Abstract B057: Base-error model (BEM) for improved detection of ctDNA in lymph samples of HPV-negative head and neck cancer patients

Abstract

Abstract Introduction: Previously we demonstrated that ctDNA present in lymphatic exudate collected via surgical drains (“lymph”) outperformed plasma for detecting minimal residual disease (MRD) in head and neck squamous cell carcinoma (HNSCC) patients through a targeted sequencing (TS) approach. However, artifacts introduced during library preparation and sequencing remain challenges to sensitive low allelic fraction mutation detection. To enhance the performance of the TS approach, we characterized the background noise in lymph samples and developed a base-error model (BEM) that reduces sequencing artifacts and maximizes ctDNA signal. Methods: Lymph and blood were collected from 44 HPV-negative HNSCC patients postoperatively at 24 hours along with resected tumor. Cell-free DNA was extracted from lymph and sequenced using a custom TS panel with unique molecular identifiers (UMI). Somatic mutations were identified by tumor and blood exome sequencing (200x). Five patients had <2 mutations in tumor and were excluded. Two patients were censored due to lack of clinical data, yielding 17 patients with disease recurrence (REC) and 20 with no evidence of disease (NED) with >1 year of follow-up. Tumor-specific variants were force-called in lymph using a custom pipeline. A BEM of each tumor variant was built using a series of high-quality lymph reference samples to quantify the background noise. BEM was fit by Weibull distribution if more than 5 non-zero non-reference alleles were observed at a tumor variant position across the reference samples. Otherwise, Gaussian distribution was used to fit BEM. Force-called variants were considered artifacts if the variant allele fraction (VAF) was not greater than BEM cutoff controlled by false discovery rate. Patients were considered MRD positive if the mean VAF was greater than the estimated limit of detection. The Kaplan-Meier (KM) estimator with log-rank test and Cox proportional-hazards model were used for survival analyses. Results: Background noise was observed in reference lymph samples across all 12 nucleotide substitution classes. Out of 299 single nucleotide variants (SNVs), 94 were determined as artifacts by BEM. Among the artifacts, 77 were C/G to A/T mutations, indicating that the majority of recurrent background artifact in TS is likely the result of oxidative damage. The artifacts had VAFs ranging from 0.004% to 0.12% with median of 0.02%. Incorporation of BEM in lymph TS showed significantly improved specificity (sensitivity (SN) = 71%, specificity (SP) = 65%; p = 0.027, Hazard ratio (HR) = 3.07) compared to the same cohort without BEM (SN = 76%, SP = 30%, p = 0.60, HR = 1.34). Conclusions: BEM is a sensitive and robust method to improve ctDNA detection accuracy in TS. We demonstrated its utility in improving recurrence prediction in a HNSCC cohort. Given its advantages, we envision that BEM will prove useful as a general strategy for deep sequencing applications requiring precise digital quantification of low frequency alleles in TS. Citation Format: Zhuosheng Gu, Maciej Pacula, Seka Lazare, Noah Earland, Marra S. Francis, Adam Harmon, Megan Long, Aadel A. Chaudhuri, Jose P. Zevallos, Wendy Winckler. Base-error model (BEM) for improved detection of ctDNA in lymph samples of HPV-negative head and neck cancer patients [abstract]. In: Proceedings of the AACR Special Conference: Liquid Biopsy: From Discovery to Clinical Implementation; 2024 Nov 13-16; San Diego, CA. Philadelphia (PA): AACR; Clin Cancer Res 2024;30(21_Suppl):Abstract nr B057.