Human prostate luminal cell differentiation requires NOTCH3 induction by p38-MAPK and MYC

The human prostate gland contains an epithelial bilayer of basal and luminal cells. Within these layers resides a combination of uni- and bi-potent progenitors important for normal gland homeostasis (Kwon et al., 2016; Ousset et al., 2012; Uzgare et al., 2004). Basal and luminal cells display distinct markers, such as androgen receptor (AR) and keratin 8 (K8; also known as KRT8) in the luminal layer, and laminin-binding integrins and keratin 5 (K5; also known as KRT5) in the basal layer (Lamb et al., 2010). Human prostate tumors co-express some of the basal and luminal markers, suggesting a defect in differentiation (Tokar et al., 2005). Moreover, many of the commonly altered genes in prostate cancer (e.g. MYC, AR, ERG and PTEN) are also implicated in differentiation (Frank and Miranti, 2013). We previously demonstrated that manipulation of differentiation regulators (MYC, PTEN and ING4) in normal human prostate epithelial cells results in tumor formation when grafted into a mouse prostate (Berger et al., 2014). To better understand tumor initiation in prostate epithelium, we sought to investigate specific genes and mechanisms required for normal basal to luminal cell differentiation.

The p38 MAPK family (hereafter p38-MAPK) is a known driver of epithelial differentiation in various tissues including skin and lung (Cuadrado and Nebreda, 2010). p38-MAPK regulates a wide range of targets, including other kinases/phosphatases, transcription factors and RNA-binding proteins (Cuadrado and Nebreda, 2010). Moreover, p38-MAPK is a downstream target of FGFR2b, a crucial receptor for epithelial differentiation in the skin and prostate (Belleudi et al., 2011; Heer et al., 2006; Lamb et al., 2010). Despite these findings, how p38-MAPK expression in prostate epithelial cells drives differentiation, including its relevant targets, remains poorly defined.

MYC positively regulates normal skin and prostate differentiation, and is a major prostate cancer oncogene (Berger et al., 2014; Gebhardt et al., 2006; Koh et al., 2010). MYC potentially targets thousands of genes via its activity as a transcription factor, and many of its targets are tissue and context specific (Conacci-Sorrell et al., 2014; Lüscher and Vervoorts, 2012). In normal prostate, transient upregulation of MYC is required for loss of cell adhesion and stimulation of chromatin remodeling (Berger et al., 2014). Moreover, regulation of MYC itself is complex, occurring at many different levels including pre- and post-transcription and through post-translational modification (McKeown and Bradner, 2014). Overexpression of AR in human primary basal prostate epithelial cells is sufficient to cause growth arrest via transcriptional downregulation of MYC (Antony et al., 2014; Vander Griend et al., 2014). Thus, MYC plays a crucial role in multiple aspects of both normal prostate differentiation and cancer.

NOTCH controls cell fate, including stemness, survival and differentiation (Deng et al., 2015). Mammals express four NOTCH transmembrane receptors (NOTCH1– NOTCH4), five canonical transmembrane ligands (JAG1 and JAG2, and DLL1, DLL3 and DLL4) and ten classic downstream targets (HES1–HES7, HEY1, HEY2 and HEYL). Cell–cell contact joins ligand and receptor, triggering proteolytic cleavage of  NOTCH by the γ-secretase complex which releases the active intracellular domain (ICD) of the receptor into the nucleus to activate transcription (Kopan and Ilagan, 2009). NOTCH can promote cell cycle arrest and de-adhesion from the matrix, both of which are essential for luminal differentiation (Hodkinson et al., 2007; Mazzone et al., 2010; Rangarajan et al., 2001). Furthermore, NOTCH1 signaling can promote survival of human basal cells (Dalrymple et al., 2005; Litvinov et al., 2006). In a mouse model, constitutively active NOTCH1 driven by a luminal promoter causes prostatic intraepithelial neoplasia (PIN) and increases survival of a subset of luminal cells in 3D culture (Kwon et al., 2014; Valdez et al., 2012). However, there are conflicting reports as to whether the NOTCH pathway is oncogenic or tumor suppressive, and the specific role for the other NOTCH receptors remains undefined (Carvalho et al., 2014; Kwon et al., 2016).

We sought to understand how p38-MAPK, MYC and NOTCH work together in normal prostate differentiation. We utilized an established model of in vitro differentiation of human basal prostate epithelial cells (PrECs) (Berger et al., 2014, 2017; Lamb et al., 2010). By using pharmacologic and genetic manipulation, we tested the hypothesis that p38-MAPK upregulation of NOTCH3, via MYC, is required for efficient induction and maintenance of the suprabasal layer during prostate differentiation. We identify two mechanisms of NOTCH3 regulation by p38-MAPK, both at the transcriptional and post-transcriptional level. This knowledge improves our understanding of prostate epithelial differentiation by tying together multiple pathways and elucidating new mechanisms for key differentiation regulators.

