Androgen deprivation–mediated activation of AKT is enhanced in prostate cancer with TMPRSS2:ERG fusion

The TMPRSS2:ERG gene fusion (T:E fusion) puts ERG under the androgen receptor–regulated (AR-regulated) expression of TMPRSS2, and it is an early truncal alteration found in about half of prostate adenocarcinoma (PCa) cases in men of European ancestry. It can occur through an interstitial deletion of genes located between TMPRSS2 and ERG on chromosome 21 or through a genomic rearrangement that preserves the intervening genes. Some early studies had indicated that tumors with the T:E fusion (or possibly with T:E fusion associated with an interstitial deletion) were more aggressive, but further studies have not found any clear differences in prognosis or responses to therapy (1, 2). The importance of ERG expression in driving PCa is supported by studies in the T:E fusion–positive VCaP cell line, where RNAi-mediated downregulation of ERG impairs cell growth and invasion (3, 4). Moreover, transgenic overexpression of ERG in mouse prostate causes increased proliferation and, in combination with loss of one PTEN allele, results in prostatic intraepithelial neoplasia or invasive PCa in aged mice (5–8). In varying contexts, ERG in PCa has been found to activate proteins/pathways including EZH2, EMT (through ZEB1, ZEB2, and ILK), Wnt signaling, NF-κB, SOX9, and YAP1 (3, 4, 8–16). More recently, it has become clear that ERG expression has marked global effects on gene expression in PCa cells and, in particular, on the AR cistrome and transcriptome, where it directly interacts with AR and functions to maintain or expand AR signaling and luminal epithelial lineage (16–23).

The requirement for PTEN loss to drive PCa progression in mice overexpressing ERG is consistent with an association between T:E fusion and PTEN loss in human PCa, with T:E fusion presumed to be an initiating event and subsequent PTEN loss driving progression. One suggested basis for this association is that ERG is needed to maintain AR functions related to luminal differentiation in the context of PTEN loss (17–23). Indeed, one study found that ERG overexpression in PCa from Pten-deficient mouse prostate mitigated responses to both AR-targeted therapy and combination therapy with a phosphatidylinositol 3-kinase (PI3K) inhibitor (BEZ3235), and this was associated with maintenance of AR target gene expression (19). Conversely, another study found that ERG could suppress PI3K signaling and subsequent AKT activation and identified ERG repression of insulin receptor substrate 2 (IRS2) expression as a possible mechanism (24). In this case, PTEN loss may be required as a secondary event to increase PI3K signaling and compensate for the repressive effects of ERG.

Inositol polyphosphate-4-phosphatase type II B (INPP4B) is another phosphatase that negatively regulates PI3K signaling and has been identified as a tumor suppressor in triple-negative breast cancer (25, 26). Notably, analysis of The Cancer Genome Atlas (TCGA) data shows that T:E fusion in PCa is also strongly associated with decreased expression of INPP4B (24). In addition to negative regulation of PI3K signaling, INPP4B has further effects on signaling pathways through regulation of endosomal trafficking of proteins, including receptor tyrosine kinases (27–29). However, the role of ERG in its regulation and the functional significance of its decreased expression in T:E fusion tumors remain to be determined.

Previous studies had shown that AR can suppress AKT activity by increasing expression of FKBP5, which is a scaffold for the AKT phosphatase PHLPP1 (30, 31). This provides one mechanism through which AR-targeted therapies may increase PI3K/AKT signaling, although a recent study indicates that PHLPP1 and PHLPP2 do not function as AKT phosphatases (32). AR can also positively regulate expression of INPP4B, so that AR inhibition may further increase PI3K signaling by decreasing INPP4B (33). These stimulatory effects on PI3K may partially mitigate responses to AR inhibition and have provided the rationale for clinical trials combining inhibition of AR and PI3K/AKT signaling (34). Notably, as ERG in T:E fusion PCa is driven by AR, its expression is decreased in response to AR inhibition, which may further enhance PI3K signaling and mitigate therapeutic responses. In this study, we further assessed ERG regulation of IRS2, INPP4B, and PI3K signaling and tested the hypothesis that AR inhibition further enhances PI3K signaling in T:E fusion–positive PCa.

T:E fusion is associated with decreased INPP4B expression independently of PTEN status. While PTEN loss is associated with T:E fusion, PTEN was not altered in approximately half of cases (Supplemental Figure 1A; supplemental material available online with this article; https://doi.org/10.1172/JCI192368DS1), and its expression was not reduced in these cases where the gene is intact (Figure 1A and Supplemental Figure 1B). In contrast, although the INPP4B gene was rarely altered in primary PCa (Supplemental Figure 1A), its mRNA levels were globally reduced in T:E fusion–positive cases (Figure 1B). Consistent with this reduction, reversed phase protein analysis of PCa tumors in TCGA showed that INPP4B is the most reduced protein in T:E fusion–positive versus –negative PCa (Supplemental Figure 1C).

Figure 1

INPP4B mRNA is negatively correlated with T:E fusion independently of PTEN status. (A) Correlation between ERG and PTEN mRNA in TCGA primary PCa. (B) Correlation between ERG and INPP4B mRNA in TCGA primary PCa. (C) Correlation between ERG mRNA and INPP4B protein, and PTEN genomic alterations, in TCGA primary PCa. (D) Correlation between PTEN and INPP4B mRNA, and PTEN genomic alterations, in TCGA primary PCa. (E) Correlation between ERG and IRS2 mRNA in TCGA primary PCa. All analyses were done on cBioPortal. (F) Duplicate cultures of LNCaP and VCaP cells were lysed and immunoblotted as indicated.

