Genome-wide mutational integration for ultra-sensitive plasma tumor burden monitoring in immunotherapy

PROJECT SUMMARY A major gap in cancer diagnostics is that state-of-the-art imaging and other existing methods fail to reliably detect low levels of cancer known as minimal residual disease (MRD), which remain following surgical resection of early-stage tumors or treatment of advanced disease. Left untreated, MRD can proliferate and result in lethal

cancer recurrence. Hence, there is a critical need to sensitively detect MRD in order to optimize adjuvant therapies or precision immunotherapy. Liquid biopsy offers the ability to noninvasively monitor MRD by detecting circulating tumor DNA (ctDNA) originating from cancer cells. Nonetheless, detection of ctDNA is challenging due

to extremely low levels of ctDNA in low-burden disease. The prevailing paradigm argues for deep targeted sequencing of informative loci. However, we have shown that this approach faces fundamental barriers to sensitivity due to the low amount of available DNA in typical plasma samples, which imposes a physical ceiling

on depth of sequencing. To overcome this challenge, our interdisciplinary team of geneticists, computer scientists, and oncologists developed MRDetect, an orthogonal approach for ctDNA detection based on genome-wide mutation aggregation of single-nucleotide variants (SNVs) and copy number variants (CNVs) using

whole-genome sequencing (WGS) of plasma. MRDetect enables ultra-sensitive MRD detection down to one part in a hundred thousand, and we have demonstrated its ability to detect MRD shortly after surgery or treatment in colorectal cancer, melanoma and non small-cell lung cancer (NSCLC). Our objective in this project is to develop

crucial advances that will foster broad-based adoption of this technology across cancer settings. First, we propose to incorporate advanced machine learning (ML) framework known as ‘deep learning’ (DL) into the MRDetect platform to enable SNV identification in plasma WGS in low tumor burden settings (Aim 1). This will

yield MRDetect-DL, which we anticipate will significantly improve cancer detection at low tumor levels through a >100-fold improvement in signal to noise enrichment compared to MRDetect. MRDetect-DL performance will be tested in high-risk post-operative melanoma to define the need for adjuvant therapy, as well as in advanced

melanoma treated with immunotherapy for precision immunotherapy applications. Critically, MRDetect-DL will obviate MRDetect’s need for a matched tumor sample, ensuring broad adoption across different clinical settings. Second, we posit that in addition to SNV-based advances, MRDetect’s sensitivity can be increased by enhanced

detection of CNVs, as these are broadly observed in solid tumors. We propose to develop MRDetect-CNV, an ML-denoising technique to ultra-sensitively detect small CNVs using plasma WGS (Aim 2). We will test MRDetect-CNV on NSCLC plasma samples from patients undergoing neoadjuvant immunotherapy to define its

ability to predict treatment response. Impact: Pairing MRDetect-DL with MRDetect-CNV will significantly improve low burden cancer detection in adjuvant, neoadjuvant, and systemic immunotherapy, enabling broad clinical application in oncology.