EZH2-mediated loss of miR-622 determines CXCR4 activation in hepatocellular carcinoma

The CXC chemokine receptor 4 (CXCR4) exerts a variety of functions at different steps of hepatocellular carcinoma (HCC) progression. The molecular mechanisms and therapeutic value of CXCR4 in the development of HCC remain undefined. Here we show that aberrant CXCR4 overexpression is associated with poor prognosis and aggressive characteristics of HCC. Suppression of CXCR4 activity via CXCR4 knockdown, AMD3100 or neutralizing antibody administration inhibits hepatoma cell tumorigenesis in vitro and in vivo. CXCR4 overexpression displays the opposite effects. Using Mir library screening we identify miR-622 as a regulator of CXCR4. Further studies show that miR-622 directly target the 3′ untranslated region of CXCR4 and is transcriptionally repressed by EZH2-induced H3K27 trimethylation and promoter methylation. EZH2/miR-622 promotes tumorigenesis through CXCR4. EZH2-mediated loss of miR-622 is found to correlate with CXCR4 overexpression and unfavourable prognosis in HCC patients. This study establishes EZH2/miR-622/CXCR4 as a potential adverse prognostic factor and therapeutic target for HCC patients.

T he CXC chemokine receptor 4 (CXCR4) and its chemokine ligand 12 (CXCL12) have implicated in many key steps of cancer, including angiogenesis, epithelial-mesenchymal transition, invasion, dissemination and cancer cell stemness 1,2 . After CXCL12 binding, CXCR4 activates different pathways, notably calcium release and cellular migration, PI3K/AKT and cellular survival, Ras-MAPK and cell proliferation, b-catenin and cancer cell stemness 3,4 . The CXCR4/CXCL12 has multiple functions at various points in the progression of hepatocellular carcinomas (HCCs). Both autocrine and/or paracrine effects of this pathway have been shown to maintain cancer growth, induce angiogenesis and aid escape of immune surveillance 5 . The molecular mechanisms that explain CXCL12 and CXCR4 expression in HCC remain undefined.
Numerous studies have demonstrated immunohistochemical staining of CXCR4 in HCC tissues but not in normal hepatic tissues 6,7 . CXCR4 mRNA expression are contrasting, Liu et al. 7 found overexpression in HCC tumour tissues, while others report reduced expression in HCC tissues or no differences 6 . Nonetheless, the majority of studies showed correlations between CXCR4 expression and aggressive tumour behaviour and poor clinical outcome 6,8 . Understanding the regulation network of CXCR4 would give us a deeper insight into the mechanisms underlying hepatocarcinogenesis.
The microRNAs (miRNAs) constitute small non-coding RNAs (19)(20)(21)(22) nucleotides) that provoke mRNA degradation or blockade of mRNA translation by interacting with the 3 0 untranslated regions (3 0 -UTRs) of target mRNAs 9 . Importantly, downregulation of some miRNAs can motivate tumorigenesis through the upregulation of oncogenes and silencing of tumour suppressor genes, respectively 10 . miR-622 functions as a tumour suppressor by targeting K-RAS 11 . It is downregulated in human gastric carcinoma tissues, pancreatic adenocarcinoma and ampullary adenocarcinoma 12,13 . In the context of HCC, aberrant expression of specific miRNAs are closely associated with tumour cell proliferation, migration and invasion by targeting proteins involved in these cellular functions [14][15][16] . However, the expression and the role of miR-622 in HCC have not been clearly demonstrated.
In this work, we identify CXCR4 overexpression in a subset of HCCs, which contributing to hepatoma cell proliferation, colony formation, migration and survival. Suppression of CXCR4 activity either by shRNA or pharmacological inhibition suppresses hepatoma cell growth in vitro and in vivo. A comprehensive miRNA analysis reveals that CXCR4 expression regulated by miR-622, which is epigenetically downregulated by enhancer of zeste homologue 2 (EZH2). Moreover, EZH2/miR-622 pathway is significantly associated with CXCR4 expression and poor prognosis of HCC patients. Thus, the alteration in the EZH2/miR-622/CXCR4 pathway contributes to tumour development and represents therapeutic targets.

