MiR-4638-5p inhibits castration resistance of prostate cancer through repressing Kidins220 expression and PI3K/AKT pathway activity

MicroRNAs (miRNAs) are short, conserved segments of non-coding RNA which play a significant role in prostate cancer development and progression. To identify miRNAs associated with castration resistance, we performed miRNA microarray analysis comparing castration resistant prostate cancer (CRPC) with androgen dependent prostate cancer (ADPC). We identified common underexpression of miR-4638-5p in CRPC compared to ADPC samples, which were further confirmed by quantitative PCR analysis. The role of miR-4638-5p in prostate cancer androgen-independent growth has been demonstrated both in vitro and in vivo. We also identified Kidins220 as a target gene directly regulated by miR-4638-5p and shRNA-mediated knockdown of Kidins220 phenocopied miR-4638-5p restoration. Subsequently, we revealed that Kidins220 activates PI3K/AKT pathway, which plays a key role in CRPC. Loss of miR- 4638-5p may lead to CRPC through the activity of Kidins220 and PI3K/AKT pathway. Furthermore, we found that miR-4638-5p, through regulating Kidins220 and the downstream activity of VEGF and PI3K/AKT pathway, influences prostate cancer progression via angiogenesis. The identification of miR-4638-5p down-regulation in CRPC and the understanding of the functional role of miR-4638-5p and its downstream genes/pathways have the potential to develop biomarkers for CRPC onset and to identify novel targets for novel forms of treatments of this lethal form of PCa.


INTRODUCTION
Prostate cancer (PCa) is the commonest malignancy in males in the western world and accounts for approximately 10% of all male cancer-related deaths in the USA [1]. Androgen deprivation therapy (ADT) is an effective therapeutic option for advanced PCa but castration resistance invariably develops [2]. In this context, chemotherapy and other systemic therapies have limited efficacy [3]. An improved understanding of the molecular mechanisms of castration resistance can guide rational development of novel therapeutic strategies. miRNAs play critical roles in the regulation of cell biological processes [4,5] and are involved in the development and progression of various human malignancies including PCa [6][7][8][9][10][11][12]. In PCa, miRNAs have been reported to regulate cancer cell growth, survival and sensitivity to chemotherapy [6,7,9,11,13]. Although differential expression of certain miRNAs between androgen-dependent prostate cancer (ADPC) and castration-resistant prostate cancer (CRPC) clinical samples has been reported, each of those studies was performed on only a few CRPC clinical samples. The cellular functions of these differentially expressed miRNAs identified in the clinical samples were also only investigated generally for PCa cell growth instead of the development of castration resistance [14][15][16]. We aimed to identify novel miRNAs associated with castration resistance in PCa and to investigate the downstream pathways and molecular mechanisms. We found decreased expression of miR-4638-5p in CRPC and revealed functionally that miR-4638-5p inhibited the growth of androgen independent cancer cells and that the inhibition of miR-4638-5p promoted androgenindependent cell growth for initially androgen dependent cells. Subsequently we identified Kidins220, VEGF and the PI3K/AKT pathway as the downstream mediators of miR-4638-5p in suppressing CRPC development as well as PCa angiogenesis in vitro and in xenografted nude mice.

MiR-4638-5p is significantly down-regulated in CRPC compared to ADPC samples
To explore the expression change of miRNAs associated with castration resistance, we performed miRNA microarray experiments using RNA samples extracted from three CRPC and three ADPC samples (Supplementary Table S1). We found 30 underexpressed (Supplementary Table S2) and 32 overexpressed miRNAs (Supplementary Table S3) with >2 fold difference of mean value in the three CRPC samples compared with the ADPC samples ( Figure 1A). 25 underexpressed and 16 overexpressed miRNAs have been previously reported in human cancers, including PCa, such as underexpression of miR-135a [17,18], miR-143 [17,19], miR-145 [17,20], miR-205 [17,[21][22][23][24], miR-24-1 [17,25], miR-146a [26,27], miR-3607 [28], and overexpression of miR-1247 [29], miR-197 [20], miR-592 [30], miR-150 [31,32]. Among the 30 underexpressed miRNAs in CRPC samples, miR-4638-5p, a miRNA that has not been previously reported in any human cancers, was expressed 2.4-fold higher in the ADPC than CRPC samples. Although it has not been reported in a previous publication [33], we found that miR-4638-5p was expressed 4.7-fold higher in early-stage compared to advanced disease in the miRNA microarray data from that study. In a separate miRNA microarray analysis of PCa stem cells, which are closely associated with CRPC development [34], miR-4638-5p was the most significantly down-regulated miRNA compared with the unselected population of DU145 cells, with 12.6-fold difference (unpublished data). All these data support that reduced expression of miR-4638-5p may be responsible for a more aggressive phenotype. We further examined its expression in 30 ADPC tissues and 18 CRPC tissues by qRT-PCR. In addition to confirming the CRPC-associated downregulation of miR-4638-5p in the six samples used for microarray analysis, we found that miR-4638-5p expression level was significantly lower in CRPC compared to ADPC samples (P = 1.44 × 10 -8 ) ( Figure 1B) in this cohort of samples. We also determined the expression levels of miR-4638-5p in LNCaP, PC3, DU145 and LNCaP C4-2B PCa cell lines, consistent with our observation in the clinical samples, miR-4638-5p expression at a much higher level in LNCaP androgen sensitive cells than the three androgen-independent PCa cell lines ( Figure 1C).

