*2.4. Cross Comparison of Stromal Transcriptome among Di*ff*erent PDXs Identifies ECM and Cell Adhesion Pathways in the LAPC9 Androgen-Independent Model*

To assess the similarity between the stromal transcriptome of the androgen-independent LAPC9 and the BM18, a differential expression analysis was performed. In a panel of the 50 top-most variable genes comparing the tumors at their intact conditions, we identified several genes that follow the same pattern of modulation in intact tumors (Figure 5A) and in castrated tumors (Figure 5B). Of interest were the ECM-related genes downregulated in LAPC9 versus BM18; the Fibroblast Growth Factor receptor (*Fgfr4*), elastin microfibril interface (*Emilin3*) and upregulated collagen type 2 chain a1 (*Col2a1*).

The differential expression of LAPC9 castrated versus BM18 castrated highlighted genes that were identified in the analysis among LAPC9 castrated, replaced versus LAPC9 intact, such as Apelin (*Apln*), *Col2a1* and Tenascin C (*Tnc*).

To identify the biological processes ongoing in the LAPC9 compared to BM18, a pathway analysis was performed on the differentially expressed murine genes of the LAPC9 versus the BM18. Enrichment maps of the top 20 enriched GO biological pathways highly overlap pathways, such as ECM, focal tadhesion and cell adhesion/migration in the intact and castrated LAPC9 (Figure S3A,B). Similarly, among the KEGG pathway sets, there was an enrichment of stroma regulation (e.g., actin cytoskeleton, focal adhesion and cell adhesion) and bone and immune-related processes (e.g., osteoclast differentiation) (Figure S3C–F and Table S2). The enrichment of cancer-related pathways (e.g., PI3K/AKT, proteoglycans in cancer, pathways in cancer) was commonly found in the LAPC9 intact and castrated stroma transcriptomes (Figure S3E,F and Table S2).

Given that genes activated in a castrated state might be indicative of androgen resistance mechanism activation, we postulated that genes upregulated in the androgen-resistant LAPC9 over the androgen-dependent BM18 might be relevant for understanding the aggressive phenotype of LAPC9 and, therefore, of the advanced metastatic phenotype of similar tumors. One of those genes, Tenascin, is an ECM protein that is produced at the (myo)fibroblasts that is virtually absent in normal stroma in the prostate and other tissues and has been associated with the cancerous reactive stroma response in different cancers. We interrogated the expression of *Tnc* in the RNA-Seq data and found that it was highly upregulated in LAPC9 compared to BM18 both in intact (logFC 4.23, *p* < 0.001) and among the castrated conditions (logFC 6.9, *p* < 0.001) (Figure 5C). However, in both models, the *Tnc* levels significantly decreased upon castration (BM18, *p* < 0.001 and LAPC9, *p* < 0.05), indicating the potentially AR-mediated regulation of *Tnc* expression. In LAPC9 tumors, the TNC protein is expressed in the tumor-adjacent ECM and in the proximity of vessels (Figure 5D, intact and castrated) and co-expressed by smooth muscle actin (αSMA)- and collagen type I-positive myofibroblasts (Figure 5E). Instead, the intact BM18 tumors show TNC and collagen type I deposition in the ECM, but there is no overlap with αSMA-positive myofibroblasts (Figure 5D,E, BM18 intact). Castrated BM18 tumors have minimal TNC expression, found only in cells proximal to the remaining epithelial glands, yet with no typical fibroblast/stromal morphology (Figure 5D,E, BM18 intact), suggesting an altered phenotype of TNC upon androgen deprivation.

**Figure 3.** Proteomic analysis of human (tumor) versus mouse (stroma) of BM18 and LAPC9 tumors. (**A**) Experimental separation of human from mouse cell suspensions from fresh tumor isolations by MACS mouse depletion sorting. Cell fractions from intact/replaced (*n* = 3 each), castrated (*n* = 4) biological replicates were pooled into a single replicate (*n* = 1) to achieve an adequate cell number for the proteomic analysis (1 <sup>×</sup> <sup>10</sup><sup>6</sup> cells). Protein lysates from the different fractions of BM18/LAPC9 (intact, castrated and replaced) were subjected to Tandem Mass Tag (TMT) labeling (all-mouse or all-human samples were multiplexed in one TMT experiment each), followed by mass spectrometry. (**B**) Detected peptides from human and mouse fractions were searched against a combined human and mouse protein database. Number of species specific or shared proteins is indicated in different colors. (**C**) KLK3 (PSA; Prostate Serum Antigen) protein levels (log2 normalized TMT signal sum values) in human cell isolations (left) and in mouse cell isolations (right), and the protein sequence was predicted as human-specific (spheres indicate Homo Sapiens sequence). Seven-up Ob-BMST signature markers POSTN, PDGFRB and MCAM protein levels were absent in human cell isolations (left) and present in mouse cell isolations (right), while all the protein sequences were mouse-specific (triangles indicate Mus Musculus sequences).

