Loss of FOXO1 Cooperates with TMPRSS2–ERG Overexpression to Promote Prostate Tumorigenesis and Cell Invasion

Prostate cancer is the most commonly diagnosed cancer and the third leading cause of cancer death in American men (1). Multiple genetic alterations including gene mutations, deletions, and amplifications have been revealed by next-generation high-throughput sequencing in both primary and advanced prostate cancer (2, 3). However, the pathologic consequence of many alterations detected in human prostate cancer specimens has not been rigorously examined in cell culture systems and mouse models.

TMPRSS2–ERG gene rearrangements are one of the most frequent genetic alterations detected in human primary prostate cancer (2, 4). ERG is an oncogenic protein that belongs to the E26 transformation-specific transcription factor family (5). TMPRSS2–ERG gene fusions juxtapose the androgen-responsive TMPRSS2 gene promoter with the ERG gene coding region. Such fusions result in aberrant overexpression of ERG in approximately 50% of both primary and advanced human prostate cancer, suggesting a causal role of ERG in prostate tumorigenesis and progression.

Since the first report of the TMPRSS2–ERG gene fusion by Tomlins and colleagues (4), more than 20 different types of gene rearrangements have been identified in prostate cancer patient samples (6, 7). Among these, fusion of TMPRSS2 exon 1 (a noncoding exon) to ERG exon 4 or 5 (designated as T1/E4 and T1/E5, respectively) are the two most frequently detected TMPRSS2–ERG rearrangements in patients (6, 7). Notably, mice with prostate-specific transgenic expression of T1/E4 ERG develop very minor cancer precursor-like lesions, but not prostate neoplasia (8–11), suggesting that ERG requires other lesion(s) to drive prostate tumorigenesis. Indeed, further studies demonstrate that T1/E4 ERG cooperates with Pten deletion or AKT activation to induce prostate cancer in mice (8, 9, 12). Conversely, ERG overexpression is also linked with aspects of later-stage tumors, such as cell invasion. ERG regulates expression of matrix metalloproteinase genes MMP3, MMP9, ADAM19, plasminogen activator pathway genes PLAT and PLAU, and chemokine receptor CXCR4 in prostate cancer (11, 13). However, it remains unclear which other prostate cancer–relevant lesions may cooperate with ERG overexpression to promote both prostate tumorigenesis and tumor progression.

Forkhead box transcription factors FOXO1, FOXO3, FOXO4, and FOXO6 (the human orthologs of Caenorhabditis elegans DAF-16 and Drosophila melanogaster dFOXO) are often recognized as tumor suppressors (14). Activation of these factors results in transcriptional upregulation of genes involved in apoptosis (e.g., Bim and FasL), cell-cycle arrest (e.g., p27^KIP1 and p21^CIP1), and oxidative stress detoxification (e.g., MnSOD and Catalase; refs. 15–19). Mouse genetic studies demonstrate that somatic deletion of Foxo genes promotes formation of a cancer-prone condition characterized by thymic lymphomas and hemangiomas, suggesting that FOXOs are bona fide tumor suppressors (20). Moreover, FOXO1 and other FOXO proteins function as key downstream effectors of the PTEN tumor suppressor (21). Interestingly, the PTEN tumor suppressor gene is frequently deleted in human prostate cancer and occurs concomitantly with ERG overexpression (9). PTEN loss leads to AKT and CDK1- or CDK2-mediated phosphorylation of FOXO1, exclusion of FOXO1 from the nucleus, and loss of the tumor suppressor functions in the nucleus (15, 22–24). Increasing evidence suggests that AKT-phosphorylated FOXO1 also possesses tumor suppressor functions in the cytoplasm (25–27). In agreement with these findings, proteosome degradation of AKT-phosphorylated FOXO1 is critical for PI3K-AKT-mediated cell transformation and oncogenesis (28–30). In addition to PTEN defect-caused inactivation of FOXO1, the FOXO1 gene is also frequently deleted at the genomic level or downregulated at the transcriptional level in human prostate cancer (31).

