Cancer cells differentiate along specific lineages that largely determine their clinical and biologic behavior. Distinct cancer phenotypes from different cells and organs likely result from unique gene expression repertoires established in the embryo and maintained after malignant transformation. We used comprehensive gene expression analysis to examine this concept in the prostate, an organ with a tractable developmental program and a high propensity for cancer. We focused on gene expression in the murine prostate rudiment at three time points during the first 48 h of exposure to androgen, which initiates proliferation and invasion of prostate epithelial buds into surrounding urogenital sinus mesenchyme. Here, we show that androgen exposure regulates genes previously implicated in prostate carcinogenesis comprising pathways for the phosphatase and tensin homolog (PTEN), fibroblast growth factor (FGF)/mitogen-activated protein kinase (MAPK), and Wnt signaling along with cellular programs regulating such ‘hallmarks’ of cancer as angiogenesis, apoptosis, migration and proliferation. We found statistically significant evidence for novel androgen-induced gene regulation events that establish and/or maintain prostate cell fate. These include modulation of gene expression through microRNAs, expression of specific transcription factors, and regulation of their predicted targets. By querying public gene expression databases from other tissues, we found that rather than generally characterizing androgen exposure or epithelial budding, the early prostate development program more closely resembles the program for human prostate cancer. Most importantly, early androgen-regulated genes and functional themes associated with prostate development were highly enriched in contrasts between increasingly lethal forms of prostate cancer, confirming a ‘reactivation’ of embryonic pathways for proliferation and invasion in prostate cancer progression. Among the genes with the most significant links to the development and cancer, we highlight coordinate induction of the transcription factor Sox9 and suppression of the proapoptotic phospholipid-binding protein Annexin A1 that link early prostate development to early prostate carcinogenesis. These results credential early prostate development as a reliable and valid model system for the investigation of genes and pathways that drive prostate cancer.
The discovery of proto-oncogenes in the 1970's (Stehelin et al., 1976) and tumor suppressor genes a decade later (Friend et al., 1986) launched an effort to identify a common genetic basis for all cancers. This approach has yielded significant insights into common molecular machinery regulating tumor initiation and growth (Hanahan and Weinberg, 2000), but fails to account for dramatic differences in the behaviors of tumors arising in different sites. As revealed by recent genomic approaches, the site of origin is the dominant influence on gene expression in cancers (Ramaswamy et al., 2001) and specifies particular oncogenic mutations (Garraway and Sellers, 2006). Tissue-specific behaviors of cancers almost certainly reflect lineage-specific epigenetic programs that operate during embryogenesis in cells and tissues from which cancers arise. Indeed, reawakening of embryonic programs has long been posited for cancer (Bailey and Cushing, 1925), and could underlie tissue specific modes of regulating critical aspects of the malignant phenotype such as survival, angiogenesis, invasion and migration. The advent of comprehensive genomic profiling techniques now permits this hypothesis to be tested, and to link cancer to embryogenesis through cellular pathways that define specific lineages and organs (Garraway and Sellers, 2006). Identification of these pathways and their functions should identify targets for more specific and less toxic cancer therapies.
The prostate gland represents a prime target for such an analysis as the genetic causes of prostate cancer are poorly understood and because there are excellent models of prostate development and homeostasis. While other cancers have canonical cancer initiating mutations (for example, K-ras for pancreatic carcinoma and Adenomatosis Polyposis Coli for colon carcinoma), the driving force for prostate cancer initiation in humans remains uncertain. Answers may lie in cellular programs activated by androgen receptor (AR) signaling, which strictly controls prostate epithelial cell fate in embryogenesis and in cancer.
Upon binding androgen, AR signals through genomic and nongenomic modes (Manolagas et al., 2002). Genomic responses entail nuclear translocation of liganded AR and activation of transcription at regulatory regions containing AR-binding sites. Nongenomic signaling, in contrast, comprises protein–protein interactions in the cytoplasm and is exceedingly rapid, with measurable responses within minutes of androgen exposure (Kousteni et al., 2001). AR signaling, through genomic and/or nongenomic routes, is necessary and sufficient for prostate organogenesis. In response to circulating testosterone and its local conversion to the more potent androgen, dihydrotestosterone, AR signaling induces epithelial-mesenchymal interactions in the urogenital sinus (UGS), an embryonic rudiment present in both sexes. These interactions lead to outgrowth of buds from the UGS epithelium (UGE) that proliferate and invade surrounding urogenital sinus mesenchyme. AR expression has been described in both UGE and urogenital sinus mesenchyme (Drews et al., 2001; B Simons, EM Schaeffer and DM Berman, unpublished observations), although epithelial AR is dispensable for induction of prostate development (Cunha et al., 2004).
