E-cadherin-mediated survival of androgen-receptor-expressing secretory prostate epithelial cells derived from a stratified in vitro differentiation model

Epithelial cells serve several vital functions. For instance, all epithelial cells act as a barrier to protect organs from external environmental assault, as exemplified by the skin. Intestinal epithelial cells are required for the absorption of nutrients, and mammary and prostate epithelial cells are primarily secretory. Proper regulation of epithelial differentiation is crucial for the development and maintenance of barrier and organ function. Differentiation of epithelial cells has been extensively characterized in the epidermis. The basal layer of the epidermis consists of proliferating keratinocytes that adhere to a basement membrane via integrins. Loss of basal cell adhesion through integrin β1 initiates terminal differentiation, resulting in flattening of the cells, expression of differentiation proteins, and subsequent cornification, which ultimately produces several distinct stratified cell layers that make up the epidermis (Levy et al., 2000; Lippens et al., 2005).

The epithelium of the human prostate consists of two cell layers, a basal layer and a secretory layer. Similar to other stratified epithelium, prostate basal cells are mitotic and adhere to a basement membrane (Knox et al., 1994; Uzgare et al., 2004; van Leenders and Schalken, 2003). Prostate basal cells give rise to terminally differentiated secretory cells (Knox et al., 1994; Uzgare et al., 2004; van Leenders and Schalken, 2003). However, unlike other epithelia, prostate epithelial cell differentiation is regulated by androgen signaling (Berger et al., 2004; Cunha et al., 1987; Heer et al., 2007; Ling et al., 2001; Whitacre et al., 2002). The androgen receptor (AR) is a nuclear transcription factor activated in response to the steroid hormone androgen (Lamb et al., 2001). AR is expressed only in the differentiated secretory cells and not in the basal cells (Lamb et al., 2001). It is unclear exactly how androgen regulates epithelial differentiation. However, tissue combination studies from AR-null mice suggest that androgen stimulation of AR in the early developing mesenchyme, and not the epithelium, is solely responsible for the induction of epithelial morphogenesis in vivo (Cunha et al., 2004).

Androgen also appears to be important for secretory cell survival, in that anti-androgen therapies specifically kill the secretory cells, leaving the basal cells intact (Denis and Griffiths, 2000). Furthermore, restoration of androgens results in regeneration of the secretory cell compartment. However, tissue recombination experiments, as well as studies using conditional knockout mice that lack AR only in prostate epithelium, suggest that AR does not directly regulate epithelial survival (Cunha et al., 2004; Wu et al., 2007). Instead, androgen stimulation of the AR-positive stromal cells of the prostate might induce secreted factors that regulate secretory cell survival. Keratinocyte growth factor (KGF) and FGF10 are two factors secreted by the stromal cells, though not in an androgen-dependent manner (Alarid et al., 1994; Cunha et al., 2004; Sugimura et al., 1996; Thomson, 2001). KGF and FGF10 are both involved in murine prostate organogenesis and can induce differentiation of isolated prostate epithelial cells (Alarid et al., 1994; Cooke et al., 1991; Cunha, 1996; Donjacour et al., 2003; Heer et al., 2006; McKeenhan, 1991; Sugimura et al., 1996). In some cases, KGF can substitute for androgens and it is likely that KGF and AR signaling pathways interact (Thomson et al., 1997). KGF has also been reported to promote differentiation and survival of the epithelium of the skin, lung and eye (Geiger et al., 2005; Marchese et al., 1997; Ray et al., 2003). KGF acts specifically on epithelial cells and has been reported to activate p38 MAPK signaling (Heer et al., 2006).

Clarification of the roles of androgen and KGF in prostate epithelial differentiation and survival has been hampered by our inability to culture normal differentiated AR-expressing secretory cells in vitro. Prostate epithelial cells (PECs) cultured from normal human prostate tissue consist primarily of AR-negative basal cells and their transient amplifying derivatives. Previous studies in our lab have demonstrated that survival of cultured PECs is specifically mediated through α3β1-integrin-dependent adhesion (Edick et al., 2007). Similarly, basal keratinocytes are dependent on α3β1 integrin for their survival (Manohar et al., 2004). During keratinocyte differentiation, basal cells lose integrin expression as well as adhesion to matrix as they are extruded to the upper layers of the skin (Watt, 2002). In suprabasal keratinocytes, as well as in other epithelia, cell-cell adhesion structures such as E-cadherin appear to promote survival through phosphoinositide 3-kinase (PI3K) signaling, and when PI3K signaling is lost these cells die (Calautti et al., 2005; Espada et al., 2009; Rivard, 2009). Whether the same survival mechanisms are operative in differentiated secretory prostate epithelial cells is unknown, and the role of KGF or androgen in prostate epithelial cell survival remains unresolved.

In this study, confluent cultured primary prostate basal epithelial cells were induced to differentiate following treatment with KGF and androgen. After 2 weeks, differentiated AR-expressing secretory cells appeared as a secondary cell layer above the basal cells. This model was used to identify the signaling pathways important for prostate secretory cell survival. This new model will serve as a valuable tool for understanding the biology of prostate secretory epithelial cells, a cell population previously not available for extensive analysis.

