*2.6. Stromal Tenascin Expression as a Prognostic Factor of Disease Progression in High-Risk PCa*

The detection of key mouse stromal genes in PCa PDXs gives the opportunity to evaluate the role and potential prognostic value of the human orthologs of these stromal genes. To validate the localization and stromal specificity of TNC protein expression, we performed immunohistochemistry on the primary PCa tissue sections. TNC is localized in the extracellular space (Figure 7A, primary cases). Next, we evaluated the TNC expression in a tissue microarray of 210 primary prostate tissues, part of the European Multicenter High Risk Prostate Cancer Clinical and Translational research group (EMPaCT) [14–16] (Figure 7B–G). Based on the preoperative clinical parameters of the TMA patient cases (Table 1, Table 2) and the D'Amico classification system [17], they represent intermediate (clinical T2b or Gleason *n* = 7 and PSA >10 and ≤ 20) and high-risk (clinical T2c-3a or Gleason score (GS) = 8 and PSA ≥ 20) PCa. The number of TNC-positive cells (Figure 7B) were quantified and averaged for all cores (four cores per patient case) in an automated way, including tissue selection, core annotation and equal staining parameters set. To investigate the association between the number of TNC-positive cells and patient survival or disease progression, we calculated the optimal cut-point for the number of TNC-positive cells by estimation of the maximally selected rank statistics [18]. Association between TNC-expressing cells and pT Stage indicated that the majority of cases cluster towards stages 3a and 3b (Figure 7C). A multiple comparison test among all groups showed no statistically significant association between the TNC expression and pathological stage (Table S3, *p* > 0.05). The overall survival probability between two patient groups with, respectively, high and low numbers of TNC-positive cells was indifferent (*p* = 0.29, Log-rank test) (Figure 7D). We focused on the probability of TNC expression in primary tumors to be a deterministic factor for clinical progression to local or metastasis recurrence. Clinical progression probability was higher in the TNC-low group compared the TNC-high group (*p* = 0.04 \*, Log-rank test) (Figure 7E). Next, we examined the clinical progression in patients with pT Stage ≥3 (groups 3a, 3b and 4). The high T-Stage cases did separate into two groups based on the TNC expression, with the TNC low-expressing group exhibiting earlier a clinical progression (local or metastatic recurrence, *p* = 0.013 \*, Log-rank test) (Figure 7F). The PSA progression probability in patients with pT Stage ≥3 indicated an association trend of a TNC-low group with earlier biochemical relapse events (*p* = 0.07, Log-rank test) (Figure 7G). Similarly, the TNC-low

group correlated with a higher probability for PSA progression after radical prostatectomy among cases with carcinoma-containing (positive) surgical margins (Figure S4A**,** *p* = 0.031 \*, Log-rank test) or positive lymph nodes (Figure S4B, *p* = 0.092, Log-rank test). A low number of TNC-expressing cells coincides with a poor prognosis in terms of metastasis progression, similarly to its downregulation upon castration in the bone metastasis PDXs (Figure 4B) based on the RNA-Seq analysis. To further evaluate the clinical relevance of this finding, in multiple clinical cohorts with available transcriptomic data and clinical information, a CANCERTOOL analysis was performed [19]. Similar to the protein TMA data (Figure 7), the *TNC* mRNA levels were significantly downregulated during the disease progression from primary to PCa metastasis, compared to the expression in the normal prostatic tissues in all five datasets tested (Figure 8A). The *TNC* expression shows a pattern of inverse correlations, with the Gleason score among GS6 to GS9; however, it significantly discriminated patient groups for the Gleason score in one out of three datasets tested (Figure 8B, TCGA dataset \* *p* = 0.049, Glinsky *p* = 0.06, Taylor *p* = 0.192), with the highest expression found in a high GS10 group and indifferent among GS6-GS9. A disease-free survival analysis indicated that a low TNC expression is associated with a worse prognosis based on the Glinsky dataset (Q1 Glinsky et al. [20], \* *p* = 0.02), while no statistically significant association was observed in the Taylor and TCGA dataset (Figure 8C). Overall, the TNC expression in tumor samples, both at the RNA and protein levels, becomes progressively less abundant in primary and metastasis PCa specimens, while a low TNC expression is significantly associated with the disease progression and poor disease-free survival (DFS) outcome.