Hormonal manipulation of mouse prostate development
In mouse, epithelial expression of the androgen-sensitive homeobox family member Nkx3. 1 begins by embryonic day 16 (e16), and is the earliest reported marker of prostate development (Bhatia-Gaur et al., 1999). We comprehensively profiled gene expression in the UGS between e16 and e17.5, when prostate buds first emerge. The developmental fate of the UGS is bipotential in both sexes and depends soley on androgens to drive prostate formation in males (Jost, 1968). The absence of androgens leads to vaginal/urethral formation in females (Wilson et al., 1980). Taking advantage of this dynamic, we induced prostate development in the androgen-naïve yet androgen-sensitive female UGS with precisely timed 6 and 12h intrauterine exposure to pharmacological levels of dihydrotestosterone (Figure 1a). For the latest time point, approximating 48 h of physiologic androgen exposure, we compared UGS gene expression in unmanipulated male and female e17.5 littermates. In all cases, we isolated high quality RNA, prepared labeled cRNA pools and competitively hybridized androgen-exposed pools against androgen-naïve pools on Agilent 44 000 probe whole genome mouse microarrays. High quality expression data was confirmed as described in Supplementary (S) methods.
Distinct and prostate-specific phases of gene expression in early prostate development
The genomic response to androgen varied dramatically with the duration of exposure to the hormone. After 6, 12 and 48 h of androgen exposure, respectively, 693, 177 and 829 genes were differentially expressed (adj. P<0.03) (Figure 2a). Complete lists are provided in the Supplementary material (Supplementary Tables 1–3). Pairwise comparisons between any two time points revealed an overlap between 10–42% of differentially expressed genes. Despite some diversity in the most highly differentially expressed genes across different time points, we successfully validated the androgen-mediated developmental program using a variety of methods. Starting with the latest time point, reverse transcriptase–PCR analysis of 20 transcripts confirmed differential expression that was concordant with the array results (Supplementary Figure 1). As expected, Nkx3.1 was highly induced (adj. P<10−4) among the transcripts confirmed by reverse transcriptase–PCR. We further validated these results with separate biologic replicates and with RNA from embryos treated with pharmacologic AR blockade (Supplementary Figure 1). Using the 48 h time point as a standard, we performed ‘correspondence at the top (CAT)’ analysis (Irizarry et al., 2005) to compare identities of top ranking (most differentially expressed) genes in lists from different time points. CAT analysis showed that both pharmacologic and physiologic androgen exposure induced highly related gene expression programs in the UGS, (Figure 2d). This was particularly true when comparing gene expression at 12 h and 48 h; gene lists from these two time points showed 40% identity for top ranking genes. By connecting physiologic and pharmacologic androgen responses in the UGS, these results validated our experimental approach.
Several genes were consistently androgen regulated; either induced (Aspn, Klf9, Synpr, Gadd45 g, Sox9, Adamdec1 and Tle1) or suppressed (Prrx1 and Inhba) at all three time points (Figure 2b). This cohort may represent particularly important genes in establishing and maintaining prostate lineage.
The genomic response to androgen is dynamic
The overall pattern of genomic response to androgen was suppression, mild induction and robust induction of gene expression at 6, 12, and 48 h, respectively (Figure 2c). To better understand this pattern of regulation, we analyzed functional annotations of differentially expressed genes (Supplementary Methods). At 6 h, an overwhelming majority (76%) were suppressed (Figure 2c), and most (58%) of the suppressed genes contained one or more predicted binding site for miRNAs, far exceeding the number expected by chance (P<1 × 10−5; χ2-test). Considering previously described roles for miRNAs in developmental biology, and specific functions in targeting complementary transcripts for destruction (Stefani and Slack, 2008), our results suggest a mechanism for rapid gene suppression in response to androgen. After longer exposure to androgen, the majority of genes were induced, with 71 (69%) and 403 (54%) genes showing increased expression at 12 and 48 h, respectively.
