Enhanced cfDNA fragmentation-based treatment monitoring on the Ultima Genomics platform

Introduction: Whole-genome analysis of cell-free DNA (cfDNA) fragmentation, such as DELFI-Tumor Fraction (DELFI-TF), has emerged as a powerful tool for monitoring therapeutic response in patients with late-stage cancer. In this study, we assessed the technical feasibility of the Ultima Genomics sequencing platform for a cfDNA fragmentation-based monitoring application. Methods: The study consisted of two cohorts: 1) 48 longitudinal plasma samples collected from 16 stage IV lung cancer patients and 2) 4 tumor, normal, and matched plasma samples of varying tumor types from treatment-naive stage IV patients. cfDNA was sequenced using three methods: the Illumina NovaSeq6000 (mean depth: 11.8x) and the Ultima UG 100 using the standard approach (8 replicates per sample, 11.0x) or ppmSeq (174.6x). Fragmentation features and DELFI-TF were computed, and DELFI-TF values were compared against somatic variant allele frequencies (VAF) from a 500-gene targeted sequencing panel. WGS somatic variants were identified by filtering ppmSeq variant calls and used to categorize reads into two groups based on their likely origin: tumor or white blood cell (WBC). Fragmentation features were then calculated per group as well as for all reads combined. Results: Ultima sequences were high quality, with 88.3% of bases >Q30 base quality and 99.0% of reads aligned on average. Fragment length distributions (FLDs) within the 100-220bp range (Pearson R > 0.998, p < 1e-5) and genome-wide fragmentation profiles (Pearson R > 0.939, p < 1e-5) were highly correlated between platforms. Limits of blank, with overlapping bootstrapped confidence intervals, and ctDNA detection (100% positive, 95.7% negative, and 97.4% overall percent agreement) were also consistent. Ultima-DELFI-TF correlated strongly with VAF (Pearson R=0.979, p<1e-5), showed high precision across replicates (median robust CV: 2.6%), and reflected longitudinal ctDNA patterns in treatment response. Classifying read origin by somatic variants from ppmSeq resulted in tumor-derived FLDs that displayed a higher proportion of short fragments than WBC-derived FLDs (p-value < 1e-5, KS test). When compared to a reference FLD derived from samples with undetectable ctDNA, the tumor-derived FLD deviated more from the reference distribution than the WBC-derived and combined FLDs (median KL-divergence of 0.227 [tumor], 0.078 [WBC], and 0.089 [combined]). The relative frequencies of 10 fragment end motifs were consistently elevated in the tumor-derived fragments across the whole cohort (all p < 1e-5, Chi-sq). Conclusions: These data support the technical feasibility of implementing DELFI-TF on the Ultima platform. Combining DELFI-TF with Ultima ppmSeq enables the detection of tumor-originated reads and provides a path toward more sensitive disease detection.