Previous studies have demonstrated that KGF might be an important epithelium differentiation factor in many tissues, including prostate epithelium (Alarid et al., 1994; Cunha, 1996; Heer et al., 2006; Peehl et al., 1996; Sugimura et al., 1996). Androgen, acting via the androgen receptor, also plays an important role in prostate epithelial cell differentiation (Berger et al., 2004; Cunha et al., 1987; Heer et al., 2007; Ling et al., 2001; Whitacre et al., 2002). To determine if the combination of KGF and androgen is sufficient to induce differentiation of prostate cells grown in culture, human primary basal prostate epithelial cells (PECs) grown to confluency in monolayer cell cultures were treated with 10 ng/ml KGF and 5-10 nM androgen (DHT). Culturing the cells for 10-15 days with KGF and DHT resulted in the formation of stratified cell patches consisting of at least two cell layers, resembling the bilayer of basal and secretory cells observed in the prostate epithelium in vivo (Fig. 1A-C).

To determine if the stratified cells expressed differentiation markers specific to prostate secretory cells, expression of AR and the AR-target protein prostate-specific antigen (PSA) were examined by fluorescence confocal microscopy. Cells in a higher z-plane than the bottom cells, stained positive for AR and PSA (Fig. 1B). AR expression was both nuclear and cytoplasmic, whereas the secreted protein PSA had the expected cytoplasmic localization (Fig. 1B). AR expression was uniform throughout the top cells, whereas PSA expression was often concentrated at the upper membrane of the top-most cells, consistent with that of a secreted protein (not shown). Neither AR nor PSA was found in the bottom cells (Fig. 1B). Additionally, the AR-regulated proteins Nkx3.1 and TMPRSS2 (Bowen et al., 2000; Lin et al., 1999a; Murtha et al., 1993; Young et al., 1992) were expressed in top cells and not in bottom cells (Fig. 1C).

To determine the extent to which androgen stimulation contributes to PEC differentiation, PECs were treated for 10-14 days with KGF in the presence or absence of DHT, and the expression of AR, AR-target proteins, and differentiation cell markers was monitored. PSA, Nkx3.1 and TMPRSS2 were only expressed when DHT was present (Fig. 1D, PSA and TMPRSS2 not shown). Intriguingly, cytokeratin markers, K18 and K19, were also expressed only in the presence of androgen (Fig. 1D, K18 data not shown). Furthermore, there was a dramatic increase in AR expression itself when DHT was present.

KGF, in the absence of DHT, was sufficient to induce formation of stratified cells, with maximal formation occurring between10 and 15 days. PECs treated with KGF in the presence of KGF-blocking antibody did not stratify. Confluency of the cultures was essential. Subconfluent cells treated with KGF and DHT did not form stratified clusters. KGF-induced stratification occurred equally efficiently, with or without the supplementary bovine pituitary extract (BPE) and EGF in the culture medium. Occasionally, a few small stratified clusters appeared in BPE-containing medium without KGF treatment, suggesting the presence of low levels of KGF and/or an additional unknown factor(s) in BPE that can promote differentiation at a low efficiency. KGF-blocking antibodies prevented the appearance of these occasional clusters. The optimal concentration of KGF was 10 ng/ml. Lower doses (1-5 ng/ml) resulted in fewer clusters and higher doses (20-50 ng/ml) did not generate more clusters. DHT alone was not sufficient to induce stratification. DHT plus KGF treatment dramatically increased the number of top cells seen after 15 days. DHT was required for expression of androgen-dependent markers in the top cells. FGF10, a functionally related FGF family member shown to be important for prostate development in vivo (Donjacour et al., 2003; Igarashi et al., 1998), could also induce PEC differentiation in the presence of DHT. Differentiation was reproducibly observed in cells derived from two different patients at three different passage numbers (passages 2, 3 and 4). It was observed however, that once cells reached passage 5, the efficiency of differentiation was dramatically reduced. Furthermore, we were able to induce differentiation in an immortalized cell line derived from a third patient. We observed that these more proliferative immortalized cultures took a few days longer to reach maximal differentiation.

Markers specific to basal and differentiated epithelial cells populations were examined in the stratified cultures. The basal markers Bcl-2, K5 and K14 (McDonnell et al., 1992; Wang et al., 2001) were expressed predominantly in the bottom cells; occasionally a few K5- and K14-positive cells were seen in the top cells (Fig. 2A,B, K14 not shown). Basal marker p63 (Parsons et al., 2001; Signoretti et al., 2000) was associated only with bottom cells (Fig. 2A). EGFR, which is predominately expressed in basal cells (Sherwood and Lee, 1995), was associated primarily with bottom cells (not shown). Epithelial cell markers K19 and PMSA were expressed only in the top cells and not in the bottom cells (Fig. 2B,C). K18, as well as the cell cycle inhibitor p27 (Kip1) (Peehl et al., 1994; Tsihlias et al., 1998; Wernert et al., 1987; Yang et al., 1998), was expressed predominately in the top cells (Fig. 2C).

