Abstract 418: Highly efficient duplex DNA tagging strategy improves accuracy of detecting ultra-low-frequency mutations through consensus read reconstruction

Abstract

Abstract Introduction: Molecular diagnostics and precise personalized care continue to increase the sensitivity and specificity requirements for detecting low (~5%) to ultra-low (<1%) frequency actionable mutations by next generation sequencing (NGS). These low-frequency variants often occur below the reliable limit of detection of standard NGS as they are confounded by errors introduced during the NGS workflow. We have developed adapters containing unique molecular identifiers (UMIs) that permit tagging of double-stranded DNA and statistical reconstruction of reads sequenced as duplicates. These novel adapters are compatible with standard library preparation and enrichment methods and reagents, yet provide significantly enhanced error correction. Methods: Libraries were prepared from the 3 most commonly used oncology sample types: genomic, FFPE, or plasma-derived, cell-free DNA, using a standard, commercially available kit combined with standard adapters or the novel duplex UMI adapters. The libraries were enriched with a custom xGen Lockdown Panel targeting a 75 kb polymorphic region, and then deep sequencing and variant calling were performed. A consensus read building tool was developed to collapse PCR duplicates based on molecular barcodes, and the tool was used to evaluate the utility of UMIs in error correction. Results: Compared to standard adapters, the duplex adapters presented comparable or better library yield and mean deduplicated sequencing depth for all sample types and input masses tested. To evaluate variant-calling accuracy, we established mixtures of DNA of known SNP genotype to mimic ultra-low-frequency variant samples. For genomic DNA samples, DNA from NA12878 and NA24385 were mixed to generate minor allele frequency (MAF) down to 0.1%, and variant calls were evaluated against annotations in Genome in a Bottle. Commercial FFPE and cell-free DNA samples were genotyped and mixed to present MAF at 0.5%. When standard adapters were used, variants present at the MAF were detected with 90% sensitivity for all sample types, but only under conditions that also called >4,000 false positives, resulting in a positive predictive value (PPV) of <3%. Using the duplex UMI adapters in conjunction with consensus read construction, >90% detection sensitivity was achieved for genomic and cell-free DNA samples with 0 false positives, resulting in 100% PPV, a >40-fold error suppression. For FFPE DNA samples, the duplex adapters provided >30-fold PPV improvement compared to standard adapters. Conclusions: The duplex UMI adapters eliminate background NGS errors by collapsing duplicated reads using their unique tags. This leads to unprecedented accuracy in detecting true low-frequency variants, regardless of the input DNA source. Such advances are key to refining diagnostics and improving precision cancer care. Citation Format: Jiashi Wang, Kevin Lai, Madelyn Light, Layla Katiraee, Kristina Giorda, Mirna Jarosz, Yun Bao, Criss Walworth, David Kupec, Caifu Chen. Highly efficient duplex DNA tagging strategy improves accuracy of detecting ultra-low-frequency mutations through consensus read reconstruction [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2018; 2018 Apr 14-18; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2018;78(13 Suppl):Abstract nr 418.