PrECs were induced to differentiate by treating with keratinocyte growth factor (KGF; also known as FGF7) and synthetic androgen (R1881) for 2 weeks (Lamb et al., 2010). This results in a stratified epithelium consisting of suprabasal luminal cells sitting on top of basal cells. p38-MAPK is a known downstream target of KGF-to-FGFR2 signaling and is implicated in epithelial differentiation in several tissue types, including prostate (Belleudi et al., 2011; Lamb et al., 2010). Four different genes encode p38-MAPK isoforms: MAPK14 (p38α), MAPK11 (p38β), MAPK12 (p38γ) and MAPK13 (p38δ). p38α is ubiquitously expressed, while the other isoforms are typically more tissue specific (Cuadrado and Nebreda, 2010). RNA-seq and immunoblotting identified p38α and p38δ to be the predominantly expressed isoforms in basal PrECs (Fig. 1A,B).

Lysates from differentiating cells were collected over a 2-week time course, and p38α activity was measured by immunoblotting with an antibody specific for its activated phosphorylated form (p-p38α). In primary cells (PrECs), elevated p-p38α was detected at day 4 and remained elevated (Fig. 1C). In immortalized cells (iPrECs), which take 4 days longer to differentiate, p-p38α was elevated at day 8 (Fig. 1D). Semi-quantification of a set of biological triplicate experiments indicates that both total p38α and p-p38α levels increase ∼2-fold at day 4 and ∼3-fold by day 12 (Fig. S1A,B).

To determine whether p38-MAPK is necessary for differentiation, iPrECs were differentiated in the presence of two p38-MAPK inhibitors (SB202190 and BIRB796) or Dox-induced shRNA against p38α (sh-p38α), p38δ (sh-p38δ) or both (sh-p38α/δ). Inhibitor concentrations were selected based on their ability to block CREB1 phosphorylation mediated by constitutively active MKK6 [MKK6(CA); MKK6 is also known as MAP2K6] (Fig. S1C). Effective knockdown of p38α and/or p38δ by shRNA was verified by immunoblotting (Fig. 1B). After 16 days of differentiation, control cells (Dox plus DMSO) differentiated normally, as measured by loss of integrin α6 and gain in AR, with a 54% coverage of the culture by suprabasal cells (averaged from three fields) (Fig. 1E). Treatment with 1 µM SB202190 or 0.1 µM BIRB796 completely prevented formation of an AR-positive suprabasal layer. Unexpectedly, integrin α6 (ITGα6) expression was also decreased by these inhibitors. However, this was not due to basal cell toxicity (as judged by the lack of cleaved caspase 3) nor decreased proliferation (as demonstrated by measuring BrdU incorporation) (Fig. S1D,E). Dox-induced shRNA knockdown of p38α did not prevent AR-positive cells from appearing, but it did prevent formation of a distinct suprabasal layer. On the other hand, knockdown of p38δ reduced the production of cells that were both AR⁺ and ITGα6^– (29% suprabasal coverage, reduced from 71%), but did not completely block it. However, double p38α/δ knockdown drastically prevented suprabasal layer formation (6% suprabasal coverage, reduced from 67%) (Fig. 1E). Thus, both p38α and p38δ are required for normal luminal cell differentiation, and the differing effects of their loss suggests they may control different steps in suprabasal layer formation.

A hallmark of normal luminal cell differentiation is the downregulation of integrins including α6, α3, β4 and β1. NOTCH can negatively regulate integrin expression and is generally required for epithelial differentiation (Frank and Miranti, 2013; Koh et al., 2010; Mazzone et al., 2010). Additionally, MYC suppresses integrin α6 and β1 expression (Gebhardt et al., 2006), and was previously demonstrated to be required for prostate differentiation (Berger et al., 2014). In some contexts, MYC is a direct downstream target of NOTCH (Weng et al., 2006). To decipher the roles of MYC and NOTCH, lysates from differentiating iPrECs (Fig. 2A) or primary PrECs (Fig. S2A) were collected over a 2-week time course and protein expression measured by immunoblotting. MYC expression and activation (phosphorylation; denoted p-MYC) was initially elevated but waned as basal cell proliferation subsided and transiently elevated again at around day 8 (Fig. 2A). A similar response was observed in primary cells but it occurred 4 days earlier, as expected due to their faster differentiation (Fig. S2A).

Of the four NOTCH receptors, we were only able to detect significant expression of NOTCH1, NOTCH2 and NOTCH3 (Fig. 2A). Expression of NOTCH2 remained essentially unchanged during differentiation. NOTCH1 protein was initially high, then decreased slightly. In contrast, NOTCH3 protein expression was very low in basal cells, then increased with time during differentiation; moreover, a marked increase occurred at around day 8, when p38α and MYC activity were also maximal (Fig. 2A). A similar pattern was observed in primary PrECs at day 4 (Fig. S2A).

NOTCH1 and NOTCH3 mRNA expression, as measured by quantitative real-time PCR (qRT-PCR), paralleled protein expression; NOTCH1 dipped and recovered to baseline levels, while NOTCH3 increased dramatically and remained higher in the suprabasal layer (Fig. 2B). NOTCH3 mRNA appeared to increase in two phases; a steady climb increasing ∼10-fold over the first 8 days followed by a more dramatic spike, up ∼220-fold by day 14 in the suprabasal cells (Fig. 2B). NOTCH ligands also displayed two distinct expression profiles; JAG1 (Fig. 2B) and DLL4 (Fig. S2B) showed initial decreases but then recovered by day 10, following the pattern of NOTCH1 expression. Meanwhile, DLL3 remained flat and began to increase after day 10, paralleling the increase in NOTCH3 mRNA expression (Fig. 2B). HEY2, HEYL (Fig. 2B), HES1, HES6 and HEY1 (Fig. S2B) all increased during differentiation, with day 8 being a key inflection point. HEY2 mRNA was unique in that it segregated into the suprabasal population (up 45-fold versus day 1) similar to NOTCH3. These data indicate that the day 8–10 window is critical for activation of the NOTCH pathway, and correlates with the appearance of an emerging suprabasal layer and integrin α6β1 mRNA downregulation (Fig. S2B).