Consistent with previous observations of differentiating epithelium in vitro and in vivo (Gustafson et al., 2006; Heer et al., 2006; Levy et al., 2000; Li et al., 2008), epifluorescence and confocal imaging revealed that the subpopulation of the cells undergoing differentiation lost expression of many integrins, including α2, α3, α6, β1 and β4 (Fig. 3A,B). Basal cells also expressed αv-, but not β3- or β5-integrin subunits. None of these integrins were present in the differentiated cells (not shown). Cultured PECs secrete and organize a laminin 5 (LM5)-rich matrix (Yu et al., 2004); the differentiating cell population that lost integrin expression also no longer produced LM5 (Fig. 3A,B). Although it appears, by confocal imaging, that the cells directly below the top cells do not express integrin or LM5, it is possible that there is incomplete antibody penetrance into the lower cells. To address this, a timecourse study was performed. We observed a decrease in LM5 expression as early as 3 days after KGF and DHT treatment and a complete loss after 8 days. At 8 days decreased β1 integrin expression was observed in LM5-negative cells prior to formation of the second cell layer (supplementary material Fig. S1A). Therefore, cells directly underneath the top layer also lose LM5 and integrin expression. LM5 loss might be the trigger that initiates differentiation.

AR expression could be detected by immunoblotting of cell lysates from whole cultures treated with KGF and DHT (Fig. 4A). Expression of the androgen-dependent secreted proteins, KLK2 and PSA, was monitored in differentiated cultures by RT-PCR. KLK2 and PSA mRNAs were present only when DHT was present in the culture (Fig. 4B). Furthermore, secreted PSA, up to 0.8 ng/ml, could be detected by ELISA (Fig. 4C). PSA secretion required androgen and increased with increasing DHT concentration. The expression and secretion of an androgen-regulated protein in an androgen-dependent manner indicates the presence of differentiated prostate secretory cells in the culture, and that AR is functional and regulates expression of differentiation markers.

Overall, this in vitro differentiation model recapitulates many aspects of in vivo differentiation as assessed by the specific markers (Fig. 4D). In addition to the induction of markers common to most differentiating epithelial cells, the presence of DHT markedly stimulates the expression of markers unique to prostate secretory epithelial cells. Hereafter when referring to this model, the AR-expressing top cells will be referred to as secretory-like cells and the AR-negative bottom cells as basal cells.

Treatment of differentiated cultures with dissociation buffer preferentially dislodges the secretory-like cells. FACS analysis indicates that 96.6% (±0.8%) of the isolated dislodged population is negative for cell surface α6 integrin, whereas 97.19% (±1.70%) of the cells not dislodged are positive for α6 integrin (Fig. 5A). Further FACS sorting based on surface staining of α6 integrin and TMPRSS2 revealed that on average 87.92% (±3.71%) of the α6-integrin-negative cells were positive for TMPRSS2. A representative example is provided in Fig. 5B. Immunoblotting of separated cells indicated that some remaining basal cells expressed AR as well as full-length TMPRSS2 protein; however, only the secretory-like cells expressed the cleaved and activated form of TMPRSS2 (Fig. 5C) (Wilson et al., 2005). Conversely, only the basal cells expressed Bcl-2 and EGFR, whereas K5 was predominately found in the basal cells (Fig. 5D).

In previous studies, we demonstrated that integrin-mediated activation of EGFR and downstream signaling to ERK, but not PI3K signaling, is required for the survival of basal PECs (Edick et al., 2007). However, the differentiated secretory-like PECs have lost integrin expression, no longer adhere to the LM5 matrix, and have significantly lower levels of EGFR, suggesting that other survival pathways must be important for secretory cell survival. It has been suggested that secretory cell survival might be dependent on stromal-derived growth factors, including KGF (Kurita et al., 2001). One possibility is that the KGF used to induce differentiation, might also be necessary for survival. To test this, the KGF receptor FGFR2IIIb (Giri et al., 1999) mRNA levels were analyzed in the isolated secretory-like cells and basal cells by RT-PCR. Only the basal cells expressed FGFR2IIIb mRNA (Fig. 6A). Furthermore, removal of KGF after 15 days of differentiation did not induce cell death (not shown). Thus it is unlikely that KGF is regulating cell survival in the secretory-like cells.

Dissociated secretory-like cells and the remaining basal cells were screened for ERK and AKT activation by immunoblotting. Active ERK was present only in the basal cells, but not in the secretory-like cells (Fig. 6B). Activated AKT was present in both types of cells (Fig. 6C). Thus, ERK signaling probably does not regulate survival in differentiated cells, whereas the PI3K pathway could. Since the differentiated cells remain adherent to the bottom basal cells, we also investigated whether there is an increase in expression of the cell-cell adhesion molecule E-cadherin in the secretory-like cells. Compared with the basal cells, E-cadherin levels were elevated in the secretory cell population that also does not express α6β1 integrin (Fig. 6D). E-cadherin can lead to activation of PI3K signaling in skin and colonic epithelium as well as in some tumor cell lines (Calautti et al., 2005; Hofmann et al., 2007; Pang et al., 2005). Blocking antibodies to E-cadherin suppressed AKT activity in both the secretory-like (Fig. 6E) and the basal cells (not shown).

