Abstract
Abstract Purpose: Although improvements in mammographic screening, including digital mammography and 3D tomosynthesis, have occurred in recent years, the positive predictive value (PPV) of mammography remains low. Screening mammography has an average PPV for biopsy of 32.6 % (range 22.2-54). Furthermore,for women with abnormally dense breasts, mammography may be hard to interpret or may provide a false negative. Here, we report a blood diagnostic for women with breast lesions that is capable of detecting and discriminating epigenetic biomarkers associated with either invasive or more benign disease. Our approach is based on the premise that circulating lymphocytes undergo epigenetic changes upon exposure to the tumor microenvironment and these changes can be followed to track tumor progression from a non-invasive to an invasive state. We hypothesized that DCIS samples with high risk for progression would maintain a predominant epigenetic signature that would match those observed in invasive disease, while those DCIS samples considered to have low risk or benign disease would resemble controls. To test the utility of these epigenetic biomarkers, we employed a novel diagnostic model which uses a combined machine learning approach integrated with the recombinatorial strategy of a genetic algorithm, to screen a large response space of thousands of CpG sites to identify a small set of CpG sites (< 20) with the highest combined predictive power. Methods: Blood was obtained from women with histologically confirmed DCIS (n=14), or invasive ductal carcinoma (n=10) and women with no evidence of breast lesions on mammography (n=10). PBMCs were isolated using a modified Ficoll-Paque gradient, and DNA was extracted using standard commercial kits. Epigenetic profiling of DNA was performed using a highly sensitive and quantitative analytics platform, which utilizes methylation sensitive restriction endonucleases to detect changes in methylation of CpG sites from standard NGS data. Risk of invasiveness potential was determined based on pathologic criteria. Samples were scored as low risk (Cribroform subtype, no necrosis, low mitotic index) or high risk (Comedo or solid subtype, high mitotic index, necrosis). Results: Non-metric multidimensional scaling ordination analysis of the CpG sites revealed highly distinct methylation patterns between Normal, DCIS and invasive samples. Using a Likelihood Ratio Test with defined ANOVA contrasts, over 14,000 significantly different methylated CpG sites were identified (p<0.05 after false discovery rate correction). A proprietary machine learning diagnostic model was employed to reduce this high-dimensional variable space to the most effective set of CpG sites. DCIS and invasive samples were discriminated in blinded tests (n=17) with 69% accuracy. Prognostic ability was determined in a blinded study of 10 DCIS samples in which we obtained an accuracy of 90% for risk assessment based on our pathological criteria. Conclusions: Preliminary studies show strong ability of the identified DNA methylation metrics to detect and discriminate invasive and non-invasive breast lesions. The quantitative sensitivity and selectivity of this new diagnostic/prognostic blood test will be determined in larger patient cohorts. Citation Format: Sims-Mourtada J, Arnold KM, Marsh A. A blood based diagnostic/prognostic for early stage breast cancer [abstract]. In: Proceedings of the 2018 San Antonio Breast Cancer Symposium; 2018 Dec 4-8; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2019;79(4 Suppl):Abstract nr P4-01-22.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have