-off of 16 mut/Mb, corresponding with 42 mut/Mb for bTMB. The positive per cent agreement (PPA) of bTMB was relatively low at 32%. In stage IV samples with at least 20 ng of cfDNA input, PPA of bTMB improved to 63% and minimizing the panel to a subset of 577 kb was possible while maintaining 63% PPA.

Conclusion: Plasma samples with high bTMB values are highly correspondent with tTMB, whereas bTMB low results may also be the result of low tumor burden at earlier stages of disease as well as poorly shedding tumors. For advanced stages of disease, PPA (sensitivity) of bTMB is satisfactory in comparison to tTMB, even when using a panel of less than 600 kb, warranting consideration of bTMB as a predictive biomarker for patients with NSCLC eligible for immunotherapy in the future.

Keywords: genetic markers; immunotherapy; lung neoplasms; tumor biomarkers.

Conflict of interest statement

Competing interests: ZH, SS, MJ, and DK are employed by Roche Sequencing Solutions (Pleasanton, CA, USA), which manufactures the AVENIO Tumor Tissue and ctDNA Analysis Kits. AL is a former employee of Roche Sequencing Solutions (Pleasanton, CA, USA) and is employed by Freenome (San Francisco, CA, USA). LdV is employed by Roche Diagnostics International, CH. DvdB received payment or honoraria for lectures, presentations, speakers bureaus, manuscript writing, or educational events and for expert testimony by Roche to the institution. KM received a research grant from Astra Zeneca, speakers fees from MSD, Roche, Astra Zeneca, Benecke, consultant fees from Pfizer, BMS, Roche, MSD, Abbvie, AstraZeneca, Diaceutics, Lilly, Bayer, Boehringer Ingelheim, and non-financial support from Roche, Takeda, Pfizer, PGDx, and Delfi. MvdH received sponsorship or research funding by Astrazeneca, BMS, Janssen Pharmaceutica, Stichting Treatmeds, Merck, MSD, Novartis, Pamgene, Pfizer, Roche, Roche diagnostics and fees or other from Abbvie, Astrazeneca, BMS, Lilly, MSD, Novartis, Pfizer, and Roche. MS: none declared.

© Author(s) (or their employer(s)) 2022. Re-use permitted under CC BY-NC. No commercial re-use. See rights and permissions. Published by BMJ.

Figures

Figure 1

Tissue TMB assessment using a 2.2 Mb tumor tissue panel. (A, B) Histogram of TMB (mut/Mb) in tissue samples sequenced by tumor tissue panel filtered for germline variants (A; n=160) is compared with TMB (mut/Mb) from TCGA WES lung cancer adenocarcinoma and squamous cell carcinoma samples (B; n=1058). (C) TMB (mut/Mb) in tissue samples filtered for germline variants separated by stage of disease. (D) Comparison of TMB (mut/Mb) by in-silico validation of our tumor tissue panel in the Rizvi et al dataset which included patients with DCB (n=14) and NDB (n=16), p=0.04. Variant calls are restricted to variants overlapping with the tumor tissue panel that were present at >5%. The diamond symbol (♦) in the boxplot represents outliers. (E) Kaplan-Meier curve showing progression-free survival when using our tumor tissue panel for patients from the Rizvi et al cohort. The dataset is split in high vs low using a cut-off of 9.4 mutations/Mb as determined by variants present in the tumor tissue panel regions. DCB, durable clinical benefit; NDB, no durable benefit; NSCLC, non-small cell lung cancer; TMB, tumor mutational burden; WES, whole exome sequencing.

Figure 2

In-silico lung cancer TMB panel assessment. (A) In-silico comparison of TCGA lung cancer data as determined from WES on the x-axis vs designed lung cancer TMB panels on the y-axis. Three panel subsets are shown with a size of 358 kb (red), 577 kb (blue), and the full 1130 kb (green). (B) Comparison of TMB by in-silico validation of the variants present in lung TMB panel in the Rizvi et al dataset which included patients with DCB (n=14) and NDB (n=16), p=0.002. The diamond symbol (♦) in the boxplot represents outliers. (C) Kaplan-Meier curve showing progression-free survival for patients from the Rizvi et al cohort. The dataset is split by high vs low TMB, using a cut-off of 19.6 mutations/Mb as determined by the lung TMB panel. DCB, durable clinical benefit; NDB, no durable benefit; NSCLC, non-small cell lung cancer; TMB, tumor mutational burden; WES, whole exome sequencing.

