Systems pathology for building predictive models: The androgen receptor as a prototype biomarker in prostate cancer progression and targeted therapeutic response assessment

Abstract

4504 Background: A functional androgen receptor (AR) signaling axis plays a critical role in prostate cancer (PCA) development and progression across the clinical spectrum of the illness. Following diagnosis, most prostate cancers respond to treatments that block circulating androgen levels or block AR action. The measurement of AR levels in tumor tissue samples has the potential to provide prognostic information, to treatment selection, and a measure of the pharmacodynamic effect of a therapeutic agent(s) designed to reduce AR levels or block AR action. Existing methods to assess AR antigen levels in tissue are subjective, we have developed a systems pathology strategy for interrogating biomarker assessment in a predictive model by integrating clinical data with histological and quantitative antigen profiles. Methods: Tissue microarrays from 366 MSKCC patients were stained with H&E, images captured, analyzed and quantitative cellular features produced. Immunohistochemistry (IHC) was performed for AR and a staining index generated. A multiplex immunofluorescent (IF) assay using DAPI, CK18 and AR was performed on a subset of patients. IF mages were acquired and specific IF scripts were used to generate quantitative features of AR which were compared with AR IHC data. Results: Androgen Receptor levels by IHC in PCA demonstrated that a high-level of expression was associated with a greater risk of PSA-relapse within 5 years. (P < 0.0001; cut point 100). The correlation of AR IF with AR IHC established that all derived AR-IF measurements were statistically associated with the AR-IHC data. Furthermore, in a very preliminary model using machine learning and feature selection to predict PSA recurrence, 1AR-IF feature (epithelial and stromal AR) along with 2 clinical variables was selected with a concordance index of 0.80. Conclusion: AR levels in newly diagnosed localized prostate cancer are associated with clinical outcome. The levels can be assessed accurately and in a standardized manner using quantitative multiplex antigen methods. Such approaches are critical for evaluating biomarkers, especially when determining therapeutic response and clinical endpoints. [Table: see text]