To examine the requirement of NOTCH1 and/or NOTCH3 for differentiation, iPrECs were differentiated and treated with either a γ-secretase inhibitor (RO4929097) or Dox to induce expression of NOTCH1 and/or NOTCH3 shRNA. Efficient knockdown of NOTCH1 and/or NOTCH3 mRNA was achieved by 48 h (Fig. S2C) and protein at 96 h (Fig. S2D). NOTCH3 loss also led to a slight decrease in NOTCH1 protein; however, this was not due to an off-target shRNA effect on NOTCH1 since NOTCH1 mRNA was not affected (Fig. S2C,D). Control and non-Dox-treated cells differentiated normally as indicated by formation of a suprabasal layer of cells (both AR⁺ and ITGα6⁻; 44–53% coverage), while treatment with RO4929097 ablated differentiation (Fig. 2C). Induced knockdown of NOTCH1 or NOTCH3 by means of shRNA each led to disruption of the suprabasal layer, with 16% and 31% coverage respectively, compared to 53% and 44% for control cells. Double knockdown of NOTCH1 and NOTCH3 more severely disrupted differentiation, giving a similar appearance to that seen upon treatment with RO4929097 (Fig. 2D). Furthermore, propidium iodide staining indicated that the suprabasal cell clumps observed upon NOTCH inhibition or knockdown were mostly dead cells (Fig. S2E). Thus, NOTCH1 and NOTCH3 are both required for survival of the suprabasal cells during luminal cell differentiation.

iPrECs were immunostained for p-p38 (all p38-MAPK isoforms) and NOTCH3 at key times during differentiation to observe expression levels and localization (Fig. 2E). Nuclear p-p38 was detected in all basal cells at day 4, when very little NOTCH3 expression was detected, except for in a few cells where it was nuclear localized. By day 8, patches of more intense p-p38 nuclear staining were detected, which corresponded to cells in which NOTCH3 levels were dramatically increased (white arrow). NOTCH3 localization was primarily nuclear in the basal cells, but both nuclear and cytoplasmic staining was apparent at days 8 and 12, when suprabasal layer formation is maximal. By day 21, more membrane and less nuclear staining was observed, with staining occurring primarily in the suprabasal cells with very low levels in the basal cells. p-p38 nuclear localization was lost as suprabasal cells became established. Thus, p-p38 nuclear activity peaks around day 8, just as NOTCH3 expression and downstream signaling increases in the suprabasal layers. Once established, NOTCH3 expression remains high in the suprabasal layer and p-p38 is lost from the nucleus.

To determine the relationship between p38-MAPK and NOTCH3, we engineered an iPrEC line with a Dox-inducible constitutively active MKK6 mutant, MKK6(CA), which directly phosphorylates and activates p38-MAPK (Alonso et al., 2000). During differentiation, p38-MAPK activation is moderately elevated over several days (Fig. S1A,B), but when MKK6(CA) is induced, the signaling events that naturally occur over days are condensed into hours (Fig. 3A). Although prolonged constitutive p38-MAPK activation leads to stress and cell death, the Dox-inducible system allows us to tightly control induction and measure downstream signaling over a short time period. A 16 h treatment of iPrEC-TetON-MKK6(CA) cells with Dox led to an ∼18-fold increase in NOTCH3 mRNA (Fig. 3B). Conversely, MKK6(CA) induction decreased NOTCH1 by ∼2.5-fold. Inhibition of p38-MAPK blocked these effects (Fig. 3B).

To establish a temporal order of events, iPrEC-TetON-MKK6(CA) cells were treated with Dox and lysates collected over time (Fig. 3C). MKK6(CA) was detectable as early as 4 h, at which time a corresponding increase in active p-p38α and MYC was observed, peaking at ∼7–8 h. NOTCH3 levels began to increase around 6 h and continued to climb. At the mRNA level, MYC induction also preceded an increase in NOTCH3 and decrease in NOTCH1 (Fig. 3D). Furthermore, a short pulse of Dox was sufficient to induce NOTCH3 to higher levels than normally seen at day 4 of differentiation (Fig. 3E); meanwhile, expression of NOTCH1 was decreased. These results show that constitutive activation of MKK6 is sufficient to induce p38α, MYC, MYC phosphorylation and NOTCH3, while downregulating NOTCH1. Thus, the MKK6(CA) model mimics the regulation of these genes observed in the standard differentiation assay. Moreover, differentiation of iPrECs for 4 days in the presence of a p38-MAPK inhibitor suppressed MYC induction and dampened NOTCH3 upregulation (∼7- vs ∼28-fold), thus confirming their roles downstream of p38-MAPK in this model (Fig. 3F).

