Analysis of cell-free DNA (cfDNA) fragmentomes and prediction scores across distinct molecular subtypes.

Abstract

e15029 Background: Liquid biopsies have the advantage of being convenient in collecting tumor-derived DNA for clinical use. Genome-wide analysis of cell-free DNA (cfDNA) fragmentomes has emerged as a promising approach to detect lung cancer based on the examination of sequencing features other than actionable mutations. Here we report initial analyses of the DELFI (DNA evaluation of fragments for early interception) technology in patients enrolled in the Dutch Lung Cancer Early Molecular Assessment (LEMA) trial (NCT02894853), a prospective, observational, and multicenter study, to evaluate fragmentation patterns in genotyped blood samples with actionable alterations irrespective of disease stage and histology subtype. Methods: Plasma samples from 576 patients with an established diagnosis of lung cancer but awaiting the start of definitive treatment were processed and examined using low-coverage whole-genome sequencing. A lasso model that incorporated genome-wide fragmentation profiles was trained on an internal Delfi Diagnostics cohort and used to generate patient-specific DELFI scores for participants in LEMA. Results: Enrollees comprised 531 individuals with non-small cell lung cancer, 38 with small-cell lung cancer, and 7 with unknown lung cancer type. Most patients were diagnosed with late-stage disease (28% stage I, 10% stage II, 26% stage III, and 36% stage IV). Current smokers (N = 238; 41%) and former smokers (N = 295, 51%) outnumbered never smokers (N = 43; 8%). Analysis showed a stepwise increase of the DELFI scores from T1 to T4 and N0 to N1-3 disease, associating aberrant fragmentation profiles with the overall tumor burden. Small-cell carcinomas and squamous cell carcinomas presented with higher DELFI scores than adenocarcinomas ( p< 1e-7, generalized linear model). Patients without a smoking history had similar scores to those who had a previous history of smoking. When adjusted for stage, mutation status in ALK, BRAF, EGFR, ERBB2, KRAS, MET, RET, and ROS were not associated with DELFI scores ( p= 0.5, likelihood ratio test). Similarly, PD-L1 staining did not significantly affect DELFI scores when a 1% cutoff was applied to cases with PD-L1 staining data ( p= 0.07, generalized linear model). Conclusions: Our results reveal that DELFI scores increase stepwise with stage and are higher in small-cell and squamous cell carcinomas. The LEMA study reinforces the potential usefulness of genome-wide cfDNA fragmentation profiling as a liquid biopsy approach for identifying patients with lung cancer regardless of smoking status, somatic driver mutation, and PD-L1 expression scores.