The relative importance of the different signaling pathways on secretory-like cell survival was investigated. Fourteen-day KGF and DHT-differentiated cultures were placed in KGF- and DHT-free basal medium without any pituitary extract or EGF supplement for 72 hours to reduce any signaling induced by the growth medium (Fig. 7A). Visually, the starved cell cultures appeared viable, and the upper secretory-like cell layer remained intact (data not shown). Then the starved differentiated cultures were treated with specific inhibitors in the presence or absence of freshly added DHT or KGF and analyzed over a 72-hour timecourse. Cell death was measured in the upper secretory-like cell layer by immunostaining for active caspase 3/7, TUNEL staining or propidium iodide (PI) uptake. Staining was quantified as described in Materials and Methods. Inhibition of PI3K signaling with LY294002 resulted in maximal secretory-like cell death at 72 hours, where 60% of the cells stained positive for PI (Fig. 7B). Furthermore, inhibition of PI3K, but not EGFR, induced a 7.0- to 7.5-fold increase in secretory cell caspase 3 activity (Fig. 7C), and a 5.5- to 5.7-fold increase in TUNEL staining (Fig. 7D; supplementary material Fig. S1B). Maximal annexin V staining was observed 66 hours after LY294002 treatment (not shown). Secretory-like cell survival was not dependent on DHT or KGF, and addition of DHT or KGF was unable to promote cell survival in the absence of PI3K signaling (Fig. 7B-D). Although KGF should not be present in the media, and prostate epithelial cells have been reported not to produce KGF, KGF-blocking antibodies were used to prevent any endogenous or remaining KGF from promoting cell survival. KGF-blocking antibodies had no effect on cell survival (data not shown). KGF has been reported to activate p38, and Jnk can promote survival during stress (Heer et al., 2006; Leppä and Bohmann, 1999; Mehta et al., 2001). Inhibiting p38 with SB202190, JNK with 420119, or ERK with PD98059 did not result in cell death, suggesting these pathways are not critical for secretory cell survival (supplementary material Fig. S1C). The lack of effect of the inhibitors on cell survival was not due to a failure to inhibit signaling, as the concentrations of drugs used here did effectively block signaling to their specific targets in basal cells.

Cell-cell adhesion via E-cadherin was inhibited by treatment of differentiated cells with two different preparations (lots) of E-cadherin-blocking antibodies. Inhibition of cell-cell adhesion with one lot of E-cadherin-blocking antibody resulted in maximal cell death at 48 hours with over 80% of the cells staining positive for PI (Fig. 7E). By 66 hours, no secretory-like cells remained in the cultures. A second lot of E-cadherin antibody resulted in a seven- to eightfold increase in TUNEL staining 72 hours after treatment (Fig. 7F). The presence of DHT or KGF could not protect cells from death due to loss of E-cadherin function. No cell death was observed in the lower basal cells. Furthermore, blocking E-cadherin lead to a decrease in AKT activation (see Fig. 6E), indicating that cell-cell adhesion mediated by E-cadherin promotes secretory-like cell survival through PI3K signaling.

Although DHT was not important for survival of the differentiated secretory-like cells, it is theoretically possible that AR, acting via an androgen-independent mechanism might still be important for cell survival. To address this, 14-day KGF- and DHT-differentiated cultures were transfected with an AR-specific siRNA pool or a scrambled siRNA sequence. Confocal imaging of the transfected cells 72 hours later demonstrated the absence of AR expression in the upper cells (Fig. 8A). Absence of AR expression also resulted in loss of androgen-dependent cell markers such as Nkx3.1 and K19 (Fig. 8A). Cell viability of the AR siRNA-treated cells was assessed by TUNEL staining. Loss of AR had no effect on secretory-like cell viability (Fig. 8B). Thus, AR and androgen signaling are not required to maintain the viability of differentiated secretory-like cells derived from our in vitro culture system.

By treating cultured primary prostate basal epithelial cells with androgen and KGF, we have established an in vitro differentiation model of the prostate epithelium. The differentiated cells in our culture system possess the important features of terminally differentiated secretory prostate epithelial cells in vivo: they do not proliferate, they adhere to a basal cell layer and not to the basement membrane, they express AR protein, and they respond to DHT by inducing AR-dependent genes. Specifically, the cells express androgen-sensitive proteins, such as KLK2, PSA, Nkx3.1, PMSA and TMPRSS2. In addition, cleaved TMPRSS2 is present in the upper, but not the lower cells and PSA is secreted into the culture medium. Furthermore, cytokeratin K18 and K19 expression was found to be dependent on androgen. K18 expression has previously been reported to be regulated by androgen (Heer et al., 2007; Ling et al., 2001), and K19 has been suggested to be responsive to estrogen (Choi et al., 2000); however, both K18 and K19 promoters lack classical androgen response elements, making the mechanism of regulation unclear.

Further evidence for terminal differentiation is that the cells did not revert to basal cells when isolated and re-plated, and they failed to reattach, probably because of continued loss of integrin and/or matrix expression. Furthermore, after 21-25 days in culture the upper cells sloughed off and a few activated caspase-3-positive cells were seen in the aging cultures (data not shown), similar to what is observed in vivo. Oddly, no more differentiated cells reappeared. Only about 20% of the cells appeared to be capable of undergoing differentiation, suggesting that the differentiated cells are derived from a distinct subpopulation of basal cells. The lack of continued differentiation after 25 days may indicate depletion of these special cells and a lack of ability to renew. The population of differentiation-competent cells is not likely to be stem cells, since 20% of the cells are capable of undergoing differentiation. However, we cannot rule out the possibility that these cells arose from some stem cell-like progenitor within the culture. Further analysis would be required to determine if the progenitors are analogous to the Nkx3.1-positive luminal stem cell recently described (Wang et al., 2009). However, whatever the progenitor, it apparently cannot renew in the context of our culture conditions.

