Integrated fragmentomics and droplet digital PCR–based methylation analysis of cell-free DNA for enhanced detection of precancerous lesions and early colorectal cancer.

95

Background: Early detection of colorectal cancer (CRC), particularly at the precancerous stage, is essential for reducing disease burden and enabling cancer interception. While colonoscopy remains the standard screening modality, its invasive nature limits population compliance. Blood-based assays offer a non-invasive alternative but show limited sensitivity for early-stage CRC and advanced adenomas. We hypothesized that a multimodal analysis of circulating cell-free DNA (cfDNA), integrating fragmentomic and methylation features, could improve detection of early colorectal neoplasia. Methods: Plasma cfDNA samples were collected from 380 patients with non-metastatic CRC (stages I–IIIA), 121 individuals with precancerous lesions, and 506 healthy controls. Fragment length profiles, nucleosome footprinting patterns, and 4-mer fragment end motifs were analyzed using shallow whole-genome sequencing to identify cancer-associated fragmentomic features. For ultrasensitive methylation detection, a multitarget droplet digital PCR (ddPCR) panel was developed by ranking CpG sites through integrative analysis of 538 colorectal tumor and 81 adjacent normal tissues from The Cancer Genome Atlas and a large in-house cfDNA methylation database. Fragmentomic features were integrated into multimodal machine-learning classifiers and subsequently combined with methylation markers to distinguish CRC and precancerous lesions from healthy individuals. Results: CRC samples showed distinct enrichment of short cfDNA fragments, whereas precancerous lesions exhibited fragment length distributions comparable to those of healthy controls. Notably, precancerous lesions also demonstrated distinct nucleosome footprinting patterns, characterized by altered coverage at ±150–275 bp from the nucleosome core. Furthermore, fragment end motifs beginning with “C” (CTGG, CTAG, TTCT, CTCC, CTCA) were enriched, while motifs beginning with “G” (GCAG, GCTG, GCAC, GCCG, GCAA) were depleted in both CRC and precancerous lesions. A fragmentomics-only model achieved 96% specificity and 77.6% sensitivity for CRC but showed limited sensitivity for precancerous lesions (46.7%). Integration of five hypermethylated CpG markers associated with SEPTIN9 and HS3ST2 significantly improved performance, increasing sensitivity to 81.6% for CRC and 73.3% for precancerous lesions while maintaining 91% specificity. Conclusions: A multimodal cfDNA approach integrating fragmentomic and methylation signatures enables sensitive, non-invasive detection of early CRC and precancerous lesions. These findings support further prospective validation in screening populations to assess clinical utility for CRC interception.