Identification of the PTEN-ARID4B-PI3K pathway reveals the dependency on ARID4B by PTEN-deficient prostate cancer

PTEN is frequently mutated in prostate cancer. The tumor suppressor function of PTEN is attributed to its lipid phosphatase activity that counters PI3K action. Here, we report a PTEN-ARID4B-PI3K axis in which PTEN inhibits expression of ARID4B, while ARID4B is a transcriptional activator of the PI3K subunit genes PIK3CA and PIK3R2 that are crucial for activation of the PI3K/AKT pathway. Reciprocal binding of ARID4B and histone H1 to the PIK3CA and PIK3R2 promoters modulates chromatin condensation, suggesting a mechanism by which ARID4B activates these promoters. Functional analyses reveals that ARID4B is required for prostate tumorigenesis when PTEN is deficient. The biological significance is further substantiated by the existence of a PTEN/ARID4B/PIK3CA three-gene signature that improves the predictive power for prostate cancer recurrence in patients. In summary, we identify ARID4B as a master regulator in the PTEN-PI3K pathway, thus providing a potential therapeutic target for prostate cancer carrying PTEN mutations.

P rostate cancer is the second leading cause of male cancerrelated death worldwide. Loss of the tumor suppressor gene PTEN is a frequent genetic alteration in prostate cancers, and many late-stage prostate cancers showed altered activity in the PTEN-PI3K pathway 1,2 . To date, most targeted anticancer drugs inhibit oncogenes that acquire gain-of-function mutations in cancer. However, treatment of cancers that harbor loss-offunction mutations on tumor suppressor genes, such as PTEN, remains a significant challenge because there are no direct targets amenable to pharmaceutical intervention. The emerging concept of synthetic essentiality in cancer describes cancers with specific tumor suppressor deficiencies that depend on expression of synthetic-essential genes. Targeting such genes may be a promising alternative treatment approach for cancers that acquire loss-of-function mutations on tumor suppressor genes 3 . Therefore, significant efforts have been devoted to identifying syntheticessential genes in PTEN-deficient prostate cancers 4,5 .
PI3K phosphorylates phosphatidylinositol (PtdIns)-4,5bisphosphate (PIP2) to generate PtdIns-3,4,5-trisphosphate (PIP3), which activates the AKT signaling pathway to mediate cellular events, including cell survival, growth, proliferation, and cell migration 6,7 . Hyperactivation of the PI3K pathway is associated with prostate cancer progression 8 . The biological effects of PTEN are dominated by its ability to dephosphorylate PIP3 back into PIP2, leading to suppression of the proto-oncogenic PI3K-AKT signaling pathway 6 . Three classes of PI3Ks (class I, class II, and class III) have been identified in mammals. Class I PI3Ks are heterodimeric enzymes that consist of a catalytic subunit p110α, p110β, p110γ, or p110δ (encoded by PIK3CA, PIK3CB, PIK3CG, or PIK3CD, respectively) and a regulatory subunit p85α, p55α, p50α (all encoded by PIK3R1), p85β (encoded by PIK3R2), p55γ (encoded by PIK3R3), p150 (encoded by PIK3R4), or p101 (encoded by PIK3R5) 8 . PIK3CA is frequently mutated in prostate cancers, and many of these mutations result in activation of PI3K 9 . In addition, the Pik3ca gain-of-function mutation accelerates cancer progression of PTEN-deficient prostate in mice 10 . Although alterations of the PTEN-PI3K-AKT pathway are strongly implicated in prostate cancer pathogenesis, their therapeutic targeting has not improved clinical efficacy 8 . Therefore, identification of effectors of this important pathway to target PTEN-deficient prostate cancers is urgently needed.
ARID4B belongs to the ARID family and is a chromatin remodeling factor 11 . ARID4B was previously identified as a component of the HDAC1/SIN3A chromatin remodeling complex 12 . Although ARID4B is a chromatin remodeling protein, it lacks methyltransferase or acetyltransferase activities. To date, how ARID4B modulates chromatin modifications to regulate gene expression is not well understood. However, ARID4B contains a Tudor domain and a chromo domain, and these domains may function as adaptors to assemble chromatin remodeling complexes 13,14 . Expression of ARID4B is elevated in many types of tumors [15][16][17][18] , and ARID4B promotes mammary tumor growth and metastasis 19 . Despite these early findings, the role of ARID4B in prostate cancer remains unclear.
In the present study, we identify ARID4B as a regulator of the PTEN-PI3K pathway by functioning as a transcriptional activator of PIK3CA and PIK3R2. Importantly, we demonstrate that ARID4B is required for initiation and progression of PTENdeficient prostate cancer, suggesting ARID4B is a potential therapeutic target for prostate cancer harboring PTEN mutations.

