Abstract 4920: A nanopore sequencing approach characterizes cell-free DNA methylation-fragmentomics profiles indicative of breast cancer in a large cohort

Abstract The detection of breast cancer from cell-free DNA (cfDNA) offers a new approach with the potential to benefit individual screening and early detection. However, identifying breast cancer from cfDNA remains a major challenge when one uses conventional short-read sequencing. Methylation and DNA fragmentation characteristics provide highly informative biomarkers for detecting and monitoring cancer. To characterize these genomic features from cfDNA, we developed an approach using single molecule Nanopore sequencing. Single cfDNA molecules are sequenced intact with the methylation and fragment size directly extrapolated from Nanopore data. Moreover, our approach requires small amounts of cfDNA to discern specific genomic features from individual molecules. Thus, the use of Nanopore sequencing provides both methylation and fragment size as directly measured from native cfDNA without the use of any chemical processing or other molecular manipulation. We studied an extended breast cancer cohort with in total 1080 of samples, consisting of 440 untreated breast cancer patients and 640 healthy controls. Nanopore sequencing analysis of these cfDNA samples identified a range of 1 to 4 million CpG sites from 1 ml of plasma. Leveraging the cohort, we developed an ensemble classifier for breast cancer detection based on cfDNA methylation and fragmentation. The classifier leveraged a leave-one-out cross-validation (LOOCV) to generate predicted probability. We created nucleosome-specific methylation scores as the input features to the classifier, which scores quantified the similarity of each sample's methylation profile to a training set on a selected set of CpG sites associated with cancers. Furthermore, we also identified DNA fragmentation features, revealing distinct mono-nucleosome and di-nucleosome patterns that distinguish cancers from healthy controls. Aggregating the DNA nucleosome-specific scores and fragmentation features, we built an ensemble LOOCV random forest model. We applied the model to the breast cancer cohort and achieved a detection specificity of 74%, a sensitivity of 71%, and an overall accuracy of approximately 73%. The area under the receiver operating characteristic curve (AUROC) reached 80%. We validated the classifier with two external independent cohorts. Validation cohort 1 demonstrated as a specificity of 85%, a sensitivity of 80%, and overall accuracy of 83% with AUROC of 90%. Validation cohort 2 without controls demonstrated that the classifier achieved a sensitivity of 94%. Both validation sets confirm that our model can achieve excellent performance for early breast cancer detection. In summary, our findings demonstrate that nanopore sequencing provides cfDNA methylation and fragment profiling that enables breast cancer detection. This approach holds the potential to improve cancer diagnosis and treatment outcomes. Citation Format: Xiangqi Bai, Billy T. Lau, Alison Almada, Sue Grimes, Tianqi Chen, Hojoon Lee, Hanlee P. Ji. A nanopore sequencing approach characterizes cell-free DNA methylation-fragmentomics profiles indicative of breast cancer in a large cohort abstract. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts) ; 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84 (6Suppl): Abstract nr 4920.