MiR-4638-5p suppresses androgen-independent PCa cell growth in vitro and in vivo
We then explored the potential biological functions of miR-4638-5p in cancer cell growth in relation to androgen independence. We constructed androgen independent PC3 and DU145 PCa cell lines stably overexpressing or underexpressing miR-4638-5p (with pCDH-miR-4638-5p sponge) ( Figure 1D). As expected, miR-4638-5p expression were increased by approximately 25-and 18-fold respectively in PC3 and DU145 cells transfected with pCDH-miR-4638-5p or reduced by 5-and 3-fold, respectively, in PC3 and DU145 cells transfected with pCDH-miR-4638-5p sponge, compared with each cell line transfected with empty-vector control (all P < 0.05) (Supplementary Figure S1). In addition, miR-4638-5p evidently inhibited the reporter activity of pGL3-miR-4638-5p sensor reporter, further indicating that the miR-4638-5p lentiviral expression constructs in PCa cells were functional (Supplementary Figure S2). Overexpressing miR-4638-5p significantly reduced cell growth rate in comparison with cells transduced with the empty pCDH-vector control (P = 5.56 × 10 -3 and P = 4.42 × 10 -2 for PC3 and DU145 cells, respectively) ( Figure 2A), while the growth rate of PC3 and DU145 cells stably expressing miR-4638-5p sponge was significantly enhanced compared to cells transduced with the pCDHvector control (P = 4.34 × 10 -2 and P = 3.37 × 10 -2 for PC3 and DU145 cells, respectively) ( Figure 2A) as detected by CCK-8 assay. We also performed colony formation assay using those transfected cells and found that miR-4638-5p expressing PC3 and DU145 cells had lower colony formation abilities than cells transduced with the pCDH-vector control (P = 4.75 × 10 -2 and P = 3.27 × 10 -2 for PC3 and DU145 cells, respectively) and these cells stably expressing miR-4638-5p sponge showed opposite findings (P = 2.18 × 10 -2 and P = 4.62 × 10 -2 for PC3 and DU145 cells, respectively) ( Figures 2B  and 2C). To further explore whether the decline in cell growth rate of the above miR-4638-5p overexpressing CRPC cells is caused by the induction of apoptosis in addition to the reduced proliferative capacity, we analyzed the proportion of apoptotic cell by flow cytometry. The results demonstrated that the percentage of total apoptotic cells were significantly increased after miR-4638 mimic transfection in comparison to the scramble RNA negative control (P = 2.93 × 10 -2 and P = 1.14 × 10 -2 for PC3 and DU145 cells, respectively). In addition, PC3 and DU145 cells transfected with miR-4638 inhibitor significantly reduced cell apoptosis compared with the cells transfected with the scramble RNA negative control (P = 2.68 × 10 -3 and P = 1.27 × 10 -2 , respectively) and miR-4638 mimic ( Figure 2D and 2E). Subsequently, we investigated the effect on cell proliferation by reducing miR-4638-5p expression level in androgen sensitive LNCaP cells. LNCaP cells cultured in non/very low androgen medium with charcoal-stripped serum have very limited proliferation capacity. However, when miR-4638-5p was inhibited by stably expressing miR-4638-5p sponge, LNCaP cells continuously proliferate and at day 5 there were significantly (P = 1.24 × 10 -2 ) more cells than LNCaP transfected with the pCDH-vector control in androgenstripped cell culture medium ( Figure 2F), indicating that miR-4638-5p inhibits certain cellular pathways which are required for PCa cell proliferation when androgen is not available.
We then investigated the impact of miR-4638-5p expression change on androgen independent tumor growth in vivo. We injected the PC3 cells overexpressing miR-4638-5p into nude mice and observed a delay in tumor onset and a marked decrease in tumor growth when compared with the pCDH-vector control transfected cells that both the tumor volume and weight were significantly different at day 28 (P = 1.48 × 10 -3 and P = 1.60 × 10 -3 , respectively). The opposite result was observed in cells transfected with pCDH-miR-4638-5p sponge compared to the pCDH-vector control (P = 4.36 × 10 -2 and  Cell proliferation results of PC3 and DU145 cells transduced with the pCDH-miR-4638-5p (miR-4638) or the pCDH-miR-4638-5p sponge (miR-4638 sponge, to inhibit miR-4638-5p), determined by a Cell Counting Kit-8 assay (CCK-8). Data represent mean ± SD determined from three independent experiments (n = 3), each with four technical replicates. Representative images from CCK-8 assay, showing the reduced or increased proliferative capacity of both PC3 and DU145 cells after stably overexpressing miR-4638-5p or miR-4638-5p sponge, respectively, in comparison with cells transduced with the pCDH-vector controls (vector). (B) Colony formation ability of PC3 and DU145 cells transduced with the pCDH-miR-4638-5p (miR-4638) or the pCDH-miR-4638-5p (miR-4638 sponge). Bars represent the mean ± SD from three independent experiments (n = 3), each with three technical replicates. (C) Representative images showing the increased or decreased colony formation ability of both PC3 and DU145 cells after stably overexpressing miR-4638-5p or miR-4638-5p sponge, respectively, in comparison with cells transduced with the pCDH-vector controls (vector) after 2 weeks of seeding (original magnification, ×100). (D) Cell apoptosis assay was performed by flow cytometry analysis. The forced expression of miR-4638-5p (miR-4638 mimic) significantly increased the sensitivity of PC3 and DU145 cells to apoptosis, compared with the scramble RNA negative control (NC). The percentage of total apoptotic cells decreased in response to miR-4638-5p inhibitor (miR-4638 inhibitor) transfection. Bars represent the mean ± SD from three independent experiments (n = 3), each with three technical replicates. (E) Representative FACS analysis images. (F) LNCaP cells stably expressing miR-4638-5p sponge proliferate significantly faster than the pCDH-vector control cells in androgen stripped cell culture medium. Bars represent the mean ± SD from three independent experiments (n = 3), each with four technical replicates. * indicates P < 0.05 and ** indicates P < 0.01 by Student's t-test.

