Abstract 4448: Machine-learning assisted histopathology (HistoMAP) links nuclear membrane instability to disease progression in prostate cancer

Abstract

Abstract Introduction: The manifestation of cancer malignancy occurs at the cellular level where individual cells escape tissue confinements and disseminate. Identifying this malignant behavior at a cellular level is necessary in order to deconvolve the tumor heterogeneity, identify disease progression and predict cancer-related morbidity and mortality. To achieve that goal, we established a unique computer-assisted single-cell histopathology analysis of prostate tissue that evaluates nuclear-membrane instability and the associated extracellular vesicle (EV) biogenesis to identify malignant potential and assess future disease progression. In recent years cancer EVs have been identified as important mediators of intercellular communication. The potential for using EVs as a liquid biopsy has promoted research on profiling EVs in biofluids. However, the direct analysis of EV biogenesis in tumor tissue has been largely omitted, even thought identifying such cells is likely to be a specific and sensitive measure of disease malignancy. We have previously demonstrated that highly migratory and metastatic cancer cells shed atypically large EVs, known as large oncosomes (LO, Di Vizio et. al., 2012). LO play distinct functions and contain a specific repertoire of molecules that can be used for detection of tumor-derived cargo in plasma (Minciacchi et. al., 2015). The recent discovery that the biogenesis of large oncosomes (LO) in prostate cancer is associated with nuclear membrane instability (Reis-Sobreiro et. al., 2018) offered an opportunity to investigate such biogenesis in patient specimens. We have developed a machine-learning assisted histopathology (HistoMAP) that quantitatively identifies nuclear-membrane instability in prostate cancer tissue and demonstrate a direct correlation between this molecular biogenesis of vesicles and biochemical recurrence after prostatectomy. Experimental procedures: We developed a multiplex immunofluorescent technique to detect and quantify nuclear-derived LO in formalin fixed paraffin embedded prostatectomy tissues. Using computer-assisted image segmentation and machine-learning we distinguished these LO from all other intracellular compartments and assess malignancy in a cohort of Vanderbilt prostate cancer patients. Results: LO production was significantly elevated in prostate cancer in comparison to benign tissue and particularly evident in lymph node metastases. Moreover, LO production was associated with the risk of future metastatic disease which re-enforced its relevance in cancer progression. Conclusions: Our findings demonstrate that, given knowledge of the EV machinery, EV biogenesis can be detected in patient tissues and provide critical information related to patient disease status and assist with the prediction of future clinical performance. Citation Format: Andries Zijlstra, Tatiana Novitskaya, Dolores Di Vizio, Mariana Reis- Sobreiro, Michael Freeman. Machine-learning assisted histopathology (HistoMAP) links nuclear membrane instability to disease progression in prostate cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 4448.