At equal statistical stringency (log odd>1.1, adjusted P<0.03), we found more differentially expressed genes at the 6 and 48 h time points than at 12 h. A portion of this effect stems from additional biologic replicates available for 6 and 48 h (Supplementary Methods). However, biological factors likely explain most of the variability in numbers of differentially expressed genes. These include expression levels passing through neutral as they change from repressed at 6 h to induced at 48 h, and an amplification effect of transcription factor expression. Accordingly, 50% of genes with enrichment at both 6 and 48 h were repressed at 6 h and induced at 48 h. Several transcription factors were induced at 12 h and their predicted target genes were differentially expressed at 48 h (Supplementary Tables 2, 12). Examples include induction of Sox9 followed by differential expression of 33 of its predicted targets (Supplementary Table 12), including fibroblast growth factor R3 (FgfR3), Hoxb9 and Tle1 (see below). Altogether, these data indicate that androgen exposure reorganizes the genomic repertoire of the UGS toward prostatic growth and differentiation–first by suppressing a bipotential gene expression program, and then by inducing transcription factors that initiate and carry out prostate-specific gene expression.
Prostate-specific responses to androgen in the embryo
To further define the specificity of expression programs in embryonic prostate, we used publicly available gene expression data to compare early prostate development to embryonic and androgen responsive gene expression in other tissues. As the prostate utilizes programs of branching morphogenesis during development, we employed CAT analysis to compare top ranking genes in prostate development with top ranking genes associated with embryonic branching in the lung (Lu et al., 2004). Similarly, we compared gene expression in the developing prostate and another androgen-responsive organ, the adult salivary gland (Treister et al., 2005). We observed distinct clustering of top ranking genes by organ (Figures 2d and e). Rather than generic programs for branching morphogenesis and androgen response, these analyses highlight a distinct developmentally and hormonally restricted lineage program governing prostate development.
Discrete and overlapping features of prostate development and prostate homeostasis
Prostate homeostasis in adults can be studied by hormonal manipulation of the mature gland and consists of three phases: active cell death upon androgen withdrawal (regression), castrate steady state and androgen induced growth (regeneration; English et al., 1987). Cross platform analysis of the 1536 and 2197 genes differentially expressed (adj. P<0.001) in development or regeneration (Wang et al., 2007b), respectively, revealed only a limited number of differentially expressed genes enriched in both processes (333 genes, 9%) (Supplementary Figure 2). Annotation analysis reported previously for prostate regeneration noted significant enrichment of adhesion, proliferation and metabolic themes (Wang et al., 2007b). We hypothesized that prostate organogenesis would involve a wider variety of themes that govern additional processes, such as lineage differentiation, extracellular matrix remodeling and migration.
Androgen-activated embryonic gene expression invokes the ‘hallmarks’ of cancer
By analysis of functional annotation, we identified significant themes in early prostate development (Table 1, Supplementary Tables 4–12). A self renewing population of stem cells has recently been described in the adult murine prostate (Lawson et al., 2007), and stem cell-associated genes were enriched (P<0.05) among androgen response genes (Supplementary Table 19). Sca1 expression, which in part characterize stem cells in the adult prostate (Xin et al., 2005) was highly induced at the 48 h time point and may signal the emergence of an androgen-regulated progenitor population at this stage. All three time points exhibited enrichment (adj. P<0.001) of AR targets as predicted by transcription factor-binding site analysis. We also identified significant (adj. P<0.05) enrichment of genes associated with ‘hallmark’ characteristic activities of cancer (Hanahan and Weinberg, 2000) including angiogenesis, apoptosis, migration, motility and proliferation (Table 1). Similarly, and despite the dominant role of post-translational modifications in their regulation, several extracellular and intracellular signaling pathways showed significant androgen-induced responses. These included oncogenic pathways, such as FGF/mitogen-activated protein kinase (MAPK), phosphatase and tensin (PTEN)-PI3K-mTOR (mammalian target of rapamaycin) and Wnt (Table 1). These findings indicate that cancers may reactivate cellular programs initiated in development, a concept we test more formally below.
We also captured evidence of transcription factor networks acting downstream of the AR. Transcription factors including AR, Foxf2 and Sox9 were induced within 12 h of androgen exposure and associated with coordinate expression of their predicted target genes (Table 2, Supplementary Tables 11,12). This observation not only identifies new androgen-regulated activities of these transcription factors, but also cross validates the effects of androgens across different time points in our analysis. Altogether, this work, including detailed annotated gene lists (see Supplementary material), provides the first comprehensive map of the androgen response in prostate development and highlights androgen-mediated programs for regulated invasion, proliferation, survival and other cellular behaviors that become aberrantly regulated in cancer.