MiR-4638-5p directly targets Kidins220
Since miRNAs usually inhibit the translation of proteins via non-perfect pairing of six to eight nucleotides in length [35] or induce the degradation of target mRNAs in the case of perfect complementarity with the potential binding sites of target genes [36], bioinformatics analysis with several programs including TargetScan, RNAhybrid, Findtar, and Pita, was performed to predict the putative miR-4638-5p targets. Based on non-perfect complementarity with the seed sequences of miR-4638-5p, some putative binding sites were predicted in the 3′UTRs of six potential target genes, USP49, CSAD, SEC61A2, TATDN3, C10orf12 and Kidins220. Subsequent luciferase reporter assay using 3′-UTR region of the six candidate genes confirmed that the promoter activity of four out of the six genes were significantly inhibited by miR-4638-5p (P < 0.05 for all) ( Figure 4A) in HEK293T cells. However, only Kidins220 protein expression was significantly inhibited by miR-4638-5p in PC3 and DU145 cells as determined by Western blotting analysis ( Figure 4B), while the expression changes of the other three proteins were not significant. As expected, knockdown of miR-4638-5p increased the expression of Kidins220 protein in those cells ( Figure 4B). Moreover, luciferase reporter assay showed that miR-4638-5p inhibited the activity of Kidins220 3′-UTR WT in a dose-dependent fashion in HEK293T cells ( Figure 4C). As expected, overexpression of miR-4638-5p using 50 nM and 100 nM miR-4638 mimics also suppressed Kidins220 protein expression in PC3 and DU145 cells in a dose-dependent manner ( Figure 4D), further supporting that Kidins220 is the gene regulated by miR-4638-5p. To confirm the specificity of miR-4638-5p in targeting Kidins220, we performed 3′-UTR mutagenesis analysis, where we mutated the putative miR-4638-5p binding site in the Kidins220 3′-UTR WT ( Figure 4E) and found that the mutant mimic lacking the seed sequences reduced the inhibitory effect of miR-4638-5p on luciferase expression (P < 0.05; Figure 4F). In addition, the mutant miR-4638 mimic designed to match the Kidins220 3′-UTR mut ( Figure 4E) exhibited a strong inhibitory effect on the mutant reporter (P = 1.28 × 10 -2 ; Figure 4F). These results indicated that miR-4638-5p directly targets Kidins220 with specific binding sites at the seed sequences of 3′-UTR.