Although many aspects of the differentiated cells recapitulate what is observed in vivo, there still remain some differences. For instance, the distribution of AR demonstrates a significant amount of cytoplasmic expression in the in vitro culture system, whereas in vivo AR is primarily nuclear. Another difference is the absence of columnar cells. In addition, a few K5- and/or K14-positive cells were sometimes seen in the upper layer, which has also been reported in another differentiation model (van Leenders et al., 2000). Hence, we cannot unequivocally say whether our secretory-like cells represent completely terminally differentiated prostate cells and there are still some distinctive morphological differences between our cultures and what is seen in the prostate gland in vivo.

Other studies have reported on prostate epithelial differentiation in vitro. Although these studies were informative, they were limited since AR and AR-regulated proteins were not expressed (Dalrymple et al., 2005; Danielpour, 1999; Garraway et al., 2003; Gu et al., 2006; Gustafson et al., 2006; Yasunaga et al., 2001). A few studies have reported seeing stratified layering similar to ours after treating prostate epithelial cells in vitro with retinoic acid, FGF and/or insulin (Gustafson et al., 2006; Peehl et al., 1994; Robinson et al., 1998; van Leenders et al., 2000); however, in these models the top layer of cells either failed to express AR or still expressed basal markers. In our model, the top secretory-like cells expressed AR and lost basal marker expression. In one case, gland-like buds and extensions were observed to form from confluent cell cultures, reminiscent of acini structures in overall shape but without lumens (van Leenders et al., 2000). We have also observed cases where cells appear to form mounds. By confocal imaging, some of them appear to have formed a hollow mound (data not shown). A recent study demonstrated that co-treatment of prostate basal cells with the monoamine oxidase A inhibitor clorgyline, 1,25-dihydroxyvitamin D₃, all-trans retinoic acid and TGF-β1 induced AR expression and loss of basal marker K14 (Zhao et al., 2008), suggesting that there might be alternative mechanisms to inducing prostate epithelial cell differentiation.

In contrast to other published systems, we have demonstrated that our model can be utilized for biochemical and genetic manipulation. It is amenable to treatment with pharmacological inhibitors or siRNA to study signaling and biological pathways. Furthermore, exploitation of differential cell surface markers and adhesion properties can be used to separate basal from secretory-like cells to separately analyze RNA and protein expression.

It is unknown whether AR represses integrin expression or whether loss of integrin expression must precede expression of AR. Unpublished data from our laboratory and others demonstrates that re-expression of AR in prostate cancer cell lines results in decreased integrin expression (Bonaccorsi et al., 2000; Nagakawa et al., 2004). However, in our model we observed that not all integrin-negative cells were AR positive, suggesting that integrin loss might precede AR expression. Furthermore, LM5 matrix loss preceded integrin loss, which preceded stratification and robust AR expression in our timecourse studies. Heer et al. have demonstrated that blocking integrin β1 is sufficient to induce partial differentiation; however, cells do not reach terminal differentiation since the cells do not express AR-regulated genes (Heer et al., 2006). This suggests that loss of adhesion can initiate early differentiation and may even be required, but that integrin loss alone is not sufficient for terminal differentiation. By contrast, unbound integrin β1 is sufficient to initiate terminal differentiation in keratinocytes (Levy et al., 2000; Watt, 2002). In mammary epithelium, however, loss of integrin β1 suppresses differentiation (Naylor et al., 2005).

Interestingly, in most of the reported prostate differentiation models (including ours), confluent cultures were necessary for stratification. In addition, previous studies suggest that cell cycle inhibition is a prerequisite for expression of secretory cell markers K18, K19 and AR (Danielpour, 1999; Garraway et al., 2003; Gustafson et al., 2006; Litvinov et al., 2006). We similarly saw a loss in cell proliferation in the differentiating cell population (data not shown). This led us to develop the following model for prostate differentiation (Fig. 9). Basal cells are proliferative and a subset begins to undergo growth arrest once the cells are confluent. Treatment with KGF causes a select population of cells, perhaps those that express higher levels of the KGF receptor FGFR2IIIb (Giri et al., 1999), to lose LM5 and then integrin expression, causing the cells to detach. Integrin loss and detachment might then trigger low AR expression. AR expression was not detectable by immunostaining in cultures treated with only KGF, in which integrin expression was lost; however, some AR expression was detectable in the basal cells from the differentiated cultures by immunoblotting. The presence of androgen in the culture appears to be necessary to allow the integrin-deficient cells to express AR at a higher level, which then turns on AR-dependent differentiation-specific genes.

Work by Heer et al. suggests that AR might be expressed at low levels in primary prostate epithelial cells and is rapidly degraded by the proteosome (Heer et al., 2007); hence androgen treatment might stabilize and/or help drive production of AR protein. In fact AR mRNA has been detected in some cultured prostate epithelial cells (Litvinov et al., 2006). However, in our studies and those of others, androgen alone is not very effective in inducing AR expression (Litvinov et al., 2006). Thus, additional events are required to induce stable AR expression even in the presence of androgen. Reduced cell proliferation caused by strong growth suppression or loss of cell adhesion, which is also growth suppressive, might be necessary. Significant increases in AR expression can be detected in isolated suspended cells in the presence of androgen (Heer et al., 2007), thus supporting cell detachment as a potential mechanism required for stabilizing AR.