Signaling in prostate development: FGF/MAPK
Secreted fibroblast growth factor (FGF) ligands signal through their receptors to activate intracellular pathways controlling proliferation. FGFs have multiple known roles in development and constitutive activation of the pathway in mouse prostate leads to cancer (Acevedo et al., 2007; Memarzadeh et al., 2007). In development, Fgf7 and Fgf10 have been intensively studied as candidate paracrine signals that facilitate epithelial budding in response to androgen (Donjacour et al., 2003). However, despite dramatic effects of these peptides on prostate ductal outgrowth, Fgf7 and Fgf10 demonstrate largely equivalent expression patterns in male and female UGS tissue. Thus, other factors must be required to induce prostate development, possibly by modulating responsiveness to Fgfs (Thomson, 2001; Donjacour et al., 2003). In the developing prostate, we identified sexually dimorphic expression of Fgf receptors and ligands as part of the MAPK pathway through which they signal (Table 1). Fgf18, which binds to FgfR3, (reviewed in (Mohammadi et al., 2005) is induced at 6 h and then is highly suppressed later in prostate development. Fgf12 was induced in males, although it reportedly does not bind FgfRs (Olsen et al., 2003). Two Fgf receptors showed increased expression in androgen-induced tissue; FgfR3, which can bind a wide range of Fgf ligands, and Fgfr2 which encodes the preferred receptor for Fgfs 7 and 10 (Mohammadi et al., 2005). These results implicate Fgf12 and Fgf18 in UGS differentiation and suggest that previously observed sexually dimorphic responses to Fgf7 and Fgf10 could stem from male-specific expression of FgfR2.
Signaling in prostate development: PTEN/PI3K/mTOR signaling
Aberrant signaling through PTEN and downstream effectors is highly implicated in prostate cancer, but its expression in early prostate development is uncharacterized. PTEN function suppresses a downstream signaling cascade whose effectors PI3 K, AKT (a proto-oncogene), mTOR and ribosomal S6 kinase (S6K) promote a wide variety of cellular functions, including growth, survival, migration and angiogenesis. PTEN loss is common in advanced prostate cancer (McMenamin et al., 1999; Wang et al., 2003) and PTEN loss is oncogenic in the mouse prostate, perhaps due to its promotion of epithelial cell growth and survival in the face of androgen withdrawal (Shen and Abate-Shen, 2007). Intriguingly, androgen rapidly induced several components of this pathway, including Pik3r, a PI3K regulatory subunit Rheb, a small GTPase Ras homolog required for mTOR activity (Long et al., 2005), and Rps6kb1, a subunit of the mTOR target, ribosomal protein S6 kinase. By the 48-h time point, PTEN transcripts were suppressed, along with the Pik3r2 regulatory subunit of PI3K. These results indicate dynamic androgen-responsive regulation of the Pten signaling pathway. Future studies in prostate development may therefore elucidate normal Pten/PI3K/mTOR regulation, its functional roles and more effective strategies for pathway measurement and manipulation.
Signaling in prostate development: the Wnt pathway
Extending an observation made previously by Serial Analysis of Gene Expression in the urogenital sinus (Zhang et al., 2006), analyses of functional annotation revealed Wnt signaling as one of the most androgen regulated pathways in early prostate development (adj. P<0.001), with regulation of multiple Wnt ligands and soluble inhibitors (Table 1). Wnt ligands are highly conserved secreted molecules that play critical but pleiotropic roles in cell-cell signaling during embryogenesis (Nusse, 2005). Studies of this pathway in prostate cancer show up- and downregulation of Wnt ligands, both stimulatory and inhibitory, with little or no agreement as to their net effect (Verras and Sun, 2006). Likewise, nearly all Wnts are expressed in the developing prostate (not shown) along with Wnt inhibitory factor (Wif-1), and secreted frizzled proteins, which can modulate UGS development in culture (Joesting et al., 2005; Table 1). Androgen-responsive expression was also observed for Wnt ligand coactivator, R-spondin1 (Rspo1; Nam et al., 2007), and several intracellular components of the pathway, including disheveled 1 (Dvl1), lymphoid enhancer-binding factor 1 (lef1) and transducin-like enhancer of split 1 (Tle1; Table 1). Of special interest, the Wnt target gene matrix metalloproteinase 7 (Mmp7; Crawford et al., 1999) has a well-established role in cancer invasion (Ii et al., 2006), and is induced during prostate budding (Table 1). These results suggest a general role for Wnt signal modulation in prostate development that may underlie the mechanics of epithelial invasion in prostate development and prostate cancer.