INPP4B dephosphorylates phosphatidylinositol 3,4-bisphosphate [PI(3,4)P₂], which similarly to PI(3,4,5)P₃ mediates the binding to and activation of AKT and additional proteins with pleckstrin homology domains. Therefore, similarly to PTEN loss, the downregulation or loss of INPP4B can enhance AKT activation. However, they are not redundant. While PTEN primarily inhibits AKT activation at the plasma membrane, INPP4B targets endosomal AKT activation and also regulates the endosomal trafficking of multiple proteins including receptor tyrosine kinases (27–29). Notably, consistent with nonredundant functions, this reduced INPP4B expression was independent of PTEN status, as it is similarly reduced in the T:E fusion–positive tumors whether PTEN is lost or intact (Figure 1C).

Also consistent with nonredundant functions, there was a positive correlation between PTEN and INPP4B expression that was independent of T:E fusion status (Figure 1D). As shown previously (24), IRS2 expression was lower in T:E fusion–positive tumors (Figure 1E), although this decrease was not as marked as the decrease in INPP4B. We also examined INPP4B in the VCaP PCa cell line, which was T:E fusion–positive and PTEN intact. Compared with LNCaP PCa cells (T:E fusion negative, PTEN deficient), VCaP has markedly lower INPP4B (Figure 1F). As expected, AKT activation, as assessed by phosphorylation at T308 and S473, was greater in the LNCaP cells.

ERG suppresses PI3K signaling in PCa cells in vitro. To directly assess effects of ERG on PI3K signaling we used RNAi to decrease expression of ERG in T:E fusion–positive VCaP PCa cells. Consistent with a previous study (24), shRNA-mediated suppression of ERG increased PI3K signaling based on increased phosphorylation of AKT and S6 (Figure 2A). There was also a small increase in ERK phosphorylation. Acute suppression of ERG with siRNA similarly increased AKT and S6 phosphorylation (Figure 2B).

Figure 2

ERG suppresses AKT activity and IRS2 expression in PCa cells. (A) VCaP cells stably expressing an ERG shRNA versus a nontargeting control shRNA were assessed by immunoblotting as indicated. All proteins were assessed from 1 shCtrl or shERG lysate, and the vinculin loading control is just shown for 1 of the gels. (B) VCaP cells were treated for 2 days with siRNA targeting ERG or a nontargeted control and assessed by immunoblotting as indicated. (C) LNCaP cells stably expressing a DOX-inducible HA-tagged ERG were treated with DOX over a time course and assessed by immunoblotting as indicated. (D) LNCaP cells stably expressing a DOX-inducible HA-tagged ERG were treated with DOX for 48 hours and assessed by immunoblotting. (E) VCaP cells were treated for 2 days with siRNA targeting ERG or nontargeted control and assessed by immunoblotting as indicated. All data shown are representative of at least 3 experiments.

Conversely, we assessed the effects of inducing ERG expression in T:E fusion–negative LNCaP PCa cells. For this we generated a tet-operator-regulated (tetO-regulated) vector (pTET-Splice) expressing N-terminal HA-tagged ERG, using human ERG with deletion of amino acids 1–39 to mimic the product generated from the most common T:E fusion (pTET-ERG) (Supplemental Figure 2, A and B). This was stably transfected into LNCaP cells expressing pcDNA6/TR for doxycycline-regulated (DOX-regulated) expression. The LNCaP cells are PTEN deficient and therefore have high basal PI3K/AKT signaling (Figure 2C). DOX-mediated induction of ERG for up to 29 hours caused a small decrease in AKT and S6 phosphorylation and a clear decrease in ERK phosphorylation (Figure 2C). The modest decrease in phosphorylation of AKT was more apparent after 48 hours of DOX induction and was observed in both the AKT1 and AKT2 isoforms (Figure 2D).

Notably, ERG induction in LNCaP cells did not decrease expression of IRS2 (Figure 2D and Supplemental Figure 2C), indicating that the modest acute effects of ERG on AKT and ERK were not mediated via IRS2. We similarly examined effects of acutely inducing ERG in the 22RV1 PCa cell line, which is T:E fusion–negative and PTEN intact. In two independent lines, we also observed only small and inconsistent effects of ERG induction on AKT activation and IRS2 expression (Supplemental Figure 3).

ERG suppresses IRS2 directly and INPP4B indirectly in T:E fusion–positive cells. We next examined effects of depleting ERG on IRS2 and INPP4B in T:E fusion–positive VCaP cells. Consistent with findings from a previous report (24), ERG knockdown increased the expression of IRS2 protein (Figure 2E). In contrast, there was not a clear increase in INPP4B protein. RNA-seq analysis confirmed a approximately 2.5-fold rise in IRS2 mRNA upon ERG siRNA treatment, whereas INPP4B was not changed (Supplemental Table 1). Analysis of available ERG ChIP-seq data in VCaP cells showed an ERG binding site at the promoter of IRS2, but not INPP4B, further indicating that ERG directly regulates IRS2 and not INPP4B (Supplemental Figure 4, A and B). AR ChIP-seq further showed that this ERG binding site overlapped a broad AR binding site present in VCaP cells but not in T:E fusion–negative LNCaP cells (Supplemental Figure 4A). This is consistent with previous results showing that ERG opens cryptic AR binding sites in many genes, so these come under joint ERG and AR regulation (16).

As noted above, the acute induction of ERG in LNCaP and 22Rv1 cells did not cause a clear decrease in IRS2, indicating that the ERG suppression of IRS2 in T:E fusion–positive tumors requires further adaptations. Interestingly, although ERG knockdown increases IRS2 mRNA in VCaP cells, it did not increase H3K27Ac at the IRS2 promoter (Supplemental Figure 4A), suggesting that ERG repressed IRS2 through a mechanism distinct from modulation of histone acetylation.