stromal genes as response to androgen deprivation. (**A**) Heatmap represents a differential expression analysis of the most variable genes from the mouse transcriptome of BM18 castrated compared to BM18 intact tumors. Genes modulated by androgen deprivation due to castration in the up/downregulation compared to intact tumors are indicated in red or blue colors, respectively. (**B**) Heatmap represents Z-score of the differential expression analysis of most variable genes in the mouse transcriptome of LAPC9 castrated (with and without androgen replacement) compared to LAPC9 intact tumors. (**C**) Description of mouse genes found upregulated in BM18 intact tumors and the biological processes they are involved in, according to the Gene Ontology (GO) terms. (**D**) Description of the mouse genes found upregulated in LAPC9 intact tumors and the biological processes they are involved in, according to the GO terms.

**Figure 5.** Cross-comparison of LAPC9 versus BM18 suggests stromal gene Tenascin C expression being associated with advanced PCa and regulated by androgen levels. (**A**) Heatmap represents the differential expression analysis of the top 100 most variable genes from the mouse transcriptome of LAPC9 intact tumors compared to BM18 intact tumors and (**B**) of LAPC9 castrated tumors compared to BM18 castrated tumors. (**A**) Subset of genes in LAPC9 samples have zero counts, leading to the same z-score, while the same genes are highly expressed in BM18 samples. (**C**) *Tnc* RNA expression (log2CMP counts) in the stroma transcriptome. LogFC (fold change) enrichment of *Tnc* in LAPC9 over BM18 is indicated. Ordinary two-way ANOVA with Tukey's multiple comparison correction was performed, *p* < 0.05 (\*) and *p* < 0.0001 (\*\*\*\*). (**D**) Tenascin protein expression and stromal specificity assessed by immunohistochemistry in LAPC9 and BM18 tumors, both at the intact and castrated states. Scale bars: 20 μm. (**E**) Tenascin protein (indicated in red) colocalization with stromal markers, smooth muscle actin (αSMA, green) and collagen type I (gray) assessed by immunofluorescence in LAPC9 and BM18 tumors, both at the intact and castrated states. DAPI marks the nuclei. Scale bars: 50 μm.

#### *2.5. Protein Expression of Tenascin and Its Interaction Partners*

To assess whether the transcriptomic changes of *Tnc* in the PDX models corresponds to the functional protein and, thus, a relevant role in bone metastatic PCa, we performed a proteomic analysis. A mass spectrometry analysis of the human and mouse fractions indicated that the Tnc protein was expressed specifically in the mouse (stromal) fractions in BM18 and LAPC9 (Figure 6A). The isoform Tenascin X was also expressed at the protein level (Figure 6B). The interaction network of the mouse protein Tnc is based on experimental observations and prediction tools (STRING) and consists of laminins (Lamc1 and Lamb2); fibronectin (Fb1); integrins (Itga2, a7, a8 and a9) and proteoglycans (Bcan and Vcan) (Figure 6C). The human interactome is less-characterized, yet most of the interactome is conserved: laminins (LAMC1 and LAMB2); proteoglycans (NCAN and ACAN) and others such as interleukin 8 (IL-8), BMP4, ALB and SDC4 (Figure 6D). However, integrin interaction-binding partners in a human setting have not been confirmed. Given the importance of integrins for cell adhesion and migration known to be found in mesenchymal/stromal and epithelial tumor cells, we focused on the expression of human- and mouse-derived integrins. The *ITGA9*, *ITGA6* and *ITGA2* were all found to be expressed in both the RNA-Seq and proteomic data (Figure 6E, ITGA6, respectively); however, only the *ITGA2* protein was specifically found in the human counterpart and not overlapping with the mouse stroma (Figure 6F). Co-labeling both proteins indicated adjacent spatial localization with TNC deposition in close proximity to ITGA2-positive epithelial cells (Figure 6G); however, whether those cell populations acquired different properties compared to other epithelial cells has yet to be investigated. Overall, the tumor *ITGA2* and stromal *Tnc* is a potential molecular interaction, possibly part of the dual cellular communication among a tumor and its microenvironment cellular types and ECM.