Figure 3

Comparison of plasma TMB with tumor tissue TMB. (A) A subset of tumor tissue samples sequenced by the tumor tissue panel were recaptured with the lung TMB panel. The TMB values by the tumor tissue panel correlated with the lung TMB panel: TMB/Mb=2.28 * tissue panel TMB/Mb+5.84. (B) Tissue TMB measured by tumor tissue panel (mut/Mb) vs plasma TMB measured by lung TMB panel (mut/Mb) for all samples (n=143). The linear relationship between the tissue and lung TMB panels determined in (A) is shown in blue, and TMB cutoffs corresponding to 16 mutations/Mb in the tissue panel are shown with red lines. PPA of plasma TMB high calls (percentage of tissue TMB high samples called high in plasma) and positive predictive value of plasma TMB high calls (percentage of plasma TMB high samples called high in tissue) are shown. PPA, positive per cent agreement; PPV, positive predictive value; TMB, tumor mutational burden.

Figure 4

cfDNA input, tumor burden and stage of disease impact PPA of blood-based TMB. (A) Impact of cfDNA input mass, divided into three categories (0.1% (blue). Only samples with >10 ng cfDNA input mass and with variants detected by the tumor tissue panel which overlap with regions of the lung TMB panel are included in this plot. (C) Impact of stage of disease on the correlation between tissue and plasma TMB. Only samples with >10 ng cfDNA input mass are included in this plot. (D) The effect of a combination of cfDNA input mass and stage of disease on PPA for TMB high calls in plasma based on a 16 mut/Mb cut-off in tTMB is shown on the Y-axis. The cell-free DNA input mass was categorized into >0, 10, 20, 30, 40, 50 ng and is shown on the X-axis. Stage of disease was categorized as I–IV, II–IV, III–IV, or IV and is represented by the color of the bars. The number of samples included in the analysis is listed above each bar. (E) Tissue TMB vs plasma TMB for stage IV samples with >20 ng cfDNA input mass (n=39). The linear relationship between the tissue and lung TMB panels is shown in blue, and TMB cutoffs corresponding to 16 mutations/Mb in the tissue panel are shown with red lines. PPA and PPV of plasma TMB high calls (as defined in figure 3B) are shown. AF, allele frequency; PPA, positive per cent agreement; PPV, positive predictive value; TMB, tumor mutational burden.

Figure 5

An NGS panel as small as 577 kilobases enables accurate plasma TMB calls. (A) Correlation metrics (R squared in red; Spearman’s rho in blue) between tissue TMB and plasma TMB using different lung TMB panel subsets. Only samples with >10 ng cfDNA input mass and at least one variant captured were included. In smaller sub-sized panels, less variants could be captured, resulting in exclusion of 4% of the samples in the 358kb panel and 2% in the 423 to 577 kb panels. (B) Tissue TMB vs plasma TMB with the 577 kb panel for all samples (n=143). The linear relationship between tissue and lung TMB panels is shown in blue, and TMB cutoffs corresponding to 16 mutations/Mb in the tissue panel are shown with red lines. (C) Tissue TMB vs plasma TMB with the 577 kb panel for stage IV samples with >20 ng cfDNA input (n=39). The linear relationship between tissue and lung TMB panels is shown in blue, and TMB cutoffs corresponding to 16 mutations/Mb in the tissue panel are shown with red lines. (D) Comparison of TMB by in-silico validation considering only variants in the 577 kb lung TMB panel subset from the Rizvi et al dataset which included patients with DCB (n=14) and NDB (n=16). (E) Kaplan-Meier curve showing progression-free survival for patients from the Rizvi et al cohort. The dataset is split by high vs low TMB, considering only variants in the 577 kb lung TMB panel subset (TMB high: >20.8 mutations/Mb; TMB low: <20.8 mutations/Mb). DCB, durable clinical benefit; NDB, no durable benefit; TMB, tumor mutational burden.