To assess more broadly the effects of ERG knockdown we used gene set enrichment analysis (GSEA). PI3K/AKT/MTOR signaling was only modestly increased in ERG knockdown cells (Supplemental Figure 5, A and B), while the most enriched gene set after ERG knockdown was androgen response followed by fatty acid metabolism (which may be androgen regulated) (Supplemental Figure 5, A, C, and D). This result is somewhat surprising as ERG can maintain or expand AR signaling in the setting of PTEN loss, and it may reflect that VCaP is PTEN intact. Conversely, Hallmark gene sets, including MYC targets and E2F targets, were markedly decreased in response to ERG knockdown (Supplemental Figure 5, A and D), consistent with an oncogenic function of ERG in these cells. Together, these results support the conclusion that ERG directly suppresses IRS2 expression in T:E fusion tumors. In contrast, these results also indicate that the decreased expression of INPP4B in T:E fusion tumors is not direct and that it is an adaptation that is selected, at least in part, to compensate for suppressive effects of ERG on PI3K signaling and potentially ERK signaling.

AKT activation is inversely associated with ERG expression in a mouse model. We next used a mouse model to assess ERG function in an in vivo context. We used the insert from the pTET-ERG vector to generate transgenic mice expressing HA-tagged human ERG (amino acids 1–39 deleted) under the control of the tetracycline operator. We had previously developed transgenic mice with a probasin-driven reverse tetracycline transactivator (rtTA, tet-on) that can drive prostate specific expression of tetracycline operator-driven transgenes (35). These mice were crossed with the pTET-ERG transgenic mice to obtain DOX-stimulated expression of ERG in the prostate. We initially sacrificed a small cohort of mice at approximately 4–5 months after DOX induction and isolated protein from each prostate lobe. By immunoblotting we found human ERG expression in the ventral lobe, which we confirmed by qRT-PCR (Supplemental Figure 6A). By IHC, we further confirmed DOX-driven human ERG expression in ventral prostate (Supplemental Figure 6B) and in two pTET-ERG lines confirmed that expression was dependent on the rtTA (Supplemental Figure 6C). Consistent with previous studies showing a modest effect of ERG overexpression in mouse prostate (5, 6), induction of ERG for as long as 16 months caused only mild hyperplasia (Supplemental Figure 6D).

Next, we crossed these mice onto a Pten-haploinsufficient background (Pten^–/+) (5). In these mice, we then induced ERG expression by feeding mice with both DOX-containing food and water to further increase ERG expression (Supplemental Figure 7, A and B). We identified areas of PCa after approximately 11 months of DOX induction (Figure 3A). Notably, ERG expression was heterogeneous, with much lower ERG expression in the areas with invasive tumor (Figure 3B). Moreover, in those tumor areas with low ERG, we found elevated expression of pAKT(S473), pS6(S235,236), and to a lesser extent pERK, reflecting activation of the PI3K/AKT pathway (Figure 3C). These observations were consistent with ERG suppression of PI3K/AKT signaling. The decrease in ERG in these areas may be due in part to a decrease in AR, which drove the rtTA in these mice (Supplemental Figure 8A). Expression of another AR-driven gene, Tmprss2, was also decreased in these tumor areas with low ERG (Supplemental Figure 8B). These findings indicate that, while ERG is oncogenic, higher levels may be tumor suppressive, possibly due to suppression of PI3K/AKT signaling.

Figure 3

PCa development in probasin-rtTA; tetO-ERG; Pten^-/+ mice. (A) Histology in mice treated for 1–11 months with DOX, showing foci of PCa in mouse treated with DOX for 11 months. (B) FFPE sections from representative areas with or without tumor were stained for ERG or the HA tag on ERG. Lower ERG expression is seen in areas with invasive tumor versus areas showing only dysplasia. (C) Higher pAKT and pS6 in areas with invasive tumor versus areas showing only dysplasia. Original magnification, ×20.

ERG-regulated genes in probasin-rtTA; tetO-ERG; Pten^-/+ mice. To identify ERG-regulated genes that may be contributing to PCa development in this model, we carried out time course experiments with both short-term ERG induction and long-term induction plus short-term DOX withdrawal in the probasin-rtTA; tetO-ERG; Pten^-/+ (ERG;Pten^+/–) mice. For the short-term induction, mice were treated with DOX or vehicle for 3 or 6 days. Ventral prostate epithelium was then laser-capture microdissected, and transcriptome profiles were obtained using Affymetrix microarrays. The level of ERG induction based on the microarrays was similar at 3 and 6 days (~25% increase), so expression data from both days were pooled. There were 5 Hallmark gene sets that were significantly (FDR < 0.25) increased and 16 that were decreased (Supplemental Figure 9A). MTORC1 signaling and PI3K AKT MTORC signaling were among the highly suppressed gene sets (NES –3.34 and –2.64, respectively) (Supplemental Figure 9, B and C). Notably, this suppression could not be attributed to IRS2, as it was not decreased by ERG induction (ratio induced/uninduced ~1.18). INPP4B was increased (ratio induced/uninduced ~1.16), indicating it may acutely contribute to decreased PI3K signaling in these mice.

Interestingly, the most altered gene involved in phosphatidylinositol metabolism was PIKFYVE (phosphoinositide kinase, FYVE-type zinc finger containing) (ratio induced/uninduced 0.71). This kinase phosphorylates the D-5 position in phosphatidylinositol and phosphatidylinositol-3-phosphate on endosomes, the latter generated by INPP4B, to promote lysosomal targeting. The second most altered gene in phosphatidylinositol metabolism was phosphoinositide-3-kinase regulatory subunit 3 (PIK3R3, ratio induced/uninduced 0.74), which may contribute directly to a decrease in PI3K activity in these mice. Analysis of TCGA PCa data showed that PIK3R3 was significantly decreased in T:E fusion tumors (Supplemental Figure 9D), indicating this may also contribute to ERG-mediated suppression of PI3K signaling in patients.