Prostate carcinogenesis and progression reactivates androgen-induced embryonic gene expression programs
Having demonstrated that androgen-responsive embryonic gene expression programs regulate processes that are fundamental to cancer, we examined the expression of embryonic prostate genes in human prostate cancer tissues. Like prostate organogenesis, initiation and progression of prostate cancer depends on AR signaling (Scher and Sawyers, 2005). Neoplastic transformation in the prostate is first recognized as high grade prostatic intraepithelial neoplasia (PIN), a noninvasive lesion that precedes invasive cancer (Bostwick and Qian, 2004). Invasive prostate cancer can be indolent, particularly in lower grade tumors (Gleason 6 or less). Higher grade tumors (Gleason 8–10) more often metastasize and cause death. Using public domain gene expression data from a large and well-characterized series (Lapointe et al., 2004), we assembled lists of genes that characterize transitions to more invasive lesions (that is, normal vs locally invasive cancers; lower grade vs higher grade locally invasive cancers and localized vs metastatic cancers). Comparing these lists to lists of differentially expressed genes at each of the developmental time points revealed significant (adj P<0.001) enrichment of developmental genes in invasive transitions, especially in more aggressive prostate cancers (Figure 3a, top). In contrast, gene expression in adult prostate regeneration showed little enrichment in cancer (Figure 3a, bottom). The latter result was somewhat surprising, given the previously reported enrichment of cell cycle, cell adhesion and metabolic genes in the prostate regeneration study (Wang et al., 2007b).
Using laser capture microdissection to isolate and profile pure epithelial cell populations from the prostate, Tomlins et al. (2007) have assigned much of the variation in previously published prostate cancer gene expression studies to varying ratios of cancer epithelia to benign stroma. We therefore performed a second analysis, looking for enrichment of developmental genes in the cancer contrasts from Tomlins et al. (2007). Surprisingly, these analyses (Figure 3b) showed increased enrichment of embryonic genes compared with macrodissected bulk cancer samples (Figure 3a). Enrichment was significant for all time points (adj P<0.001), and included transitions for categories indicating malignant transformation (normal vs PIN), invasion (PIN vs cancer) and aggressiveness (Gleason grade; Figure 3b). The genes driving these processes are listed in Supplementary Tables 13–18. Thus, invasive prostate cancer adopts embryonic programs particularly during neoplastic transformation and invasion (PIN vs cancer) as indicated by enrichment across all three developmental time points (boxes 2, 5 and 6 in Figure 3b). In addition, genes differentially expressed in prostate cancer progression overlap most significantly with genes differentially expressed at the earliest (6 h) time point in prostate development, when prostate identity is first becoming established. Although most of these genes are repressed at the 6-h time point in development (Figure 2c), the same genes show a more equal balance between induction and repression (Supplementary Tables 13–18) in cancer. These observations identify androgen-responsive gene networks that establish prostate identity and operate in prostate cancer. Comparing and contrasting normal and malignant operation of these networks will likely yield useful insights into the molecular basis of prostate cancer.
Embryonic genes drive neoplastic transformation and invasion
Several shared functional themes from early prostate development also operate in human prostate cancer tissues, although at distinct stages of disease progression (Figure 4). Selected differentially expressed genes (P<0.05) and themes enriched in development and cancer are listed in Table 2. Complete lists are provided in Supplementary Tables 13–18. The transition between normal prostate and PIN is characterized by enrichment in pathways regulating transcription and apoptosis (Figure 4a). The latter observation is in keeping with striking suppression of programmed cell death previously observed in PIN (Zeng and Kyprianou, 2005). In contrast, the transition from noninvasive PIN to invasive cancer is enriched for extracellular matrix remodeling, cell motility and angiogenesis, as well as signaling pathways including Wnt and bone morphogenetic proteins (Figure 4a). These observations outline a program for prostate invasion that serves prostate development and cancer.