MiR-4638-5p may suppress CRPC cell growth by inhibiting Kidins220 expression
To explore whether miR-4638-5p may suppress CRPC development by inhibiting Kidins220 expression, we firstly examined the expression of Kidins220 in the 30 ADPC and 18 CRPC samples as well as PCa cell lines using qRT-PCR, we found that Kidins220 expression was significantly much higher in CRPC tissues compared with ADPC tissues (P = 3.10 × 10 -7 ; Figure 4G), which was reversely correlated with miR-4638-5p in those two groups of clinical samples. In addition, the mRNA of Kidins220 was also detected at higher levels in the androgen independent LNCaP C4-2B, PC3 and DU145 cells than in the androgen sensitive LNCaP cells (P < 0.05; Figure 4H). To further determine the cellular effect of Kidins220 in CRPC cells, we knocked down Kidins220 in PC3 and DU145 cells. We tested three Kidins220 shRNAs, which all showed knockdown effect at the protein levels, and selected shRNA (S-2), which achieved the best knockdown effect ( Figure 5A) to establish stable Kidins220 knockdown cells ( Figure 5B). As expected, compared with the pCDH-vector control cells, Kidins220knockdown PC3 and DU145 cells showed a reduction in cell proliferation and the increase of apoptotic cells in vitro (P < 0.05; Figures 5C-5E) and reduced tumor growth in xenographed nude mice (P < 0.05; Figure 5F-5H), mimicking the effects of miR-4638-5p overexpression.

Kidins220, mediating the effect of miR-4638-5p downregulation, promotes PCa neo-angiogenesis
As Kidins220 have been reported to activate VEGFR [37], which play a key role in cancer neoangiogenesis, we also investigated the role of Kidins220 in PCa angiogenesis. Microtubule formation assay was performed with HUVECs and conditioned medium collected from culture supernatant fluid of PC3 cells 48 hours after Kidins220 knockdown by shRNA. We found that knockdown of Kidins220 in PC3 cells decreased microtubule formation ability of HUVECs in vitro (P = 1.14 × 10 -2 ; Figure 6A). Vasculogenic mimicry analysis was also performed with PC3 cells with and without Kidins220 knockdown and we found that knockdown of Kidins220 reduced 62% mimicry formation of PC3 cells (P = 2.83 × 10 -3 ; Figure 6B). We also determined whether knockdown of Kidins220 in the androgen-independent PC3 cells would inhibit angiogenesis in vivo by measuring the final hemoglobin content using Drabkin's reagent kit in the xenografts and found that hemoglobin content in the tumor xenografts of cells with Kidins220 knockdown was significantly (P = 1.96 × 10 -2 ) decreased compared with control cells transfected with non-targetting shRNA ( Figure 6C). As expected, we detected significantly decreased VEGF expression in xenografts of Kidins220 knockdown PC3 cells by immunostaining (P = 6.06 × 10 -3 ; Figure 6D).

PI3K/AKT pathway may mediate KIDINS220induced androgen independent PCa growth and angiogenesis
To further investigate cellular signaling pathway regulated by miR-4638-5p and shKidins220, we performed Western blotting for candidate downstream genes responsible for CRPC and angiogenesis, in the PC3 and DU145 cells transfected with the pCDH-miR-4638-5p, pCDH-miR-4638-5p sponge or empty pCDHvector control. We found that stable transduction with the pCDH-miR-4638-5p or pCDH-miR-4638-5p sponge resulted in the reduction or increase, respectively, of VEGF as well as phosphorylated forms of PI3K and AKT (p-PI3K and p-AKT, respectively) ( Figure 8A). Consistent with the results of miR-4638-5p manipulating, Kidins220 knockdown by lentivirus mediated shKidins220 transfection similarly inhibited the VEGF, p-PI3K and p-AKT protein expression ( Figure 8B). To further investigate whether PI3K/AKT pathway mediates Kidins220-induced tumor growth and angiogenesis, we knocked down AKT with shRNA ( Figure 8C and 8D). As expected, compared with the pCDH-vector transfected control cells, AKT-knockdown PC3 and DU145 cells  showed reduction in cancer cell growth (P < 0.05 for all; Figure 8E and Figure 9A-9B) and angiogenesis (P < 0.01 for all; Figure 8F and Figure 9C-9D) in vitro and in vivo, similar to the phenotype observed upon Kidins220-knockdown. Meanwhile, vasculogenic mimicry formation ability of PC3 cells transduced with shAKT was significantly decreased (P = 4.22 × 10 -3 ), compared to the cells transduced with the pCDH-vector control ( Figure 8G). Taken together, these results suggest that PI3K/AKT pathway may mediate Kidins220-induced angiogenesis and vasculogenic mimicry in addition to PCa cell proliferation and castration resistance.