Results
ARID4B regulates the PI3K-AKT pathway. To determine genomic alterations of ARID4B in prostate cancer, we analyzed the Cancer Genome Atlas (TCGA) and other prostate cancer genomic datasets 2,[20][21][22][23][24][25] . Amplification of ARID4B was detected in up to 20% of human prostate cancer cases ( Supplementary  Fig. 1a, left), while deletion and loss-of-function mutation of PTEN are frequent ( Supplementary Fig. 1a, right). Using two prostate cancer gene expression datasets in the Oncomine database 23,26 , we found that expression of ARID4B was significantly increased in prostate carcinoma (PCa) compared to normal prostate glands ( Supplementary Fig. 1b). In addition, the human prostate cancer cell lines PC3, DU145, LNCaP, and C4-2B expressed higher levels of ARID4B protein than the normal (nonmalignant) prostate epithelial cell lines RWPE-1, PZ-HP-7, and HPrEC (Supplementary Table 1 and Supplementary Fig. 1c). Furthermore, patients expressing higher levels of ARID4B had a significantly increased risk of prostate-specific antigen recurrence in two datasets 27,28 (Supplementary Fig. 1d). These results indicate that expression of ARID4B positively correlates with prostate cancer development, and tumor recurrence is sensitive to higher levels of ARID4B expression.
To determine how ARID4B plays a role in prostate cancer, we performed whole transcriptome analysis (RNA-Seq) to compare gene expression profiles of PC3 cells with or without knockdown of ARID4B (Supplementary Data 1). Expression of the PI3K subunit genes PIK3CA and PIK3R2 was significantly reduced in PC3 cells with knockdown of ARID4B (Table 1 and Supplementary Data 1). Independent quantitative RT-PCR (qRT-PCR) analysis confirmed decreased expression of PIK3CA and PIK3R2 (Fig. 1a). In contrast, E-cadherin (CDH1), an invasion suppressor in many cancers, was upregulated (Table 1 and Fig. 1a). Furthermore, luciferase reporter gene assays showed that ARID4B activated the promoters of PIK3CA and PIK3R2 (Fig. 1b). Consistent with these results, we found a positive correlation between ARID4B and PIK3CA expression within the prostate cancer cohort in the TCGA database (Fig. 1c).
Furthermore, western blot analysis revealed that knockdown of ARID4B reduced protein levels of p110α (encoded by PIK3CA) and p85β (encoded by PIK3R2) in PC3 cells (Fig. 1d). Accordingly, phosphorylation of the PI3K downstream effector AKT at Ser473, required for the full activation of AKT 29 , was reduced in PC3 cells after knockdown of ARID4B (Fig. 1d). In addition, phosphorylation of mTOR at Ser2448 was also diminished (Fig. 1d), consistent with reports that phosphorylation of AKT at Ser473 is linked to phosphorylation of mTOR at Ser2448 and activation of the mTOR pathway 30,31 . Concomitantly, expression of downstream target genes in the PI3K/AKT signaling network, such as cyclin D1 (CCND1), was decreased (Table 1, Fig. 1a, d  Supplementary Fig. 2, and Supplementary Table 2). Consistent with these results, knockdown of ARID4B in the DU145 and LNCaP human prostate cancer cell lines reduced expression or phosphorylation of the core regulators and downstream effectors of the PI3K-AKT pathway ( Supplementary Fig. 3). Together, these results suggest that ARID4B positively regulates expression  Fig. 4a, b), and identified PIK3CA and PIK3R2, whose promoters were associated with ARID4B ( Fig. 2a, b, orange lines). The CRISPR-Cas9 gene editing system was used to achieve knockout of ARID4B (ARID4BKO) in PC3 cells ( Supplementary Fig. 6a). ChIP-qPCR analysis using the antibody against ARID4B further confirmed recruitment of ARID4B to the PIK3CA and PIK3R2 promoters in control PC3 cells, but showed minimal ARID4B binding in ARID4BKO PC3 cells (Fig. 2c). In contrast, ARID4B was not recruited to the promoters of myogenic differentiation 1 (MYOD1), glyceraldehyde 3-phosphate dehydrogenase (GAPDH), or ARID4B itself (Fig. 2c, Supplementary Fig. 4c, and Supplementary Data 2), indicating that recruitment of ARID4B to the PIK3CA and PIK3R2 promoters is specific. Together with the fact that ARID4B regulates promoter activities of PIK3CA and PIK3R2 (Fig. 1b) and controls their expression (Table 1 and Fig. 1a), these results indicate that PIK3CA and PIK3R2 are direct target genes of ARID4B.
ARID4B modulates chromatin condensation on PIK3CA and PIK3R2. Since histone H1 stabilizes the condensed chromatin structure that contributes to gene silencing 33 , reciprocal binding of ARID4B and histone H1 to the PIK3CA and PIK3R2 promoters led us to examine whether ARID4B impacts chromatin structure on these promoters. Using micrococcal nuclease (MNase) that preferentially digests loosened chromatin followed by qPCR analysis, the promoter regions of PIK3CA and PIK3R2 (but not MYOD1) in control PC3 cells were more sensitive to MNase digestion than ARID4BKO PC3 cells (Fig. 2e), indicating that the chromatin was less condensed in the presence of ARID4B. Collectively, these results suggest that reciprocal binding of ARID4B and histone H1 to the PIK3CA and PIK3R2 promoters modulates chromatin condensation. Binding of ARID4B excludes histone H1 from the PIK3CA and PIK3R2 promoters, leading to relaxed chromatin architecture and gene activation (Fig. 2f). Upon ARID4B ablation, histone H1 binds to the promoters and condenses the chromatin structure to suppress gene expression (Fig. 2f). Together, these results suggest a mechanism by which ARID4B activates transcription of PIK3CA and PIK3R2.
