**4. Discussion**

The reduced AR dependency of PCa cells is an important clinical development because of its association with progression to CRPC. In this study, we found that fibroblast-secreted AR-activating factors preserved AR signaling in E9 cells after ADT, indicating that these PCa cells could be controlled by ADT. In contrast, loss of fibroblast-dependent AR activation in F10 cells may be responsible for the development and progression of CRPC.

Development and progression of CRPC after ADT are mediated by multiple molecular mechanisms, classified as adaptation to a low-concentration androgen environment caused by ADT, or clonal selection [28–31]. Androgen-insensitive PCa cells can be generated by adaptation of androgen-sensitive PCa cells to a low androgen environment. In contrast, ADT results in the expansion of androgen-insensitive PCa cells, which coexist with androgen-sensitive PCa cell populations in PCa tissue, i.e., clonal selection of androgen-insensitive PCa cells [28]. PCa tissue consists of heterogeneous cell populations. Tumor heterogeneity is reflected in the increased subclonal populations in PCa tissue [32]. These subclones may interact in complex ways with each other or with the surrounding microenvironment. Thus, we hypothesize that tumor heterogeneity in PCa tissue is an extremely important phenomenon not only for understanding tumor progression but also for developing truly personalized treatment regimens for patients with PCa.

To compare the biochemical characteristics of androgen-sensitive and -insensitive PCa cells, we generated three sublines from androgen-sensitive AR-positive LNCaP cells: E9 and F10 cells (showing low androgen sensitivity) and AIDL cells (showing androgen insensitivity) [22–24]. E9 cells are less sensitive to androgen-related responses, such as growth stimulation and PSA production, than parental LNCaP cells [22]. Moreover, E9 cells have a more aggressive tumorigenic phenotype than parental LNCaP cells in vivo. We have previously investigated the mechanisms underlying the low androgen sensitivity of E9 cells and found that decreased phosphorylation of Akt was associated with low androgen sensitivity of E9 cells [33]. The Akt and p44/42 MAPK pathways are known to be associated with the regulation of androgen responses [34,35]. We also demonstrated that PSA production was significantly decreased in parental LNCaP cells when Akt phosphorylation was suppressed by phosphatidylinositol 3-kinase or Akt inhibitors [33]. Thus, E9 cells may be a useful model to reflect high-grade Gleason tumors with low phosphorylation of Akt. Similar to E9 cells, F10 cells are also less sensitive to androgen-related responses than parental LNCaP cells and have a more aggressive tumorigenic phenotype than parental LNCaP cells in vivo [23]. Interestingly, F10 cells can survive under low-pH/low-nutrient conditions, whereas parental LNCaP cells show significant cell death under such conditions. The intratumor environment is characterized by low-pH, low-nutrient, and chronic hypoxic conditions owing to poor vascular development [36,37]. Thus, we sugges<sup>t</sup> that F10 cells may be a useful model to determine the mechanisms underlying their adaptation to a low-pH/low-nutrient environment. In contrast to E9 and F10 cells, AIDL cells are insensitive to androgen-related responses [24]. Parental LNCaP cells harbor an AR mutation at codon 877 (T877A). In addition to the T877A mutation, we found that AIDL cells harbored a point mutation at codon 741 (W741C), suggesting that the T877A/W741C double mutation may be responsible for the androgen insensitivity of AIDL cells [38]. Thus, AIDL cells may be a useful model to investigate the mechanisms underlying the mutated AR in PCa cells.

In PCa, the tumor microenvironment is highly complex and heterogeneous and is composed of carcinoma-associated fibroblasts (CAFs) as well as epithelial cancer cells that infiltrate into the surrounding tumor stroma, referred to as the reactive stroma [39]. This heterogeneous stromal component of PCa tissues contains multiple populations of fibroblasts that are associated with tumorigenesis [40,41]. CAFs contribute to the malignancy of PCa cells by enhancing the proliferation and invasion of cancer cells and promoting angiogenesis in tumors [42]. Thus, inhibition of CAF generation and function in PCa tissue could be a new target for controlling primary cancer progression. Importantly, most fibroblasts in the prostate stroma are negative for AR [15–17], and the phenotypes of human PCa fibroblasts are strongly heterogeneous [13]. CAFs secrete abundant growth factors and

cytokines, which enhance the proliferation of PCa cells. However, the proliferation of PCa cells is regulated by AR signaling, suggesting that stromal paracrine factors derived from CAFs can activate AR signaling in PCa cells. In patients with CRPC, PCa cells can grow in the absence of androgen, indicating that AR signaling in PCa cells is activated by CAF-derived growth factors and cytokines instead of androgen. Therefore, CAFs could be an important target to prevent androgen-independent outgrowth.

In the clinical setting, serum PSA is the most useful biomarker to detect PCa. However, increased levels of serum PSA are also observed in cases of benign prostatic hyperplasia or inflammation of the prostate. PSA is a serine protease and member of the tissue kallikrein family of proteases and is produced in both normal luminal epithelial cells and well-differentiated PCa cells [43]. Transcription of the *PSA* gene is normally regulated by androgens through activation of AR signaling. In addition to androgens, PSA expression is induced through activation of AR signaling by CAF-derived growth factors and cytokines. In our laboratory, Sasaki et al. reported that fibroblasts directly affected PSA expression in LNCaP cells cocultured in vitro [17]. Among various CAF-derived growth factors and cytokines, we confirmed that EGF, IGF-1, and IL-6 directly increased PSA expression in LNCaP cells, suggesting that soluble factors derived from fibroblasts may function as AR-activating factors in the absence of androgen.