DISCUSSION
CRPC is the main cause of mortality from PCa. Our current knowledge on CRPC development is mainly focused on androgen metabolism and androgen receptor pathways [38]. Identification of other molecular changes associated with CRPC will greatly help us to control this common and lethal form of male malignancy. Alterations of many miRNAs have been reported in PCa development, progression as well as response to therapies [39]. The role of miRNAs in androgen signaling and CRPC development has also been investigated in cell lines and animal models [40]. Differential expression of miRNAs between CRPC and ADPC samples have also been reported [41][42][43][44]. Here, using miRNA microarray analysis, we identified a number of differentially expressed miRNAs between Chinese ADPCs and CRPCs. While many of them have been previously studied, miR-4638-5p has not been investigated previously in any human cancers. The investigation of the down-regulation of miR-4638-5p in PCa was also supported by its down-regulation in advanced disease [33] and PCa stem cells from separate miRNA microarray data. MiR-4638-5p was the mostly underexpressed miRNA in isolated stem cells compared to non-stem DU145 cells. PCa stem cell are closely associated with castration resistance, that PCa stem cell markers have been associated with the development of CRPC, such as CD166 [45], Nanog [46,47], Bmi-1 [48] and Sox2 [49,50]. We also further confirmed the downregulation of miR-4638-5p in CRPC samples using a larger series of clinical samples and demonstrated both through in vitro and in vivo studies that miR-4638-5p play a significant role in suppressing androgen independent PCa cell growth.
The function of miR-4638-5p is currently unknown; the only published report on miR-4638-5p was from a bioinformatics analysis of potential intronic and 3′UTR targeted miRNAs from known cardiac marker genes, where miR-4638-5p has been supposed to regulate cardiac marker gene MYH-7, which encodes beta-myosin heavy chain and is expressed in cardiac and skeletal muscle tissues [51]. Here we demonstrated that miR-4638-5p is a putative tumor suppressor, which suppresses PCa growth, angiogenesis and castration resistance development through regulating Kidins220, VEGFR and PI3K/AKT pathway. As miRNA has been developed for therapeutic use [52], miR-4638-5p may represent a novel target for CRPC treatment. The miR-4638-5p level change, as a driver, may occurs before the clinical features of CRPC, thus it has the potential to be developed into a biomarker for the onset of CRPC before PSA increase or other clinical progressions.
In the search for genes/proteins regulated by miR-4638-5p, we identified Kidins220 as a gene whose expression was downregulated post-transcriptionally by miR-4638-5p through targeting the Kidins220 3′-UTR seed sequences. In consistent to miR-4638-5p underexpression, we demonstrated that CRPC samples overexpressed Kidins220 compared to ADPC samples and knockdown of Kidins220 generated similar effects as miR-4638-5p overexpression in CRPC cells.
Kidins220 (kinase D-interacting substrate of 220 kDa), also known as ankyrin repeat-rich membrane spanning protein (ARMS), is a recently identified tetraspanning-transmembrane protein abundantly expressed in the developing and adult neural tissues [53,54]. Previous studies of Kidins220 were mainly focused on its roles in neural development and Kidins220, as a downstream substrate of protein kinase, neurotrophin and ephrin, regulates neuronal differentiation and survival [55][56][57][58]. Kidins220 also expresses outside the nervous system. It interacts with the microtubule and actin cytoskeleton, modulating cell plasticity and migration and is important for cardiovascular development and certain immune function [58][59][60]. Although most of the above cellular features are associated with cancer development and progression, research of Kidins220 in human malignancies is very limited. Kidins220 overexpression has been reported in melanoma and it has been shown to contribute to tumor formation by activating MEK/ERK signaling pathway and consequently preventing transformed melanocytes from the stressinduced apoptosis [61]. Expression of Kidins220 has also been reported in neuroblastoma tumours, where it stabilizes nerve growth factor-induced survival signalling through MAPK/ERK signalling [62]. Both of these studies demonstrated that Kidins220 protects cells for survival through the MAPK/ERK signalling pathway. However, the role of Kidins220 in PCa has not been reported. In this study we demonstrated that Kidins220 promoted CRPC cell growth and tumor angiogenesis and enhanced VEGFR, PI3K and AKT activity in CRPC cancer cells. These findings suggest that, in addition to the above reported roles, Kidins220 impacts on PI3K/AKT and VEGF/VEGFR pathways to promote PCa angiogenesis and the development of castration The mimicry formation ability of PC3 cells transduced with the pCDH-shAKT (shAKT) was significantly decreased, compared to the cells transduced with the pCDH-vector controls (vector) with representative images of mimicry captured at 6 hour post seeding (original magnification, ×100) on the top. The data represent the mean ± SD from three independent experiments (n = 3), each experiment containing four technical replicates. * indicate P < 0.05 and ** indicate P < 0.01 by Student's t-test, respectively. www.impactjournals.com/oncotarget resistance. MiR-4638-5p, to our knowledge, is also the first identified miRNA to regulate Kidins220. Overall, these novel findings increase our understanding of the cellular and biological functions of miR-4638-5p and Kidins220 as well as reveal new mechanisms underlying the development of CRPC.