Previous work from our laboratory has demonstrated that integrin-mediated survival of primary prostate basal cells requires integrin-induced EGFR signaling to ERK, but not PI3K signaling (Edick et al., 2007). In this study we have expanded our analysis of survival mechanisms to secretory-like prostate epithelial cells and demonstrated that secretory-like cells depend on a non-integrin-dependent mechanism for cell survival that involves cell-cell interactions through E-cadherin. Interestingly, there is switch from ERK-dependent survival in the basal cells to PI3K-dependent survival in the secretory-like cells. In the secretory-like cells EGFR levels dropped dramatically and EGFR-dependent signaling to PI3K was not required for survival (blocking EGFR had no effect on secretory cell survival). Interestingly, in prostate cancer, there appears to be a strong dependence on PI3K signaling for survival, as these cells tend to acquire mutations in Pten, a negative regulator of PI3K signaling (Bertram et al., 2006; Edick et al., 2007; Lin et al., 1999b; Wen et al., 2000). This suggests that prostate cancer might arise from a more differentiated cell that has already acquired dependence on PI3K for its survival.

In our studies, secretory cell survival was not dependent on the presence of androgen, and knockdown of AR with siRNA in differentiated cells did not induce their death. The lack of dependence on androgen or AR for secretory cell survival in our human culture system is in agreement with genetic and tissue recombination studies in mice. Conditional knockout of AR in mature mouse prostates results in decreased numbers of secretory cells without inducing cell death, suggesting that AR functions to increase secretory cell numbers by promoting differentiation rather than cell survival in mature glands (Wu et al., 2007). Tissue recombination experiments using mesenchyme and epithelium from AR-negative or wild-type mice demonstrate that AR expression in the epithelium is not required for early prostate development, indirectly ruling out a role for AR in epithelial cell survival in newly formed glands (Cunha et al., 2004). Thus, in both models, as well as ours, androgen is responsible for the synthesis of secretory proteins and the secretory function of the prostate.

If androgen and AR do not act cell autonomously to control epithelial cell survival, then why do only the AR-expressing epithelial cells die upon castration-induced androgen deprivation (Evans and Chandler, 1987; Mirosevich et al., 1999)? One possibility is that AR signaling in the stromal cells promotes survival by paracrine factors that act on the epithelial cells (Verhoeven and Swinnen, 1999). In our model the paracrine function of KGF, known to be expressed by stromal cells in vivo, was required for differentiation; however, it was dispensable for cell survival in committed differentiated cells. Thus, the nature of the paracrine survival factor(s) remains undetermined. In our in vitro model, survival was highly dependent on E-cadherin-based cell-cell adhesion and signaling to PI3K. Whether paracrine factors in vivo are responsible for maintaining survival via E-cadherin or whether they act on other pathways remains to be determined.

Our study supports a simpler concept that the role of stromal-derived paracrine factors is to act primarily on the stem and/or basal cells, whose proliferation and regenerative capacity is driven by these factors. As terminally differentiated cells are sloughed into the lumen, basal cells are triggered to proliferate and differentiate to replace the lost cells. Under androgen-ablative conditions, the loss of paracrine factors in the stroma prevents stem cell and/or basal cell renewal and the terminally differentiated cells eventually slough off and are not replaced. Re-administration of androgen restores basal cell proliferation and differentiation, and subsequent restoration of secretory cells. This model would preclude the need for stromal factors acting directly on the secretory cells.

An alternative model to explain castration-induced loss of prostate secretory cells involves the observation that castration reduces blood flow and microvasculature collapse in the gland, inducing a state of hypoxia (Buttyan et al., 2000). It would appear that secretory cells are much more sensitive to such stress than the basal or stromal cells. This might be related to a lack of extracellular matrix support that provides additional survival signaling cues to the basal and stromal cells. Alternatively, hypoxia might affect the production of the paracrine factors required for maintenance of epithelial differentiation or survival.

In summary, we have established an in vitro differentiation model of human prostate epithelium composed of stratified cells that recapitulates many in vivo characteristics of basal and secretory cells, including AR-dependent differentiation and function. This model can be treated with pharmacological inhibitors and siRNA to study biochemical and genetic effects and the differentiated secretory-like cells can be isolated for further analysis. We have further established that although KGF, AR and androgen are important for initiating the differentiation process and AR is important to maintain the androgen-dependent phenotype of secretory-like cells, these factors are not required for survival of the committed differentiated cells. The primary critical mechanism driving cell survival is E-cadherin-based cell-cell adhesion and subsequent activation of the PI3K signaling pathway.