CXCR4 is upregulated in HCC and correlated with survival.
To investigate the potential significance of CXCR4 in the development and progression of HCC, we first evaluated CXCR4 expression by immunohistochemical analysis in 127 HCC specimens. CXCR4 was located diffusely in the cytoplasm and nucleus of tumour cells (Fig. 1a). However, the level of cytoplasmic and nuclear CXCR4 expressions was significantly higher in tumour than peritumour tissues (Fig. 1b). Immunoblot analysis of lysates obtained from surgical samples of 13 HCC patients confirmed increases of CXCR4 expression in tumour relative to peritumour tissues (Fig. 1c). As shown in Supplementary Table 1, the level of cytoplasmic CXCR4 expression closely correlated with tumour size (P ¼ 0.003, w 2 -test), venous invasion (P ¼ 0.006, w 2 -test), high Barcelona Clinic Liver Cancer stage (P ¼ 0.003, w2-test) and Tumour, Node, Metastasis (TNM) stage (Po0.001, w 2 -test). The level of cytoplasmic CXCR4 expression was correlated with tumour progression from TNM stage I to III (Fig. 1d). Importantly, patients with high cytoplasmic CXCR4 intensity showed significantly worse overall survival (mean of 33.0 versus 61.3 months, log-rank test Po0.001; Fig. 1e) and recurrence-free survival (RFS; mean of 26.9 versus 42.6 months, log-rank test P ¼ 0.001; Fig. 1f) than those with low cytoplasmic CXCR4 expression, while nuclear CXCR4 expression was found not to be significantly related to overall survival and RFS ( Supplementary  Fig. 1). The Cox proportional hazards model revealed that high cytoplasmic CXCR4 expression was an independent prognostic factor with respect to overall survival (hazard ratio ¼ 1.889 (95% confidence interval, 1.040-3.431), P ¼ 0.038) and RFS (hazard ratio ¼ 1.695 (95% confidence interval, 1.008-2.853), P ¼ 0.048) ( Table 1 and Supplementary Table 2). Taken together, it was clearly indicated that cytoplasmic CXCR4 expression was a significant and independent index for HCC outcomes.
CXCR4 is required for tumour growth, migration and survival. To explore the biologic function of CXCR4, we first determined the endogenous CXCR4 expression in hepatoma cell lines, and identified Huh7 and SK-Hep1 cells with the lowest and highest CXCR4 expression for subsequent experiments (Fig. 2a). We confirmed stable knockdown of CXCR4 in SK-Hep1 cells, and stable CXCR4-transfected Huh7 cells ( Fig. 2a and Supplementary  Fig. 2a). While ectopic CXCR4 expression promoted the proliferation of Huh7 cells, stably CXCR4 knockdown, AMD3100 (CXCR4 antagonist) and CXCR4-neutralization antibody inhibited the proliferation of SK-Hep1 cells (Fig. 2b). Furthermore, CXCR4 overexpression enhanced anchorage-independent growth, migration and survival of Huh7 cells, which were impaired by either CXCR4 knockdown, AMD3100 or neutralization antibody in SK-Hep1 cells (Fig. 2c-e).
Suppression of CXCR4 inhibits tumorigenesis in vivo. We next investigated the effect of CXCR4 on tumorigenesis of hepatoma cells in vivo. Quantification of tumour size and weight showed that Huh7 cells with CXCR4 overexpression generated larger tumours than control cells (Fig. 3a). Conversely, SK-Hep1 cells with CXCR4 knockdown generated smaller tumours than control cells (Fig. 3b). To assess the therapeutic potential of targeting CXCR4 in vivo, we tested subcutaneous exnografts of SK-Hep1 cells. Transplant recipient nude mice with palpable tumours were treated with AMD3100 for 3 weeks. AMD3100 treatment potently suppressed tumorigenesis (Fig. 3c, left panel). Intratumoural injection of AMD3100 had extended survival (mean of 52.7 versus 41.2 days, log-rank test P ¼ 0.032; Fig. 3c right panel). To further validate the role of CXCR4 in tumorigenesis, we turned to a CXCR4-neutralization antibody. CXCR4-neutralization antibody treatment suppressed tumour size, and had extended survival (mean of 54.4 versus 42.1 days, log-rank test P ¼ 0.004; Fig. 3d). This was accompanied by commensurate reduction in ki67 staining (Fig. 3e). These data indicate that CXCR4 promoted tumorigenesis in vivo and in vitro, and targeting CXCR4 is a potential candidate for clinical application to the treatment of HCC.