In this study, we demonstrate that FOXO1, but not FOXO3 or FOXO4, binds to prostate cancer–associated ERG and inhibits ERG-mediated gene transcription, as well as subsequent ERG-mediated invasion of human prostate cancer cells. Importantly, we also find that concomitant deletion of the Foxo1 gene and transgenic expression of prostate cancer-associated ERG promotes prostate tumorigenesis in mice, whereas each individual lesion alone has no such effect.

Mammalian expression vectors for HA-tagged full-length wild-type ERG, HA-tagged T1/E4 ERG (Δ39), and HA-tagged T1/E5 ERG (Δ99) were described previously (32). Mammalian expression vectors for FLAG–FOXO1, FLAG–FOXO1–HR-537, HA—AKT–CA, PTEN WT, PTEN C142S, PTEN G129R, PTEN G129E, and GST–FOXO1 fusion protein vectors were described previously (33, 34). GST–ERG fusion protein vectors were also described previously (32). Empty vector pcDNA3.1 was purchased from Thermo Fisher Scientific. MMP9 promoter-based firefly luciferase reporter was acquired from Addgene. Antibodies used for Western blotting were: anti-ERG (sc353 and sc354) and anti-ERK2 (sc1647; Santa Cruz Biotechnology); anti-FLAG M2 (Sigma Aldrich); anti-HA 1.1 (Covance); anti-FOXO1 (9462L), anti-AKT (9272S), anti-PTEN (9559L; Cell Signaling Technology); anti-light chain specific rabbit IgG secondary antibody (211-032-171; Jackson ImmunoResearch Laboratories); ECL anti-rabbit (or anti-mouse) IgG horseradish peroxidase–linked whole antibody (GE Healthcare UK Limited). Antibodies used for coimmunoprecipitation (co-IP) were: mouse IgG and rabbit IgG (Vector Laboratories, Inc.); anti-Myc (2276) and anti-FOXO1 (9462L; Cell Signaling Technology). Antibodies used for chromatin immunoprecipitation (ChIP) were: rabbit IgG (Vector Laboratories, Inc.); anti-ERG (ab92513; Abcam).

The cell lines VCaP, PC-3, and LAPC-4 were purchased from ATCC and authenticated via STR profiling. VCaP cells were cultured in DMEM (Corning cellgro) supplemented with 13% FBS (Thermo Fisher Scientific 10437028). PC-3 cells were cultured in RPMI1640 medium (Corning cellgro) supplemented with 10% FBS (Thermo Fisher Scientific 10437028). LAPC-4 cells were cultured in Iscove's Modified Dulbecco's Medium (Corning cellgro) supplemented with 15% FBS (Thermo Fisher Scientific 10437028). Potential contamination mycoplasma was often checked using the Lookout Mycoplasma PCR Detection Kit purchased from Sigma-Aldrich. Cell culture medium was routinely supplied with Plasmocin (InvivoGen) to prevent mycoplasma contamination. Transfections were performed following manufacturer's instructions with Lipofectatmine2000 (Thermo Fisher Scientific) or by electroporation using Electro Square Porator ECM 830 (BTX) with Mirus Ingenio solution. Approximately 75–90% transfection efficiencies were achieved.

Cells were transfected with siRNA following manufacturer's instructions, by electroporation. Nonspecific control siRNA was purchased from RIBOBIO (siN05815122147). siRNA for PTEN was purchased from Thermo Fisher Scientific (M00302302), for ERG was purchased from Dharmacon (M003886010005), and for FOXO1 was purchased from Dharmacon (D003006060020).

Cell invasion was quantified by Crystal violet staining using the Corning Matrigel invasion chamber assay according to manufacturer's instructions (Corning). Cells were transfected with indicated plasmids or siRNA and cultured in medium with normal serum concentrations before plating in Corning Matrigel invasion chambers in 24-well plates. Once in the invasion chambers, cells were cultured in serum-free medium inside the chamber, with medium containing 10% FBS outside the chambers. After 24 hours, cells were fixed in methanol for 15 minutes and then stained with 1 mg/mL Crystal violet in 10% ethanol for 30 minutes. After rinsing with water three times, the membranes of the chambers were mounted and covered on slides and observed using a light microscope. Eight fields of each view from three independent replicates were recorded and analyzed.