We next used RNA-seq to assess the effects of discontinuing DOX in mice with established tumors. There were no Hallmark gene sets altered at FDR < 0.25. Nonetheless, IRS2 was approximately 4-fold higher in tumors after DOX discontinuation, consistent with it being suppressed by ERG in the established tumors. Levels of INPP4B prior to and after DOX discontinuation were too low for reliable assessment. Interestingly, the long noncoding RNA (lncRNA) H19 was the most increased transcript in the DOX-induced tumors compared with transcripts after DOX withdrawal (Figure 4A). H19 was also increased by short-term DOX induction (although only about 2-fold versus the >100-fold increase in the long-term DOX induced tumor samples) (data not shown), indicating that ERG is inducing H19. Notably, H19 and IGF2 are coregulated in human and mouse by genomic imprinting through methylation at the locus between H19 and IGF2, and we also observed increased IGF2 in the DOX-induced tumors versus the samples after DOX withdrawal (Figure 4A). Marked ERG-mediated increases in IGF2 and H19 mRNA were also observed in a previous independent study using stable probasin-ERG transgenic mice, but the significance of this increase with respect to decreased PI3K/AKT signaling is not clear (13) (Figure 4B).

Figure 4

H19 lncRNA regulation by ERG in mouse model and human PCa. (A) H19 and IGF2 expression in mice with long-term DOX induction versus after DOX withdrawal. (B) H19, IGF2, and ERG expression in mouse prostate expressing probasin-ERG transgenes versus WT (from ref. 13). (C) Correlation between ERG and H19 mRNA in TCGA primary PCa. (D) Correlation between ERG and H19 mRNA in T:E fusion–positive tumors in TCGA primary PCa. (E) Correlation between ERG and H19 mRNA in T:E fusion–negative tumors in TCGA primary PCa. Correlations were assessed in cBioPortal.

While H19 appears to be a major target of ERG in mice, analysis of gene expression data in human primary PCa showed that H19 was instead decreased in T:E fusion–positive tumors (Figure 4C). Interestingly, analyzing T:E fusion–positive and T:E fusion–negative tumors separately, we found that H19 expression was positively correlated with ERG in T:E fusion–negative (but not –positive) tumors, but the significance of this association is not clear (Figure 4, D and E). Together these findings indicate that H19 is a major transcriptional target of ERG in this mouse PCa model, but not in human PCa. Overall these observations support the conclusion that ERG is repressing PI3K/AKT signaling in this mouse PCa model and in human PCa, although some distinct mechanisms may be involved.

Castration enhances AKT activation in a T:E fusion–positive PDX model. As expression of ERG in T:E fusion–positive tumors is driven by AR, it is markedly decreased in response to AR-targeted therapies. Therefore, we next assessed the extent to which AR inhibition increases PI3K/AKT signaling in VCaP cells. As expected, treatment with increasing concentrations of an AR antagonist (enzalutamide) caused a progressive decrease in ERG and PSA and further decreased INPP4B, which is AR regulated (Supplemental Figure 10). This was associated with an increase in phosphorylation of AKT2 but no clear change in AKT1. Interestingly, there was also a decrease in PHLPP1 and PHLPP2 that may dephosphorylate AKT (30, 31), although a recent study indicates these do not function as AKT phosphatases (32). Finally, PTEN expression was unchanged.

To assess effects of AR inhibition in vivo, we examined a T:E fusion–positive/PTEN negative patient-derived xenograft (PDX) (BIDPC4) we derived from an omental metastasis in a patient with castration-resistant prostate cancer (CRPC). PDXs were established subcutaneously in male scid mice and analyzed prior to castration and at approximately 2 weeks after castration. Castration decreased ERG (as expected) and caused a marked increase in phosphorylation of AKT and S6 (Figure 5A). To explore the role of AKT activity in this model, we generated short-term ex vivo cultures from BIDPC4 PDX cells. These cultures were treated with an AKT inhibitor (ipatasertib) in either medium containing FBS or androgen-depleted charcoal-stripped serum (CSS). Notably, cells in the CSS medium showed heightened sensitivity to AKT inhibition (Figure 5B). A similar effect was observed with a second AKT inhibitor, MK2206 (Figure 5C).

Figure 5

Androgen deprivation increases AKT activity and dependence in T:E fusion–positive PDX. (A) T:E fusion–positive BIDPC4 PDXs were established in male scid mice as subcutaneous tumors. Mice were sacrificed prior to or at approximately 2 weeks after castration. FFPE sections were stained for ERG, pAKT(473), and pS6 as indicated, and representative slides are shown. Original magnification, ×10. (B) Cells from BIDPC4 PDX were cultured in medium with 10% FBS or in medium with 10% charcoal stripped FBS (CSS). Ipatasertib was then added at the indicated concentrations, and cell recovery was assessed by CellTiter-Glo luminescent cell viability assay after 7 days. Luminescent readouts were normalized to the average luminescent signal of DMSO under the respective conditions, with 6 replicates per dose of drug. Data are shown as mean ± SD. The graph illustrates the half-maximal inhibitory concentration (IC₅₀) in the presence of FBS compared with CCS. Nonlinear regression curves were generated by a variable slope model. (C) Responses to MK2206 were assayed as in B.

AR signaling inhibition enhances AKT activation in T:E fusion–positive clinical samples. We next examined available clinical data to assess the extent to which AR-targeted therapies increase AKT activation in T:E fusion–positive versus –negative PCa. For this analysis we used genomic data and ERG mRNA levels to identify T:E fusion–positive tumors in the TCGA data set of primary PCa (36) and in two CRPC data sets (37, 38) (Supplemental Figure 11A). We then carried out single-sample GSEA (ssGSEA) for each tumor using the Hallmark PI3K_AKT_MTOR gene set to determine whether AKT signaling was increased in the T:E fusion–positive CRPC cases relative to the TCGA primary tumors, but we did not find a significant difference in the T:E fusion–positive or –negative tumors (Supplemental Figure 11B). We similarly examined a small series of cases that had matched RNA-seq data from CRPC tumors before and after treatment with AR inhibitors (39). While there was enrichment for the PI3K_AKT_MTOR gene set in some T:E fusion–positive cases, the increase was not significant and was similar to that in the T:E fusion–negative cases (Supplemental Figure 11C).

