*3.2. WGS Variant Calling*

The GATK pipeline version 3.5-0 was used for small variant calling [107]. We defined small variants as single nucleotide variants (SNVs) and insertions or deletions (indels) ≤ 50 bases and structural variants (SVs) as events ≥ 50 bases. Analysis-ready alignment per sample was called for SNVs and indels (GVCF mode) using GATK HaplotypeCaller (GVCF mode; [107]). Per-sample GVCFs were used for joint genotyping across genomes (GATK GenotypeGVCFs). Joint-called SNVs and indels were filtered via machine learning variant quality score recalibration and passed loci were kept. High-confidence somatic variants were called for each tumour-blood pair using MuTect2 [108]. A combination of GRIDSS and LUMPY was used for the detection of germline and somatic SVs [109,110]; potentially relevant SVs were manually inspected using Integrative Genomics Viewer (IGV) [111]. For somatic copy number alterations (SCNAs), binned copy number and segmentation profiles were determined using the copy number calling pipeline in the CNVkit package; gains (CN > 2) and losses (CN < 2) were assessed on calls adjusted for tumour purity [112].

#### *3.3. WGS Variant Annotation*

Germline and somatic SNVs and indels were annotated using Annovar [113] and pathogenic variants were manually inspected using IGV [111]. Missense mutations were further classified as potential oncogenic drivers using CanDrA [114] with PCa-specific databases.

The 30 SNV-derived Catalogue Of Somatic Mutations In Cancer (COSMIC) Mutational Signatures were annotated using the SomaticSignatures package in R [115]. Estimation of clonality and clonal segregation of somatic mutations were computed using PhyloWGS [116] and TITAN program [117].

#### *3.4. Tumour Mutational Burden and Percentage Genome Alteration*

TMB was calculated by counting the total number of small somatic mutations and dividing by genome size (3088 megabases (Mb)). PGA was calculated based on the cumulative number of base pairs altered for each gain or loss in the autosome (Chromosomes 1–22) per patient divided by the reference autosomal genome size (2875 Mb).

#### *3.5. Whole Genome Optical Mapping*

HMW DNA were fluorescently-labelled using either nicking enzyme Nt.BspQI (New England Biolabs) or non-nicking enzyme DLE-1 (BNG, Part #20351), according to the Bionano Prep Labeling NLRS Protocol (Document #30024) or Direct Label & Stain protocol (Document #30206), respectively. Samples prepared with BspQI (1135, 147, A153) were imaged using the Bionano Genomics (BNG) Irys system (San Diego, CA), while those prepared with DLE-1 (80002, 19651, 12543, 13179, 5545, 5684) were imaged using the BNG Saphyr system, to generate single molecule optical maps.

De novo assembly of single molecules into consensus genome maps was performed with the Bionano Solve (≥v3.2) software with aligner RefAligner (≥7437) [118–120]. Custom sets of parameters were used for this purpose and are included as File S1.