Migration assays were performed using a 24-well Transwell chamber system (Corning Inc.). Cells were transfected with indicated plasmids and cultured in medium with normal serum concentrations before plating in Corning migration chambers in 24-well plates. Cells were seeded in the upper chamber at 1.5 × 10⁴ cells/mL in 0.1 mL serum-free culture media. Media supplemented with 10% FBS was placed in the bottom well in a volume of 0.8 mL (used as a chemoattractant). After incubation for 24 hours at 37°C in an atmosphere containing 5% CO₂, migrated cells on the lower surface were stained with Crystal violet stain and counted under a light microscope. All experiments were repeated six times over the days.

Foxo1 loxp/loxp (Foxo1^p/p) conditional mice were originally generated in the laboratory of Dr. Ronald DePinho and reported previously (20). Transgenic ERG mice were purchased from The Jackson Laboratory (010929), originally generated in the laboratory of Dr. Valeri Vasioukhin at Fred Hutchinson Cancer Research Center, Seattle, WA (10). Probasin (Pb)-Cre4 transgenic mice were acquired from the National Cancer Institute (NCI) Mouse Repository, originally generated in the laboratory of Dr. Pradip Roy-Burman at University of Southern California, Los Angeles, CA (35). Foxo1^pc−/−;Pb-ERG mice were obtained by cross-breeding Pb-Cre4 males with Foxo1^pc−/− and Pb-ERG females. All mice were maintained under standard conditions of feeding, light, and temperature with free access to food and water. All experimental protocols were approved by the Institutional Animal Care and Use Committee at Mayo Clinic.

Genotyping of wild-type and conditional alleles of Foxo1 genes as well as the Cre and ERG transgenes was performed following standard PCR protocol with primers listed in Supplementary Table S1.

Four-micrometer-thick sections were cut from formalin-fixed paraffin-embedded (FFPE) mouse prostate tissues and mounted onto slides. Slides were deparaffinized with xylene and rehydrated through graded ethanol washes. Slides were then stained with hematoxylin, and washed with water followed by ethanol before counterstaining with 1% eosin. Finally, slides were dehydrated through graded ethanol washes and xylene washes before coverslips were sealed over the tissue sections.

Four-micrometer-thick sections were cut from FFPE tissues and mounted onto slides. Antigen retrieval and immunostaining was performed as described previously (34). Antibodies used for IHC were: Abcam: anti-Ki-67 (ab15580), anti-ERG (ab92513); Cell Signaling Technology: anti-FOXO1 (29H4). Hematoxylin was applied for counterstaining. Cleaved caspase-3 IHC detection was performed as described previously using cleaved caspase-3 IHC Detection Kit (Cell Signaling Technology).

Prostate cancer tissue microarrays (TMA) were purchased from US Biomax, Inc. As indicated by US Biomax, Inc. (https://www.biomax.us/FAQs#q10), all tissues were collected under the highest ethical standards with the donors being formally informed and with their consent obtained. Given that US Biomax, Inc., has the patient consent form in place already, the investigators of this study did not acquire additional informed written consent from the subjects. However, the patient studies were conducted in accordance the ethical guidelines of Declaration of Helsinki and approved by Mayo Clinic Institutional Review Board. TMA specimens were used for antigen retrieval and immunostaining as described previously (36). Primary antibodies used were anti-FOXO1 (Bethyl) and anti-ERG (Abcam). Staining intensity and staining percentage for each tissue was graded using a set of criteria. Intensity was graded 0 to 3, with 0 being no staining, 1 low staining, 2 medium staining, and 3 strong staining. A final staining index (SI) score for each staining was obtained by multiplying values obtained from staining percentage and intensity and used for correlation analysis.

Cell culture experiments were carried out with three or more replicates. Statistical analyses were performed by the Student t test for cell culture and mouse tissue studies. Heatmap and correlation for ERG and FOXO1 IHC were generated by R software (version 2.15.0 from http://www.r-project.org).

