Association of ultra-sensitive ctDNA assay to identify actionable variants and response to immune checkpoint inhibitor (ICI) therapy in metastatic melanoma.

9562 Background: Detection of molecular residual disease (MRD) via circulating tumor DNA (ctDNA) can identify therapeutic response/resistance months in advance of imaging, and monitoring clinically actionable variant dynamics in ctDNA may be important for guiding treatment. Despite this, adoption has been slow due to the reduced sensitivity and reproducibility of current approaches. Here, we profile melanoma patients receiving ICI over several years using an ultra-sensitive, tumor-informed ctDNA platform, and correlate the findings with clinical outcome. Methods: Over 150 plasma samples from 23 advanced melanoma patients (stage IV) were collected during ICI treatment (up to 40 months) and profiled with NeXT Personal, a tumor-informed ctDNA assay that leverages whole-genome sequencing of tumor/normal samples to generate personalized liquid biopsy panels. Each bespoke panel consists of up to 1,800 selected variants which enable sensitive MRD detection, and a fixed set of 2,100 known clinically actionable and resistance loci for detection of variants emerging under therapeutic pressure. MRD signal and variant dynamics were then correlated with RECIST assessments and outcome. Results: In this cohort, ctDNA was detected across a broad dynamic range (2.3-100,000 PPM; median limit of detection = 1.97 PPM) with 37% of detections occurring below 100 PPM. Baseline ctDNA was detected in 94% (17/18) of patients, and was significantly correlated with S100B and metastatic burden (Spearman’s rho, 0.73 and 0.5; p, 0.0002 and 0.025). The mutation-rate normalized count of melanoma-specific driver genes detected at baseline predicted reduced overall survival (OS) (p = 0.03). 94% (17/18) of ctDNA+ patients displayed dynamic variant allele frequency (VAF) shifts, impacting 149 loci across 37 ICI-related genes including BRAF, CDKN2A, KIT, MAP2K1, and NRAS. Delta ctDNA from baseline to the first on-treatment time point correlated with PFS (log rank p = 0.004). Treatment elicited greater than 3-fold reduction in average PPM, and the magnitude was significantly correlated with progression (p = 0.0005). VAF increases in known ICI-related genes (JAK1, ALK, etc) predicted shortened OS (p = 0.007) and fewer variants in genes conferring sensitivity to (NRAS, BRAF, etc) ICI were detected on-treatment in progressing patients (p = 0.025). Conclusions: We tracked both tumor-informed MRD and tumor-agnostic evolution of ICI related variants over the course of treatment using a single liquid biopsy platform. We achieved ultra-sensitive ctDNA detection down to 2.3 PPM and dynamics in ctDNA signals were predictive of clinical outcome. Furthermore, we demonstrated that ctDNA measurements and VAF dynamics of clinically actionable and resistance variants correlated with response as well as resistance to ICI.