The PI3K/AKT signaling is a critical pathway in cell proliferation, survival, neovascularization and tumor growth [63,64]. Previous studies have identified that PI3K/AKT signaling plays a key role in the development and maintenance of CRPC [65,66]. Deregulation of PI3K/ AKT signaling occurs in almost all cases of advanced CRPC [67]. Recent studies have revealed a direct link between PI3K/AKT and AR signaling pathways that they interplay during the development of castration resistance [68,69]. In addition, the activation of PI3K/AKT signaling pathway promotes castration resistance and the inhibition of AR function accelerates the development of CRPC in PTEN deficient mice [70]. As mentioned above, the known function of Kidins220 in cancer is to protect cells for survival when they are under stress [61,62]. In PCa, Kidins220 may activate PI3K/AKT signaling to sustain cancer cell growth under the stress of androgen deprivation and lead to CRPC. Inhibiting a single target in the PI3K/AKT signaling pathway have shown limited clinical efficacy and simultaneously inhibition of multiple targets in series or parallel have demonstrated greater promise in the treatment of CRPC [65]. As Kidins220 regulates both PI3K and AKT activities, targeting miRNA-4638-5p and/or Kidins220 offers another promising therapeutic approach for the management of CRPC.
In this study, we provided evidence that Kidins220 modulated angiogenesis of PCa cells via regulation of both the VEGF/VEGFR2 and PI3K/AKT signaling pathway. It has been reported that Kidins220 interacts with VEGFR2 and VEGFR3 in mouse endothelial cells and modulates VEGF signaling through promoting the interaction between VEGF and VEGFR for cardiovascular development [37]. VEGF is the main angiogenic growth factor [71] and PI3K/AKT signaling interplay with VEGF level to regulate angiogenesis [64], which is critical for tumor development when cancer reaches to a certain size. Numerous studies have demonstrated that Hif-1α activation could increase the expression and secretion of VEGF via PI3K/AKT/mTOR signaling pathway [72][73][74]. Meanwhile, VEGF itself can effectively activate PI3K/ AKT signaling pathway, stimulating blood vessel in autocrine and paracrine manner. Therefore, the observation that Kidins220 regulated angiogenesis in human cancers via activation of VEGF and PI3K/AKT signaling is not surprising, but this has not been investigated before. In this study, we also showed for the first time that Kidins220 through VEGF and PI3K/AKT signaling regulates both neo-vascularization from endothelial cells and the formation of vasculogenic mimicry of CRPC cells for tumor angiogenesis. Vasculogenic mimicry (VM) phenomenon provides a new pathway for tumor blood supply independent of endothelial cells [75]. The tumor cells possessing mimic vessel display highly malignant characteristics of invasive growth, dedifferentiation and metastasis [76]. The ability of vasculogenic mimicry formation of cancer cells may be an additional mechanism for them to access sufficient nutrition and growth factors for cell survival and proliferation under adverse stressful environment, such as androgen deprivation in the case of CRPC. Based on our results, a model of miR-4638-5p and Kidins220 functions for cell survival and angiogenesis has been proposed ( Figure 10).
While angiogenesis is critical for cancer progression and anti-angiogenesis therapy has been extensively tested with certain success in treating CRPC [77], the role of angiogenesis in CRPC development has not been reported. Therefore, the role of accelerated angiogenesis by miRNA-4638-5p down-regulation and the consequently deregulation of the above molecular pathways in castration development is unclear. Angiogenesis may provide oxygen, nutrition and other growth factors to help the survival of cancer cells under stress of androgen depletion during CRPC development. Neuroendocrine PCa is AR negative and with poor prognosis, therefore, the development of neuroendocrine PCa has been considered as a mechanism of CRPC [78]. Kidins220 is critical for neuronal differentiation and survival [37,55,[57][58][59]. As we have demonstrated in vitro and in vivo that Kidins220 promote tumor angiogenesis through activation of PI3K/AKT and VEGF signalling, Kiddins220 may also involved in both neuroendocrine differentiation and angiogenesis of PCa cells. Both neuroendocrine differentiation and vasculogenic mimicry formation are cellular changes of losing epithelial features of PCa cells. Kiddins220 may coordinate neuroendocrine differentiation and angiogenesis, together with regulating PI3K/AKT signaling, promoting the development of castration resistance. Further investigations are required.
In summary, we identified a novel tumor suppressor miRNA, miR-4638-5p, which is downregulated in CRPC, and dissected its target genes and downstream cell signaling pathways. We revealed for the first time that miR-4638-5p regulates Kidins220 mRNA by targeted degradation and consequently suppression of VEGF/ VEGFR and PI3K/AKT in CRPC cells, and impact on tumor growth and angiogenesis using in vitro and in vivo models. While the role of PI3K/AKT pathway in CRPC initiation is well established, the role of VEGF signaling and angiogenesis in CRPC development warrants further investigations. These findings provide new molecular insights into the development of CRPC and can guide the development of novel approaches targeting the miR-4638-5p/Kidins220/VEGF/VEGFR/PI3K/AKT axis to prevent or treat CRPC. regulated by miR-4638-5p, binds to the VEGFR2 and VEGFR3 and promotes interaction between VEGF and VEGFR as a co-receptor (auxiliary receptors) of VEGFR, so that VEGF/PI3K/AKT signaling pathway is activated to promote proliferation and mimicry formation and suppress apoptosis of PCa cells. AKT activation also consequently increases the expression level and secretion of VEGF, which stimulates angiogenesis of endothelial cells. Therefore, miR-4638-5p inhibits these cellular process by downregulating its target gene Kidins220. www.impactjournals.com/oncotarget