In our previous work, Sasaki et al. found that the PSA kinetics after ADT were not an accurate prognostic marker when considering serum PSA levels after ADT to determine the number of viable PCa cells [44,45]. Compared with rapid decreases in PSA after ADT, prolonged PSA decreases after ADT can predict favorable progression-free survival and overall survival. In this study, AR signaling in E9 cells, but not in F10 cells, was activated by paracrine signals derived from fibroblasts, suggesting that the androgen sensitivity of PCa cells may not reflect the AR dependency of PCa cells. Preservation of AR signaling after ADT may have an important role in maintaining the AR dependency of PCa cells. Thus, fibroblast-dependent AR activation after ADT may cause persistent activation of AR signaling in PCa cells, preventing loss of AR dependency after ADT. Notably, Sasaki et al. demonstrated that fibroblasts could enhance the treatment efficacy of ADT during in vivo tumorigenesis, resulting in a more favorable prognosis, e.g., prolonged serum PSA decline and maintenance of the efficacy of ADT [17]. Other researchers also demonstrated that normal human fibroblasts could inhibit the proliferation of tumor cells [46,47]. We still know very little about the tumor-promoting CAFs and the factors that distinguish CAFs from other fibroblasts found in the same tissue. Future studies are needed to identify the specific profiles of fibroblast-derived factors responsible for disease progression. Additionally, several AR variants (ARVs), derived from alternative splicing of the AR transcript, have been identified [48–51]. ARVs may emerge as more common mediators of androgen-independent and AR-dependent tumor progression, although their functions are still unclear. AR-V7 is a major splice variant expressed in human PCa that is associated with the development and progression of CRPC [52]. In human PCa cell lines, expression of AR-V7 mediates resistance to a new generation of AR-targeted therapies, such as enzalutamide and abiraterone [53]. In addition, Shimizu et al. recently reported that knockdown of AR-V7 in LNCaP95-DR cells did not restore sensitivity to docetaxel and cabazitaxel, suggesting that AR-V7 may be not involved in taxane resistance [54]. Thus, expression of AR-V7 has been proposed for the assessment of suitability for taxane chemotherapy [55].

With regard to the androgen sensitivity of PCa cells, our data showed the following results, in the presence or absence of fibroblasts: (1) tumor growth of E9 cells was significantly diminished after ADT compared with that in the sham group; (2) tumor growth of F10 cells was temporally reduced after ADT, but was restarted under androgen deprivation; and (3) tumor growth of AIDL cells was not decreased after ADT. These results established some important clinical concepts. For example, to treat certain PCa cells (e.g., E9 cells), fibroblast-target therapy should be avoided because of the preservation of AR signaling after ADT. Additionally, for certain PCa cells expressing AR-V7 (e.g., F10 cells), ADT may not be effective for treating CRPC. In these cells, the responsiveness to fibroblasts may not be associated with tumor growth, and the efficacy of ADT may be limited. Finally, in androgen-insensitive PCa cells (e.g., AIDL cells), ADT is completely useless because of the AR independence of PCa cells.

Identification of improved, personalized treatments will be supported by recent major progress in the molecular characterization of early- and late-stage PCa. Indeed, such advancements have already led to novel classifications of prostate tumors based on gene expression profiles and mutation status and should greatly facilitate the choice of novel targeted therapies best tailored to the needs of patients [56,57], particularly for individual subgroups of patients, representing a major step towards personalized medicine adapted to the individual needs of patients with PCa. Selecting the optimal drug or drug sequence and combination for PCa treatment will be improved by the identification of molecular biomarkers predictive of response and progression. Similar to breast cancer, subdivisions of luminal A, luminal B, and basal subtypes, which exhibit di fferent clinical prognoses and responses to ADT, have been proposed for PCa [58]. Such a classification may greatly support treatment choices for early- and late-stage disease and ultimately improve the overall survival rates and quality of life in patients with PCa. As we learn more about the genetic heterogeneity of PCa cells and mechanisms of treatment resistance, we expect that markers for treatment choice and response will be developed and validated to better guide treatment decisions [59,60].

Using three sublines of androgen-sensitive LNCaP cells, we demonstrated that loss of fibroblast-dependent AR activation in PCa cells (e.g., F10 cells) may be responsible for the development and progression of CRPC. In the absence of androgen, the AR dependence of PCa cells interacting with fibroblasts reflected the e fficacy of ADT; for example, E9 cells could be controlled by ADT. To choose appropriate patients with advanced PCa for ADT, it is necessary to evaluate the degree of AR dependence in PCa cells interacting with fibroblasts before ADT is started. Further investigations are needed to identify clear molecular markers using biopsy tissue samples or bodily fluid samples derived from patients with advanced PCa.