ARID4B is required for PTEN-deficient prostate cancer. As a tumor suppressor, PTEN counters the actions of PI3K via its lipid phosphatase activity to govern cellular processes, including cell growth, proliferation, and cell migration 7 . PTEN is frequently disrupted in prostate cancer 1,2 . Loss of PTEN function leads to derepression of the PI3K-AKT pathway that dictates tumor susceptibility and favors cancer progression 7 . To determine the functional significance of our findings that ARID4B is a transcriptional activator of the PI3K subunits PIK3CA and PIK3R2 (Table 1 and Fig. 1a, b), we investigated the role of ARID4B in the development of prostate cancer harboring PTEN deficiency. By analyzing the Cancer Genome Atlas (TCGA) and other prostate cancer genomic datasets 2,20,24,25,34 , we identified that ARID4B is consistently retained in prostate carcinoma cohorts with deep deletion of PTEN, although deletion of ARID4B also occasionally occurs in prostate cancer (Fig. 3a).
Because there tends to be mutual exclusivity between ARID4B and PTEN deletions in prostate cancer genome (Supplementary  Table 3), ARID4B might be a synthetic-essential gene for prostate tumorigenesis in the context of PTEN deficiency. Therefore, we examined whether inhibition of ARID4B impairs the tumorpromoting actions in prostate cancer cells driven by loss of PTEN. In PC3 cells, knockout of ARID4B not only efficiently constrained cell growth (Fig. 3b), but also significantly suppressed aggressiveness of cancer cells, including cell migration, cell invasion, and tumorsphere formation (Fig. 3c). In addition, ablation of ARID4B sufficiently inhibited cell proliferation and tumor formation in tumor xenograft experiments (Fig. 3d, e). On the other hand, knockout of ARID4B in the PTEN-intact prostate cancer cell line DU145 only moderately suppressed tumor growth and cell proliferation, and not to the same extent as in PC3 cells ( Supplementary Fig. 6). Together, these results indicate that ARID4B is required for tumor growth of PTEN-null prostate cancer cells.

Knockout of Arid4b compromises prostate cancer progression.
To demonstrate the effects of ARID4B ablation on prostate tumorigenesis driven by PTEN deficiency in mice, we used the well-characterized prostate cancer mouse model that carries prostate-specific deletion of Pten (Pten PC−/− ), and generated mice carrying prostate-specific deletions of both Pten and Arid4b (Pten PC−/− Arid4b PC−/− ) ( Supplementary Fig. 7a). While deletion of Pten resulted in enlarged prostates in the Pten PC−/− mice, this effect was reduced in the Pten PC−/− Arid4b PC−/− mice ( Fig. 4a and Supplementary Fig. 7b). The Pten PC−/− mice develop lowgrade prostatic intraepithelial neoplasia (PIN) that may progress to high-grade PIN or adenocarcinoma 35 , which faithfully mimics prostate cancer progression in humans. Consistent with the earlier report, we showed that most Pten PC−/− mice developed highgrade PIN or adenocarcinoma at 5 months of age (Fig. 4b, c and Supplementary Fig. 7c). In contrast, tumor development in prostates of most of the 5-month-old Pten PC−/− Arid4b PC−/− mice was arrested at the normal, hyperplastic, or low-grade PIN stages (Fig. 4b, c and Supplementary Fig. 7c), suggesting that ablation of ARID4B attenuated prostate cancer progression. Correspondingly, there were fewer Ki67-labeled nuclei (a marker of proliferation) in prostates of the Pten PC−/− Arid4b PC−/− mice ( Fig. 4d), suggesting that ablation of ARID4B compromised the proliferation advantage elicited by loss of PTEN. These effects were not due to defects in prostate development since the Arid4b PC−/− mice had prostates of normal size and histology ( Supplementary Fig. 8a, b).
Consistent with the results from PC3 cells with ablation of ARID4B (Fig. 1d), expression or phosphorylation of the core regulators and downstream effectors of the PI3K-AKT pathway were reduced in prostates of Pten PC−/− Arid4b PC−/− mice (Fig. 4e), suggesting a mechanism by which ARID4B is required for executing actions that promote prostate cancer in the context of PTEN deficiency.
To analyze whether ablation of ARID4B delays the onset of prostate cancer caused by PTEN deficiency, we then examined the Pten PC−/− Arid4b PC−/− mice at 9 months of age. While the Pten PC−/− mice exhibited enlarged prostates, prostate size was still comparable in the Pten PC−/− Arid4b PC−/− and control mice ( Supplementary Fig. 9a). Similar to the results from 5-month-old mice, only normal/hyperplasia and PIN were found in the Pten PC−/− Arid4b PC−/− mice ( Supplementary Fig. 9b, c). These results suggest that in the Pten PC−/− Arid4b PC−/− mice, PIN did not progress to adenocarcinoma over time. Collectively, these data demonstrate a synthetic essentiality of ARID4B in prostate cancer elicited by PTEN deficiency.  relationship between ARID4B and PTEN, we queried the dataset of transcriptome comparison between wild-type and Pten PC−/− prostates at 15 weeks of age 36 , and found that expression of Arid4b was elevated in prostates of the Pten PC−/− mice (fold change: 4.859; P = 0.0094; statistical analysis: t test). We confirmed the increase of Arid4b mRNA in the Pten PC−/− prostates by qRT-PCR analysis (Fig. 5a). Immunohistochemistry and western blot analyses verified a robust increase of ARID4B protein levels in prostates of the Pten PC−/− mice (Fig. 5b, c). Consistent with these results, expression of ARID4B was significantly increased in prostate cancer patients harboring PTEN deletion compared to those with unaltered PTEN in the TCGA provisional dataset (Fig. 5d). These results suggest that deficiency of PTEN increases expression of ARID4B.