Because ERG cooperates with PTEN loss in prostate tumorigenesis (8, 9, 12, 37) and FOXO1 is a key downstream effector of PTEN (21, 38, 39), we first examined whether the transcriptional activity is regulated by FOXO1. To this end, we performed MMP9 promoter-based luciferase assay because MMP9 is a well-studied ERG transcriptional target gene (40). As expected, expression of wild-type ERG (ERG-WT) and T1/E4 and T1-E5, two most frequently detected TMPRSS2-ERG rearrangements (Fig. 1A), increased the activity of the MMP9 reporter gene in PC-3 cells, a TMPRSS2-ERG fusion-negative prostate cancer cell line (40) (Fig. 1B). Importantly, ectopic expression of wild-type FOXO1 (FOXO1-WT) inhibited the transcriptional activity of ERG-WT, T1/E4, and T1/E5 ERG (Fig. 1B). This observation is consistent with previous reports that while a significant portion of ectopically expressed FOXO1–WT proteins retain in the cytoplasm of PTEN-deficient prostate cancer cells like PC-3, some FOXO1 proteins are localized and functional in the nucleus (33). The ERG inhibitory activity of FOXO1 was largely enhanced by wild-type PTEN, but not tumor-associated, lipid phosphatase-deficient mutants (C124S, G129R, and G129E) in PTEN-null PC-3 cells (Fig. 1C), which is consistent with the previous findings that restored expression of PTEN largely increased nuclear localization and activity of FOXO1 in PTEN-deficient prostate cancer cells (21, 33). Moreover, expression of the constitutively active FOXO1-A3, in which three AKT phosphorylation sites (Thr24, Ser256, and Ser319) were converted to alanine residues, largely inhibited ERG transcriptional activity and importantly, this effect was not enhanced by PTEN (Fig. 1D). Conversely, ectopic expression of constitutively active AKT (AKT-CA) in PC-3 cells completely reversed ERG inhibition induced by FOXO1-WT expression in a dose-dependent manner, but not by FOXO1-A3 expression (Fig. 1E and F). Next, we examined the effect of endogenous FOXO1 on expression of transcriptional target genes of endogenous TMPRSS2–ERG fusions. We demonstrated that knockdown of endogenous FOXO1 (by a pool of three independent siRNAs) increased expression of ERG target genes (e.g. PLAT, PLAU, ADAM19, MMP3, and MMP9; ref. 11) in T1/E4 ERG fusion-positive human prostate cancer cell line VCaP (4), and similar results were obtained in PTEN knockdown cells (Fig. 1G and H). Notably, FOXO1 and PTEN co-knockdown failed to further increase the expression of these genes (Fig. 1G and H), suggesting that PTEN and FOXO1 regulate ERG activity through the same pathway. Taken together, these data identify FOXO1 as a negative regulator of ERG transcriptional activity, and this effect is inhibited due to PTEN loss or AKT activation.

To elucidate how FOXO1 regulates ERG transcriptional activity, we next explored whether ERG and FOXO1 can bind one another. co-IP assays demonstrated that endogenous FOXO1 forms a protein complex with endogenous T1/E4 ERG in VCaP cells (Fig. 2A), indicating an interaction at the physiological condition. Ectopically expressed FOXO1 also interacted with expressed T1/E4 ERG, the most commonly occurring TMPRSS2–ERG rearrangement, in ERG-fusion negative PC-3 cells (Fig. 2B). This interaction is specific to FOXO1, as there was no detectable interaction of T1/E4 ERG with FOXO3 and FOXO4, two other FOXO family members expressed in prostate cancer (Fig. 2B; refs. 23, 41). In vitro protein binding assays with purified GST-tagged fragments of ERG and endogenous FOXO1 from LAPC-4 cell lysate revealed that FOXO1 binds to the N-terminus of ERG (amino acids 1–200; Fig. 2C and D). Conversely, using GST-tagged fragments of FOXO1 purified from bacteria and in vitro transcribed and translated ERG proteins, in vitro protein binding assays demonstrated that ERG binds directly to a central region of FOXO1 (amino acids 149–419; Fig. 2E and F). Given that FOXO1 interacts with T1/E4 ERG, which lacks N-terminal 39 amino acids (Figs. 1A and 2A and B), these data indicate that FOXO1 binds to the region of ERG constituting of amino acids 40–200.