Human primary prostate epithelial cells (PECs) derived from prostectomy specimens were isolated, cultured, and verified to be free of stromal contamination as described previously (Edick et al., 2007; Gmyrek et al., 2001). Specific patient samples used in this study were again verified to be negative for the stromal cell marker smooth muscle actin by immunostaining. PECs were grown in keratinocyte-SFM medium (Invitrogen) supplemented with bovine pituitary extract (BPE) and epidermal growth factor (EGF). Experiments were reproducibly performed in cells derived from two different patients at three different passage numbers (passage 2, 3 and 4). In addition, at least three separate primary cultures from each patient were used. Experiments were verified at least three times for each of the two patients. We were also able to induce differentiation in an immortalized cell line derived from a third patient. The AR-positive prostate cancer cell line LNCaP was purchased from ATCC. LNCaP cells were grown in RPMI 1640 medium (Gibco) supplemented with 10% fetal bovine serum, 2 mM glutamine, 50 IU penicillin, 50 μg/ml streptomycin, 0.225% glucose, 10 mM HEPES, and 1 mM sodium pyruvate.

To induce differentiation, a 10-cm culture dish of confluent PECs was divided equally between three eight-chambered slides (Lab-Tek). Cells were grown in keratinocyte-SFM supplemented with BPE, EGF, 10 ng/ml keratinocyte growth factor (KGF; Calbiochem), and 5-10 nM dihydrotestosterone (DHT; Sigma) for 10-18 days. KGF and DHT were replenished three and five times a week, respectively. For larger-scale experiments, three 10-cm plates of confluent PECs were combined onto one 10-cm dish and treated with KGF and DHT for 21-30 days.

KGF-FGF7 blocking antibody (clone 29522) was purchased from R&D Systems. 2 μg/ml KGF blocking antibody or IgG control was added immediately prior to KGF addition. Differentiation of PECs was then assessed by immunofluorescent staining for differentiation markers.

Whole cultures of differentiated PEC cultures were placed in suspension by washing the cells twice with PBS, treating with cell dissociation buffer (Gibco, Invitrogen) for 5 minutes, then adding TrypLE Express trypsin (Gibco, Invitrogen). Cells were then washed with wash buffer (1% sodium azide, 2% FBS-PBS) and incubated with primary antibodies or control IgG molecules for 1 hour at 4°C. Cells were washed twice and incubated with fluorescently labeled secondary antibodies for 1 hour at4°C in the dark. Cells were washed twice more, and fluorescence was detected using a Becton-Dickinson FACSCalibur four-color flow cytometer with CellQUEST Pro Software v5.2.1 (Becton-Dickinson).

Differentiated PEC cultures were washed with 1 mM EDTA in PBS without calcium or magnesium, and then incubated for 5 minutes with 1 mM EDTA-PBS. Cells were then incubated with cell dissociation buffer (Gibco, Invitrogen) for 6-8 minutes. The top layer of cells could then be removed by pipetting the cell dissociation buffer over the cells; the bottom confluent cell layer remained attached to the culture vessel. The isolated cells were used directly or undifferentiated α6-integrin-expressing cells were separated from the differentiated cells using α6 integrin antibodies and FACS as described above using fluorescently conjugated integrin α6 antibody (BD Pharmingen). Cells were sorted on a Becton-Dickinson FACSAria special order system 12-color flow cytometer using FACSDiVa software v5.2 (Becton-Dickinson).

Total cell lysates were prepared for immunoblotting as previously described (Edick et al., 2007; Miranti, 2002). Briefly, cells were lysed with Triton X-100 lysis buffer and 45-75 μg of total cell lysates in 2× SDS sample buffer were boiled for 10 minutes. Samples were run on SDS polyacrylamide gels following standard SDS-PAGE protocols and transferred to PVDF membrane. Membranes were blocked in 5% BSA in TBST for 2 hours at room temperature, then were probed with primary antibody overnight at 4°C. Membranes were washed three times, and incubated with horseradish peroxide-conjugated secondary antibodies (Bio-Rad) in 5% BSA in TBST for 1 hour at room temperature. After washing an additional three times, signals were visualized using a chemiluminescence reagent with a CCD camera in a Bio-Rad Chemi-Doc Imaging System using Quantity One software v4.5.2 (Bio-Rad).

Antibodies for phospho-specific AKT (S473) or phospho-specific ERK1/2 (T202/Y204) were purchased from Cell Signaling. Antibodies for total ERK were from Becton-Dickinson Transduction Labs and total AKT antibodies have been described previously (Bill et al., 2004). α6 integrin and TMPRSS2 antibody were gifts from Anne Cress (University of Arizona, Phoenix, AZ) and Peter Nelson (Fred Hutchinson Cancer Research Institute, Seattle, WA) (Lucas et al., 2008), respectively. Androgen receptor antibody (441) was purchased from Santa Cruz Biotechnology. E-cadherin antibody (clone HECD1) was purchased from Zymed. Tubulin antibody (clone DM1A) was purchased from Sigma.

Differentiated PEC cultures were fixed with 4% paraformaldehyde (Mallinckrodt Chemicals) for 10 minutes and permeabilized for 4 minutes with 0.2% Triton X-100 (EMD) at room temperature. Cells were then blocked with 10% normal goat serum (Pierce) for 2 hours at room temperature before incubation with primary antibodies overnight at 4°C. Cells were incubated with appropriate secondary antibodies for 1 hour at room temperature. DNA was visualized by staining with Hoechst 33258 (Sigma) for 10 minutes at room temperature. Cells were washed three times with PBS between all steps. Coverslips were mounted on the slides using Gel-Mount (Biomeda).