Identification of endogenous miRNA directly target CXCR4. Changes in the expression of miRNAs appear to be a common characteristic of cancers including HCC 17  CXCR4 in HCC. Therefore, we used comprehensive bioinformatics analysis as a filter to generate a selective miRNA library for subsequent screening. A total 64 miRs were successfully identified as candidate miRs ( Fig. 4a and Supplementary Fig. 3). Screening with the candidate miR library was carried out by quantitative PCR for the downregulation of miRs in HCC samples compared with peritumour tissues (Fig. 4b). We found five miRNAs (miR-302c, miR-139-5p, miR-9, miR-206 and miR-622) were downregulated in HCC compared with peritumour tissues (Fig. 4c and Supplementary Table 3). To determine whether CXCR4 expression is selectively regulated by the five aforementioned miRs, we transfected these selected miR-mimics or anti-miRs in hepatoma cells. We found that miR-622 mimic suppressed CXCR4 expression in SK-Hep1 cells, whereas anti-miR-622 increased CXCR4 expression in Huh7 cells, which suggested that miR-622 is a specific regulator of CXCR4 in hepatoma cells (Fig. 4d). The TargetScan algorithm showed that the bases from 71 to 77 in the CXCR4 3 0 -UTR have perfect complementarity to the seed sequence of miR-622. To substantiate the site-specific repression of miR-622 on CXCR4, we constructed a mutated CXCR4 3 0 -UTR luciferase reporter (Fig. 4e), which completely restored luciferase activity induced by miR-622 mimic (Fig. 4f), and suppressed luciferase activity induced by anti-miR-622 (Fig. 4g). These data suggest that CXCR4 is a novel direct target of miR-622 in hepatoma cells.
CXCR4 mediate the effects of miR-622 on tumour promotion. CXCR4 upregulation by anti-miR-622 was prevented using siRNAs before assessment of cell growth and migration. Hepatoma cells were transfected with CXCR4 or control siRNA before transfection with anti-miR-622 followed by assessment of cell growth and migration, respectively. CXCR4 knockdown with siRNA was confirmed by immunoblotting ( Fig. 5a and Supplementary Fig. 4a,b). Inhibition of miR-622 significantly promoted growth and migration of Huh7 and PLC/PRF/5 cells, however, CXCR4 knockdown prevented the increased growth and migration induced by anti-miR-622 expression (Fig. 5b,c and Supplementary Fig. 4d). Similar rescue to the above was obtained in SK-Hep1 and SNU448 cells transfected with miR-622 (Fig. 5b,c and Supplementary Fig. 4c,d). The above data show that inhibitory effects of miR-622 are partially mediated by targeting CXCR4.
Changes in DNA methylation could also be associated with the subsequent acquisition of other histone modifications. We further performed chromatin immunoprecipitation (CHIP) to evaluate repressive histone hallmarks, including trimethylated H3K9 (H3K9me3) and trimethylated H3K27 (H3K27me3). The results showed higher levels of methylation at H3K9 and H3K27 in a broad area upstream of the miR-622 coding region in SK-Hep1 cells ( Supplementary Fig. 5a,b). These data allowed us to hypothesize that histone methylation, especially those of EZH2dependent H3K27me3, may contribute to miR-622 repression. To confirm our hypothesis, we performed EZH2 overexpression and knockdown in hepatoma cells. As expected, overexpression of EZH2 led to an increase in CXCR4 expression and decrease in the levels of miR-622 in Huh7 cells (Fig. 7a). Furthermore, CHIP assays showed that EZH2 occupied in the upstream region of miR-622, which is concomitant with the increase in H3K27me3 and H3K9me3 levels (Fig. 7b). We confirmed an EZH2 knockdown using a specific shRNA ( Supplementary Fig. 2c). Knockdown of EZH2 resulted in a great increase in the levels of miR-622 and decrease in the CXCR4 expression in SK-Hep1 cells (Fig. 7c). CHIP assay indicated that decreased EZH2, H3K27me3 and H3K9me3 occupancy in the upstream regions of miR-622 (Fig. 7d). Similarly, effects were observed during inhibition of EZH2 activity in SK-Hep1 cells by DZNep (Fig. 7e,f). These data indicate a link between epigenetic regulation and miR-622 transcription in hepatoma cell lines.