On the other hand, re-expression of PTEN in PC3 cells (PTENdeficient) decreased expression of ARID4B (Fig. 5e), accompanied by lessened abundance of p110α and p85β (encoded by PIK3CA and PIK3R2, respectively) and inactivation of the PI3K-AKT signaling pathway ( Fig. 5e and Supplementary Fig. 10a). Furthermore, ARID4B promoter-driven luciferase reporter gene assays showed that knockdown of PTEN activated the ARID4B promoter in the PTEN-intact prostate cancer cell line DU145 and cervical cancer cell line HeLa ( Fig. 5f and Supplementary  Fig. 10b), whereas overexpression of PTEN suppressed the ARID4B promoter activity (Fig. 5g). Together, these results suggest that PTEN negatively regulates expression of ARID4B.
PTEN lipid phosphatase activity dephosphorylates PIP3 into PIP2 to negatively regulate the PI3K signaling cascade. Thus, loss of PTEN promotes PI3K-AKT signaling 6,7 . To demonstrate that PTEN deficiency correlates with increased ARID4B expression via activation of the PI3K-AKT pathway, we showed that knockdown of AKT in the PTEN-deficient prostate cell line PC3 decreased expression of ARID4B (Fig. 5h). Consistent with this result, adding an AKT inhibitor MK-2206 downregulated ARID4B in PC3 cells (Fig. 5i). Furthermore, the ARID4B promoter-driven luciferase reporter gene assay showed that a constitutively active form of AKT (CA-AKT) activated the promoter of ARID4B (Fig. 5j). These results indicate that AKT positively regulates expression of ARID4B. Together with the fact that ARID4B is required for activation of the PI3K-AKT signaling pathway (Fig. 1), our results suggest an AKT-ARID4B-positive feedback loop (Fig. 5k).
Functional interactions between PTEN, ARID4B, and PIK3CA. To investigate the clinical relevance of the PTEN-ARID4B-PI3K axis in prostate cancer, we performed immunohistochemical staining with antibodies against ARID4B, PTEN, and p110α on tissue microarrays (TMAs) consisting of 118 prostate adenocarcinoma samples from 85 patients and 73 prostate hyperplasia samples from 36 patients. Results revealed that expression of ARID4B was higher in prostate adenocarcinoma compared to prostate hyperplasia (Fig. 6a, left), shown as an increase in the number of ARID4B-positive cells in prostate adenocarcinoma samples (Fig. 6c, left and Supplementary Fig. 11a, left). This is consistent with results from our Oncomine database analysis, showing increased mRNA levels of ARID4B in human prostate carcinoma compared to normal prostate glands ( Supplementary  Fig. 1b). In prostate adenocarcinoma, PTEN-positive cells were diminished (Fig. 6a, c, middle and Supplementary Fig. 11a, middle), whereas the number of p110α-positive cells was increased (Fig. 6a, c, right and Supplementary Fig. 11a, right). Furthermore, expression of ARID4B and PTEN was negatively correlated in human prostate adenocarcinoma (Fig. 6b, c and Supplementary Fig. 11a), while expression of ARID4B and p110α was positively correlated (Fig. 6b, c and Supplementary Fig. 11a).
To complement our TMA results, we determined how expression of these three genes correlates with prostate cancer recurrence in patients. A multivariate Cox proportional hazards model was used to combine expression levels of PTEN, ARID4B, and PIK3CA with time to tumor recurrence from the dataset GSE40272 28 into an integrated risk score model. We found that the PTEN/ARID4B/PIK3CA three-gene signature had an increased predictive power for prostate cancer recurrence compared to any individual gene alone or two genes combined ( Table 2 and Supplementary Table 4). This three-gene signature reliably stratified patients into high-and low-risk groups for tumor recurrence (Fig. 6d). The enhanced prognostic value of the three-gene signature for recurrent prostate cancer was similarly confirmed by analysis of another two independent datasets GSE21032 2 and TCGA provisional (Supplementary Tables 5, 6 and Supplementary Fig. 11b, c). Together, these data support the importance of functional interactions between PTEN, ARID4B, and PIK3CA in human prostate cancer.

Discussion
Although ARID4B does not have intrinsic methyltransferase or acetyltransferase activities, it contains a Tudor domain and a chromo domain 11 . These two domains may promote assembly of chromatin remodeling complexes 13,14 . ARID4B is a component of the SIN3A/HDAC1 repressor complex that inhibits expression of target genes 12 . In this report, we showed that ARID4B was recruited to the promoters of PIK3CA and PIK3R2 and activated their expression. These results are consistent with our previous findings that ARID4B functions as coactivator (not a co-repressor) for androgen receptor in the testis and is essential for testis development 37 .