Both FOXO1 and ERG are highly expressed in normal endothelial cells (42, 43). Endothelial cells are useful to determine whether FOXO1 interacts with ERG under physiological conditions in cells that normally express both proteins. Endothelial cells also express androgen receptor (AR; ref. 44). Similar to the finding in prostate cancer cells (45), co-IP assays demonstrated that ERG interacts with AR in HUVEC cells (Supplementary Fig. S1A). In contrast, no FOXO1–ERG interaction was detected under the same condition in HUVEC cells (Supplementary Fig. S1A). Although the exact molecular basis underlying the binding difference in different cell types is unclear at present and warrants further investigation, these findings are highly significant because it suggests that due to the presence of their interaction ERG functions are constrained to certain degree by FOXO1 in prostatic cells, but not in endothelial cells. As a result, FOXO1 deletion can be oncogenic in the prostate, but not in endothelial cells since loss of FOXO1 inhibition of ERG occurs specifically in prostatic cells.

FOXO1–HR-537 is a DNA binding- and transactivation-dual deficient mutant of FOXO1, in which histidine 215 (a key residue for DNA binding) is mutated to arginine (H215R) and the transactivation domain (TAD, amino acids 538–655) is deleted (Fig. 3A; refs. 33, 34). To determine whether FOXO1 inhibition of ERG activity is mediated through protein–protein or the transcriptional activity of FOXO1, PC-3 cells were cotransfected with T1/E4 ERG in combination with FOXO1–WT or FOXO1–HR-537. Ectopic expression of each construct inhibited expression of ERG target genes PLAU, ADAM19, and MMP9 to a similar extent (Fig. 3B). Luciferase assays also showed that ectopic expression of FOXO1–HR-537 inhibited the transcriptional activity of T1/E4 ERG to the same extent as FOXO1–WT (Fig. 3C). These data suggest that FOXO1 inhibits T1/E4 ERG in a manner independent of its DNA binding and transactivation functions. We further demonstrated that FOXO1 inhibition of ERG-mediated transcription does require the ERG-binding region in FOXO1 encompassing amino acids 167–354 (Fig. 2E, F and 3D). Thus, FOXO1 inhibits ERG-mediated gene expression via protein–protein interaction, but independently of FOXO1 DNA binding and transactivation functions.

Because protein–protein interaction inhibited ERG target gene expression, we examined the effect of FOXO1–ERG interaction on ERG recruitment to chromatin. ChIP assays in VCaP cells demonstrated that ectopic expression of either FOXO1-WT or transactivation-deficient mutant FOXO1–HR-537 invariably inhibited ERG recruitment to target genes PLAU, MMP3, and MMP9 to a similar degree (Fig. 3E and F). Conversely, knockdown of endogenous FOXO1 by a pool of three independent siRNAs in VCaP cells increased endogenous ERG binding to its target genes (Fig. 3G and H). These data further support our hypothesis that FOXO1 can inhibit ERG recruitment to target genes through protein–protein interaction. Given that FOXO1 does not directly binds to the DNA binding motif of ERG, the exact mechanism by which FOXO1 interaction reduces ERG recruitment to its downstream targets is unclear. However, our data cannot rule out the possibility that FOXO1 interaction with the N-terminal half of ERG may affect ERG DNA binding by causing overall conformation changes in ERG protein as demonstrated in IQGAP1, another FOXO1 interacting protein (27).

