Lynch syndrome (LS), an inherited predisposition syndrome associated with an increased risk of colorectal, endometrial, and other cancers, is characterized by germline mutations in mismatch repair pathway genes, which typically lead to microsatellite instability (MSI) in the resulting tumors. The FDA approval of pembrolizumab for all advanced MSI-H solid tumors has led to increasing MSI assessment. The presence of MSI in LS-associated tumors provides a unique and transformative opportunity for early detection and disease monitoring in these patients. Here we describe an approach to detect MSI from plasma cfDNA using MSK-ACCESS, a custom capture “liquid biopsy” approved for clinical use by the NY State Department of Health. In addition to frequently mutated exons of 129 genes, MSK-ACCESS also includes 165 highly informative microsatellite loci, selected from over 1,000 microsatellite regions based on >25,000 tumors sequenced using MSK-IMPACT, an FDA-authorized tumor sequencing panel. A key challenge in detecting MSI from cfDNA is the lack of ground truth in these samples, as cfDNA obtained from patients with MSI-high tumors may not always exhibit sufficient tumor-derived DNA fragments. To address this, we developed a machine learning approach for cfDNA analysis trained on orthogonally validated tumors sequenced via MSK-IMPACT. We present Allelic Distance-based Microsatellite Instability Estimator (ADMIE), an approach to translate deviation in tumor/cfDNA from normal/buffy coat DNA at individual microsatellite loci to a binary MSI call. ADMIE achieved a cross-validation precision of 1.00 +/- 0.02 and recall of 0.99 +/- 0.07. We ran this on 44 plasma samples collected from over 30 patients with MSI tumors including colorectal, prostate, and gastric cancers across multiple time points. We also evaluated plasma from 70 patients with known MSS tumors and 46 healthy controls. None of the cfDNA from healthy controls or patients with MSS tumors were found to be MSI positive, indicating high specificity. To establish our limit of detection, we performed in silico dilution experiments leveraging patient samples and MSI signal of biologic origin to simulate different tumor fractions, establishing our limit of detection at 1%. Among patients with MSI-high tumors, we found the presence and magnitude of MSI in the cfDNA to be correlated with measurable response to treatment with immunotherapy. In these patients, we detected MSI in the cfDNA of 6/8 samples where at least one mutation was detectable in plasma above 0.2% at baseline. Among the 4/6 patients for whom we had additional time points post treatment, we did not detect any mutations or evidence of MSI. In one patient, MSK-ACCESS indicated the presence of a second primary tumor based on the detection of MSI and mutations in cfDNA completely independent from those identified in the previously sequenced tumor. Our results suggest that MSI can be reliably detected in cfDNA using MSK-ACCESS and the MSI signature can represent a marker of occult metastatic disease in LS.

This abstract is also being presented as Poster A54.

Citation Format: Preethi Srinivasan, Alicia Latham, Zalak Patel, John Ziegler, Maysun Hasan, Juber A. Patel, Ian Johnson, Ronak Shah, Fanli Meng, Xiaohong Jing, Grittney Tam, Rose Brannon, Andrea Cercek, Ahmet Zehir, Brian Houck-Loomis, Dana Tsui, Zsofia Stadler, Michael F. Berger. MSI detection in plasma cfDNA: MSI as a marker of disease burden [abstract]. In: Proceedings of the AACR Special Conference on Advances in Liquid Biopsies; Jan 13-16, 2020; Miami, FL. Philadelphia (PA): AACR; Clin Cancer Res 2020;26(11_Suppl):Abstract nr PR07.