*2.1. Simultaneous Transcriptome Analysis of Human and Murine Signatures in PDXs Can Distinguish Androgen-Dependent Expression Changes in Tumor and Host-Derived Stroma*

We analyzed the transcriptome of bulk PDX tumors grown subcutaneously in immunocompromised murine hosts by next-generation RNA-sequencing (RNA-Seq). Bone metastasis (BM)18 and LAPC9 PDXs were used in three different states: intact, post-castration (day 8 LAPCa9 and day 14 BM18) and androgen replacement (24 h) (Figure 1A). Tumor growth kinetics revealed the androgen-dependent phenotype of BM18, which regressed completely in two weeks post-castration (Figure 1B), and the androgen-independent phenotype of LAPC9 PDX tumors, which grew exponentially even after castration (Figure 1C), thus confirming the differential aggressiveness of the two models. The reduction of epithelial glands and proliferating Ki67+ cells in the BM18 castrated conditions (Figure 1D) was in contrast to the LAPC9 tumors (Figure 1E), which were morphologically indistinguishable among intact and castrated hosts. Bulk tumor tissues, which contain human tumor cells and mouse infiltrating stroma cells, were simultaneously analyzed from the same samples by RNA-Seq. To distinguish the transcriptome of the different organisms, the mouse and human reads were separated by alignment to a mouse and a human reference genome, respectively. Principal component analysis (PCA) of the human (tumor) 500 most variable genes showed that both castrated and replaced groups have altered expression profiles among each other and compared to the intact tumors. This was the case for the BM18 (Figure 2A) and the LAPC9 human transcriptomes (Figure 2B). The response to short-term androgen replacement showed a larger degree of variability in the BM18 (Figure 2A). However, the expression levels of direct AR target genes (*KLK3, NKX3.1* and *FKBP5*) identified by the RNA-Seq confirmed that androgen levels affected the activation of androgen receptor signaling in both BM18 (Figure 2C) and LAPC9 (Figure 2D, *KLK3* and *NKX3.1*). Differential expression analysis of the most variable human (tumor) genes, showed high variability among the castrated and intact groups, for both BM18 (Figure S1A) and LAPC9 (Figure S1B) transcript levels, while the LAPC9 replaced and castrated groups had similar profile among each other, discriminating them from the intact condition (Figure S1B).

PCA analysis of the BM18 mouse (stroma) transcriptome indicated that the majority of castrated samples (with and without 24-h androgen replacement) diverged from the intact tumor (Figure 2E). The LAPC9 mouse (stroma) transcriptome instead did not show specific clustering within or between the sample groups when plotting the top 500 most variably expressed genes (Figure 2F). The Ob-BMST signature of all seven genes (*Aspn*, *Pdgrfb*, *Postn*, *Aspn*, *Sparcl1*, *Mcam*, *Fscn1* and *Pmepa1),* which were upregulated in the bone stroma, as previously identified [5], were indeed expressed in the primary PCa TCGA cohort, as well as in both BM18 and LAPC9 PDXs (Figure S2A). *Pdgrfb*, *Postn*, *Aspn* and Sparcl1, specifically in the mouse RNA-Seq data, thus, are stroma-specific. Collectively, the Ob-BMST gene signature is expressed at equal levels in the BM18 and LAPC9 (intact) (Figure S2A). Some of these genes were differentially expressed upon castration in the BM18 (Figure 2G) but not in the LAPC9 (Figure 2H). A bone microenvironment-specific stroma signature induced by osteoblastic cell lines was conserved in bone metastasis PDXs maintained in other microenvironments and found in primary prostatic tissues.