Human prostate tissue specimens
Fresh frozen tissue from 30 ADPC and 18 CRPC patients were obtained from Wuxi No. 2 People's Hospital affiliated with Nanjing Medical University. The 18 CRPC patients were selected based on their increased serum prostate-specific antigen (PSA) level during androgendeprivation therapy and the cancer tissues were obtained through transurethral prostatic resection to relieve the symptom of urinary obstruction. The 30 ADPC samples were derived from radical prostatectomy in treatmentnaive patients. Clinical stage of disease was determined by histopathology, magnetic resonance image and radio-nucleotide bone scans according to the 2002 TNM classification of prostate carcinoma. The clinicpathological features of all study patients are summarized in Supplementary Table S1. All these samples contained more than 60% tumor by pathological examination. The use of these human PCa tissues for this study was approved by the institutional ethics review board of Nanjing Medical University.

Cell lines
Human umbilical vein endothelial cells (HUVECs) were maintained in EBM-2 culture media (LONZA, Allendale, NJ, USA) as previously described [79]. DU145 human androgen independent PCa cell line and HEK293T were maintained in DMEM, and androgen dependent LNCaP, androgen independent LNCaP C4-2B subline and PC3 human PCa cell lines was cultured in RPMI 1640, all supplemented with 10% fetal bovine serum (FBS), 100 units/ml of penicillin and 100 µg/ml of streptomycin.

MiRNA microarray analysis
Total RNA was isolated using TRizol (Invitrogen) and purified with RNeasy mini kit (QIAGEN) according to manufacturer's instructions. RNA quality and quantity was measured using NanoDrop spectrophotometer (ND-1000, NanoDrop Technologies) and RNA Integrity was determined by gel electrophoresis. 2.5 µg total RNA samples were labeled using the miRCURYTM Array Power Labeling kit (Exiqon, Vedbaek, Denmark) following the manufacturer's instruction. Microarray images were acquired using a Genepix 4000B scanner (Axon Instruments, Union City, CA, USA), processed and analyzed with Genepix Pro 6.0 software (Axon Instruments).
To stably knockdown Kidins220, the following three shRNAs for Kidins220 were designed using bioinformatics software including siDirect, Integrated DNA Technologies, Learning with Clontech and SDS (siRNA Design Software).

Quantitative real-time PCR analysis of miRNA and mRNAs
Total RNA was extracted from PCa tissues and cultured cells using the TRizol method (Invitrogen) and subjected to the TaqMan® MicroRNA Reverse Transcription Kit (Applied Biosystems, Carlsbad, CA, USA) or the Promega Reverse Transcription Kit (Promega, Madison, WI) to obtain cDNA. TaqMan MicroRNA (Assay ID: 462195_mat for has-miR-4638-5p) and gene expression (Assay ID: Hs01057000_m1 for Kidins220) assays (Applied Biosystems) were performed in accordance with the manufacturer's instructions using a 7500 Real-time PCR System (Applied Biosystems). The expression of U6 snRNA (Assay ID: 001973) was used as a control for miRNA expression and GAPDH (Assay ID: Hs02758991_g1) was used as a control for gene expression. The relative expressions were calculated using the 2 -∆∆Ct method.

Cell proliferation assay
Cell Count Kit-8 (CCK-8) (Dojindo Molecular Technologies, Tokyo, Japan) was used to evaluate cell proliferation. PC3 and DU145 cells stably transfected with pCDH-miR-4638-5p, pCDH-miR-4638-5p sponge, pCDH-shKidins220, pCDH-shAKT and pCDH-vector controls were seeded into 96-well plates at density of 2 × 10 3 cells per well and cultured for 1 day, 2 days, 3 days, 4 days, 5 days and 6 days. Each well of cells were then cultured in 20 µl of CCK-8 solution in addition to the 200 µl of culture medium for 1.5 hours at 37°C. Absorbance at 450 nm was measured using a V max microplate spectrophotometer (Molecular Devices). LNCaP cells transfected with pCDH-miR-4638-5p sponge and pCDH-vector control were seeded into 96-well plates (also 2 × 10 3 cells per well) in the RPMI 1640 medium supplemented with 10% charcoal-stripped FBS and cultured for 1 day, 2 days, 3 days, 4 days and 5 days. Then the cell proliferation ability was evaluated using CCK-8 solution as described above.

