Cancer-like fragmentomic characteristics of somatic variants in cell-free DNA

Abstract

Cell-free DNA (cfDNA) in plasma consists of short DNA fragments resulting from a non-random fragmentation process, with distinct "fragmentomic" characteristics that are related with their cellular origins. Despite the wealth of low-pass whole genome sequencing data from recent cancer liquid biopsy studies, somatic variants within this data have often been overlooked without detailed characterization. Here, we reported that in low-pass sequencing data, somatic variant signatures in cfDNA markedly differ between non-cancerous controls and cancer patients, indicating that tumor-associated signals are retained in these variants. Surprisingly, our investigation into cfDNA fragmentomics showed that even in controls, cfDNA molecules harboring somatic variants exhibit cancer-like traits, such as reduced size, decreased DNA methylation, and altered end motif usages and distributions in the nucleosome structure. Importantly, these somatic variants-associated fragmentomic aberrations are more pronounced in cancer patients, enabling cancer diagnosis. In a large pan-cancer cohort, we utilized artificial intelligence to integrate genomic, fragmentomic, and epigenomic features, developing diagnostic models named FreeSV (Fragmentomic and epigenomic examinations of Somatic Variants in cfDNA) and FreeSV-m. Leveraging somatic variant-associated features alone, the FreeSV model achieved area under the ROC curves (AUCs) between 0.81-0.92 across cancer types; however, when genomewide features were also included, the AUCs of FreeSV-m model substantially increased to 0.93-0.99 across cancer types, highlighting the significance of integrative genomic and fragmentomic analyses in cfDNA for cancer liquid biopsy.