Discovery of targetable mutational signatures in advanced prostate cancer (aPC) using machine learning and next-generation sequencing (NGS) of circulating tumor DNA (ctDNA).

Abstract

226 Background: Alterations (alts) of the AR gene are common in aPC and are associated with reduced responses to AR-axis inhibitors; optimal treatment is unknown. Genomic sequencing of ctDNA from aPCs may recapitulate truncal and branching alts across metastatic sites in heavily pretreated progressive aPC and facilitates identification of AR and other alts that contribute to aPC progression. Such data could lead to better understanding of aPC molecular biology and guide development of novel treatments. Methods: Two unsupervised machine learning algorithms, hierarchical clustering and principal components analysis, were used to evaluate mutational profiles of 2,679 plasma samples from 2,309 men with aPC. Samples were assessed by a validated ctDNA NGS panel that sequences 73 clinically relevant cancer genes (Guardant360, Redwood City, CA). Indels, amplifications, and fusions are evaluated in a subset of genes with high sensitivity and specificity. Chi-squared residual analysis and gene ontology enrichment were used to identify significantly co-altered molecular pathways. Results: Distinct, non-overlapping mutational signatures were observed in tumors with only AR amplification (amp) vs. only AR mutation (mut). Tumors with only AR amp, and those with both AR amp and mut, were strongly enriched for amps in the MAPK pathway (P = 3.62E-10, 4.92E-05, resp.). Tumors with only AR mut, and those with no AR alts, were strongly enriched for muts in negative regulators of cell proliferation (P = 1.09E-05, 3.30E-09, resp.). Muts further stratified to oxidative stress response in AR mut (P = 4.37E-05), and cell aging in no AR alts (P = 8.31E-07). Conclusions: In men with aPC, machine learning of ctDNA NGS data revealed recurrent mutational signatures that stratified according to the type of AR alts. These hypothesis-generating data require external validation but suggest that aPC may progress due to distinct mechanisms dependent on AR status. Future efforts to develop novel therapeutic approaches for aPC may need to consider the impact of the type of AR alt present and explore the interplay of AR with other altered pathways identified in this analysis.