All biomarker studies were prospectively planned. Whole-exome sequencing and transcriptome studies were performed on DNA from fresh-frozen tumor-biopsy samples obtained before treatment; germline whole-exome sequencing was performed on DNA from saliva samples. These studies were conducted at the University of Michigan as previously described.3 (link),17 (link) Targeted next-generation sequencing studies were conducted at the Institute of Cancer Research18 (link),19 ; libraries were constructed with the use of the GeneRead DNAseq Panel (Qiagen) and run on a MiSeq sequencer (Illumina). Copy-number data were validated by means of droplet digital polymerase-chain-reaction (PCR) testing with the use of the QX100 Droplet Digital PCR System (Bio-Rad).19 Circulating tumor-cell counts were performed with the use of CellSearch (Janssen Diagnostics).20 (link) For the purpose of correlating the results of next-generation sequencing with the response to treatment, patients were classified as positive or negative for genomic defects in DNA-repair genes. A patient was considered to be biomarker-positive if a homozygous deletion or deleterious mutation was identified in a gene reported to be involved either in DNA damage repair or sensitivity to PARP inhibition.21 (link)–23 (link) PTEN and ERG protein expression was determined by means of immunohistochemical assessment24 (link) (for details, see the Supplementary Appendix).