A Novel Prostate Cell Type-Specific Gene Signature to Interrogate Prostate Tumor Differentiation Status and Monitor Therapeutic Response

Abstract

In this study, we extracted prostate cell-specific gene sets (metagenes) to define the epithelial differentiation status of prostate cancers and, using a deconvolution-based strategy, interrogated thousands of primary and metastatic tumors in public gene profiling datasets. We identified a subgroup of primary prostate tumors with low luminal epithelial enrichment (LumElow). LumElow tumors were associated with higher Gleason score and mutational burden, reduced relapse-free and overall survival, and were more likely to progress to castration-resistant prostate cancer (CRPC). Using discriminant function analysis, we generate a predictive 10-gene classifier for clinical implementation. This mini-classifier predicted with high accuracy the luminal status in both primary tumors and CRPCs. Immunohistochemistry for COL4A1, a low-luminal marker, sustained the association of attenuated luminal phenotype with metastatic disease. We found also an association of LumE score with tumor phenotype in genetically engineered mouse models (GEMMs) of prostate cancer. Notably, the metagene approach led to the discovery of drugs that could revert the low luminal status in prostate cell lines and mouse models. This study describes a novel tool to dissect the intrinsic heterogeneity of prostate tumors and provide predictive information on clinical outcome and treatment response in experimental and clinical samples.

Circulating extracellular vesicles release oncogenic miR-424 in experimental models and patients with aggressive prostate cancer

“…broadinstitute.org/) and 75 CRPCs (Fred Hutchinson Cancer Research Center) was used. Sequencing reads were aligned as previously described 28 . Heat map plots were generated in R environment.…”

A Tale of Two Cancers: A Current Concise Overview of Breast and Prostate Cancer

“…Highly differentiated luminal secretory cells and a small fraction of basal cells can be found in the prostate gland epithelium and, therefore, adenocarcinomas may arise from both luminal and basal tumor progenitor cells [ 94 ]. Although the origins of human PC cell type is controversial, frequently diagnosed prostate tumors tend to have a luminal phenotype [ 94 ]. Analogous to BC, luminal and basal subtypes of PC have been explained using a slightly changed PAM50 algorithm [ 33 ].…”

Derivation and Application of Molecular Signatures to Prostate Cancer: Opportunities and Challenges

“…This often entails the integration ChIP-seq and gene expression datasets to identify true regulatory (TREG) TF–gene interactions in terms of interaction probability aiding in the generation of complex disease transcriptional profiles [ 49 ]. Prostate cell-specific signatures were recently generated through a discriminant function analysis, accompanying the identification of a subset of primary PCa tumors with a low luminal epithelial enrichment and a high mutational burden to assess the degree of differentiation of luminal and basal epithelial cells in primary and metastatic PCa tissue samples [ 50 ].…”

“…The main cluster represents analytics performed directly to dissect data for the purpose of signature generation. These take data from epigenetic, transcriptomic, Assays for Transposase-Accessible Chromatin using sequencing (ATAC), and genomic data as inputs and include calculating interaction probability scores between TFs and targets [ 49 ]; correlating transcriptional signatures of multiple driver pathways [ 27 ]; cross validating gene expression signatures using leave one out cross validation (LOOCV) and principal component analysis (PCA) plots [ 65 ]; assessing tumoral degree of differentiation through discriminant function analysis [ 50 ]; combining genome-wide methylation with microarray data to generate polycomb signatures [ 59 ]; and utilizing artificial neural networks, Bayesian networks, support vector machines and decision trees to combine existing data and generate cancer prediction models [ 66 ]. Moreover, they use mixed integer linear programming (MILP) to discern downstream cumulative effects of TFs [ 67 ] and incorporate nonnegative matrix factorization (NMF), latent Process Decomposition (LPD), and gene set enrichment (GSEA) analyses to combine signatures to attempt to generate combination signatures involving multiple biochemical pathways [ 43 ].…”