Many ERG target genes of interest in this study regulate cellular invasion and migration in prostate cancer (5), including PLAT, PLAU, MMP3, MMP9, and ADAM19. Previous reports have demonstrated that ERG overexpression in immortalized, nonmalignant TMPRSS2–ERG fusion negative prostatic epithelial cells (e.g., BPH-1 and RWPE cells) promotes cell invasion (8, 10, 46). We further explored how knockdown of endogenous FOXO1 or ERG in TMPRSS2–ERG fusion positive VCaP cells affected cell invasion. Knockdown of FOXO1 by a pool of three independent RNAs increased invasion of VCaP cells (Fig. 4A–C). This effect was reversed by coknockdown of endogenous ERG (Fig. 4A–C). Similarly, the increase in cell invasion observed after ectopic expression of ERG in ERG fusion–negative PC-3 cells was abrogated by coexpression of either FOXO1–WT or transactivation-deficient mutant FOXO1–HR-537 (Fig. 4D and E). These results are consistent with previous findings from us and others that TMPRSS2–ERG is a known predominant factor in control of cellular invasion of fusion-positive cells such as VCaP (11), and that FOXO1 can inhibit ERG-mediated cell invasion independently of FOXO1 transcriptional activity. Next, we examined the effect of the ERG-binding region (amino acids 167–354) in FOXO1 on ERG-induced prostate cancer cell migration and metastasis. Similar to the effect on cell invasion, we demonstrated that ERG expression in ERG fusion-negative PC-3 cells increased cell migration (Fig. 4F and G). This effect was reversed by coexpression of constitutively active FOXO1-A3, but not the ERG binding region-deletion mutant FOXO1–A3Δ167-354 (Fig. 4F and G). These data suggest that the ERG binding region in FOXO1 is important for FOXO1 inhibition of ERG-induced prostate cancer cell migration.

It has been shown previously that expression of FOXO1 is often completely or partially lost due to genomic deletion, promoter methylation, or transcriptional downregulation in human prostate cancer cell lines and patient specimens (31, 41, 47). These observations are further supported by the result obtained from The Cancer Genome Atlas (TCGA) data set (Supplementary Fig. S1B; ref. 2). In contrast, ERG proteins are frequently upregulated through mechanisms of gene fusions and posttranslational modifications. We employed a TMA strategy to examine expression of these two proteins in a cohort of prostate adenocarcinoma specimens (n = 184 TMA elements) obtained from 92 patients. IHC was performed using protein-specific antibodies and staining was evaluated by measuring both staining intensity and percentage of positive cells. Representative IHC images showing high and low/no staining of FOXO1 and ERG and corresponding hematoxylin and eosin (H&E) staining are shown in Figure 5A. In this cohort, 80 of 184 (43.5%) TMA specimens were ERG positive, suggesting that the rate of ERG alterations is similar to that reported previously (2). By excluding the cases that were negative for both ERG and FOXO1 staining (n = 66 TMA elements), our further analysis showed that FOXO1 protein expression inversely correlated with ERG expression in the rest of specimens examined (Spearman's correlation r = −0.3425, P = 0.000147; Fig. 5B and C). In support of this finding, TCGA data analysis showed that approximately 46% (21 of 46) FOXO1-deleted specimens harbor ERG fusions (Supplementary Fig. S1B). Thus, both IHC and genomic data indicate that FOXO1 loss and ERG overexpression co-occurred at least in a subset of human prostate cancers.