Induction of NOTCH3 mRNA by p38-MAPK could be due to direct activation of an existing transcription factor or indirect, requiring synthesis of a new factor. iPrEC-TetON-MKK6(CA) cells were treated with Dox for 12 h and cyclohexamide (CHX) was added at 6, 8 or 10 h to measure the requirement for new protein synthesis. Addition of CHX at 6 h blocked NOTCH3 mRNA upregulation, while addition at 8 h or later did not (Fig. 4A; Fig. S3A). Thus, there is a requirement for the synthesis of an intermediate, which must be translated between 6 and 8 h after Dox; this matches the time of maximal MYC induction and activation (see Fig. 3C).

To test whether NOTCH3 induction requires MYC, iPrEC-TetON-MKK6(CA) cells were transfected with siRNA against MYC (denoted si.MYC) or a non-targeting control sequence and induced with Dox for 12 h. MYC mRNA was knocked down ∼80% and NOTCH3 mRNA induction was half that seen in the control cells (5- vs 10-fold) (Fig. 4B). Similar results were observed at the protein level as assessed by immunoblotting (Fig. 4C). To further address the dependency of NOTCH3 induction on MYC, we utilized an antagonist of the MYC–MAX complex, 10058-F4 (Huang et al., 2006). iPrEC-TetON-MKK6(CA) cells were treated with Dox and increasing concentrations of 10058-F4 for 16 h. Treatment with as little as 5 µM 10058-F4 suppressed the induction of NOTCH3 protein (Fig. 4D), whereas 20 µM was required to suppress NOTCH3 mRNA (Fig. S3B). These doses are at or below common usage for 10058-F4 (Guo et al., 2009; Wang et al., 2014). As an alternative approach, we used JQ-1, a BET bromodomain inhibitor, to block transcription of MYC (Delmore et al., 2011). JQ-1 prevented MKK6(CA)-induced MYC and NOTCH3 expression at 100–500 nM (Fig. 4E). Taken together, these results demonstrate that MYC is required for maximal p38-MAPK-mediated induction of NOTCH3.

To determine whether MYC is sufficient for NOTCH3 induction, we generated a Tet-inducible MYC-expressing cell line: iPrEC-TetON-MYC. MYC induction occurred within 2 h of Dox treatment and NOTCH3 protein increased slightly by 6 h (Fig. 4F). However, there was no change in NOTCH3 mRNA (Fig. S3C). We also induced MYC after first differentiating cells for 5 days and still observed only a slight increase in NOTCH3 protein expression (Fig. S3D). Thus, MYC is not sufficient in this context to transcriptionally induce NOTCH3, although it may have some slight effect on NOTCH3 protein expression.

The NOTCH3 2 kb upstream proximal promoter contains a CpG island and no TATA sequence (Kent et al., 2002). The 2 kb region of the NOTCH3 promoter was not sufficient to induce a luciferase reporter after 6 days of differentiation (Fig. 5A), a time when endogenous NOTCH3 was elevated over 16-fold. We used two approaches to identify candidate enhancer regions. First, we labeled newly initiated transcripts at the NOTCH3 transcriptional start site and enhancer elements by using BruUV-Seq (Magnuson et al., 2015). Dox induction in iPrEC-TetON-MKK6(CA) cells dramatically increased NOTCH3 reads from the coding (−) strand accumulating near the transcription start site (Fig. 5B). Strikingly, there was also a peak of reads from the non-coding (+) strand within the second intron, a locus previously reported to contain a NOTCH3 enhancer (Gagan et al., 2012; Romano et al., 2012). The gene for MKK6 (MAP2K6) served as a positive control; it was induced only upon Dox treatment and with reads mapping only to the exons generated from the cDNA construct (Fig. S4A). Other controls included CALB1 and TRIM22, which were increased and decreased, respectively, upon MKK6 induction (Fig. S4A).

Our second approach used a combination of DNase hypersensitivity, histone acetylation and methylation patterns (H3K27Ac+H3K4me1/2), and ChIP-seq data from ENCODE to identify potential enhancer elements (The ENCODE Project Consortium, 2012; Kent et al., 2002). Five different elements were cloned into a pNL1.1-miniTK luciferase reporter (Fig. S4B). En2.1, En2.2 and the NOTCH3 promoter showed no induction by Dox in the MKK6(CA) model (Fig. 5C). However, two elements (En1 and En3) were upregulated by 5- and 3-fold, respectively. En1 is ∼10 kb upstream while En3 is in the second intron and corresponds to the site with bidirectional transcripts identified by BruUV-seq. A deletion in En1, Δ1–360, that eliminated most of the predicted MYC-binding sites (Fig. S4C) completely ablated the ability of the En1 reporter to be induced by MKK6(CA) (Fig. 5D). A small En3 deletion, Δ1–350, that removed two-thirds of the predicted MYC sites did not significantly decrease expression of the reporter while a larger deletion, Δ1–655, that removed all three predicted MYC sites significantly blocked induction (Fig. 5E).

To further determine whether MYC is required for induction of these enhancer elements, MKK6(CA) cells were induced in the presence of the MYC inhibitor 10058-F4. Both En1 and En3(Δ1–350) (the core En3 responsive element) were sensitive to MYC inhibition (Fig. 5F). Induction mediated by En1 was partially decreased (2.7- vs 4.5-fold) while En3(Δ1–350) induction was more thoroughly blocked (0.7- vs 1.7-fold). Thus, both En1 and En3 are sensitive to MYC inhibition and both contain MYC-binding sites, which when deleted significantly reduced reporter induction in response to MKK6(CA).

