*4.4. Signature Validation on TCGA and Other Publically Available Datasets*

TCGA gene expression, Gleason scores and outcome data were obtained from the PanCanAtlas publications supplemental data site (https://gdc.cancer.gov/about-data/publications/ pancanatlas) [56,57]. For the Gene Set Variation Analysis (GSVA) analysis, RSEM expected counts in the upper quartile normalized to 1000 (i.e., the same normalization as TCGA) were used for BM18/LAPC9 gene expression. Mouse genes in gene signature lists were mapped to human homologs using the biomaRt R package (Table S5), using the "mmusculus\_gene\_ensembl" dataset and selecting only homologs with hsapiens\_homolog\_orthology\_confidence = 1. Signature scores were calculated using the GSVA R package using the GSVA method [58].

Validation of the C1-C4 stroma signatures on publicly availably cohorts was performed using the Taylor (GSE21034) and the Grasso (GSE35988) datasets. Gene expression and sample information, including Gleason scores, were obtained via the GEOquery Bioconductor package. Mouse genes in the C1-C4 gene signature lists were mapped to human homologs using the biomaRt R package, using the "mmusculus\_gene\_ensembl" dataset and selecting only homologs with hsapiens\_homolog\_orthology\_confidence = 1. Other gene sets are either human genes or include info on human homologs. Signature scores were calculated using the GSVA R package using the GSVA method [58].

## *4.5. Tissue Dissociation and MACS*

Tumor tissue was collected in a basis medium (advanced Dulbecco Modified Eagle Medium F12 serum-free medium (12634010, Thermo Fisher Scientific, Basel, Switzerland) containing 10-mM Hepes (15630080, Thermo Fisher Scientific, Basel, Switzerland), 2-mM GlutaMAX supplement (35050061, Thermo Fisher Scientific, Basel, Switzerland) and 100 μg/mL Primocin (ant-pm-1, InVivoGen, LabForce AG, Muttenz, Switzerland). After mechanical disruption, the tissue was washed in the basis medium (220 relative centrifugal force (rcf), 5 min) and incubated in the enzyme mix for tissue dissociation (collagenase type II enzyme mix (17101-015, Gibco, Thermo Fisher Scientific, Basel, Switzerland) 5 mg/mL dissolved in the basis medium and DNase: 15 μg/mL (10104159001, Sigma-Aldrich, Buchs, Switzerland) and 10-μM Y-27632-HCl rock inhibitor (S1049, Selleckchem, Zürich, Switzerland). Enzyme mix volume was adjusted so that the tissue volume did not exceed 1/10 of the total volume, and tissue was incubated at 37 ◦C for 1 to 2 h, with mixing every 20 min. After the digestion of large pieces was complete, the suspension was passed through a 100-μm cell strainer (21008-950, Falcon®, VWR International GmbH, Dietikon, Switzerland) attached to a 50-mL Falcon tube, then using a rubber syringe to crash tissue against the strainer and wash in 5-mL basic medium (220 rcf, 5 min). Cell pellet was incubated in 5-mL precooled red blood cell lysis buffer (150-mM NH4Cl, 10-mM KHCO3 and 0.1-mM EDTA), incubated for 10 min and washed in equal volume of basis medium, followed by centrifugation (220 rcf, 5 min). Pellet was resuspended in 2–5 mL accutase™ (StemCell Technologies, 07920), depending on the sample amount; biopsies versus tissue and incubated for 10 min at room temperature. The cell suspension was passed through a 40-μm pore size strainer (21008-949, Falcon®, International GmbH, Dietikon, Switzerland), and the strainer was washed by adding 2 mL of accutase on the strainer. Single-cell suspension was counted to determine the seeding density and washed in 5 mL of basis medium and spun down 220 rcf, 5 min. Magnetic cell sorting was performed to separate purified human versus mouse cell fractions using the Mouse Cell Depletion Kit (130-104-694, Miltenyi Biotek, Solothurn, Switzerland). For the proteomic experiments, cell fractions from tumor tissues (*n* = 3 to 4 per condition) were pooled together in order to suffice for 106 cells, representing one technical replicate per sample.