Colony formation assay
Cells were seeded into 12-well plates at a low density of 100 cells per plate and cultured for 14 days. The cells were then fixed with 95% methanol and pigmented with 0.1% crystal violet. The number of colonies with >10 cells observed under low magnification microscope were counted.

Flow cytometry analysis of cell apoptosis
For the cell apoptosis analysis, PC3 and DU145 cells, 48 hour after transfection with miR-4638 mimic, miR-4638 inhibitor, siKidins220 and the scramble RNA negative controls (RIBOBIO), were collected and resuspended to 1 × 10 5 cells in 195 µL of buffer. Then 5 µl Annexin V-FITC and 10 µl PI staining (Beyotime, Shanghai, China) were added into the cell mix and incubated for 15 minutes in the dark at room temperature. The apoptosis rates were determined using the flow cytometry-BD FACS Calibur (BD Biosciences) and Flowjo software one hour after staining.

Matrigel tube formation assay
Microtubule formation assay was performed as previously described [81]. Briefly, 96-well plates were coated with 60 µl Matrigel solution (BD Bioscience, New Bedford, MA, USA) diluted at 1:1 in cell culture medium at 4°C. The plate was allowed to solidify for one hour at 37°C before cell seeding. The conditioned medium was collected from supernatant fluid of 48-hour cultured PCa cells with different transfections and filtered using 0.45 µm filter. HUVECs in 150 µl conditioned medium diluted at 2:1 in EBM-2 culture media without FBS, were seeded at density of 1 × 10 4 cells per well. After incubation at 37°C for 6 hours, four randomly selected fields of each well were photographed. The angiogenesis index was calculated using the formula as previously described [81].

Vasculogenic mimicry analysis
96-well plates were coated with 60 µl Matrigel solution diluted as a 1:1 mixture of High Concentration Matrigel (BD Bioscience) and culture medium without FBS, penicillin and streptomycin solution at 4°C. The plate was allowed to polymerize for one hour at 37°C. PC3 cells, stably transfected with pCDH-miR-4638-5p, pCDH-miR-4638-5p sponge, pCDH-shKidins220, pCDH-shAKT and pCDH-vector controls in RPMI 1640 medium without FBS, were seeded at density of 1.5 × 10 4 per well. After incubation at 37°C for 6 hours, three randomly selected fields of each well were photographed. The vasculogenic mimicry index was calculated using the same method as microtubule formation assay described above.

Tumorigenicity analysis and Matrigel plug assay for angiogenesis analysis in nude mice
BALB/C nu/nu male athymic mice (3-4 weeks old) were purchased from Model Animal Research Center of Nanjing University (Nanjing, China). All animal experiments were evaluated and approved by the institutional ethics review board of National Institute of Health Guide for the Care and Use of Laboratory Animals and Wuxi No. 2 People's Hospital. The treated cells (PC3 cells infected with lentivirus) were harvested at a subconfluent density, washed with phosphate-buffered saline (PBS) and re-suspended in serum-free medium. 2 × 10 7 cells in 500 µl of serum-free medium were mixed with 500 µl of High Concentration Matrigel (BD Biosciences) and 100 µl mixture (2 × 10 6 cells) was immediately injected subcutaneously into the left or right flanks of nude mice. The tumor size was measured every 4 days with a two-dimensional caliper and tumor volume was calculated using the formula: 0.52 × length × width 2 . At day 28 after injection, mice were sacrificed and the xenograft tumors were surgically removed. Tumor samples were collected to make fresh frozen as well as formalinfixed and paraffin-embedded sections for gene expression analysis. The hemoglobin content of the Matrigel plug was determined using Drabkin's reagent kit according to the manufacturer's instruction (Sigma-Aldrich) for tumor angiogenesis analysis. The final hemoglobin level was determined by spectrophotometric analysis at 540 nm.

Western blot analysis
Proteins were separated by 8% and 10% SDS polyacrylamide gels and electrophoretically transferred onto polyvinylidene fluoride membranes (Millipore, Billerica, MA, USA). The membranes were blocked overnight with 5% non-fat dried milk and incubated overnight with antibodies. Anti-Kidins220 rabbit polyclonal antibody (PAb) (Cat No. ab97345) was purchased from Abcam Inc (Abcam, Cambridge, UK).

Statistical analysis
All experiments were performed at least in triplicates unless otherwise mentioned. Numerical data were expressed as mean ± SD. Two group comparisons were analyzed by two-sided Student's t-test. P < 0.05 was considered significant.