EZH2 regulates CXCR4 by controlling miR-622 expression. On the basis of our findings, we considered an aspect of the biological communication between epigenetic silencing and  ARTICLE CXCR4 expression through miR-622 regulation. EZH2 knockdown in SK-Hep1 cells resulted in reduction of CXCR4 expression, which is consistent with those of miR-622 overexpression. Then, we tested whether exogenous manipulation of miR-622 could inhibit the effects of EZH2. We restored miR-622 expression in Huh7 cells, but inhibited miR-622 expression in SK-Hep1 cells (Fig. 8a). Indeed, enhancement of proliferation and migration induced by EZH2 overexpression was partially inhibited by treatment with miR-622 mimic in Huh7 cells. On the other hand, repression of proliferation and migration induced by knockdown of EZH2 was partially restored by treatment with miR-622 inhibitor in SK-Hep1 cells (Fig. 8b,c and Supplementary  Fig. 6a,b). Additional expression of CXCR4 rescued the inhibition of proliferation and migration induced by EZH2 knockdown in SK-Hep1 cells (Fig. 8d-f and Supplementary Fig. 6c). These results suggest that EZH2-mediated miR-622 suppression leads to CXCR4 activation.
Clinical correlation of CXCR4 with miR-622 EZH2. Given that miR-622 might regulate CXCR4 expression in HCC, miRNA in situ hybridization and immunohistochemical analysis were done to evaluate the relationship between miR-622 and CXCR4 expression in HCC (n ¼ 127). miR-622 level was inverse correlated with CXCR4 expression in HCC tissues (Pearson's coefficient test r ¼ À 0.391 (Po0.001), Fig. 9a,b). At the    Fig. 9a,b). Considering the inverse correlation between CXCR4 and miR-622 in the HCC tumours, we further evaluated their combined influence on patient outcome. Patients were classified into four groups, using their median as the cutoff. I, CXCR4 low and miR-622 low (n ¼ 19); II, CXCR4 low and miR-622 high (n ¼ 45); III, CXCR4 high and miR-622 low (n ¼ 44); and IV, CXCR4 high and miR-622 high (n ¼ 19). Patients in group III showed significantly worse overall survival and RFS than those in groups I and II (both log-rank test Po0.050, Fig. 9c). These data confirmed that EZH2/miR-622 pathway correlated with CXCR4 expression and was clinical relevant in HCC (Fig. 9d).

Discussion
Inflammation drives different mechanisms involved in tumorigenesis and progression, including proliferation of tumour cells, angiogenesis and metastasis 18 . These mechanisms are, in part, driven by secreted molecules such as CXCL12, which plays multiple roles in tumour pathogenesis 19 . Although they were first described to be produced by bone marrow stromal cells, they are also secreted by tumour cells of different origin, including hepatocellular carcinoma cells 19 . The CXCR4/CXCL12 has multiple roles in the pathogenesis of HCC, and can modulate cell growth, migration and survival via both autocrine and/or paracrine mechanisms 5 . A number of studies have demonstrated correlations between high CXCR4 expression and aggressive tumour behaviour and poor prognosis 6,8,20 . Therapeutic intervention with CXCR4 signal activation could be used as a promising strategy against hepatocellular carcinoma after curative resection. Administration of CXCR4 antagonist has been found to inhibit tumour growth and metastasis 4,[21][22][23] . Discrepancies between our results and the report by Duda et al., which suggest that AMD3100 was ineffective in HCC tumorigenesis, probably rely on that the CXCR4 antagonist intervention on endogenous CXCR4 high-expression HCC cell line, SK-Hep1, in tumorigenesis in vivo in the present study 24   currently being tested in early-phase clinical trials 4,25 . Our studies suggest that HCC patients should be included in these trials. In HCC patients, CXCR4 was detected in HCC tissues, but not in normal hepatic tissues. CXCR4 expression significantly correlated with progressed local tumours (T-status), lymphatic metastasis and distant dissemination, as well as with a decreased survival 6,7,26,27 . Nonetheless, conflicting data were reported that the CXCL12-CXCR4 are detected in sinusoidal endothelial cells in HCC tissue, and their expression are significantly higher than in non-HCC tissues 20 . Furthermore, Zhou reported that CXCR4 nuclear localization can be used to identify patients with HCC at high risk for developing lymph node metastasis 8 . The discrepancy may lie in the heterogeneity of HCC and different detection methods applied.