#### *2.7. Stroma Signatures from Androgen-Dependent and -Independent States Correlate with Disease Progression*

In order to comprehensively map the stroma responses related to the disease severity, we analyzed the stroma gene signature lists associated to androgen dependency and aggressive androgen-independent states. The stroma signatures are categorized in clusters (C1–C4, Table S4) based on a differential expression analysis (Figures 4 and 5): C1 (50 highly upregulated genes in BM18 intact that get downregulated upon castration), C2 (27 highly upregulated genes in LAPC9 intact that get downregulated upon castration), C3 (32 highly upregulated genes in LAPC9 intact compared to BM18 intact) and C4 (24 highly upregulated genes in LAPC9 castrated compared to BM18 intact). Clusters C1 and C2 aim to identify the most responsive genes to androgen deprivation. C3 and C4 are designated to identify the genes/pathways enriched in the stroma of castration-resistant prostate cancer (CRPC) compared to the androgen-dependent tumor model. The TNC gene was among the signature list: C1, C3 and C4. The prognostic potential of the C1-C4 signatures in comparison to the bone signature Ob-BMST was tested on the TCGA cohort based on the Gleason score, gene expression and outcome data (Figure 9 and Table S5). The high signature scores of Ob-BMST, C1, C2 and C4 had statistically significant positive correlations with the high GS groups (Figure 9A, Ob-BMST and C1 (*p* < 0.001), C2 and C4 (*p* < 0.01)). In terms of gene expression, the C1 signature was significantly higher in primary tumors versus normal tissues (Figure 9B, *p* < 0.001), while the C2, C3 and C4 have lower signature scores in the tumor samples compared to normal (Figure 9B, C2 and C3 (*p* < 0.001) and C4 (*p* < 0.01)). Kaplan-Meier plots of progression-free survival (PFS) stratified as the bottom 25% (Q1), middle 50% (Q2 and 3) and top 25% (Q4) showed significant correlations among the high signature scores (Q4) of the C1 gene set and PFS (Figure 9C, *p* < 0.001), while none of the other gene lists showed significant correlations.

**Figure 6.** Tenascin C and its predicted interaction partners analyzed by mass spectrometry. (**A**) Tenascin C (TNC) and (**B**) alternative isoform Tenascin X (TNXB) protein relative abundance (log2 ratios; single replicates per sample from a pool of *n* = 3 to 4) in human cell isolations (left) and present in mouse cell isolations (right). The variance stabilization normalization (vsn)-corrected TMT reporter ion signals were normalized by the intact conditions of either BM18 or LAPC9. The protein sequences were predicted as mouse-specific (green). (**C**) Protein interaction network of the mouse TNC protein based on the STRING association network (https://string-db.org/). (**D**) Protein interaction network of the human TNC protein based on the STRING association network https://string-db.org/. (**E**) Predicted TNC-binding partner integrin A6 (ITGA6) was detected by mass spectrometry in both the human and mouse protein lysates and matching the organism-specific protein sequence based on the bioinformatics analysis (red for human and green for mouse). (**F**) Predicted TNC-binding partner integrin A2 (ITGA2) was detected by mass spectrometry, specifically in the human protein lysates, and matched the human-specific protein sequence. (**G**) Spatial localization of the Tenascin protein (TNC, indicated in red) and integrin A2 (ITGA2, green) assessed by immunofluorescent co-labeling in LAPC9 intact and castrated tumors. DAPI marks the nuclei. Scale bars: 50 μm.

**Figure 7.** TNC protein expression is a negative metastasis prognostic factor in primary, high-risk PCa. (**A**) Validation of the protein expression and stromal specificity of TNC by immunohistochemistry in primary PCa cases. (**B**) Representative cases of TNC staining on primary PCa Tissue Microarray (TMA) from European Multicenter Prostate Cancer Clinical and Translational Research Group (EMPaCT). (**C**) TNC expression levels in terms of the no. of positive cells in the pT Stage classification. Statistical multiple comparison test, the Wilcoxon rank sum test, was performed; *p* > 0.05 (**D**) Overall survival probability in patient groups of TNC-high and TNC-low (no. of positive, TNC-expressing cells) (*p* = 0.29, ns—non significant). Average value represents the mean of four cores per patient case. (**E**) Clinical progression to the local recurrence or metastasis probability in patient groups of TNC-high and TNC-low expressions (*p* = 0.04 and \* < 0.05). (**F**) Clinical progression to the local recurrence or metastasis probability among patients of pT Stages 3a, 3b and 4 based on TNC-high and TNC-low expressions (*p* = 0.013 and \* < 0.05). (**G**) PSA progression probability among patients of pT Stages 3a, 3b and 4 based on the TNC-high and TNC-low expressions (*p* = 0.074, ns).


