The Circulating Cell-free Genome Atlas (CCGA) Study: Size selection of cell-free DNA (cfDNA) fragments.

Abstract

3103 Background: Detection of somatic copy number aberrations in individuals with cancer via cfDNA whole-genome sequencing (WGS) is challenging at low tumor fractions. Given that tumor-derived cfDNA fragments are shorter than those from healthy tissues, this exploratory analysis evaluated the potential effect of size selection on the ability to detect cancer. Methods: CCGA WGS libraries were in silico and in vitro size selected to estimate the change in tumor fraction by tumor types (breast, lung, and colorectal [CRC]) and stage (I-III vs IV). In silico analyses used clinically evaluable training set samples with WGS assay results (n = 1422: 560 non-cancer [NC], 862 cancer [C] stages I-IV); classification (cancer/non-cancer) performance was estimated using fragments within the 90-150 bp range. In vitro analyses used a subset of samples (n = 93: 28 NC, 65 C stages I-IV), including C cases sampled within a range of tumor fractions; tumor fraction was also measured at each progressive removal of maximum-length fragments (intervals of 10 bp: 150 bp down to 50 bp). Results: In silico and in vitro analyses, respectively, resulted in median 2.00±0.58-fold (at 6.91±2.64X depth) and 2.00±0.52-fold (at 23±4.45X depth) increases, in overall tumor fraction (compared to non-size-selected 36X depth). This was consistent across tumor types ( in silico: 1.78±0.73 breast, 2.00±0.58 CRC, 2.00±0.41 lung; in vitro: 2.00±0.82 breast, 2.51±0.52 CRC, 2.53±0.94 lung) and stages ( in silico: 2.00±0.74 I-III, 1.78±0.52 IV; in vitro: 2.00±0.55 I-III, 1.68±0.29 IV). Tumor fraction increased with initial fragment length titrations, but not following size selection to shorter lengths ( < 140 bp). Classifier trained on in silico size-selected data had increased sensitivity at 98% specificity compared to those trained on non-size-selected data (p < 1e-5). Conclusions: In silico and in vitro size selection consistently increased tumor fraction across cancer types and stages, and this increase was maximized by tuning the length range of size selection. Relative to full-depth data, classification performance improved significantly. These data suggest that size selection targeting cfDNA under 140 bp may enhance cfDNA-based cancer detection. Clinical trial information: NCT02889978.