Abstract P5-01-18: A non-invasive diagnostic assay of early stage breast cancer

Abstract

Abstract Objective: To develop a non-invasive diagnostic assay for breast cancer. Methods: Breast cancer patients and non-cancerous donors were recruited. Cancer or benign non-cancerous tissues and peripheral blood were collected. The study was designed to have a training group and a validation group. The training group included cancer tissue and blood samples from 6 patients with breast cancer and 34 healthy donors, and the validation group including 40 patients with breast cancer and 11 with benign breast disease. Methylation capture sequencing and exome capture sequencing with an oncology panel were performed on the tissue samples. Based on these result, a liquid phase capture assay was designed to cover coding sequences of 340 genes as well as over 1000 common SNPs and more than 2000 differentially methylated sites in breast cancer tissues. NGS was performed with cfDNA from patients with breast cancer and healthy donors. Mutation abundance measurement based on population genetic statistics, methylation haplotype analysis based on machine learning model, and Gaussian decomposition based on spatial localization of nucleosomes were developed to calculate cancer specific genetic and epigenetic features. Results: We identified a major epigenetic landscape shift of breast cancer tissues from normal tissues or benign non-cancerous hyperplasia. Breast cancer tissues processed specific methylated haplotypes those absent in benign noncancerous or normal breast tissues. Prevalent nucleosomal shifts across the genome in breast cancer were identified, suggesting possible scenario of a global epigenetic switch. Using Gaussian mixture model and non-negative matrix factorization, we found a large number of breast cancer specific nucleosome spatial localization signals carried by cfDNA. Based on the uncovered epigenetic features, we trained a machine learning computational model to identify patients with breast cancer from those with benign disease and healthy donors. The model predicts nucleosomal fraction of malignancy for breast cancer originated signal. With the model, we uncovered nucleosomal spatial distribution in cfDNA associated with tumor burden and staging. In a Her2-positive early stage breast cancer patient, the tumor specific shift of nucleosome in cfDNA on a germline pathogenic mutation of ATR became unmeasurable after treatment with trastuzumab. The model archieved 100% accuracy (6/6 positive, and 34/34 negative) in the training group consisting 6 cancer pateints, 3 of which were patients with early stage (T1) breast cancer. In the validation group, the sensitivity, specificity and accuracy were 0.810 (33/40), 1.000 (11/11) and 0.862 (44/51). Conclusions: In this study, we developed a new non-invasive diagnostic assay for breast cancer based on decomposition of nucleosome spatial distribution in cfDNA sequencing. The model showed satisfactory diagnostic sensitivity and specificity in distinguishing early stage breast cancer patients from benign, non-cancerous donor. Further large cohort evaluation is necessary to confirm their potential in clinical applications. Key Words: Breast cancer; cfDNA; cancer specific nucleosome prediction; machine learning Citation Format: Wanyan Tang, Xin Zhou, Haiwei Zhang, Chuang Ge, Huiqing Yu, Weiqi Nian, Yi Zhang. A non-invasive diagnostic assay of early stage breast cancer [abstract]. In: Proceedings of the 2019 San Antonio Breast Cancer Symposium; 2019 Dec 10-14; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2020;80(4 Suppl):Abstract nr P5-01-18.