The mechanism underlying CXCR4 overexpression in HCC is unclear at present. A number of studies have demonstrated upregulation of CXCR4 in HCC tissues, while CXCR4 mRNA expression reduced or remain no differences 6,28 . This means that CXCR4 may be regulated by post-transcriptional level in HCC. We provide definitive evidence for the notion qthat miR-622 negatively regulates CXCR4 expression. Here miR-622 is verified to be frequently decreased in HCC tissues, and inversely correlated with the survival of HCC patients, outlining a potential marker for predicting the prognosis of HCC patients. Furthermore, the therapeutic role of miR-622 in HCC remains to be elucidated. In the meanwhile, we cannot exclude other potential miRNA participating in CXCR4 regulation.

ARTICLE
The miRNA regulation involves multiple steps, including miRNA maturation, genetic deletion and epigenetic deregulation 29,30 . In particular, polycomb group proteins have central functions in cellular development and regeneration by controlling histone methylation, especially at histone H3 Lys27 (H3K27), which induces chromatin compaction 31 . Alterations of PcG genes directly modulate the trimethylation of H3K27 and thus affect the epigenome of HCC, which is crucial for controlling the HCC cell phenotype 32 . Oncogene polycomb group protein EZH2 in HCC is highly correlated with tumour progression 33 . We show that EZH2 epigenetically represses miR-622 expression by facilitating H3K27me3 trimethylation. Consistently, miR-101 was epigenetically silenced by EZH2 overexpression in HCC 34 .
Recent studies have suggested unique expression profiles of miRNAs in HCC 17,35 , but loss of miR-622 has not been focused.
Besides the variations of technological methodologies and sample origin, one main reason for this inconsistency is possibly that the liver is composed by a heterogeneous population of parenchymal cells, kupffer cells, stellate cells, bile duct cells, fibroblasts and inflammatory cells 36 . HCC samples of patients with different aetiologies usually with different cell activities and proportions may result in artifact of miRNA profiles. In the present study, we also discovered many dysregulated miRNAs in HCC that were consistently reported before (such as miR-139-5p and miR-1) (refs 15,16).
In conclusion, the coordinated expression of EZH2/miR-622/ CXCR4 may be predictive of worse prognostic in patients with HCC. Our findings also highlight the therapeutic potential of CXCR4 in HCC treatment, and support the development of effective therapeutic strategies that target CXCR4 by a pharmacological approach. Therefore, CXCR4 could be a therapeutic target and a valuable prognostic marker for HCC.  Plasmid construction and transfection. Expression plasmid encoding wild-type CXCR4 was kindly provided by Dr Ann Richmond (Vanderbilt University, Nashiville, TN). Expression plasmid encoding wild-type EZH2 was generated as previous described 39 Supplementary Fig. 7.
PCR products were cloned using the pGEM-T Easy Vector system (Promega, Madison, WI). Four individual clones were sequenced. The region assessed by BSP included 14 CpG sites from the miR-622 promoter and average methylation from individual clones was calculated as a percentage of the number of methylated CpG sites over the number of total CpG sites sequenced.
