A novel dual-strand whole-genome sequencing (WGS) method for ultra-sensitive molecular residual disease detection (MRD)

Whole genome sequencing (WGS) of circulating cell-free DNA (cfDNA) enables ultra-sensitive circulating tumor DNA (ctDNA) detection by leveraging genome-wide mutations. However, the detection sensitivity remains limited by errors. Here we describe the performance of a new, high-Q WGS library prep method, which enables >Q60 cfDNA sequencing, and demonstrate its application in molecular residual disease (MRD) testing. The method enables increased accuracy by sequencing both top and bottom strands of double stranded cfDNA fragments without requiring unique molecular identifiers (UMIs). Briefly, oligonucleotide adaptors are attached to cfDNA samples and are subsequently processed to generate a library structure that contains both top and bottom strands. Samples are appended with barcodes and pooled for sequencing on a standard NovaSeq X TM instrument. Errors originating from DNA damage, library preparation or sequencing are informatically detected and masked, creating duplex consensus reads. A novel q-score prediction algorithm is employed to predict the probability a reported base is an error, as well as substitution-specific error probabilities. Empirical q-score was measured using HG002 cell line DNA. To measure performance in tumor informed MRD application, cfDNA samples extracted from healthy donors were sequenced and analyzed against patient-specific somatic variants derived from tissue WGS using the Illumina Oncology WGS prep. Limit of detection (LoD) was assessed using 5 replicates at each of the 7 target VAF levels from 0-50 parts-per-million (ppm) dilutions of the SeraCare MRD reference material, with analysis by the DRAGEN TM MRD pipeline.Libraries prepared from 10 ng cfDNA delivered ~50x deduplicated duplex consensus coverage with >80% of passed-filter clusters delivering high quality duplex consensus reads. The quality of the resulting base calls was empirically measured at >Q60 accuracy, and certain substitution-specific error rates were measured at >Q70. Healthy plasma was used to establish statistical scores for determining ctDNA presence at different analytical specificity goals. These values were subsequently used to determine the detection in the LoD study. In aggregate, the results showed the method can achieve a LoD95 of 2.5 ppm with analytical specificity of 99.8%. The method presented herein demonstrates the potential of highly sensitive and specific ctDNA detection for MRD testing.