NOTCH3 contains an AU-rich element in its 3′ UTR and p38-MAPK is known to regulate RNA-binding proteins (Cuadrado and Nebreda, 2010). Actinomycin D was used to halt transcription, and measurements of mRNA decay were taken at nine time points (Harrold et al., 1991) at day 1 and day 4 of differentiation (Fig. 6A; Table 1). The MYC half-life, of 0.8 h, was similar to that found in previous reports (Herrick and Ross, 1994). MYC and NOTCH1 half-lives remained essentially the same at day 4 (P>0.2). However, NOTCH3 mRNA half-life nearly doubled (11.5 vs 5.9 h), along with an 8.5-fold increase in total mRNA levels. We similarly compared iPrEC-TetON-MKK6(CA) cells stimulated with Dox for 16 h to non-Dox-treated cells (Fig. 6B; Table 2). Both NOTCH1 and NOTCH3 mRNA half-lives more than doubled: 3.3 to 8.8 h for NOTCH1, and 7.6 to 17.6 h for NOTCH3. However, the overall mRNA level of NOTCH1 decreased ∼4-fold while NOTCH3 increased ∼9-fold (Table 2). Thus, differentiation and acute p38-MAPK activation both lead to increased NOTCH3 mRNA half-life, indicating that NOTCH3 is regulated post-transcriptionally through mRNA stabilization.

NOTCH1 expression has been reported to primarily be present in basal cells of mouse and human prostate, while NOTCH3 has been reported (with some disagreement) to be more luminal (Pedrosa et al., 2016; Shou et al., 2001; Valdez et al., 2012). We detected abundant NOTCH1 and NOTCH2 and very low NOTCH3 in undifferentiated human basal cells. NOTCH4 protein was detectable but at a very low level and did not increase during differentiation (not shown). Owing to their dynamic regulation during differentiation, we focused on NOTCH1 and NOTCH3. We observed a dramatic induction of NOTCH3 mRNA and protein during differentiation, which coincided with the appearance of suprabasal cells. Therefore, NOTCH3 appears to be a primary driver of luminal cell differentiation, while NOTCH1 serves its previously described role in maintaining the basal population (Pedrosa et al., 2016; Shou et al., 2001; Valdez et al., 2012).

Previous studies have shown that low-Ca²⁺ medium, such as the KSFM in which we culture our cells, selects for basal transit-amplifying prostate epithelial cells and promotes their survival via constitutive activation of NOTCH1 (Dalrymple et al., 2005; Litvinov et al., 2006). However, inhibition or knockdown of NOTCH1 or NOTCH3 did not affect basal cell survival in our assays (Fig. S2E). In the previous studies, constitutive NOTCH1 signaling was most important in subconfluent cultures. We only inhibited NOTCH in completely confluent cells, which may account for the observed differences.

The function of NOTCH3 has been controversial, but recent reports show that it drives luminal differentiation of airway basal cells and mammary epithelium (Baeten and Lilly, 2015; Bhat et al., 2016; Gomi et al., 2015; Mori et al., 2015; Ohashi et al., 2010). Moreover, of the four NOTCH receptors only NOTCH3 is sufficient to drive hepatocyte differentiation in embryonic mouse liver cells (Ortica et al., 2014). Though NOTCH1 seems to drive prostate basal cell commitment, our data supports the idea that NOTCH3 is required to generate the suprabasal cell layer required for prostate luminal cell differentiation.

Part of the mechanistic insight from this work demonstrates that p38-MAPK can regulate NOTCH3 transcription in part via MYC. Although a relationship between p38-MAPK and NOTCH has previously been suggested, mechanistic details were not clearly established (Brown et al., 2009; Gonsalves and Weisblat, 2007; Kiec-Wilk et al., 2010; Park et al., 2009). We found that the full ability of p38-MAPK to induce NOTCH3 is dependent on MYC. We previously demonstrated MYC is required for PrEC differentiation (Berger et al., 2014; Marderosian et al., 2006). Thus, NOTCH3 appears to be one of the MYC targets downstream of p38-MAPK. MYC has typically been considered a downstream target of NOTCH (Weng et al., 2006), whereas we found that it is upstream of NOTCH3. Although MYC was required for full NOTCH3 induction, blocking its activity did not fully block NOTCH3 induction suggesting that there are likely other factors involved. Furthermore, overexpression of MYC was not sufficient to induce NOTCH3 mRNA. Thus, p38-MAPK is likely activating additional unidentified factors that are also required for NOTCH3 mRNA induction.