Notably, there is substantial restoration of AR activity in most CRPC cases, and we reported previously that this includes increased ERG expression T:E fusion–positive CRPC (40). Indeed, in the matched untreated versus treated CRPC samples, ERG remained markedly elevated in the T:E fusion–positive versus –negative tumors (Supplemental Figure 11D). Therefore, to further assess effects of AR-targeted therapy on tumors prior to emergence of CRPC, we examined samples from a clinical trial of neoadjuvant intensive AR-targeted therapy (NCT02430480) (41).

In this trial, patients with high-risk primary PCa were treated for 24 weeks with leuprolide in combination with enzalutamide, followed by radical prostatectomy (RP). Notably, IHC for ERG showed that ERG was markedly decreased in the residual tumor in the RP specimens versus the pretreatment biopsies (Figure 6, A and B). ERG mRNA was similarly greatly decreased in the posttreatment T:E fusion–positive tumors (Figure 6B). Analysis of the RNA-seq data by ssGSEA then showed that the PI3K_AKT_MTOR gene set was significantly increased (P = 0.020) in the posttreatment versus matched pretreatment tumors (Figure 6C). This gene set was also modestly increased in some T:E fusion–negative cases, but this was not significant (Figure 6C).

Figure 6

PI3K_AKT_MTOR activity in primary PCa before and after neoadjuvant AR-targeted therapy. (A) FFPE sections of tumor from matched pairs of baseline biopsy and posttreatment radical prostatectomy specimens stained with anti-ERG antibodies, from 3 representative T:E fusion–positive cases. Scale bar: 500 μm; 100 μm (inset). (B) ERG levels in matched pairs as determined by (left) anti-ERG IHC quantified by machine-guided image analysis or (right) RNA-seq. (C) Single-sample GSEA for the Hallmark PI3K_AKT_MTOR gene set in T:E fusion–positive (left, n = 14) and T:E fusion–negative (right, n = 8) tumors. Statistical significance was determined using Wilcoxon’s matched-pairs signed-rank test.

As expected, PTEN loss was greater in the T:E fusion–positive (13/16) versus –negative (9/21) tumors (Supplemental Figure 11E). This high frequency of PTEN loss in both groups may reflect selection for patients with high-risk tumors. ERG expression by IHC was markedly decreased by the neoadjuvant therapy in both the PTEN-intact and -deficient T:E fusion tumors (Supplemental Figure 11E). Due to limited residual tumor in the PTEN-intact T:E tumors we were not able to carry out RNA-seq to assess effects of the treatment on PI3K/AKT signaling. However, effects in the T:E fusion–negative tumors did not appear related to PTEN status (Supplemental Figure 11F).

We also examined samples from a clinical trial of neoadjuvant leuprolide, abiraterone, and apalutamide administered for 24 weeks prior to RP in men with high-risk primary PCa (NCT02903368) (42). ERG IHC on diagnostic biopsies was used to identify cases that were T:E fusion–positive versus –negative. Diagnostic core biopsies and corresponding RP specimens containing residual tumor in the T:E fusion–positive and –negative cases were then stained for phospho-AKT (Figure 7A). Notably, phospho-AKT staining was variable in the biopsies. Therefore, to take into consideration factors other than ERG that may influence PI3K signaling, we quantified the difference in phospho-AKT staining between the biopsies and RP specimens for each case. Staining in the RP specimens was greater than in the matched biopsies in over half of the T:E fusion–positive cases versus only 2 of the T:E fusion–negative cases (Figure 7B), and this increase in the T:E fusion–positive versus –negative residual tumors was significant at P < 0.01 (Figure 7C). Taken together these findings indicate that AR-targeted therapies may increase PI3K/AKT pathway activity by several mechanisms, with decreased ERG expression in T:E fusion tumors being a major mechanism in this tumor subset (Supplemental Figure 12).

Figure 7

AKT phosphorylation prior to and after neoadjuvant AR-targeted therapy. (A) FFPE sections of tumor in pretreatment biopsies and in radical prostatectomy specimens from a T:E fusion–positive and –negative case were stained for pAKT(473). Scale bars: 100 μm. Samples were scored based on fraction of cells staining. Pretreatment biopsy scores for the T:E fusion–positive and –negative tumors shown here were 0 and 3, respectively. Scores for the corresponding posttreatment tumors were 3 (T:E fusion–positive) and 2 (T:E fusion negative). (B) pAKT scores for each pretreatment and corresponding posttreatment tumor in T:E fusion–positive (left) and negative (right) cases. Note that darker lines reflect multiple cases with the same pre- and posttreatment scores. (C) For each patient, the pAKT score in the pretreatment biopsy was subtracted from the score in the corresponding posttreatment tumor to give a differential pAKT score. Statistical significance was assessed by unpaired 2-tailed t test, P < 0.01.

T:E fusion is frequently associated with PTEN loss and is strongly associated with decreased INPP4B expression. The basis for these associations remains to be established, but one possibility is that ERG may suppress PI3K/AKT signaling and that decreased expression of PTEN and INPP4B may compensate for this suppression. Our finding that ERG knockdown enhances PI3K signaling in T:E fusion–positive VCaP cells supports this conclusion and is consistent with results in a previous study (24). Also consistent with this previous study, we found that ERG knockdown in VCaP cells increased IRS2, supporting ERG repression of IRS2 as a mechanism contributing to ERG suppression of PI3K signaling. In contrast, despite the strong negative correlation between ERG and INPP4B levels in clinical samples, ERG knockdown did not increase INPP4B mRNA or protein, suggesting the decreased expression of INPP4B is indirect and reflects potent selective pressure to decrease INPP4B function. Results in mice with DOX-regulated induction of ERG further support the conclusion that ERG suppresses PI3K signaling, although mechanisms driving this decrease and subsequent adaptations in mouse may in part be distinct. Interestingly, while ERG appears to be directly suppressing IRS2 in T:E fusion–positive tumors, IRS2 is not suppressed by the acute induction of ERG, indicating that the ERG suppression of IRS2 is dependent on further adaptations.

