(a) Hierarchical clustering with serial plasma samples demonstrating that samples derived from the same patient tend to cluster together (Clustering-Manhattan average). (b) Genome-wide log2-ratio plots of plasma samples from P40 obtained at a castration-sensitive stage (upper panel) and 10 months later after development of CRPC. The inset illustrates enlarged log2-ratio plots of the X chromosome, the bottom sample shows gain of chromosome X material with the highest copy-number gain on Xq12, the region that harbours the AR gene. In this and in the subsequent panels, the grey arrows indicate the time intervals between the sample collections. Copy-number gains are depicted in red and copy-number losses in blue. (c) P106’s tumour genome developed AR amplification within 12 months and an additional amplicon at Xq23-q24 within the subsequent 6 weeks. (d) The first plasma sample of P147 had two high-level amplifications, that is, on chromosomes Xq12 (AR) and 5q14.3 (EDIL3), which had not been observed in the primary tumour. Within the next 4 months, a further amplicon evolved on chromosome 10q11.21 (RET). The quality of the analysis of the primary tumour was not optimal due to the fixation conditions of the tissue. The ‘peak’ on chromosome 10 in the primary tumour does not involve the RET region and is instead most likely an artefact. (e) The plasma sample from prostate cancer patient P112 displayed an amplicon on 1q21.3 including SETDB1, which had not been present in the analysed part of the primary tumour.