We investigated potential regulatory regions of the NOTCH3 gene and found two elements capable of inducing a luciferase reporter upon MKK6(CA) induction that are sensitive to MYC inhibition. One element lies ∼10 kb upstream, denoted En1, and has not previously been linked to NOTCH3. A 5′ deletion that eliminates most of the predicted MYC-binding sites in En1 severely compromises its induction; however, it is only partially sensitive to inhibition of MYC. Thus, there are likely to be other factors that cooperate with MYC to fully activate this enhancer. A second element, En3, lies in a previously implicated locus within the second intron (Gagan et al., 2012; Romano et al., 2012). Our report is the first to show functional validation of En3 in human cells. Furthermore, we identified bi-directional eRNA from En3 upon p38-MAPK stimulation, as measured by BruUV-Seq (Kim et al., 2010; Lam et al., 2014; Magnuson et al., 2015). A small deletion (En3Δ1–350) that removed two-thirds of the predicted MYC sites retained reporter activity, thus narrowing down the core regulatory region. Likewise, a second larger deletion (En3Δ1–655) that removed all the predicted MYC-binding sites greatly diminished induction of the reporter. Both elements contain numerous other potential transcription factor-binding sites (The ENCODE Project Consortium, 2012; Mathelier et al., 2016) that may be required for NOTCH3 to cooperate with MYC. Further detailed analysis will be required to completely define all possible mechanisms of NOTCH3 transcriptional regulation.

We also demonstrate that NOTCH3 is post-transcriptionally regulated through mRNA stability during differentiation mediated by p38-MAPK. NOTCH1 expression is affected by RNA stability, which is known to be modulated through AU-rich elements in its 3′ untranslated region (UTR) and by p38-MAPK (Cisneros et al., 2008; Gonsalves and Weisblat, 2007). p38-MAPK regulates mRNA stability through phosphorylation of mRNA-binding proteins (Cuadrado and Nebreda, 2010). NOTCH3 also has predicted AU-rich elements in its 3′ UTR (Gruber et al., 2011). Interestingly, p38-MAPK activation via MKK6(CA) for 16 h increased both NOTCH1 and NOTCH3 mRNA half-life, but only NOTCH3 stability was increased after 6 days of differentiation. This may reflect differences in the extent of p38-MAPK activation in the two models or may suggest that other modes of stabilization are involved. There are reports of post-transcriptional NOTCH regulation by micro (mi)RNAs, which may also contribute to long-term stability (Furukawa et al., 2013; Gagan et al., 2012; Liu et al., 2015).

We also found that MYC enhances NOTCH3 expression independently of mRNA. For instance, it took 20 µM of MYC inhibitor (10058-F4) to suppress NOTCH3 mRNA expression, yet there were effects on NOTCH3 protein at 5 µM. Similarly, overexpression of MYC did not alter NOTCH3 mRNA, but it did increase NOTCH3 protein, suggesting that there may be a mechanism for stabilizing NOTCH3 protein or increasing its translation rate. In addition, shRNA against NOTCH3 resulted in partial loss of NOTCH1 protein, but not mRNA. Thus, there are several mechanisms that regulate both NOTCH1 and NOTCH3 during luminal cell differentiation, and further research will be required to define them all.

One of the key roles for AR in normal luminal differentiation is to inhibit proliferation, which is the opposite of its role in tumors. Previous reports have shown that AR overexpression in basal PrECs can induce growth arrest and that this requires AR (in cooperation with β-catenin/TCF-4), which transcriptionally represses MYC (Antony et al., 2014; Vander Griend et al., 2014). This is opposite to what is seen in tumors, where AR can drive MYC expression (Antony et al., 2014). Our data showed that p38-MAPK can upregulate MYC expression, which is transient in our differentiation model. Although we have not investigated it, AR may help suppress MYC expression once the suprabasal layer is established. Likewise, it may be that full luminal commitment and increased AR activity may provide a brake for NOTCH3 induction by antagonizing MYC.

Temporal regulation of NOTCH3 throughout differentiation is dynamic. We observed two phases of NOTCH3 mRNA induction: an early steady increase up to day 8 (day 4 in primary cells) followed by a more dramatic increase. Considering that NOTCH3 mRNA is stabilized by day 6, it could be that early upregulation is less dependent on transcriptional mechanisms and more on message stability. The suprabasal layer is visible at around day 8, coinciding with induction of downstream target HES and HEY genes. Additionally, it is at this transition point that p38-MAPK and MYC are activated. Thus, robust transcriptional induction of NOTCH3 appears to peak at around this time and may drive the secondary phase of NOTCH3 induction. It is also at this time that NOTCH1 mRNA begins to increase following an initial dip. Thus, day 8 is a key point for NOTCH1 and NOTCH3 induction and cell commitment to the luminal transition.

The direct effectors of NOTCH signaling include the canonical HES and HEY transcriptional repressor family. Indeed, we observed differential induction of several family members during differentiation. In ongoing studies, we are determining which of these are critical for luminal cell differentiation. Previous findings have reported that AR and GATA cooperate to regulate a set of target genes, and HEY transcriptional repressors can prevent GATA-mediated induction of AR target genes (Belandia et al., 2005; Fischer et al., 2005; Litvinov et al., 2006; Xiao et al., 2016). This would support downstream HEY activity in maintaining a basal commitment. With the NOTCH pathway, timing and dose are critical. Our attempts to drive differentiation with inducible NOTCH ICD (NICD) constructs led to cell stress and death within 24–48 h (data not shown). It may be that a low or moderate amount of NOTCH activity is needed for survival and initial differentiation but too much activity can block terminal differentiation. Whether the functional role of NOTCH3 is via HES and HEY or non-canonical downstream targets will require further investigation.