Because FOXO1 knockdown increased the transcriptional activity of ERG (Figs. 1 and 3) and FOXO1 loss and ERG protein overexpression correlated in human prostate cancer specimens (Fig. 5 and Supplementary Fig. S1B), we generated Foxo1 deletion and TMPRSS2–ERG transgenic compound mice to determine the role of the combination of Foxo1 deletion and TMPRSS2–ERG overexpression in prostate tumorigenesis. Foxo1 deletion and transgenic expression of ERG were confirmed by IHC in prostate tissues from double mutated mice (Foxo1^pc−/−;Pb-ERG) and their littermate controls: “Wild-type” (Pb-Cre negative), prostate-specific Foxo1 deletion Foxo1^pc−/− (Pb-Cre⁺;Foxo1^p/p); prostate-specific transgenic T1/E4 ERG (Pb-ERG) (Supplementary Fig. S2). IHC for ERG also demonstrated expected nuclear expression of transgenic ERG and loss of Foxo1 in the appropriate mice (Supplementary Fig. S2). Importantly, H&E staining for all four lobes of the mouse prostate revealed the development of HGPIN in the double mutant Foxo1^pc−/−;Pb-ERG mice at 10 months of age, but not in the single mutant or “wild-type” mice (Fig. 6A). It is worth noting that even up to 12 months of age, only double mutant Foxo1^pc−/−;Pb-ERG mice displayed HGPIN (Fig. 6B). IHC for the cell proliferation marker Ki-67 demonstrated that Foxo1 knockout alone moderately (three- or four-fold) increased the number of proliferating cells compared to “wild-type” in mice at 10 months of age (Fig. 6C and D). However, concomitant Foxo1 knockout and T1/E4 ERG overexpression largely (approximately 15-fold) increased the number of proliferating cells compared to “wild-type” (Fig. 6C and D), which is consistent with the observed HGPIN formation and tumorigenesis (Fig. 6A). Additional IHC for cleaved caspase-3, a marker of apoptosis in mice at 10 months showed no overt change in the number of apoptotic cells among all groups of mice examined (Supplementary Fig. S3), further supporting the idea that cell proliferation is the major contributor to prostatic intraepithelial neoplasia (PIN) formation.

Previous studies show that the periacinar fibroblast layers appear to be wider in PIN lesions, implying a local activation of periacinar fibroblasts in response to PIN development (48). IHC for smooth muscle actin (SMA) demonstrated that the prostate acini in “wild-type,” Foxo1 deletion and ERG transgene mice were surrounded by a highly condensed layer of SMA-positive stroma, which represents a natural barrier blocking the invasion of tumor cells into the adjacent stromal tissue (Fig. 6E). In contrast, loosened layers of SMA-positive stroma were detected in the prostates of Foxo1 knockout and ERG transgenic compound mice (Fig. 6E). These data suggest that loss of FOXO1 and gain of ERG overexpression cooperate to drive PIN formation and a potential invasive phenotype in mice as early as 10 months of age.

We next examined the effect of Foxo1 deletion on expression of the target genes of its own and ERG in the prostates of double mutant mice. FOXO proteins are implicated as tumor suppressors by transcriptionally upregulating cyclin-dependent kinase inhibitor Cdkn1b (called p27^KIP1) and pro-apoptotic gene BCL2L11 (also called BIM). As shown in Fig. 7A, homozygous deletion of Foxo1 largely diminished expression of p27^KIP1, which is consistent with increased cell proliferation in Foxo1-deleted prostates (Fig. 6C and D). As expected, Foxo1 deletion also reduced expression of Bim although there was no overt difference in apoptosis between Foxo1 intact and deleted mice (Fig. 7B and Supplementary Fig. S3). One plausible explanation is that the basal level of Bim expression in the prostates of “wild-type” mice is much lower than other genes such as p27^KIP1 (Fig. 7C and D), and therefore may not be insufficient to trigger apoptosis. Intriguingly, it seems that expression of transgenic ERG increased expression of both p27^KIP1 and Bim mRNAs, but the data varied greatly among replicates and they were not statistically significant (Fig. 7A and B). This result was further confirmed by RNA high-throughput sequencing (RNA-seq) data obtained from independent mice (Fig. 7C and D). Moreover, in agreement with the development of loosen stroma layers in the prostates of Foxo1 deletion and ERG overexpression compound mice (Fig. 6E), ERG target genes, known to be involved in cell invasion such as Plau and Mmp3, were upregulated in the prostates of double mutant mice (Fig. 7E and F). This finding supports our human prostate cancer cell line data that FOXO1 loss relieved inhibition of expression of ERG target genes linked with cell invasion (Figs. 3 and 4). Together, these findings provide an explanation for the increased proliferation of prostatic cells in double mutant mice. More interestingly, these findings also suggest that ERG-positive, FOXO1-negative tumors may be more invasive than ERG-negative counterparts.