**Table 1.** Clinical parameters of the EMPaCT TMA patient cases.


To further assess the prognostic performance of the signatures, we correlated the C1-C4 gene signatures with PCa-specific stroma signatures identified by Tyekucheva et al. [7] and Mo et al. [21] (Table S4) across two cohorts containing both primary and metastatic PCa that were used [22,23]. The C3 and C4 showed the strongest linear correlations with the Tyekucheva and the Mo\_up (upregulated in metastases) signatures when tested across the Grasso dataset (Figure S5A, r > 0.64), while the C4 signature also had positive correlations when tested across the Taylor et al. dataset (Figure S5B, r > 0.6). The C1 signature did not significantly correlate with the gene lists tested (Figure S5A, C1 *p* > 0.05). The low signature score of the C2 and C3 were significantly associated with metastatic disease progression (Figure S5B, *p* < 0.001) in both cohorts tested, and C4 showed a similar pattern (Figure S5B, C4 *p* = 0.062). A common pattern of the stroma signatures is a similar or enriched signature score at the primary stage compared to benign/normal tissue, and lower/depleted signature scores at the metastasis stage (Figure S5C,D; C2, C3 and C4, Tyekucheva and Mo and Figure 9B; C2-C4). Only a significant correlation with the Gleason score was observed by the C1 signature list, with a high signature score found at the high GS patient groups (Figure S5E, *p* ≤ 0.001), which is in concordance to the linear correlation with metastatic disease in all clinical cohorts tested (Figure S5C,D, *p* ≤ 0.001 and Figure 9, TCGA).

**Figure 8.** *TNC* RNA expression is inversely correlated with the disease progression, Gleason score and survival. (**A**) Violin plots depicting the expression of *TNC* among nontumoral (N), primary tumor (PT)

and metastatic (M) PCa specimens in the indicated datasets. The Y-axis represents the Log2-normalized gene expression (fluorescence intensity values for microarray data or sequencing read values obtained after gene quantification with RNA-Seq Expectation Maximization (RSEM) and normalization using the upper quartile in case of RNA-seq). An ANOVA test is performed in order to compare the mean gene expression among two groups (nonadjusted *p*-value), obtained by a CANCERTOOL analysis. (**B**) Violin plots depicting the expression of *TNC* among PCa specimens of the indicated Gleason grade in the indicated datasets. The Gleason grades are indicated as GS6, GS7, GS8, GS8+9, GS9 and GS10. An ANOVA test is performed in order to compare the mean among groups (nonadjusted *p*-value), obtained by a CANCERTOOL analysis. (**C**) Kaplan-Meier curves representing the disease-free survival (DFS) of patient groups selected according to the quartile expression of *TNC*. Quartiles represent ranges of expression that divide the set of values into quarters. Quartile color code: Q1 (Blue), Q2 plus Q3 (Green) and Q4 (Red). Each curve represents the percentage (Y-axis) of the population that exhibits a recurrence of the disease along the time (X-axis, in months) for a given gene expression distribution quartile. Vertical ticks indicate censored patients. Quartile color code: Q1 (Blue), Q2 plus Q3 (Green) and Q4 (Red). A Mantel-Cox test is performed in order to compare the differences between curves, while a Cox proportional hazards regression model is performed to calculate the hazard ratio (HR) between the indicated groups. Nonadjusted *p*-values are shown. Analysis obtained by CANCERTOOL.