Notably, INPP4B expression is strongly and broadly decreased in T:E fusion–positive PCa, including in cases with genomic PTEN loss. This indicates that PTEN loss alone does not fully compensate for effects of ERG induction and that decreasing INPP4B may have distinct effects that are required to support the growth of T:E fusion tumors. Indeed, PTEN suppresses AKT activation primarily by decreasing P(3,4,5)P₃ [converting it to (P4,5)P₂] at the plasma membrane, although it may in some settings also target P(3,4)P₂ (43). Its loss results in increased AKT recruitment to the plasma membrane and subsequent AKT phosphorylation and activation. PTEN loss is common in primary PCa, and its frequency is increased in metastatic CRPC, which is presumed to reflect increased AKT activation and subsequent increases in tumor cell survival and invasion.

In contrast, INPP4B-mediated dephosphorylation of PI(3,4)P2 primarily regulates AKT activation intracellularly, with one study indicating that INPP4B loss may selectively increase activation of the AKT2 isoform (28). Moreover, the PI(3)P generated by INPP4B on endosomal membranes can enhance lysosomal targeting of signaling proteins, including EGFR and GSK3β, to downregulate receptor tyrosine kinase signaling and to enhance Wnt signaling, respectively (27, 44, 45). To the extent that T:E fusion tumors need to compensate for decreased IRS2, this former increase in receptor tyrosine kinase signaling may be an important factor driving selection for decreased INPP4B. Further studies are needed to address the extent to which loss of these or other INPP4B functions drive selection for its downregulation in T:E fusion–positive PCa and whether this results in novel therapeutic vulnerabilities.

Previous studies have shown that AR-targeted therapies can increase PI3K signaling through mechanisms including downregulation of FKBP5 and of INPP4B (30, 31, 33). As ERG is AR regulated in T:E fusion–positive tumors, we hypothesized that AR inhibition, and subsequent decreased expression of ERG, would in particular enhance AKT activation in these tumors. Indeed, we found that AKT activity and dependence was increased after castration in a T:E fusion–positive PDX model. We then used ssGSEA to assess for increases in PI3K/AKT signaling in T:E fusion–positive versus –negative CRPC but did not find a significant increase. However, while ERG expression is decreased acutely after AR-targeted therapies, it is substantially restored in CRPC. Therefore, we next examined matched samples from two neoadjuvant trials of intensive AR-targeted therapy. We confirmed that ERG was markedly decreased in the posttreatment residual tumor and found that PI3K/AKT activity was significantly increased in the T:E fusion–positive versus –negative posttreatment cases. Based on these findings, we suggest that increased AKT activity due to decreased ERG mitigates the efficacy of AR-targeted therapies to a greater extent in T:E fusion PCa and that men with T:E fusion–positive tumors may most benefit from the addition of an AKT inhibitor to AR-targeted therapy.

Results of a phase III trial of abiraterone combined with an AKT inhibitor (ipatasertib) in CRPC (IPATential150 trial) indicated that there was radiographic progression-free survival benefit in patients with PTEN-deficient tumors (34). As PTEN loss is associated with T:E fusion, it is possible that some of this benefit may reflect T:E fusion. A subsequent analysis of the IPATential150 trial (published while this manuscript was under revision) found that while there was an increase in median overall survival in men with PTEN-deficient tumors (from 29.8 months in the placebo control group to 36.8 months in the ipatasertib arm), this did not reach statistical significance (46). Notably, in an exploratory analysis, this study found that median overall survival in T:E fusion tumors was improved from 30.8 months to 42.9 months by addition of ipatasertib, with no clear benefit in the T:E fusion–negative tumors.

These clinical trial findings further support the hypothesis that T:E fusion tumors may have increased dependence on AKT activation after AR-targeted therapy. Moreover, we suggest that this benefit of adding an AKT inhibitor may be greater in patients with castration-sensitive PCa who are initiating AR-targeted therapy. Notably, the phase III CAPItello-281 trial (NCT04493853) is assessing another AKT inhibitor (capivasertib) in combination with AR inhibition (GnRH agonist plus abiraterone and prednisone) in men with PTEN-deficient metastatic castration-sensitive PCa. It will be of interest to determine whether the PTEN-deficient tumors with T:E fusion have greater benefit in this trial. Finally, we suggest that T:E fusion tumors may have additional vulnerabilities, including those related to INPP4B downregulation, that may be exploited therapeutically.

Sex as a biological variable. Prostate cancer only occurs in males and is dependent on androgens. Therefore, the in vivo studies only used male mice.

Cell lines and reagents. LNCaP and VCaP cells were obtained from ATCC and cultured in RPMI-1640/10% FBS or DMEM/10% FBS, respectively. Cells were used within 6 months of thawing and tested negative for mycoplasma. siRNA targeting ERG and control SiCtrl were from Dharmacon (Thermo Fisher). shRNA plasmids targeting ERG and controls were from Santa Cruz. Transfections were performed using Lipofectamine 2000 (Invitrogen) according to the manufacturer instructions. The pTET-ERG plasmid was made by inserting N-terminal HA-tagged ERG (with deletion of amino acids 1–39) into the tetO-regulated vector pTET-Splice (Invitrogen). LNCaP cells with DOX-inducible ERG were then generated by transfecting the pTET-ERG vector into a LNCaP clone carrying pcDNA6/TR, followed by selection.

RNA interference. ERG knockdown was performed in VCaP cells. Briefly, VCaP cells were transfected with 20, 40 and, 60 nmol/L ON-TARGETplus siRNAs (Dharmacon) with Lipofectamine RNAiMAX (Thermo Fisher Scientific, 13778–075). At 48 hours after transfection, cells were harvested for protein purification and RNA purification.

RNA isolation and qRT-PCR. Total RNA was isolated with the RNeasy Mini Plus Kit (Qiagen, 74134). RNA purification was directly performed from 2D cultured cells. For RNA isolation from tissues, homogenization were performed using a tissue ruptor (Qiagen, 9002755). qRT-PCR was performed using standard SYBR Green reagents from the StepOnePlus Real-Time PCR system (Thermo Fisher Scientific, 11780200) or TaqMan One-Step RT-PCR reagents (Thermo Fisher Scientific, 4444434). Target mRNA expression was quantified using the ΔΔCt method and normalized to GAPDH expression.