Interestingly, other investigators have reported that metastatic potential can pre-exist in localized tumors rather than being induced by the external environment (Bernards and Weinberg, 2002; Kang et al., 2003). This concept may explain why the transition from localized to metastatic lesions did not represent themes expected to underlie metastasis, such as cell migration and adhesion. Instead, enriched processes included microtubule dynamics, mitotic spindle regulation and enhanced intracellular signaling (tyrosine kinase activity and adaptor proteins). Proteins in the latter class make excellent drug targets (Zwick et al., 2002), and to the extent that these genes might facilitate prostate cancer metastasis, they warrant further investigation.
Prostate budding recapitulates the aspects of neoplastic transformation
As a complement to statistical analyses and reverse transcriptase–PCR confirmation, we performed immunohistochemical analysis of early prostate development. Here, we show examples of two genes, Anxa1 and Sox9, which are coordinately suppressed and activated, respectively in prostate epithelial buds, mirroring their regulation in prostate carcinogenesis.
Annexin A (Anxa1) is a potent inducer of cell death in prostate cancer cell lines (Ornstein and Tyson, 2006). Reduced Anxa1 expression characterizes 91% percent of PIN lesions/cases, and is maintained in 94% of prostate cancers (Kang et al., 2002). Anxa1 expression has not been described in prostate development. In contrast to its loss in cancer, Anxa1 transcript expression was strongly induced at the 48-h time point (Supplementary Table 3). Immunohistochemical staining of developing prostates confirmed increased expression of Anxa1 in male tissue; however, there was a striking reduction of expression in emerging epithelial buds (Figures 5a–c). This pattern of expression is consistent with a role for Anxa1 repression in growth, survival, and/or invasion of prostate epithelium. Future studies involving genetic and/or pharmacologic manipulation of the developing prostate should further elucidate this role.
Sox9 is a member of the SOX (Sry-related high mobility group) family of transcription factors that binds a DNA consensus sequence and enhances steroid hormone receptor binding. Functionally, it has been linked to a number of themes shared by development and cancer (Table 2). In developing prostate, Sox9 and its predicted transcriptional targets are differentially expressed during each of the three time points after androgen exposure (Supplementary Tables 1–3, 10–12). Immunohistochemical analysis revealed that Sox9 is expressed in early prostate epithelium and is particularly concentrated in the tips of invading prostate buds (Figure 5d). This expression pattern contrasts to that of Anxa1 with its reduced staining in bud tips and differs from p63, which is diffusely expressed in UGE at this stage of development. Altogether, this suggests Sox9 is critically located to facilitate prostate outgrowth, a hypothesis supported by recent work demonstrating its essential role in prostate development (Thomsen et al., 2008; Z Huang, B Simons, EM Schaeffer and DM Berman, unpublished observations). We noted that Sox9 transcripts and predicted targets were differentially expressed across prostate cancer progression (Supplementary Tables 14–18), and others have linked Sox9 to invasion, cell growth and metastasis in prostate cancer models.
We further characterized SOX9 expression in a panel of 219 human prostate cases, using immunohistochemistry and Immunofluorescence. In benign prostate tissue, Sox9 is expressed predominantly in basal epithelial cells, the stem cell compartment for prostate epithelia. (Xu et al., 2005; Wang et al., 2007a). We discovered aberrant Sox9 expression in neoplastic luminal epithelial cells in 51 of 53 (96%) PIN lesions (Supplementary Table 20), indicating an association with neoplastic transformation (Figures 5e and f). In prostate carcinomas (n=105), basal cell gene expression is lost; however Sox9 expression is maintained in cancer cells (n=59, 56%) with a trend toward increased expression with increasing grade (P=0.1; Supplementary Table 20). Nuclear Sox9 was also noted in similar numbers (17/31, 55%) of advanced (lymph node metastasis) prostate cancer cases. In sum, alterations in the geographic and temporal location of Sox9 expression parallel the earliest events in the neoplastic transformation of prostate epithelial cells. These results suggest that Sox9 may play a critical role in the early, initiating phase of prostate carcinogenesis and contribute aspects of the basal/stem cell phenotype to prostate cancer. This pattern differs from recent work, which identified increased Sox9 in metastatic rather than localized lesions in a murine prostate cancer model (Acevedo et al., 2007). It is likely that Sox9 has several context-dependent functions in prostate epithelium and prostate cancer. The list of predicted Sox9 target genes that participate in prostate development (Supplementary Table 21) represents a useful starting point for further investigation.
