Abstract 6202: Plasma cell-free DNA fragmentation patterns combined with tumor mutation detection in diagnosis of lung cancer

Abstract

Abstract Background: Lung cancer is one of the deadliest types of cancer in China. Its 5-year survival rates at early stages are significantly higher than advanced stages. LDCT has been adopted for lung cancer screening in high-risk individuals; however, debate regarding its accuracy is still ongoing. Mutation detection on cell free DNA (cfDNA) has traditionally been used to monitor DNA molecular changes derived from lung cancer cells in blood, while recently fragmentation pattern profiling of cfDNA has been shown as a promising alternative for early cancer detection. We aimed to combine mutation detection and fragmentation pattern analysis on cfDNA to develop a non-invasive assay to screen early lung cancer. Methods: Candidate DNA mutations were curated from literature and public databases, including COSMIC and TCGA. DNA fragmentation markers were collected from literature and whole genome sequencing (WGS) datasets. A panel of 407 primers covering selected mutation and fragmentation markers was developed to distinguish lung adenocarcinoma (ADC) plasma and normal plasma. We enrolled a total of 122 plasma samples (64 normal, 58 ADC) for this study. 49 normal samples and 44 ADC samples were used to construct a mutation-based classification model and a fragmentation-based classification model separately. The tuning parameters and features were determined by inner 4-fold cross validation. For the mutation-based model, baseline was set using normal samples in training set. Maximum allele frequency was calculated for each sample in test data (15 normal, 14 ADC), which was filtered by the background baseline. For the fragmentation-based model, we used the DELFI fragment score to construct fragmentation profiles, which was the ratio between short fragments (100-150bp) and long fragments (151-220bp). After optimization, the two models were integrated by Logistic Regression to create a combined model, which was validated by 4-fold nested cross validation. Results: ADC and normal plasma were sequenced by the aforementioned panel at an average depth of 2,000X to ensure the reliability of model construction and classification results. In classifying normal and ADC plasma, the mutation model alone is only modestly accurate as it produced an AUC of 0.69. But the fragmentation model demonstrated significantly higher accuracy, achieving AUC of 0.85. Furthermore, the combined model performed better than either model along, achieving an elevated AUC of 0.87. Conclusions: We demonstrate that DNA mutation and fragmentation pattern of cfDNA can classify lung cancer and normal plasmas separately, but fragmentation pattern are more accurate than mutation in this task. Combining the two models further improved prediction accuracy, suggesting they complement each other. Although this is a pilot study of limited cases, it demonstrated the potential of combining markers for accurate lung ADC detection in plasma. Citation Format: Zhoufeng Wang, Minjie Xu, Hua Chen, Kehui Xie, Minyang Su, Qiye He, Zhixi Su, Rui Liu, Weimin Li. Plasma cell-free DNA fragmentation patterns combined with tumor mutation detection in diagnosis of lung cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 6202.