In addition, while knockdown of ARID4B decreased expression of p110α and p85β in the prostate cancer cell lines PC3, DU145, and LNCaP ( Fig. 1d and Supplementary Fig. 3), knockdown of HDAC1 increased expression of p110α and p85β ( Supplementary  Fig. 12). These results indicate that ARID4B and HDAC1 have opposite effects on regulation of PIK3CA and PIK3R2 expression. As a repressor, HDAC1 deacetylates histones H4 and H3 to inhibit transcriptionally active states of the gene promoters. While HDAC1 suppresses expression of PIK3CA and PIK3R2, ablation of ARID4B in PC3 cells did not affect levels of H4 Fig. 2 PIK3CA and PIK3R2 are direct target genes of ARID4B. a, b ChIP-Seq analysis shows enrichment of ARID4B binding to the promoters of PIK3CA (a) and PIK3R2 (b) in PC3 cells. Orange lines indicate the ARID4B binding regions. c, d Recruitment of ARID4B (c) and histone H1 (d) to the promoters of PIK3CA, PIK3R2, and MYOD1 in control and ARID4BKO PC3 cells were analyzed by ChIP-qPCR analyses using anti-ARID4B (c) or anti-H1 (d) antibodies. Normal rabbit IgG was used for comparison. e Chromatin accessibility of the PIK3CA, PIK3R2, and MYOD1 promoters to micrococcal nuclease (MNase) digestion. The PCR products on the promoter region of each gene detected by specific primer sets in qPCR are as follows: PIK3CA (PCR products 1 and 2), PIK3R2 (PCR products [3][4][5], and MYOD1 (c-e). See Supplementary Fig. 4e for details on regions of qPCR products. Data are means ± SEM from three experiments performed in triplicate (c-e). *P < 0.05; **P < 0.01; ***P < 0.001; NS no significant differences; Statistical analysis: t test (c-e). f A schematic representation of how ARID4B controls chromatin condensation to regulate the promoter activities of PIK3CA and PIK3R2. ARID4B maintains a relaxed chromatin structure on the promoters, leading to gene activation. Loss of ARID4B results in occupancy of histone H1 on the promoters and chromatin condensation, leading to suppression of gene transcription acetylation on the promoters of PIK3CA and PIK3R2 (Supplementary Fig. 5a), suggesting that ARID4B does not mediate HDAC1 activity on these two promoters. Together, our results suggest that ARID4B and HDAC1 are likely to function independently to regulate PIK3CA and PIK3R2 in prostate cancer cells. Therefore, the SIN3A/HDAC1 repressor complex does not contribute to activation of PIK3CA and PIK3R2 by ARID4B. The function of ARID4B in transcriptional regulation may depend on the gene promoter context and the proteins that associate with ARID4B, which could dictate its function as an activator Analysis of five prostate cancer genome datasets using the cBioportal data source (http://www.cbioportal.org/) suggest that ARID4B and PTEN deletions tend to be mutually exclusive in the prostate cancer genome, although the relationship did not reach statistical significance (P > 0.05) (Supplementary Table 3). In most cohorts, the ARID4B gene alteration rate is less than 5%, which makes it difficult to accrue enough samples for a significant P value. Similar to our findings, others have reported a tendency of mutual exclusivity between CHD1 and PTEN deletions in prostate cancer without reaching statistical significance 5 . In addition, although the mutual exclusivity between ARID1B and KRAS is not statistically significant in any individual type of cancer, statistical significance for mutual exclusivity was achieved when multiple cancer types were combined to increase the sample size 38 . The tendency of ARID4B-PTEN to be mutually exclusive  expression levels in human prostate cancers with unaltered or deleted PTEN from the TCGA dataset (provisional). Data are means ± SEM. e Levels of PTEN, ARID4B, p110α, p85β, and actin in PC3 cells with or without PTEN re-expression were determined by western blot analysis. f, g Luciferase reporter assays using the ARID4B promoter-driven luciferase reporter were performed in DU145 cells with or without knockdown of PTEN by two different short interfering RNAs SiPTEN 1 or SiPTEN 2 (f), and in HeLa cells with or without PTEN overexpression (g). h, i Western blot analyses determined the levels of ARID4B and AKT in PC3 cells with or without knockdown of AKT by short interfering RNAs SiAKT 1 or SiAKT 2 (h), or in PC3 cells untreated or treated with an AKT inhibitor MK-2206 (i). j Luciferase reporter assay using the ARID4B promoter-driven luciferase reporter was performed in HeLa cells with or without overexpression of CA-AKT (constitutively active AKT). Data are means ± SEM from three experiments performed in triplicate (a, f, g, j). *P < 0.05; **P < 0.01; ***P < 0.001; Statistical analysis: t test (a, d, f, g, j). k Genetically engineered mouse models demonstrate that ARID4B is a key factor in the PTEN/ PI3K pathway to promote tumorigenesis in prostate with PTEN deficiency. In normal prostate (left), PTEN opposes the PI3K/AKT signaling pathway and constrains prostate tumorigenesis. In In both PC3 and DU145 cells, depletion of ARID4B resulted in reduced expression or phosphorylation of the core regulators and downstream effectors of the PI3K-AKT pathway, while expression of PTEN in DU145 cells was unchanged ( Supplementary  Fig. 3a). We showed that depletion of ARID4B inhibited colony formation in PC3 and DU145 cells (Supplementary Fig. 13a, b), and re-expression of ARID4B in the ARID4BKO PC3 and ARID4BKO DU145 cells increased colony formation (Supplementary Fig. 13c). Depletion of ARID4B in another PTENdeficient prostate cancer cell line, LNCaP, significantly suppressed cell growth (Supplementary Fig. 14). In the PC3 (PTEN-deficient) cells, knockout of ARID4B efficiently inhibited cancer cell growth and tumor formation in xenografts, suggesting that PTENdeficient prostate cancer cells depend on ARID4B. In the DU145 (PTEN-intact) cells, knockout of ARID4B also suppressed cell growth and tumor formation in xenografts, but less than in PC3 cells. PTEN-deficient PC3 cells may depend more on activation of the PI3K-AKT pathway for tumor growth than the PTEN-intact DU145 cells. This would explain why the AKT inhibitor MK-2206 inhibited cell growth more effectively in PC3 cells than  Fig. 15a, b). In PTEN-deficient PC3 cells, ablation of PTEN activates the PI3K/AKT pathway, leading to tumorigenesis (Supplementary Fig. 15c, left). In ARID4BKO PC3 cells, knockout of ARID4B compromises activation of the PI3K/AKT pathway, thus efficiently inhibiting tumorigenesis elicited by PTEN deficiency (Supplementary Fig. 15c, right). In PTEN-intact DU145 cells, PTEN dephosphorylates PIP3 into PIP2, which opposes the PI3K/AKT signaling pathway (Supplementary Fig. 15d, left), suggesting tumorigenesis of DU145 cells is less dependent on the PI3K-AKT signaling than PC3 cells. Therefore, knockout of ARID4B that inactivates PI3K/AKT pathway only moderately suppressed tumorigenesis of DU145 cells ( Supplementary Fig. 15d, right).
In addition, ARID4B may promote cancer cell activity through additional pathways unrelated to PTEN tumor suppressor function. For example, our immunohistochemical analyses detected increased numbers of cells positive for cleaved caspase 3, an apoptosis marker, in prostates of the Pten PC−/− Arid4b PC−/− mice compared to the Pten PC−/− mice, suggesting that ablation of ARID4B promoted apoptosis ( Supplementary Fig. 16a, b). Consistent with this result, expression levels of the apoptosis regulators p53 and Fas were elevated in prostates of the Pten PC−/− Arid4b PC−/− mice compared to the Pten PC−/− mice ( Supplementary Fig. 16c). For its oncogenic function, ARID4B might directly or indirectly inhibit the apoptosis pathway mediated by the p53 tumor suppressor.
Similar to the PTEN/ARID4B/PIK3CA three-gene signature, PTEN together with SMAD, CCND1, and SPP1 form a four-gene signature that predicts recurrence and lethal metastasis in human prostate cancer 36 . SMAD4 is a key regulator of the TGFβ pathway and a suppressor of prostate tumor progression, whereas cyclin D1 and SPP1 act as mediators of the prostate cancer processes. Inactivation of PTEN and SMAD4 and activation of cyclin D1 and SPP1 promote prostate cancer. The combination of PTEN/ SMAD4 does not predict risk for prostate cancer recurrence, but the PTEN/SMAD4/CCND1/SPP1 four-gene signature does. This four-gene model distinguished patient groups at high versus low risk for prostate cancer recurrence by Kaplan-Meier analysis [hazard ratio (HR) = 2.6, P = 0.012] 36 . In addition, the PTEN/SMAD4/CCND1/SPP1 four-gene signature involves the PTEN and TGFβ pathways, which converges to regulate prostate cancer progression 36 . Similarly, we showed that the PTEN/ARID4B/PIK3CA signature has greater predictive power for tumor recurrence than any individual gene alone or two genes combined. This PTEN/ARID4B/PIK3CA signature robustly stratified patients into those at high versus low risk for tumor recurrence (HR = 4.58, P = 0.006 in the GSE40272 dataset; HR = 3.23, P = 0.002 in the GSE21032 dataset; HR = 1.51, P = 0.025 in the TCGA dataset; Statistical analysis: log-rank test). Thus, we identified that the PTEN-ARID4B-PI3K regulatory axis underlies the predictive power of the PTEN/ARID4B/PIK3CA signature for tumor recurrence in prostate cancer patients.
