Development of a Horizontal Data Integration Classifier for Noninvasive Early Diagnosis of Breast Cancer

Overview

This is a translational no-profit study. Our proposal aims at creating a noninvasive Horizontal Data Integration (HDI) classifier for early diagnosis of breast cancer, with the final goal of avoiding in most cases useless biopsies of suspect cases encountered during radiological screening. Women with radiologically identified lesions, BIRADS-3/4/5, smaller than 2 cm by radiological assessment (i.e., radiological T1), will be enrolled and invited to donate peripheral blood samples (35 ml) and urine samples (50 ml). Radiological images as well as demographic and anatomopathological data will be collected. Objective of this project is to develop a HDI classifier enabling early noninvasive diagnosis of breast cancer with similar accuracy compared to breast biopsies. Such classifier will be developed based on the correlation between the molecular profile of peripheral blood (ctDNA, proteins, exosomes) and urine (ctDNA) collected at T0 (baseline, before diagnostic biopsy) and bioptic diagnosis. The assessment of the profile of peripheral blood (ctDNA, proteins, exosomes) and urine (ctDNA) at two time points for diagnosed pT1 breast cancers (T0: baseline, before biopsy; T1: after diagnosis of pT1 breast cancer) will allow us to distinguish between tumor- and host-specific molecular alterations in connection with the presence/absence of breast cancer.

Background: Currently, early diagnosis of invasive breast cancer relies on the combined use of mammogram and ultrasound. These approaches are still suboptimal in terms of accuracy, and confirmation biopsy or recall tests are needed in case of radiological suspect. Recently, the study of noninvasive biomarkers in cancer has received enormous interest, fostered by the advancement of technologies and the potential for early detection of malignancies. However, no study has so far tried to apply the simultaneous assessment of biologically different analytes and data-characterization algorithms (radiomics approaches) to increase the accuracy of early breast cancer diagnosis. Hypothesis: Multiple biological analytes must be combined with the refinement of radiomics algorithms to overcome the current limitations of early breast cancer diagnosis. The overall goal of the project is to develop a horizontal data integration (HDI) classifier enabling early noninvasive diagnosis of invasive breast cancer with high accuracy. Objectives: Aim 1: To test the performance for the diagnosis of small invasive breast cancers of a) ultrasensitive next-generation sequencing on circulating tumor DNA (ctDNA); b) aptamer-base proteomics arrays on plasmatic proteins; c) radiomics machine-learning algorithms. Aim 2: To develop an HDI classifier based on the aforementioned methods with the aim of reducing the needs for invasive procedures in early breast cancer diagnosis. Aim 3: To improve the performance of the HDI classifier by integrating other potentially transformative methods of noninvasive diagnosis. Experimental Design: Peripheral blood samples and urine samples will be collected from a prospective cohort of 750 patients with radiologically suspect small breast lesions undergoing diagnostic biopsy at the Diagnostics Senology Unit of San Martino Hospital. Ultrasensitive Next Generation Sequencing (NGS) on plasma ctDNA will be performed using a custom tagged-amplicon panel designed by us on a cohort of 3,269 sequenced breast cancer cases from the GENIE initiative. We also will be applied a new protocol termed cell-free methylated DNA immunoprecipitation and high-throughput sequencing (cfMeDIP-seq) in collaboration with Dana Farber Cancer Institute, Boston for methylome analysis of small quantities of ctDNA from plasma and urine. Potential cancer-related plasma proteins will be analyzed using SomaScan aptamer-base protein arrays in collaboration with the Sidra Medical Center, Doha, Qatar. A radiomics classifier developed by the Senology team on an exploratory subgroup of the ASTOUND trial, sponsored by the University of Genoa, will be trained and tested on the same cohort. Other noninvasive diagnostics methods will be assessed as well. An HDI classifier will be generated on ctDNA, proteomics, and radiomics results, using advanced machine learning methods. Our HDI classifier will finally be integrated as needed with other predictors and validated on our cohort. Expected Results: 1. Assessment of the performance of cutting-edge noninvasive methodologies in the context of early breast cancer diagnosis. 2. Development of a noninvasive HDI classifier for early breast cancer. 3. Novel biological insights on small breast cancers. Impact On Cancer: 1. Increase in early breast diagnosis accuracy over current methods. 2. Reduction in the need for recall and invasive tests in breast cancer diagnosis. 3. Long-term impact on breast cancer mortality.