Abstract
Background: Whole-genome sequencing (WGS) of cell-free DNA (cfDNA) paired with methods like ichorCNA and DELFI has become a valuable approach for detecting circulating tumor DNA (ctDNA). However, the mixture of ctDNA with non-tumor cfDNA in plasma complicates accurate validation of ctDNA detection methods. In contrast, aqueous humor (AH) from treatment-free retinoblastoma (RB) eyes is uniquely enriched in ctDNA, providing an optimal source for generating reference samples. This study leverages AH-derived biological ctDNA to evaluate two established ctDNA analysis methods. Methods: Biologically derived cfDNA mixtures were created by combining WGS reads from two AH samples with pooled reads from non-tumor blood samples (N=20). One AH sample (RB_023_OS) showed >99% tumor fraction (TFx) with ichorCNA, while the other (RB_025_OD) had confirmed presence of ctDNA based on >99% variant allele frequency (VAF) for an RB1 variant, despite being somatic copy number alteration (SCNA)-negative. Mixtures were prepared across a range of TFx values (0.01-0.5) and unique sequencing coverages (0.1x, 0.5x, 1x, and 2x). TFx was calculated using ichorCNA with standard parameters and optimized parameters for low TFx levels. We also tested an enrichment strategy targeting shorter ctDNA fragments (≤150 base pairs) and generated genome-wide fragmentation profiles using the DELFI method to assess short-to-long (S:L) fragment ratios across 5-Mb bins. Results: We evaluated the impact of sequencing coverage, parameter selection, and fragment size selection on TFx estimation by ichorCNA. Standard ichorCNA parameters demonstrated limited sensitivity, detecting ctDNA only in samples with TFx of 5-6% or higher. Increasing sequencing coverage did not substantially improve the limit of detection (LoD) or the accuracy of TFx quantification. Optimized parameters reduced the LoD to 3% TFx, but only in samples with ≥0.5x coverage, indicating a dependence on sequencing depth for accurate ctDNA detection and quantification. In RB_025_OD, where no SCNAs were present, ichorCNA failed to estimate the TFx despite an RB1 VAF exceeding 99%, highlighting the platform's reliance on SCNAs for accurate TFx detection. Short fragment enrichment (≤150 base pairs) significantly increased TFx values across all samples but could be mapped back to the original TFx values accurately using polynomial curve fitting (R-squared=0.97). Genome- wide fragmentation profiles showed increasing deviation from non-tumor controls as TFx increased, with SCNA-positive samples exhibiting greater deviations than SCNA-negative samples. Conclusions: These findings demonstrate the dependence on sequencing coverage and the presence of SCNAs by current ctDNA analysis methods, highlighting their limitations for accurately detecting ctDNA in challenging scenarios.
Citation Format: Nikki Higa, Peter Kuhn, Jesse L Berry, Liya Xu. Assessment of ctDNA analysis methods using high purity ctDNA from the aqueous humor of retinoblastoma eyes [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 B059.