Analytical performance of an ultrasensitive whole genome sequencing assay for molecular residual disease detection

In non-metastatic cancers after curative intent intervention, a significant subset of patients retain tumor cells which can lead to disease recurrence. These residual tumor cells can be detected through ultra-sensitive circulating tumor DNA (ctDNA) assays and reported as molecular residual disease (MRD). The most common approaches generally utilize patient-specific, bespoke panels, which have extended turnaround times for initial testing and relatively high cell-free DNA (cfDNA) input requirements.  Here we present a non-bespoke approach based on whole genome sequencing (WGS), which leverages native duplex error correction using the Ultima Genomics (UG) sequencing platform for rapid turnaround times for initial testing with low cfDNA input. Specifically, patient tumor, white blood cell, and plasma derived DNA were sequenced to approximately 80x, 30x, and 80x depth, respectively, through a PCR-free WGS workflow on the UG 100 platform. These data were demultiplexed and aligned on-instrument to the hg38 human reference genome. Paired variant calling for matched tumor and white blood cell samples was performed with the UG-adapted DeepVariant algorithm, and subsequently filtered to retain only tumor-specific single nucleotide variants (SNVs). Plasma variant analyses at those tumor-specific SNV positions were performed to leverage the paired plus-minus sequencing (ppmSeq) approach and allowed for ≥Q60 base quality, equating to a theoretical 1x10 -6 error rate. ctDNA status and abundance was then assessed based on the level of the sample-specific machine learning model weighted and normalized signal compared to a reference population of noncancerous donor plasma samples (n=85). We assessed analytical specificity for 120 noncancerous donor plasma samples evaluated against clinical whole-genome somatic mutation profiles and demonstrated a specificity > 99.5%. Analytical sensitivity for ctDNA detection was assessed using five commercially available cell lines across ten levels between 1 - 500 parts per million (ppm) and demonstrated a 95% limit of detection < 5 ppm. Additionally, analytical concordance was evaluated in pre-surgical, treatment naïve plasma samples across a cohort of patients with bladder, breast, colon, and colorectal cancers. Taken together, these data support the significant potential for tumor-informed, non-bespoke MRD approaches for ctDNA detection across a broad range of solid tumor types and curative-intent clinical settings.