**Figure 1.** In vivo tumor growth properties of androgen-dependent BM18 versus androgen-independent patient-derived xenograft (PDX) models. (**A**) Scheme of in vivo BM18 and LAPC9 experiments, including the timeline of castration, androgen replacement (single dihydrotestosterone (DHT) administration) and collection of material for transcriptomic analysis. (**B**) BM18 PDX tumor growth progression in time. Groups: (1) intact tumors (collected at max size, *n* = 3), (2) castrated (day 14, *n* = 4) and (3) castrated, followed by testosterone readministration (castrated-testosterone) (day 15 since castration and 24 h since the androgen receptor (AR), *n* = 3). R; right tumor, L; left tumor per animal. (**C**) LAPC9 PDX tumor growth progression in time. Groups: (1) intact tumors (collected at max size, *n* = 3), (2) castrated (day 8, *n* = 4) and (3) castrated, followed by testosterone readministration (castrated-testosterone) (day 9 since castration and 24 h since AR, *n* = 3). Tumor scoring was performed weekly by routine palpation; values represent average calculations of the tumors of all animals per group (considering 2 tumors, left, L, and right, R, of each animal). Error bars represent SEM, calculated considering the no. of animals for each time point. Ordinary two-way ANOVA with Tukey's multiple comparison correction was performed, *p* < 0.01 (\*\*) and *p* < 0.0001 (\*\*\*\*). (**D**) Histological morphology of BM18 and (**E**) LAPC9 (from intact, castrated and androgen-replaced hosts), as assessed by Hematoxylin and Eosin staining (H&E, top). Scale bars: 20 μm, and proliferation marker Ki67 protein expression (bottom panel).

**Figure 2.** Separation of human (tumor) and mouse (stroma) transcriptomes of BM18 and LAPC9 tumors. (**A**,**B**) Principal component analysis plot of the gene expression of the 500 most variable genes on all samples; BM18 human transcripts (**A**) and LAPC9 human (**B**) at intact, castrated and replaced (castrated + 24 h AR) conditions. (**C**,**D**) Expression values of AR direct target genes as detected by RNA-Seq (transcript per million (TPM) counts) in the BM18 (**C**) or LAPC9 (**D**) tumors as confirmation of the effective repression of AR downstream signaling by castration. Intact (*n* = 3), castrated (*n* = 4) and replaced (*n* = 3). (**E,F**) Principal component analysis plot of the gene expression of the 500 most variable genes on all samples, BM18 mouse (**E**) and LAPC9 mouse (**F**) at intact, castrated and replaced (castrated + 24 h AR) conditions. (**G**,**H**) Expression values of the prostate-specific bone metastasis signature (Ob-BMST) seven upregulated stroma signature genes, as detected by RNA-Seq (TPM normalized counts) in the mouse transcriptome of BM18 (**E**) or LAPC9 (**F**) tumors.

#### *2.2. Proteomic Analysis Provides Functional Information over the Identified Human*/*Mouse-Specific Transcriptome*

To study the proteome of the tumor versus the stroma, human and mouse cell fractions were isolated by the magnetic cell sorting (MACS) mouse depletion method from tumor sample preparations: BM18 and LAPC9 each at the intact, castrated and replaced states. Protein lysates of either mouse or human origins (single replicate from a pool of *n* = 3 to 4 biological replicates per condition) were subjected to an in-solution tryptic digest following Tandem Mass Tag (TMT)-labeling of the resulting peptides and their mass spectrometric analysis (Figure 3A).

In addition to the initial experimental separation of the protein lysates, we further explored the species homologs of the identified proteins by computational analysis using a combined human and mouse protein sequence database. We identified 4198 proteins in the sample that were enriched for human cells. Thereof, 3154 were human-specific proteins, with 996 revealing a high homology shared among human and mouse, and only a fraction of 48 mouse-specific, peptides. (Figure 3B, left plot). For samples enriched in mouse cells, we identified, in total, 5192 proteins; thereof, 2486 mouse-specific proteins, 2379 shared homologs and 247 human-specific (Figure 3B, right plot). We searched for prostate specific markers such as KLK3, a prostate-specific antigen that is secreted by luminal cells. In the proteomic data, the human-specificity was confirmed, and the secreted protein was found also in the mouse fraction (Figure 3C). To further ensure that the proteomic data were indeed identifying real stromal-specific candidates, we searched specifically for the seven-gene Ob-BMST signature found also to be expressed in both BM18 and LAPC9. POSTN, PDGFRB and MCAM (Figure 3C) were indeed detected at the protein level, thus might have a functional role, and were found exclusively in the mouse fraction (Figure 3C, right plot) and hybridizing with mouse-specific sequences (Figure 3C, triangle indicates Mus Musculus species specificity).