Some of the reported non-canonical NOTCH targets include PTEN and CDH1 (also known as E-cadherin), both of which are critical for luminal cell survival (Bertrand et al., 2014; Lamb et al., 2010). Furthermore, NOTCH downregulates adhesion genes, including integrins such as β4, which is required for basal cell detachment from the extracellular matrix (Cress et al., 1995; Mazzone et al., 2010; Nguyen et al., 2006). We also see loss of integrin expression during differentiation. There are also reports that NOTCH can upregulate MKP1 (also known as DUSP1), a phosphatase that targets p38-MAPK, thus providing a potential feedback mechanism in terminally differentiated cells to balance p38-MAPK activity (Gagan et al., 2012; Yoshida et al., 2014). The balance of downstream NOTCH targets (both canonical and non-canonical) could help explain the conflicting roles for the pathway in promoting both basal and luminal commitments.

Previous studies suggested that the ICD from NOTCH3 is a weaker activator than other NICDs (Beatus et al., 1999; Ong et al., 2006). However, our findings and other recent reports have begun to reveal novel signaling effects and preferential targets for NOTCH3 (Baeten and Lilly, 2015; Cui et al., 2013; Wang et al., 2016). As it stands, NOTCH3 appears to be unique among the receptors. Further research will be needed to validate which downstream NOTCH3-specific targets are most relevant to luminal cell differentiation.

In this study, we report on a novel mechanism for crosstalk between p38-MAPK, MYC and NOTCH. Moreover, we identify two distinct regulatory mechanisms for NOTCH3 in the prostate: a coordination of elevated mRNA stability and increased transcription from multiple enhancers. These findings provide a better understanding for how these differentiation pathways are connected in normal prostate epithelium and opens the door to investigating how their dysregulation may impact prostate cancer development and progression.

Primary and immortalized PrECs (Berger et al., 2014) were grown in KSFM medium (Gibco) plus penicillin-streptomycin at 30 units/ml (Gibco). Differentiation was induced as previously reported with 2.5 ng/ml recombinant KGF (Cell Sciences) and 1–10 nM R1881 (Perkin Elmer) (10 nM unless otherwise specified) with fresh medium added every 24 h (Lamb et al., 2010). Suprabasal layer separation was achieved by using Ca²⁺ and Mg²⁺-free PBS with 1 mM EDTA as previously described (Lamb et al., 2010). HEK 293FT cells were used for lentivirus production (ViraPower, Invitrogen) and grown in Dulbecco's modified Eagle's medium (DMEM; 11995, Gibco) with 10% fetal bovine serum (FBS; Gemini) and 2 mM L-glutamine (Gibco). Cell lines were tested via a MycoAlert PLUS kit (Lonza) and confirmed to be mycoplasma free.

Immortalized PrECs (iPrECs) were engineered with Dox-inducible shRNAs using the EZ-Tet-pLKO-Puro and EZ-Tet-pLKO-Hygro vectors (Addgene plasmids 85966, 85972) (Frank et al., 2017). shRNA sequences are listed in Table S1. Expression cDNAs were subcloned, via PCR with Q5 polymerase (NEB), into the pENTR3C gateway vector (Invitrogen) between the SalI and NotI sites and then recombined with LR Clonase II (Invitrogen) into pLenti-CMV/TO-Puro-DEST (Addgene plasmid 17293) (Campeau et al., 2009). The constitutively active MKK6 mutant (MKK6-DD) was a gift from Angel Nebreda (Oncology Unit, Institute for Research in Biomedicine, Spain) (Alonso et al., 2000). The MYC cDNA, subcloned from pBabe-Myc, was a gift from Beatrice Knudsen (Biomedical Sciences and Pathology, Cedars Sinai, USA). TetR lines were established by using pLenti-CMV-TetR-Blast (Addgene plasmid 17492) (Campeau et al., 2009). iPrECs antibiotic selection doses were as follows: 50 µg/ml hygromycin, 5 µg/ml blasticidin and 2 µg/ml puromycin. Doxycycline (Sigma) was used at 50 ng/ml to induce shRNAs and 2–10 ng/ml to induce cDNA expression.

A mixed siRNA pool against MYC and non-targeting siRNA (siScram) were purchased from Origene (SR303025). Cells were transfected by using siLentfect reagent (Bio-Rad). Cyclohexamide was used at 10 µg/ml, and actinomycin D at 5 µg/ml (Calbiochem). SB202190, BIRB796/Doramapimod, 10058-F4, JQ-1, BrdU and staurosporine were purchased from Cayman Chemical. RO4929097 was purchased from Apex Bio.

Cell lysates were prepared in RIPA as previously described (Edick et al., 2007). Protein loading was standardized by use of the BCA assay (Pierce). 20–50 µg of denatured protein was run on Novex SDS polyacrylamide Tris-glycine gels (Life Technologies) and transferred onto PVDF membrane (Fisher). Chemiluminescence was used to image blots with a Bio-Rad Chemi-Doc imaging system with CCD camera. The quantification shown in Fig. S1A,B was performed with ImageJ software. Data were first normalized to tubulin, then to day 1 ‘i’ samples and plotted as mean±s.d. P-values were determined by paired one-way ANOVA with Turkey's multiple testing correction. Antibodies are listed in Table S2. The protein ladder was from Cell Signaling Technology (7720) or GoldBio (P007).