**Figure 9.** Stroma signatures identified from bone metastatic PDXs as prognostic biomarkers in primary PCa. (**A**) Violin plots showing Gene Set Variation Analysis (GSVA) signature scores of the Ob-BMST, C1-C4 gene sets, stratified by Gleason score from the TCGA cohort. Box-and-whisker plots illustrating median (midline), inter-quartile range (box), with the whiskers extending to at most 1.5 IQR from the box. Outliers beyond the range of the whiskers are illustrated as dots. P-values computed by Spearman correlation tests. (**B**) Violin plots showing GSVA signature scores of the Ob-BMST and C1-C4 gene sets stratified by sample types (NT: nontumor and TP: primary tumor) from the TCGA cohort. Box-and-whisker plots illustrating the median (midline) and interquartile range (box), with the whiskers extending to at most 1.5 IQR from the box. Outliers beyond the range of the whiskers are illustrated as dots. *P*-values computed by Mann-Whiney U tests. (**C**) Kaplan-Meier plots of progression-free survival (PFS) stratified as the bottom 25% (Q1), middle 50% (Q2 and 3) and top 25% (Q4) of the signature scores of the Ob-BMST and C1-C4 gene sets. *P*-values and hazard ratios computed by Cox proportional hazard regression.

#### **3. Discussion**

The role of the microenvironment upon cancer formation and progression to metastasis is supported by numerous studies [24,25]; however, the current knowledge is not sufficient to reconstruct the chain events from primary to secondary tumor progression. The normal stroma microenvironment is considered to halt tumor formation; however, after interactions with tumor cells, it also undergoes a certain "transformation" at the transcriptomic, and even at the genetic, levels [26–29]. The processes by which PCa tumor cells affect stroma and, in turn, stroma impacts primary PCa tumor growth or metastasis are complex and remain largely unclear.

We utilized well-established bone metastasis PDX models, which can be propagated subcutaneously and have different aggressiveness in terms of androgen dependency: the CRPC model LAPC9 representing complete androgen-independent advanced disease [30] and the BM18 that mimics human luminal PCa [31,32] and uniquely retains androgen sensitivity, typically seen in the primary and treatment-naïve stages. The androgen-independent stem cell populations that survive castration are well characterized in both models [31,33,34]; yet, the contribution of the stroma in those district tumor phenotypes has not been investigated. In vivo PDX models grafted in immunocompromised mice, although they lack the complexity of a complete immune system, represent the stroma compartment (endothelial cells, smooth muscle cells, myofibroblasts and cancer-associated fibroblasts). Due to the subcutaneous growth of BM PCa PDXs, the human stroma is replaced by mouse-infiltrating stromal cells and vasculature [35,36]. Mouse cell infiltration allows the discrimination of organism-specific transcripts, human-derived transcripts representing the tumor cells and mouse-derived transcripts representing the mouse stroma compartment. Using next-generation RNA-Seq, MACS-based human and mouse cell sorting, mass spectrometry and organism-specific reference databases, we have identified the tumor-specific (human) from the stroma-specific (mouse) transcriptomes and proteomes of bone metastasis PCa PDXs. The dynamics of AR signaling in the stroma are best represented in an in vivo setting [11]; therefore, to specifically examine the stroma changes dictated by PCa cells, we subjected the PDXs in androgen and androgen-deprived conditions. By imposing this selection pressure, we could identify androgen-dependent gene expression patterns.

We demonstrated that the human (tumor), as well as the mouse (stroma), transcriptomes follow androgen-dependent transcriptomic changes in the BM18 groups (intact versus castrated versus replaced). Despite the androgen-independent tumor growth of LAPC9, at the gene expression level, the LAPC9 tumor cells do follow AR-responsive patterns (human transcriptomes). However, the principal component analysis showed that, although castrated and replaced LAPC9 groups separate adequately based on the human transcriptome, they appear to have overall uniform stromal transcriptomes.

We report that transcriptomic mechanisms linked to osteotropism were conserved in bone metastatic PDXs, even in nonbone environments, and differential stroma gene expressions are induced by different tumors, indicating the tumor specificity of stroma reactivity. The Ob-BMST signature of all seven genes (*Aspn*, *Pdgrfb*, *Postn*, *Sparcl1*, *Mcam*, *Fscn1* and *Pmepa1)*, which were upregulated in bone stroma previously identified [5], were indeed expressed in both BM18 and LAPC9 PDXs, specifically in the mouse RNA-Seq and, also, expressed at the protein level, as identified by mass spectrometry. The gene expression modulation of mouse stroma is, ultimately, an important evidence of the effects of tumor cells in their microenvironment, where they induce favorable conditions for their growth.