Large human prostate cancer datasets have revealed that prostate cancer is often characterized by the presence of multiple lesions (2, 3). However, many of these combinations of lesions and the role they may play in prostate cancer initiation and progression have not been functionally tested. This study is the first report to explore a role for concomitant FOXO1 inactivation and ERG overexpression in prostate tumorigenesis and cancer progression. Through a combination of prostate cancer cell culture models and a mouse model with prostate-specific Foxo1 knockout and transgenic TMPRSS2–ERG expression, we have demonstrated that ERG overexpression cooperates with loss of Foxo1 tumor suppressive activity and loss of Foxo1-mediated inhibition of ERG to increase cell proliferation, leading to HGPIN formation in mice (Fig. 7E). In addition, increased expression of ERG target genes important for cell invasion in both the cell systems and double mutant mouse model further suggest that FOXO1 loss and ERG overexpression are also important for tumor progression (Fig. 7E). This study highlights that the FOXO1–ERG axis may be a valuable therapeutic target for prostate cancer, and expands our knowledge of how these two overlapping lesions function to promote prostate tumorigenesis.

In particular, the data presented emphasize that FOXO1 loss and concomitant ERG expression play bifunctional roles in prostate cancer. Although FOXO1 inhibited ERG-mediated cell invasion, likely through decreased ERG chromatin-binding and decreased expression of ERG gene targets, the increased proliferation of prostatic cells in Foxo1^pc-/−;Pb-ERG mice also contributes to HGPIN formation. These two phenomena (cell proliferation and HGPIN formation versus cell invasion) could contribute to both initiation and progression of prostate tumors. One remaining question that warrants further investigation is whether ERG controls a currently uncharacterized transcriptional program important for cell proliferation, which is held in check by FOXO1 in prostatic cells under normal conditions. Although loss of FOXO1 tumor suppressive activity is expected to increase cell proliferation, there may be additional ERG-mediated regulation of the cell cycle. This idea is supported by the finding that ERG overexpression and FOXO1 loss have an additive effect on proliferation in the mouse prostate, greater than the increase in proliferation observed with FOXO1 loss alone.

Our findings also highlight a previously well-characterized functional link between PTEN, downstream AKT signaling, and FOXO1 activity. We demonstrated that both PTEN and FOXO1 have an inhibitory effect on ERG-mediated transcription, which can be rescued by expression of constitutively active AKT. Here, AKT activation through PTEN loss is thought to trigger FOXO1 phosphorylation and exclusion from the nucleus where it may no longer inhibit ERG simply due to spatial separation. This result suggests our findings are not only relevant to ERG-positive, FOXO1 deletion tumors, but that a similar mechanism may occur in tumors with ERG overexpression and PTEN deletion.

The data presented in this study emphasize the importance of understanding how multiple lesions cooperate in prostate cancer initiation and progression. With the advent of “omics” data, researchers within the prostate cancer field have rapidly expanded the list of lesions that may contribute to prostate cancer development. Potential prostate cancer therapeutics must account for the network of lesions and unique molecular interplay that occurs in any one patient. Thus, studies such as ours that reveal additive, synergistic, or even antagonistic relationships between prostate cancer lesions are extremely valuable to the understanding of the origins and drivers of prostate cancer as well as of potential future therapeutic targets.

No potential conflicts of interest were disclosed.

Conception and design: H. Huang

Development of methodology: J. An

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Yang, D. Wang, J. An, Y. Pan, J. Dugdale, J. Zhang, Y.A. Wang, R.A. DePinho

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Yang, Y. Yan, T. Ma, J. Zhang, W.-G. Zhu, W. Xu

Writing, review, and/or revision of the manuscript: A.M. Blee, Y. He, J. Zhang, S.J. Weroha, R.A. DePinho, H. Huang

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Yang, D. Wang, J. An, X. Hou, W.-G. Zhu

Study supervision: W. Xu, H. Huang

This work was supported in part by grants from NIH (CA134514, CA130908, and CA193239 to H. Huang), DOD (W81XWH-14-1-0486 to H. Huang), Natural Science Foundation of China (81270022 and 81611130070 to W. Xu), and Natural Science Foundation of Heilongjiang Province of China (No. QC2014C111 to Y. Yang).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.