RNA was harvested and extracted with Trizol following the manufacturer's protocol (Invitrogen). cDNA was synthesized with M-MuLV reverse transcriptase (NEB) using a 4:1 mix of poly-d(16)T and random hexamer primers. qRT-PCR was performed using SYBR Green Master Mix (Roche) and an ABI 7500 thermal cycler (Applied Biosystems). Data were standardized to 18S plus GAPDH unless otherwise stated and were normalized (ΔΔCT) and plotted as Log₂(Fold). Primers were synthesized by Integrated DNA Technologies. Primers are in Table S3.

Cells were fixed, permeabilized, and stained as previously described (Berger et al., 2014). Antibodies against ITGα6 (555734, BD) and AR (sc-815, Santa Cruz Biotechnology) were used at 1:200 dilution. Suprabasal coverage of the underlying basal layer was determined by tracing the clusters that were both AR⁺ and ITGα6⁻ by hand using ImageJ software, and calculating the percentage area of suprabasal regions versus total image area. Three fields of view were measured for each condition. For propidum iodide staining, cells were fixed with 4% paraformaldahyde, treated with 100 ng/ml RNaseA (Thermo) for 10 min, then stained with 100 ng/ml propidum iodide (Sigma) for 5 min. Nuclei were stained with 10 µg/ml Hoescht 33258 (Sigma) for 10 min. Epifluorescence microscopy was performed on a Nikon TE300 using Nikon Elements software (v4.11.00). Fig. 2E was captured on a DeltaVision (GE) epifluorescence scope with SoftWoRx software, with processing by deconvolution and maximum intensity projection from a z-stack capture.

Putative NOTCH3 regulatory elements were PCR subcloned from the RP11-937H1 BAC library (Life Technologies) using Q5 or LongAmp polymerase (NEB). The NOTCH3 2 kb promoter element was ligated into pGL4.15-Hygro (Promega). Candidate regulatory elements were ligated into pNL1.1 (Promega) after first cloning in a miniTK promoter at the HindIII site. Deletion mutants were made using the QuickChange II Mutagenesis kit (200524, Stratagene). Cloning primers, miniTK sequence, and mutagenesis primers are in Table S4.

En1 and En3 maps in Fig. S4C were generated using SnapGene and modified with Canvas software. MYC-binding sites were determined by using the JASPAR online database (http://jaspar.genereg.net/) (Mathelier et al., 2016), with a threshold of 80% using the MA0147.2 matrix model for MYC-binding sites (Chen et al., 2008).

Luciferase assays were performed by using the NanoGlo kit (Promega) and a Synergy Neo II (Bio Tek) plate reader with Gen5 software (v2.04). Cells were transfected as a pool with XtremeGeneHP reagent (Roche) and then split for different treatments. pNL1.1-miniTK served as the negative control. Luciferase assays were run 16 h after Dox treatment and 48 h after transfection.

Cells were treated with 5 µg/ml Actinomycin D for 0–8 h. RNA and cDNA were prepared as described above. Data were standardized to 18S rRNA and normalized as ΔΔCT values versus the Day 1 or ‘no Dox’ samples at 0 h. 18S rRNA has a very long half-life (1–7 days) and thus is suitable for standardization (Defoiche et al., 2009). Linear regression curves, line equations, r² values, and P-values were calculated with GraphPad PRISM software. Half-life was calculated as 1/m, where m is the slope. Overall expression change was calculated as 2^(b2–b1), where b=y-intercept; b1 is intercept 1 and b2 is intercept 2. AU-rich elements were identified using the ARE site (v1) online tool (http://nibiru.tbi.univie.ac.at/cgi-bin/AREsite/AREsite.cgi) (Gruber et al., 2011).

iPrEC-TetON-MKK6(CA) cells were treated with 5 ng/ml Dox for 10 h or left untreated, then exposed to UV (100 J/m²) using a Stratalinker UV Crosslinker 1800 (Stratagene) and labeled with 2 mM 5-bromo-deoxyuridine (sc-256904, Santa Cruz Biotechnology) for 30 min before washing with PBS and collecting RNA with Trizol (Life Technologies). BrU isolation, library prep, sequencing and mapping was performed as previously described (Andrade-Lima et al., 2015; Paulsen et al., 2014). Data were exported (bin size=300 bp) and graphed using GraphPad PRISM software.

Unless otherwise specified, P-values were calculated using paired, one-tailed t-tests on biological triplicates, with *P≤0.05, **P≤0.01, ***P≤0.001 and n.s., not significant (P>0.05). For Tables 1 and 2, P-values were calculated by ANCOVA analysis using PRISM GraphPad software. Fig. 5D,E used one-way ANOVA with Greenhouse–Geisser correction. Fig. 5F used two-way ANOVA with Turkey's multiple testing correction.

We acknowledge Mary Winn and the VARI Bioinformatics and Biostatistics Core for assistance with statistical and RNA-Seq analysis. We thank Michelle Paulsen for technical consolation on BruUV-Seq. Additional thanks to the VARI Program for Skeletal Disease and Tumor Metastasis for feedback and suggestions. The TetON cDNA and TetR vectors were gifts from Eric Campeau, the MMK6(CA) (MKK6-DD) construct was a gift from Dr. Angel Nebreda, and pBabe-MYC a gift from Dr. Beatrice Knudsen.