Previous studies have linked primitive embryonic gene expression profiles to aggressive subsets of brain and lung cancers (Kho et al., 2004), thereby supporting balances between differentiation state and growth potential that operate similarly in organogenesis and tumorigenesis. Here, we identified prostate-specific programs for growth, survival, angiogenesis, and invasion that originate in organogenesis and are reactivated at specific steps in cancer progression. This curated list of androgen-regulated programs in prostate development and carcinogenesis provides a roadmap for understanding prostate growth and invasion. Prostate development in the mouse becomes a tractable experimental system in which to investigate the specific functions of these genes. This model bypasses the difficulty of probing gene regulation in human prostate cancer cell lines, which represent rare and possibly skewed exceptions to the rule that human prostate cancers do not adapt to growth in the laboratory. Unlike the extant models of prostate cancer involving transgenic mice, prostate development is unbiased by a preselected genetic lesion. Development is reproducible, genetically (Xin et al., 2003) and pharmacologically (Berman et al., 2004) tractable, and shown here to be reflected the entire spectrum of human prostate cancer progression.
A unique feature of these studies was the ability to induce prostatic lineage commitment and growth by controlled induction of signaling by AR, a gene with lineage-specific oncogenic properties (Garraway and Sellers, 2006). Evidence indicates the operation of other lineage-specific oncogenes in melanocytes (Garraway and Sellers, 2006), lung (NKX2-1; Weir et al., 2007) and elsewhere in the body (Garraway and Sellers, 2006), suggesting that gene expression programs relevant to other types of cancers can be identified, manipulated and modeled in the embryo.
Materials and methods
Mice and tissues
C57/Bl6J (The Jackson Laboratories, Bar Harbor, ME, USA) pregnancies were timed according to scheduled 4 h pairings. Paired pregnant females were injected intraperitoneally with dihydrotestosterone (Sigma-Aldrich, St Louis, MO, USA) at 50 mg/kg or triolein vehicle at e16.0. 6 or 12 h later, embryos were sexed and female UGS tissue dissected. For the 48-h time point tissue from unmanipulated male and female littermates was harvested at e17.5.
Frozen tissue was homogenized and total RNA purified using the RNeasy system (Qiagen, Hilden, Germany) and analyzed using a Bioanalyzer 2100 (Agilent Technologies, Santa Clara, CA, USA). Each time point yielded pools of ‘androgen exposed’ and ‘androgen naïve’ RNA. There were a total of 8,6 and 10 pools of RNA at the 6-, 12- and 48-h time points, respectively. Each pool was obtained from 3–4 individual UGS.
Probe synthesis, hybridization to oligonucleotide arrays and detection
Amplification, labeling, hybridization, and detection of 250 ng samples were carried out according to the manufacturer's directions (Agilent).
Data were processed without background subtraction with packages from R/Bioconductor (www.bioconductor.org/) (Ihaka and Gentleman, 1996; Gentleman et al., 2004). Within-array ‘loess’ and between-arrays ‘scale’ normalization methods were applied to log2 expression ratios. Moderated t-statistics (by empirical Bayes shrinkage of standard errors (Smyth, 2004)) log odds ratios of differential expression and adjusted P-values (Benjamini et al., 2001) were obtained from a linear model accounting for biological replicates’ correlation, and dye/group effects.
Functional themes were obtained from Gene ontology (Ashburner et al., 2003), KEGG (Kanehisa et al., 2004), MsigDb(Subramanian et al., 2005). Enrichment analysis was performed by one-sided Wilcoxon test and multiple testing correction performed separately (Benjamini et al., 2001). Detailed statistical methods, public microarray data set identification and MIAME (Minimal Information about Microarray Experiments) information are provided in Supplementary materials. Data will be hosted in the Gene Expression Omnibus database.
Detection was performed as described in Berman et al.(2004) using Sox9 (Chemicon, Billerica, MA, USA) and Annexin A1 (Invitrogen, Carlsbad, CA, USA) antibodies.
Detection was performed with anti-Sox9 and anti-p63 (Santacruz, Santa Cruz, CA, USA) antibodies as described in supplement (Bivalacqua et al., 2007).
We thank W Matsui and N Watkins for comments on the manuscript, A DeMarzo and J Epstein for TMAs and YQ Chen for prostate regeneration data. These studies were funded by the Evensen Family, Passano and Patrick C Walsh Prostate Cancer Foundations, and NIH5K08DK059375 (DB) NIHK08 DK081019 (EMS), NIH5P30CA06973-39 (GP) and NSF034211 (GP and LM).