In situ hybridization. Paraffin-embedded sections were deparaffinized and rehydrated by an ethanol series. Slides were quenched endogenous peroxidase activity with 3% H 2 O 2 for 30 min. Following digestion by proteinase K for 5 min, slides were fixed in 4% paraformaldehyde and rinsed in PBS. Slides were incubated in hybridization buffer at 60°C for 2 h, and incubated with miR-622 or scrambled miRNA control probes (50 nM; digoxigenin-labelled LNA probes, Exiqon, Vedbaek, Denmark) at 60°C overnight. Stringent wash buffer (50% formamide in 2 Â SSC and PBS plus Tween-20 (PBST)) was used. Alkaline phosphatease substrate (nitroblue tetrazolium/5-bromo-4-chloro-3-indolyl-phosphate was used for the alkaline phosphate reaction, and the slides were mounted in aqueous mounting medium. Scoring was measured according to the following criterion: 0 ¼ absent cell cytoplasm staining; 1 ¼ weak cell cytoplasm staining; 2 ¼ moderate cell cytoplasm staining; 3 ¼ strong cell cytoplasm staining, and the ISH score of percentage multiplying the intensity was recorded 41 . The median value of the ISH score was chosen as the cutoff criterion to dichotomize into high-and low-expression subgroups, and the median value of miR-622 ISH score is 36. Tumour xenograft experiments. Four-to five-week-old male athymic nude (Foxn1 nu/nu , BALB/c background) mice were purchased from Shanghai Laboratory Animal Center (Chinese Academy of Sciences, Shanghai, China). Mice age 5-6 weeks were injected subcutaneously in the flank on each side with 1 Â 10 7 viable SK-Hep1, SK-Hep1-shCon, SK-Hep1-shCXCR4, Huh7-Con or Huh7-CXCR4 cells. Ten mice per cell line were used. Tumour size was monitored by digital caliper. Tumour volume ¼ (L Â W 2 )/2, where L is length at the widest point of the tumour and W is the maximum width perpendicular to L. When the tumours reached B80 mm 3 in diameter, the mice were randomized into treatment groups. The mice were administered subcutaneously daily with a solution AMD3100 (10 mg kg À 1 ) or PBS for 3 weeks. Anti-CXCR4 neutralizing antibody (clones 12G5, BD Pharmingen, SanDiego, CA) or irrelevant antibody (IgG2a) (R&D System, Minneapolis, MN) was injected intraperitoneally (1 mg per injection) twice a week for 3 weeks. Mice were sacrificed if the volume of their tumour reached 2,000 mm 3 . All animal procedures were approved and performed in accordance with the guidelines of the Fudan University Animal Care and Use Committee (Permit No. 13022708).
Tissue microarray and immunohistochemical staining. Haematoxylin and eosin staining was used to define the diagnostic area, and one representative core (6 mm) was obtained from each case using a tissue arrayer (Beecher Instruments, Silver Spring, MD). Tissue microarray sections (4 mm) was immunostained with antibodies to CXCR4 (dilution: 1:200, Supplementary Table 6), EZH2 (dilution: 1:400, Supplementary Table 6). Immunohistochemical evaluation was performed independently by two researchers (Y.L. and W.L.) blinded to the clinical data, and cases with discrepant grades were re-evaluated by discussion until consensus was achieved. We classified the IHC staining results into two categories according to subcellular localization of CXCR4, for example, cytoplasmic and nuclear. A semi-quantitative H-score ranged from 0 to 300 was calculated for each specimen by multiplying the distribution areas (0-100%) at each staining intensity level by the intensities (0: negative, 1: weak staining, 2: moderate staining and 3: strong staining) 42 . The median value of the H-score was chosen as the cutoff criterion to dichotomize into high and low expression subgroup. As a result, CXCR4 median H-score is 110 and EZH2 median H-score is 75.
Statistical analysis. All quantified data represent a mean of triplicate samples ± s.d. 43 through analysing with GraphPad Prism 5 (GraphPad Software, La Jolla, CA) and assessing comparisons between different groups by the Student's ttest and one-way analysis of variance. The correlation between CXCR4 and clinicopathologic features was assessed using w 2 or Fisher's exact test with Stata software, version 12 (StataCorp, College Station, TX). Survival was calculated starting from the data of death or last follow-up. Survival curves were estimated using Kaplan-Meier method and log-rank test was used to compute differences between the curves. The correlation between CXCR4 with miR-622 and EZH2 staining by ISH and IHC, were determined using Pearson's coefficient test. Differences were considered significant at values of Po0.05. All statistical tests were two sided.