Transgenic mice. The tetO-regulated HA-ERG was excised from the pTET-ERG vector and used for pronuclear microinjection in FVB mice (Beth Israel Deaconess Medical Center [BIDMC] Transgenic Facility). Founder lines determined by genotyping were then bred with transgenic mice, which we previously generated and expressed a probasin-driven rtTA, tet-on, which we showed could drive prostate-specific expression of tetracycline operator-driven transgenes (35). Transgene expression was induced by feeding a rodent diet containing DOX chow (0.625g/kg, Harlan Tekland) alone or with drinking water containing 1 g/L DOX in 1% dextrose. Pten-deficient mice on a C57BL/6J background were provided by Pier Paolo Pandolfi (Beth Israel Deaconess Medical Center). Prostates were dissected and snap-frozen for RNA/protein extraction or formalin-fixed for paraffin embedding.

PDX tumors. The BIDPC4 PDX was generated from an omental metastasis in a patient with T:E fusion CRPC. Exome sequencing showed in addition the following: AR amplification, PIK3CA H419_L422del, PTEN loss, MYC amplification, FAS loss, TP53 G245S. Male 5- to 6-week-old ICR SCID mice were purchased from Taconic and housed in the Animal Research Facility at BIDMC. The PDX was initially established by inoculating human biopsies both subcutaneously and by renal graft with 50% Matrigel. Upon successful establishment, tumors were harvested and minced into 1–2 mm³ tumor bits and cryopreserved. Portions of the tumors were then further expanded subcutaneously in mice. All materials in this study are early passages.

Ex vivo culture establishment. The ex vivo cultures were generated from PDXs by culturing in standard DMEM medium with 10% FBS or 10% CCS and 5% Matrigel, without further additives. Briefly, the tumor was excised from subcutaneous xenografts around 500 mm³, rinsed with DMEM (gentamicin 50 mg/mL), washed with DMEM, and minced into 1–2 mm³ tumor bits. The tumor bits were digested with collagenase/dispase (Millipore Sigma, 10269638001 and 11097113001, respectively) at 37°C for 30 minutes. The digestion was then stopped by adding medium with 10% FBS, and the slurry went through 100 μM filter to remove big chunks. The tumor cells were collected by centrifugation. Resuspended cell pellets were then seeded in DMEM either with 10% FBS or 10% CCS and containing 5% (v/v) Matrigel (CORNING, 356230), and plated at 0.8~1 × 10⁴ cells per well in 96-well plate.

CellTiter-Glo luminescent cell viability assay. Ex vivo cultures generated from BIDPC4 PDX were cultured with DMEM plus either 10% FBS or 10% CCS and 5% Matrigel in 96-well plates. Cells were treated with 9 doses of pan AKT inhibitors, serial diluting from 20 μM with a dilution factor of 2. Six replicates were set up for each dose. The CellTiter-Glo luminescent cell viability assay (Promega, G7571) was performed after 7 days of treatment according to the manufacturer’s protocol. The killing curve was plotted based on the readout, which is normalized to the average luminescent signal of DMSO under the respective conditions. Two independent AKT inhibitors were used, MK2206 (MedChemExpress, HY-10358) and ipatasertib (MedChemExpress, HY-15186). Experiments were repeated at least 3 times, and representative results were shown.

Immunoblotting. Protein extracts were immunoblotted with antibodies against AKT pS473 (Cell Signaling, rabbit mAb 4058, 1:1,000) or pT308 (Cell Signaling, rabbit mAb 2965, 1:1000), S6 pS235,236 (Cell Signaling, 2211, 1:1,000), ERK pT202,Y204 (Cell Signaling, rabbit mAb 4858, 1:1,000), AR (mAb 441, Abcam, ab9474, 1:1,000), ERG (Cell Signaling, rabbit mAb 97249, 1:1,000), Vinculin (Cell Signaling, rabbit mAb 13901, 1:1,000), PRAS40 (Cell Signaling, rabbit mAb 13175, 1:1,000), INPP4B (Cell Signaling, rabbit mAb 14543, 1:1,000), IRS2 (Cell Signaling, rabbit mAb 4502, 1:1,000), HA (Covance, rabbit polyclonal PRB-101C, 1:1,000), AKT (Cell Signaling Technology, 9272, 1:1,000), AKT1 (Cell Signaling Technology, 2938, 1:1,000), AKT1 pS473 (Cell Signaling Technology, 9018, 1:1,000), AKT2 (Cell Signaling Technology, 3063, 1:1,000), AKT2 pS474 (Cell Signaling Technology, 8599, 1:1,000), PTEN (Cell Signaling Technology, 9559, 1:1,000), PHLPP1 (Abcam, ab305295, 1:1,000), and PHLPP2 (Bethyl Laboratories, A300-661A, 1:2,000).

Immunohistochemistry. Primary antibodies were anti-pAKT (pS473, Cell Signaling, rabbit mB 4058, 1:200), anti-HA (Cell Signaling, rabbit mAb 3724, 1:200), anti-pS6 (S240/244, Cell Signaling, 2211, 1:200), anti-AR (Upstate, rabbit polyclonal 06-680, 1:100), anti-TMPRSS2 (Cell Signaling, rabbit mAb 84382), and anti-ERG (Abcam, rabbit mAb EPR3864, ab92513, 1:100) in 1% BSA. These were incubated overnight at 4°C, followed by biotinylated secondary antibody and streptavidin-HRP (1:400, Vector).