The differential expression analysis of the LAPC9 stroma signature from intact, castrated and replaced hosts highlighted the most significantly variable genes, which were modulated by androgen levels, despite the androgen-independent tumor growth phenotype. Focusing on the genes that were highly activated in intact but strongly modulated by castration, we categorized these genes based on Gene Ontology terms. We found that LAPC9 stromal genes were ECM remodeling components and genes involved in smooth muscle function or even in striated muscle function. Of interest are *CD56*, *Tnc* and *Flnc*. Among the BM18 most abundant stromal transcripts are genes involved in cell cycle regulation and cell division. Interrogating the differences among the two models, we focused on the transcriptome of LAPC9 normalized versus the less aggressive, androgen-dependent BM18. In particular, *Tnc* is expressed in both PDXs, higher in LAPC9, yet downregulated upon castration, suggesting a direct AR gene regulation. The differential expression analysis among both the PDXs after castration indicated that *Tnc* is upregulated more in LAPC9 than BM18, suggesting an association with disease aggressiveness. Genes that become upregulated in castrated conditions are likely to be linked to androgen resistance; thus, we studied *Tnc* for its potential role in metastasis progression.

TNC is an extracellular glycoprotein absent in normal prostates and postnatally silenced in most tissues. TNC is re-expressed in reactive stroma in human cancers, and there is evidence of its expression in low-grade tumors (Gleason 3) of human PCa [37] and, possibly, already activated at the prostatic intraepithelial neoplasia (PIN) stage [38,39]. In particular, high molecular weight TNC isoforms are expressed in cancer due to alternative mRNA splicing [38]. We examined whether an abundance of TNC-positive cells in primary PCa TMA can predict the metastatic progression and overall survival (12 years follow-up after radical prostatectomy). A high number of TNC-positive cells did not correlate with the overall survival or histological grade, in agreement with previous data [38]. The PSA progression after radical prostatectomy occurred earlier in the TNC-low group compared to the TNC-high group when high stage cases (pT ≥ 3), surgical margin-positive or lymph node-positive cases were investigated. In terms of clinical progression, the TNC-low group in the total number of cases and among the high stage (pT ≥ 3) cases showed a worse prognosis in terms of local recurrence/metastasis. This finding is in contrast to the study of Ni et al., showing that high levels of TNC are significantly linked to lymph node metastasis and the clinical stage [40] but in agreement with another study that reported a weak TNC expression in high-grade PCa [39]. No low-risk cases or metastasis tissues were used in our study, and we focused on TNC-producing cells, not the overall TNC expression in the matrix. Therefore we can only conclude that the TNC is indeed expressed in intermediate- and high-risk primary PCa as assessed at the preoperative diagnosis based on the D´Amico criteria [17] and that a high number of TNC-positive cells is inversely correlated with clinical progression.

More evidence points to the direction that the TNC might be degraded upon local recurrence in lung cancer [41,42], while high TNC is found in lymph and bone metastases sites [38] or even in certain types of bone metastasis [43]. In the TMA of PCa bone metastasis, San Martin et al. demonstrated a high TNC expression in trabeculae endosteum, the site of osteoblastic metastasis, and yet, a low TNC expression in the adjacent bone marrow sites [43]. Osteoblastic PCa cell lines proliferate rapidly in vitro and adhere to TNC protein, while osteolytic PC3 or lymph node-derived PCa lines do not show this phenotype, suggesting an association of TNC with osteoblastic but not osteolytic metastases. One of the ligands of TNC highly upregulated in VCap cells was α9 integrin, which binds directly TNC and a modulate expression of collagen [43], providing evidence for TNC-integrins in human PCa. Our RNA-Seq data indicate, also, the expression of α9 integrin, along with α6 and α2, and based on the proteomic human–mouse separation, we found integrin α2 to be the only one human-specific and, thus, tumor-specific for the PDXs used in this study. Although the molecular mechanism among TNC-ITGA2 should be further characterized, evidence on the correlation among α2 and α6 expressions in primary PCa and bone metastasis occurrence has been previously reported [44].