In this study, our data suggest that ARID4B, by directly impinging on the PTEN-PI3K signaling pathway, functions as a synthetic-essential effector for prostate cancer initiation and progression elicited by loss of PTEN. Our study reveals the PTEN-ARID4B-PI3K axis and an AKT-ARID4B-positive feedback loop whereby PTEN antagonizes the PI3K-AKT signaling pathway. In normal prostate, PTEN is attributed to its lipid phosphatase activity that directly counters PI3K actions, whereas ARID4B is a transcriptional activator of the PI3K subunit genes PIK3CA and PIK3R2, and expression of ARID4B itself is regulated by the AKT-ARID4B-positive feedback loop (Fig. 5k, left). Deficiency of PTEN activates the PI3K-AKT pathway and the AKT-ARID4B-positive feedback loop, leading to prostate tumorigenesis (Fig. 5k, middle). In contrast, ablation of ARID4B inhibits the PI3K-AKT pathway, and compromises cancer progression in prostate harboring PTEN deficiency (Fig. 5k, right). Therefore, our study demonstrates the synthetic-essential role of ARID4B in PTEN-deficient prostate cancer, and provides the underlying mechanism by which ARID4B is required for executing prostate cancer-promoting actions in the context of PTEN deficiency. Together, our findings reveal insights into how PTEN controls prostate tumorigenesis, and support the concept that ARID4B is a potential target for interventions in patients with prostate cancer carrying PTEN mutations. Histology and immunohistochemistry. The prostates were dissected from mice and fixed in 4% paraformaldehyde in phosphate-buffered saline (PBS) for 24 h, processed, and embedded in paraffin. Histological analysis was performed on 5-μm-thick paraffin-embedded prostate sections by hematoxylin and eosin staining. For immunohistochemistry, antigen retrieval was performed by boiling the paraffin-embedded prostate sections in citric acid-based antigen unmasking solution (Vector Laboratories). Samples were blocked with blocking solution (5% donkey serum, 2% bovine serum albumin, and 0.02% Triton-X 100 in PBS), followed by incubation with primary antibodies in blocking solution. The primary antibodies used are anti-ARID4B (A302-233A, Bethyl Laboratories; dilution 1: 1000), anti-Ki67 (B56; 550609, BD Biosciences; dilution 1: 50), and anti-cleaved caspase 3 (9661, Cell Signaling; dilution 1: 400) antibodies. Sections were washed with 0.1% Triton X-100/PBS buffer, and incubated with biotinylated secondary antibodies (Jackson ImmunoResearch). Signal detection was carried out using the ImmPACT DAB Peroxidase Substrate system (Vector Laboratories). Sections were counterstained with hematoxylin (Sigma). Images were captured with a Nikon Eclipse Ti-U microscope.
Immunohistochemical staining of tissue microarrays (TMAs). Immunohistochemical staining of two TMAs (PRC1021 and PRD961) and image scanning services were provided by Pantomics (Pantomics, Inc. USA). Two TMA slides, PRC1021 and PRD961, included 118 prostate adenocarcinoma cores from 85 patients, 73 prostate hyperplasia cores from 36 patients, and 5 benign tumor cores. These TMA slides were baked at 60°C for 30 min and then deparaffinized. After antigen retrieval, slides were incubated with primary antibodies: anti-ARID4B (A302-233A, Bethyl Laboratories; dilution 1:400), anti-PTEN (A2B1; sc-7974, manufacturer's instructions. Plasmid transfection into HeLa cells was accomplished using Lipofectamine 2000 TM Transfection Reagent (Invitrogen) according to the manufacturers' instructions. Forty-eight hours after transfection, whole cell lysates were prepared and luciferase activity was determined by the luciferase assay system following the manufacturer's instructions (Promega). Transfection of SiRNA was carried out using Lipofectamine™ RNAiMAX Transfection Reagent (Invitrogen) according to the manufacturers' instructions. Seventy-two hours after transfection, total RNA was purified for RNA-Seq and qRT-PCR analysis, or whole cell lysates were prepared for luciferase reporter gene assays. MISSION® Predesigned siRNA targeting human ARID4B (SASI_Hs01_00134230 and SASI_Hs02_00349366) and human PTEN (SASI_Hs01_00196478 and SASI_Hs01_00196479) were purchased from Sigma-Aldrich.
To re-express PTEN in PC3 cells, we generated stable PC3 clones with doxycycline-inducible expression of PTEN using the T-REx system according to the manufacturer's instructions (Invitrogen). First, PC3 cells were transfected with pLenti6/TR (Invitrogen) using Lipofectamine 2000 (Invitrogen), and selected by blasticidin. The PC3-TR clones stably expressing the Tet repressor was confirmed by western blot analysis. The human PTEN cDNA was ligated into the pcDNA4/ TO/Myc-His A vector (Invitrogen) to generate pCDNA4/TO-PTEN. PC3-TR cells were then transfected with the pcDNA4-TO-PTEN vector and were selected by zeocin. Upon addition of doxycycline, stable clones with inducible expression of PTEN were confirmed by western blot analysis.
Knockout using CRISPR. To generate the knockout of ARID4B in PC3 and DU145 cell lines (ARID4BKO), the plasmids with ARID4B sgRNA were transfected into cells using Lipofectamine 2000. The ARID4B CRISPR guide RNA plasmid targeting human ARID4B was purchased from GenScript. These two ARID4B sgRNA sequences are ATCAGTGCCCACTGTCAAAT and AGTTCAGGATGACCACA-TAA. The level of ARID4B protein in each ARID4BKO clone was determined by western blot analysis. Cell proliferation assays were performed using Cell Proliferation Kit I (MTT) (Roche) according to the manufacturer's instructions.