Specific antibodies against proteins of interest were obtained as indicated in supplementary material Table S1 and used for immunofluorescent (IF) staining at the stated dilutions. Whole IgG antibodies for controls were purchased from Pierce. Species appropriate Alexa Fluor 488 or 546 antibodies (Molecular Probes, Invitrogen) were used as secondary antibodies for indirect fluorescence.

Epifluorescence images were acquired using a Nikon Eclipse TE300 fluorescence microscope with OpenLab v5.5.0 image analysis software (Improvision). Confocal images were acquired by sequential detection using a Zeiss 510 Meta NLO v4.2, or Olympus FluoView 1000 LSM using FluoView software v5.0.

Differentiated PEC cultures in eight-chambered slides were grown in the presence or absence of DHT for 72 hours in 200 μl per well of growth medium. To quantify PSA concentrations in conditioned medium, a human PSA ELISA kit (Abzyme) was used according to the manufacturer's directions with the following modifications: the entire 200 μl samples were incubated 50 μl at a time per well for 1 hour each. PSA standards were added to coated wells during the final 50 μl of sample incubation.

Human KLK2, human KLK3 (PSA), FGFR2IIIb and GAPDH mRNA levels were quantified in differentiated cells by RT-PCR. Total RNA was isolated from upper and lower cell populations of dissociated cells from differentiated cultures or from LNCaP cells using TRIzol (Gibco) and chloroform (Sigma-Aldrich). Contaminating DNA was then removed using a RNAse-free DNAse kit (Qiagen) following manufacturer's directions. RT-PCR was performed on 1-2 μg RNA with the primers listed in supplementary material Table S2 using the One-Step RT-PCR kit (Qiagen) following manufacturer's directions. RT-PCR products were analyzed on a 2% agarose-TBE gel and DNA was visualized with ethidium bromide and a CCD camera in a Bio-Rad Chemi-Doc Imaging System using Quantity One software v4.5.2 (Bio-Rad).

A pool of four small interfering RNAs (siRNA) against androgen receptor (siGENOME SMARTpool) or a non-targeting sequence were purchased from Dharmacon. Differentiated cultures were transfected with 20 nM siRNA in keratinocyte-SFM medium using siLentFect lipid reagent (Bio-Rad) and Opti-MEM (Invitrogen) medium following manufacturer's directions. The medium was changed 16 hours after transfection.

Differentiated PECs were starved of growth factor in keratinocyte-SFM medium containing no supplements, KGF, or DHT for 72 hours. Then DMSO (control; Sigma), pharmacological inhibitors 0.5 μM PD168393, 2 μM LY294002, 20 μM PD90859, 10 μM SB209102, 10 μM 420119 (all purchased from Calbiochem), 1 μM staurosporine (Promega) or 1 μg/ml E-cadherin-blocking antibody (SHE78-7, Calbiochem) or non-specific mouse IgG (Sigma) was added; in some experiments, siRNAs were used to knock down AR expression (Dharmacon). Cells were incubated for 24, 48, 66 or 72 hours after drug, antibody or siRNA addition. LY294002 was replenished 48 hours after its initial addition. To assess cell viability, cells were fixed and DNA fragmentation was monitored using terminal deoxynucleotide transferase dUTP nick end labeling (TUNEL) following the protocol of the APO-BrdU TUNEL Assay Kit (BD Pharmingen). On several occasions, cleaved caspase 3 (Asp175) staining with antibody clone 5A1 from Cell Signaling was also used to measure cell viability of fixed cells. TUNEL and caspase activity were quantified using Imagine software (Qian et al., 2006). Total TUNEL- or caspase-positive pixels were normalized to total propidium iodide-stained DNA pixels in fixed cells and expressed as relative intensity of TUNEL staining. This quantification is based on pixel counts and does not necessarily reflect the percentage of positive cells, but rather the relative intensity of TUNEL or caspase 3 staining between treated and untreated cultures. As an alternative method for measuring cell viability, unfixed cells were treated with propidium iodide (PI). High intensity PI staining of dead, i.e. permeabilized cells, was quantified on a per cell basis and expressed as the percentage PI-positive cells.

We thank Veronique Schulz, Rich West and Jim Resau at the Van Andel Research Institute (VARI) and Melinda Frame at Michigan State University for technical assistance, and we acknowledge members of the laboratories of Developmental Cell Biology, Systems Biology, and Cell Structure and Signal Integration at VARI for their constructive suggestions. We also thank Doug Green (St Jude Children's Research Hospital, Memphis, TN) for his constructive suggestions on the cell survival analyses. Integrin α2, α3 and β4 antibodies were a kind gift from William Carter (Fred Hutchinson Cancer Research Institute, Seattle, WA); the TMPRSS2 antibody was a generous gift from Peter Nelson (Fred Hutchinson Cancer Research Institute, Seattle, WA) and the α6 integrin blotting antibody was a gift from Anne Cress (University of Arizona, Phoenix, AZ). This work was supported by the Department of Defense Prostate Cancer Predoctoral Training Award (W81XWH-08-1-0058) to L.L., and the American Cancer Society (RSG-05-245-01-CSM) to C.K.M. Additional support was also provided by the generous gifts from the Van Andel Institute.