The reactivation of TNC expression is relevant for reactive stroma regulation, while TNC downregulation might be relevant for recurrence or metastasis initiation, which remains to be further investigated. Indeed, TNC is known to have pleiotropic functions in different cellular contexts, with both autocrine TNC expression in tumor cells and paracrine TNC from stroma in different stages of metastasis [45]; however, the cellular source of TNC in primary PCa was not addressed in our study. Our data demonstrate that androgens regulate stromal TNC expression, evident by the reduced TNC expression upon castration (even in the castration-resistant LAPC9) and immediate increased expression upon androgen replacement; thus, the TNC expression should be further evaluated in CRPC samples. Genomic amplification in the TNC gene associated with highly aggressive neuroendocrine PCa occurrence [46]. In a multi-omics approach study, the TNC protein was one of the panels of four

markers detected in preoperative serum samples and, collectively, predict the biochemical relapse events with high accuracy [47].

In summary, we identified the stroma signature of bone metastatic PDXs, and by analyzing androgen-dependent versus androgen-independent tumors, we could demonstrate that the tumor-specific stroma gene expression changes. We could show that there are AR-regulated stromal genes modulated upon castration, even in the androgen-independent, for tumor growth, like the LAPC9 model. The osteoblastic bone metastasis stromal seven-gene signature was induced in the mouse-derived stroma compartment of BM18 and LAPC9, indicating conserved tumor mechanisms that can induce the transcriptomic "transformation" of mouse-infiltrating stroma (even in subcutaneous sites) to bone microenvironment-like stroma. The prognostic value of stroma signatures has been also demonstrated by another study utilizing PDXs associated with the metastasis prognosis from different lesions from a single PCa case and demonstrated the strong predictability of 93-gene stroma signatures to metastasis phenotypes in different clinical cohorts [21]. We identified androgen-dependent Tenascin C expression in the stroma of PDX models, which is downregulated in the conditions mimicking an aggressive disease (upon castration), similarly to the high clinical progression probability of a low TNC group in the primary PCa TMA. The higher stromal *Tnc* mRNA levels in the aggressive LAPC9 compared to BM18 may suggest that it would be relevant to examine the TNC mRNA and protein expressions in human bone metastasis or ideally matched primary metastasis cases in order to understand the kinetics of TNC in terms of disease progression. Given that TNC expression was found elevated from 0% in benign prostatic hyperplasia (BPH) stroma to 47% in tumor-associated stroma [29], its detection in circulation [47] and its immunomodulatory role [48] indicate TNC as a promising drug target and disease-determining factor. The TNC clinical progression predictive value performs best in an earlier stage, low-risk PCa, while our data show that, in high-risk PCa, a low number of TNC-producing cells were associated with poor prognosis, possibly due to changes in tissue remodeling and, thus, variable TNC levels.

These findings were corroborated by the external clinical cohorts of patients [22,23,49,50] (Grasso et al., Lapointe et al., Taylor et al. and Varambally et al.) showing that TNC levels are downregulated during the disease progression from primary to metastasis. Based on differential expression analysis, we identified clusters of stroma signatures based on androgen-(in)dependent responses (C1-C4). TNC is a component of the C1, C3 and C4 signatures. In silico validation of the identified prostate cancer-specific stroma expression signatures on additional clinical cohorts showed the potential for patient stratification. A common feature of the majority of the four clusters of gene lists tested indicated a low stroma signature score in the advanced disease stage and a correlation with disease progression (metastasis). This was the case also for previously published stroma signatures [7,21] (Tyekucheva et al. 2017 and Mo et al. 2017) when compared to our gene sets, perhaps due to the reduced stroma content in low-differentiated, advanced PCa stage. The signature most related to the androgen-independent stage (C4) positively correlated with the Gleason score in primary tissues from TCGA but not in the metastatic cohort of the Taylor dataset. Instead, we identified a 50-gene stroma signature (C1, derived from the most androgen-responsive stroma genes), which positively correlates with the disease progression, Gleason score and poor prognosis survival, consistently on all patient cohorts evaluated, both the primary and metastasis stages.

The regime that a metastatic, stroma-specific molecular signature may be detectable in the PCa site either prior to or during metastasis will most likely require not a single marker approach but a combination of biochemical and histological markers, taking into consideration dual tumor–stroma interactions in order to provide prognostic tools for improved patient stratification after the initial PCa diagnosis and preventive surveillance for metastasis risk.