Xenograft prostate cancer model. The in vivo tumor growth of ARID4BKO PC3, ARID4BKO DU145, and their control cells was determined using subcutaneous transplant xenograft models. Cancer cells (2 × 10 6 ) in PBS/Matrigel mixture (Geltrex TM LDEV-free reduced growth factor basement membrane matrix, Gibco) were injected subcutaneously into 6-week-old male NOD/SCID mice (The Jackson Laboratory) under deep anesthesia. The sizes of tumors were measured twice a week. Once the largest tumor diameter reached 1.5 cm, all mice were sacrificed and tumors were harvested.
Migration, cell invasion, and tumorsphere formation assays. Cell migration and cell invasion of the control and ARID4BKO PC3 cells were examined using CytoSe-lect™ Cell Migration (CBA-100) and CytoSelect™ Cell Invasion Assay (CBA-110), respectively, following the manufacturer's instructions. Briefly, cells (5 × 10 4 ) in serumfree culture medium were seeded into 24-well Transwell® Inserts (8.0 μm) of the kits. Medium containing 10% FBS was added to the remaining receiver wells. After 24 h, the inside of each insert was gently swabbed, and stained with crystal violet solution for 10 min. The migratory and invasive cells were imaged with a light microscope under high magnification objective, and counted. At least three individual fields per insert were counted, and the results were from three independent experiments.
The tumorsphere formation assay was modified from a previously published protocol 43 . Briefly, the control and ARID4BKO PC3 cells were washed twice in 1× PBS, trypsinized, centrifuged and resuspended in serum-free DMEM-F12 to obtain a single cell suspension. Then, cells (1 × 10 4 ) were seeded in six-well ultra-low attachment culture plates (Corning) with DMEM-F12 supplemented with 20 ng/μl hEGF (Gibco), 20 ng/μl bFGF (Gibco), 1× B27 without vitamin A (Invitrogen), 1× insulin-transferrin-selenium A (Invitrogen), and 100 U/100 μg/ml penicillinstreptomycin. The plates were incubated in a 5% CO 2 , humidified incubator at 37°C for 2 weeks with replenishment of medium every 3 days. Tumorspheres that were 60 μm or larger in size were counted. The results were from three independent experiments performed in triplicate.
Bioinformatic analyses from human prostate cancer datasets. Data on genomic alterations of ARID4B and PTEN from human prostate cancer datasets 2,20-25,34 were acquired from The cBioPortal for Cancer Genomics (http://www.cbioportal.org) 44,45 . Z-scores for mRNA expression of ARID4B and PIK3CA with the corresponding Pearson correlation coefficient were obtained for prostate cancer samples in the TCGA database. Heat maps with a three-color scale set at −3, 0, and 3 were obtained from the cBioPortal for cancer genomics (http://www.cbioportal.org) 44,45 .
As shown in Supplementary Fig. 1d, Kaplan-Meier survival analysis was conducted using the PROGgeneV2 prognostics database 46 to seek correlations between expression of ARID4B in prostate tumors and recurrence-free survival in patients from the GSE40272 28 and the ref. 27 datasets. Levels of ARID4B expression were used to define high-and low-risk scores. The two groups (high and low expression of ARID4B) were generated by splitting patients' data based on the median value of ARID4B expression.
Prognostic index calculation of PTEN/ARID4B/PIK3CA signature. The predictive power of the PTEN/ARID4B/PIK3CA three-gene signature for tumor recurrence was calculated from three datasets: the TCGA provisional, the GSE21032 2 , and the GSE40272 28 datasets. The prognostic index (PI) was used to define high-and lowrisk scores, and the two risk groups (high and low) were generated by splitting patients' data at the median PI. The PI of multiple genes was calculated as PI = x1β1 + x2β2 + … + xpβp, where x is the gene expression value and β is a risk coefficient obtained from the Cox fitting. The fitting was performed in R (http://cran.r-project. org) using the survival package. The β represents the correlation between gene expression levels and the number of events (in our case, death from cancer recurrence). A positive β means a higher gene expression level may contribute to a higher death rate, and a negative β means a higher gene expression value may contribute to a lower death rate. In our case, PTEN has a negative β coefficient, while ARID4B and PIK3CA both have positive β coefficients. Patients with lower PTEN expression levels or with higher ARID4B and/or PIK3CA levels would have higher calculated PIs, and therefore would be classified into the high-risk group.
The multivariate Cox proportional hazards model of the ARID4B/PTEN/PIK3CA signature combined expression levels of these three genes with time to recurrence in prostate cancer patients from the TCGA provisional dataset using statistical software SPSS19.0, and from the GSE21032 2 and GSE40272 28 datasets using the SurvExpress web tool 47 (http://bioinformatica.mty.itesm.mx:8080/Biomatec/SurvivaX.jsp). In addition, Kaplan−Meier survival curves were used to test whether the ARID4B/ PIK3CA/PTEN signature distinguishes between high-and low-risk groups for the prostate cancer recurrence in the TCGA provisional (P = 0.025), GSE21032 2 (P = 0.002), and GSE40272 28 (P = 0.006) datasets. Statistical analysis: log-rank test.
Statistical analysis. All mouse experiments were performed using 3-16 mice. All experiments from cell lines were performed using at least three independent experiments in triplicate. Means were calculated from at least three independent experiments. All results were shown as the means ± standard errors of the means (SEM). Two-tailed unpaired Student's t test, log-rank test, or Wald chi-squared test were used for statistical analyses. P values less than 0.05 were considered to be statistically significant.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.