Abstract 3772: Leveraging the cfDNA fragmentome to predict immunotherapy response

Abstract Background: Assessing immunotherapy response presents unique challenges, driven by the heterogeneity of clinical responses that are inadequately captured by conventional imaging. Early changes in circulating tumor DNA (ctDNA), captured by mutation-based next generation sequencing (NGS), have been shown to serve as a reliable early predictor of response to immune checkpoint inhibitors (ICI). Nevertheless, we may be reaching a plateau in the clinical utility of mutation-based liquid biopsy approaches, which could be alleviated by non-mutation based whole genome sequencing integration. Methods: We analyzed serial plasma cfDNA (n = 323) from patients with metastatic NSCLC (n = 109) that received ICI as part of a prospective clinical protocol (NCT05995821) by low-pass coverage (∼4x) whole genome sequencing. We employed the DELFI Tumor Fraction (DELFI-TF) algorithm, a tumor- and mutation-naive approach that integrates cfDNA fragmentation patterns and aneuploidy across the genome by machine learning, resulting in a fragmentome-based tumor fraction. Limit of blank was defined as the 95th percentile of DELFI-TF in non-cancer controls. In parallel, we performed orthogonal matched white blood cell DNA and cfDNA hybrid capture targeted panel NGS utilizing biospecimens from the same patient population. DELFI-TF scores and the maximal mutant allele fraction (maxMAF) of tumor-specific variants were tracked over time and undetectable DELFI-TF, defined as below the observed LOB at 3-9 weeks on ICI indicated landmark molecular response (mRlandmark). Clinical outcomes were assessed by durable clinical benefit (DCB), progression-free (PFS) and overall survival (OS). Results: DELFI-TF was highly concordant with maxMAF (Pearson’s R = 0.93, p<0.001). Patients with a tumor fraction had shorter OS compared (log-rank p = 0.006). In assessing the clinical sensitivity of molecular response, DELFI-TF mRlandmark had 95% sensitivity and 69% specificity in capturing DCB, while the traditional maxMAF mRlandmark achieved a sensitivity of 66%. Landmark assessment allowed for a lower undetectable rate, as it allows single timepoint analysis, and DELFI-TF mRlandmark was associated with significantly longer PFS (log-rank p = 3.8e-14) and OS (log-rank p = 3.1e-09). Notably, when patients with mutation-based molecular progression were stratified by DELFI-TF mRlandmark, fragmentome-derived TF further identified a patient subset with favorable prognosis (long rank p = 0.0001 and p = 0.002 for PFS and OS respectively). Conclusions: Interrogation of the cfDNA fragmentome provides a clinically meaningful and cost-effective measure of tumor fraction that can be serially tracked to capture immunotherapy response. Tumor- and mutation-naive cfDNA fragmentome-based approaches are pushing the limits of liquid biopsies and may circumven the challenges with variant cellular origin resolution that are typically encountered with mutation-based methods. Citation Format: Noushin Niknafs, Lavanya Sivapalan, Bahar Alipanahi, Daniel Rabizadeh, Mimi Najjar, Zachary L. Skidmore, Gavin Pereira, Caitlin Schonewolf, Bryan Chesnick, Peter Bach, Nicholas C. Dracopoli, Robert Scharpf, Joseph Murray, Victor Velculescu, Valsamo (Elsa) K. Anagnostou. Leveraging the cfDNA fragmentome to predict immunotherapy response [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 3772.