A Novel Method to Detect Early Colorectal Cancer Based on Chromosome Copy Number Variation in Plasma

• Main Authors: Jun-Feng Xu, Qian Kang, Xing-Yong Ma, Yuan-Ming Pan, Lang Yang, Peng Jin, Xin Wang, Chen-Guang Li, Xiao-Chen Chen, Chao Wu, Shao-Zhuo Jiao, Jian-Qiu Sheng
• Format: Article
• Language: English
• Published: Cell Physiol Biochem Press GmbH & Co KG 2018-02-01
• Series: Cellular Physiology and Biochemistry
• Subjects: Colorectal cancer; Copy Number Variation; Cell free DNA; Early detection; Support Vector Machine
• Online Access: https://www.karger.com/Article/FullText/487571
• ISSN: 1015-8987; 1421-9778

Background/Aims: Colonoscopy screening has been accepted broadly to evaluate the risk and incidence of colorectal cancer (CRC) during health examination in outpatients. However, the intrusiveness, complexity and discomfort of colonoscopy may limit its application and the compliance of patients. Thus, more reliable and convenient diagnostic methods are necessary for CRC screening. Genome instability, especially copy-number variation (CNV), is a hallmark of cancer and has been proved to have potential in clinical application. Methods: We determined the diagnostic potential of chromosomal CNV at the arm level by whole-genome sequencing of CRC plasma samples (n = 32) and healthy controls (n = 38). Arm level CNV was determined and the consistence of arm-level CNV between plasma and tissue was further analyzed. Two methods including regular z score and trained Support Vector Machine (SVM) classifier were applied for detection of colorectal cancer. Results: In plasma samples of CRC patients, the most frequent deletions were detected on chromosomes 6, 8p, 14q and 1p, and the most frequent amplifications occurred on chromosome 19, 5, 2, 9p and 20p. These arm-level alterations detected in plasma were also observed in tumor tissues. We showed that the specificity of regular z score analysis for the detection of colorectal cancer was 86.8% (33/38), whereas its sensitivity was only 56.3% (18/32). Applying a trained SVM classifier (n = 40 in trained group) as the standard to detect colorectal cancer relevance ratio in the test samples (n = 30), a sensitivity of 91.7% (11/12) and a specificity 88.9% (16/18) were finally reached. Furthermore, all five early CRC patients in stages I and II were successfully detected. Conclusion: Trained SVM classifier based on arm-level CNVs can be used as a promising method to screen early-stage CRC.