Identification and characterization of DNA repair deficiencies with circulating tumor DNA in prostate cancer

DNA damage repair pathways are essential to maintaining genomic integrity. Somatic hypermutation resulting from DNA mismatch repair deficiency (MMRd) is rare and under-characterized in metastatic prostate cancer (mPCa). However, MMRd can result in cancer sensitivity to immune checkpoint inhibitors. Thus, practical methods for MMRd detection are urgently needed. Plasma cell-free DNA (cfDNA) offers a clinically-practical and minimally-invasive tool for genomic characterization. This Ph.D. thesis aims to: (1) demonstrate that cfDNA sequencing is a viable approach to identify MMRd, (2) to characterize the prevalence, genomic landscape, and clinical course of MMRd in mPCa, and (3) develop novel methods to improve molecular stratification of multiple classes of clinically-actionable DNA damage repair deficiencies in mPCa. To assess MMRd prevalence, I analyzed cfDNA targeted sequencing data from 433 mPCa patients. Samples with somatic hypermutation were subjected to whole exome sequencing (WES), mismatch repair gene intron sequencing, immunohistochemistry and sequencing of archival tissue. 3.7% had hypermutation with MMRd etiology, including deleterious alterations in MSH2, MSH6, or MLH1, microsatellite instability, and MMR protein loss. These patients had poor responses to AR inhibition and inferior survival compared to a control cohort. Within the MMRd group, inactivation of tumor suppressors such as PTEN, RB1, and TP53 occurred more frequently through mutation rather than copy number loss, compared with MMR-intact mPCa. Activating hotspot mutations in oncogenes such as AKT1, PIK3CA, CTNNB1 and the AR were also more frequent. High intra-patient clonal diversity was observed, evidenced by subclonal driver mutations and shifts in mutation allele frequency over time. Finally, I developed a machine learning approach called DARC Sign, that used somatic signature-based features from WES rather than gene-specific features, to identify functional deficiencies in MMR, BRCA2 or CDK12. DARC Sign produced AOCs of 0.99, 0.99 and 1.00 for identifying BRCA2d, CDK12d, and MMRd respectively, and outperformed current classifiers for WES data. The mPCa trained model also identified homozygous BRCA2 loss in bladder cancer. Together, this thesis demonstrates that cfDNA is a viable alternative to tissue biopsies for identifying and characterizing mPCa DNA damage repair status.Science, Faculty ofGraduat