For NCT02903368 clinical trial specimens, we analyzed individual pretreatment core biopsies and a series of tissue microarrays (TMAs) containing residual tumor from the posttherapy radical prostatectomies. The TMAs contained 3 mm punches taken from 2–4 representative tumor foci based on histology. The T:E fusion status of the cases was determined by immunohistochemistry using an ERG antibody (2). Staining for pAKT was scored based on the fraction of positive cells, with a score of 0 for none, 1 for 1%–5%, 2 for 5%–25%, 3 for 25%–50%, and 4 for >50%. The focus on the TMA from each case with the highest score was used for comparison with the pretreatment core biopsy. The scoring was carried out independently by two investigators who were blinded to ERG status.

For NCT02430480 clinical trial specimens, whole sections of biopsy and prostatectomy tissue stained with anti-ERG antibodies were scanned on a Carl Zeiss AxioScan.Z1 microscope slide scanner equipped with a Plan-Apochromat 20× NA 0.8 objective, 266% LED intensity, 200 μs exposure time. Tissue images were acquired using ZEN Blue 2012 (Zeiss) with objective/magnification and pixel:distance calibrations recorded within the scanned CZI file. Whole slide images were processed using HALO 3.4 (Indica Labs) with a random forest classifier for tumor detection and nuclear ERG quantification using the CytoNuclear detection module. All classified areas were reviewed by a pathologist, and raw nuclear ERG intensity values (averaged per case) were report as optical densities.

Transcriptome analyses. RNA from mouse prostate prior to and after DOX induction was analyzed by hybridization on Affymetrix Mouse Gene 1.0 ST arrays. For RNA-seq, mRNA libraries were generated using the Illumina TruSeq stranded mRNA sample kit. Raw reads were analyzed using a pipeline for RNA-seq analysis — Visualization Pipeline for RNA-seq analysis (VIPER) — based on workflow management system Snakemake (https://github.com/hanfeisun/viper-rnaseq; commit ID ffc26be). The read alignment to the hg19 reference genome was performed using STAR aligner (2.7.0f) with default parameters. Gene expression (FPKM values) was quantitated with Cufflinks (v2.2.1). After ranking according to differential expression, GSEA was performed to search for enrichment across the Molecular Signatures Database (https://www.gsea-msigdb.org/gsea/index.jsp).

ERG and gene set enrichment analyses of clinical samples. Whole-transcriptome sequencing data from matched pairs of tumors before and after neoadjuvant intense androgen deprivation therapy (GSE183100) were stratified by baseline anti-ERG IHC status (47). FASTQ pairs for GTEX prostate tissue data (phs000424.v10) were downloaded from Gen3. FASTQs for TCGA prostate cancer data (phs000178.v10) and the West Coast SU2C cohort (phs001648.v1) were downloaded from the NCI Genomic Data Commons. FASTQs for the East Coast SU2C cohort (phs000915.v2) were downloaded from dbGaP. TPM values were estimated using RSEM version 1.3.2 as a wrapper around STAR version 2.7.0f. Scaled estimates of gene expression at the TPM level in matched pairs of CRPC tumors from the West Coast SU2C cohort and Hartwig Medical Foundation were previously published and provided by Xiaolin Zhu (39). Single-sample gene set enrichment scores were computed using raw or scaled TPM values using the GSVA package for R in ssGSEA mode with tau = 0.75.

T:E fusion status from published RNA-seq cohorts lacking immunohistochemical annotation was determined using four complementary approaches. (a) If companion whole-genome sequencing data were available, structural variations involving TMPRSS2 and ERG translocations or interstitial deletions were used to call a case as fusion positive. (b) If companion whole-exome sequencing data were available, allele-specific imbalances between TMPRSS2 and ERG (including inferred copy number losses) were used to call a case as fusion positive. (c) In all cases, de novo assembly using defuse version 0.8.1 was employed to identify chimeric read-pairs mapping to the T:E mRNA. (d) Samples with outlier high levels of ERG expression lacking any other evidence of a fusion were manually inspected using the Integrative Genome Viewer to confirm it was truly T:E fusion–negative.

Gene expression correlations of prostate cancer TCGA data were performed directly in cBioPortal using the Prostate Adenocarcinoma (TCGA Firehose Legacy) dataset (48, 49).

ChIP analysis. ChIP-seq data for ERG, H3K27Ac, H3K4me3, and H3K4me1 were obtained from following sources: VCaP NTC DHT ERG, GSM2086309; VCaP SICTL H3K27AC, GSM2537220; VCaP SIERG H3K27AC CHIP-SEQ, REP2, GSM2537223; VCaP_Ethl_H3K4me3, GSM353603; VCaP_regular_medium_H3K4me1, GSM353631; LNCaP_ERG, GSM353648; H3K27AC LNCaP-DHT, GSM1249447; LNCaP_regular_medium_H3K4me3, GSM353626; and LNCaP_regular_medium_H3K4me1, GSM353634. CHIP-SEQ, REP2ChIP-seq profiles of ERG and multiple histone modifiers flanking the IRS2 gene were demonstrated using IGV reference to Hg38.

Statistics. Differences between 2 groups were determined by 2-tailed, 2-sided Student’s t tests and Wilcoxon’s rank-sum tests. Both Pearson’s and Spearman’s correlation were used for correlation analysis. Statistical significance was accepted at P < 0.05. Nonlinear regression curve fitting with a variable slope model was applied to assess the dose response of AKT inhibitors, and IC₅₀ generated for each inhibitor under both FBS and CCS conditions. Statistical analyses were performed using GraphPad Prism 9 and RStudio.

Study approval. The analysis of archival deidentified clinical samples was approved by the BIDMC IRB. All animal studies were approved by the BIDMC Institutional Animal Care and Use Committee and conformed to the NIH guidelines.

Data availability. Gene expression data generated in this study has been deposited in GEO (accessions GSE289099 and GSE289100). Supporting data values for Figures 5–7, Supplemental Figure 6, and Supplemental Figure 11 are in the Supplemental Data Values file. Supporting data for mouse gene expression values generated in this study for Figure 4 and Supplemental Figure 9 are deposited in GEO (accession GSE289100). Any further data not included in the report can be obtained from the corresponding author.