Enhanced Detection of Genitourinary Cancers Using Fragmentation and Copy Number Profiles Obtained from Urinary Cell-Free DNA

• Published in: Clinical chemistry (Baltimore, Md.) Md.), 2021-02, Vol.67 (2), p.394-403
• Main Authors: Han, Yang, Li, Xinxin, Zhang, Mingxin, Yang, Yang, Ge, Guangzhe, Wang, Kunxiang, Gong, Yanqing, Liang, Yuan, Niu, Haitao, Ci, Weimin
• Format: Article
• Language: English
• Subjects: Accuracy; Algorithms; Analysis; Cancer; Classification; Classifiers; Copy number; Copy number variations; Datasets; Deoxyribonucleic acid; Diagnosis; DNA; Fragmentation; Fragments; Gene sequencing; Genomes; Genomics; Integration; Patients; Sensitivity; Tumors; Whole genome sequencing
• ISSN: 0009-9147; 1530-8561
• DOI: 10.1093/clinchem/hvaa283

Abstract Background Recent studies have reported that examining the fragmentation profiles (FP) of plasma cell-free DNA (cfDNA) further improves the clinical sensitivity of tumor detection. We hypothesized that considering the differences of the FP of urinary cfDNA would increase the clinical sensitivity of genitourinary (GU) cancer detection. Methods 177 patients with GU cancer and 94 individuals without tumors were enrolled in the discovery cohort. An independent validation dataset comprising 30 patients without tumors and 66 patients with GU cancer was also collected. We constructed an ensemble classifier, GUIDER, to detect and localize GU cancers using fragmentation and copy number profiles obtained from shallow whole-genome sequencing of urinary cfDNA. Results Urinary cfDNA of patients with GU cancer had a higher proportion of long fragments (209–280 bp) and a lower proportion of short fragments (140–208 bp) compared to controls. The overall mean classification accuracy of the FP was 74.62%–85.39% for different algorithms, and integration of the FP and copy number alteration (CNA) features further enhanced the classification of samples from patients with GU cancer. The mean diagnostic accuracy was further improved by the ensemble classifier GUIDER, which integrated the FP and CNA profiles and resulted in a higher mean accuracy (87.52%) compared to the analysis performed without FP features (74.62%). GUIDER performed well in an independent validation dataset. Conclusions The lengthening and shortening of urinary cfDNA within specific size ranges were identified in patients with GU cancer. Integration of the FP should further enhance the ability to use urinary cfDNA as a molecular diagnostic tool.