The relationship between EZH2 expression and microRNA-31 in colorectal cancer and the role in evolution of the serrated pathway

Polycomb group protein enhancer of zeste homolog 2 (EZH2) is a methyltransferase that correlates with the regulation of invasion and metastasis and is overexpressed in human cancers such as colorectal cancer. MicroRNA-31 (miR-31) plays an oncogenic role and is associated with BRAF mutation and poor prognosis in colorectal cancer. EZH2 is functionally considered to suppress miR-31 expression in human cancers; however, no study has reported its relationship with colon cancer. We therefore evaluated EZH2 expression using immunohistochemistry and assessed miR-31 and epigenetic alterations using 301 colorectal carcinomas and 207 premalignant lesions. Functional analysis was performed to identify the association between EZH2 and miR-31 using cancer cell lines. In the current study, negative, weak, moderate, and strong EZH2 expressions were observed in 15%, 19%, 25%, and 41% of colorectal cancers, respectively. EZH2 was inversely associated with miR-31 (P < 0.0001), independent of clinicopathological and molecular features. In a multivariate stage-stratified analysis, high EZH2 expression was related to favorable prognosis (P = 0.0022). Regarding premalignant lesions, negative EZH2 expression was frequently detected in sessile serrated adenomas/polyps (SSA/Ps) (76%; P < 0.0001) compared with hyperplastic polyps, traditional serrated adenomas, and non-serrated adenomas (25–36%). Functional analysis demonstrated that the knockdown of EZH2 increased miR-31 expression. In conclusion, an inverse association was identified between EZH2 and miR-31 in colorectal cancers. Our data also showed that upregulation of EZH2 expression may be rare in SSA/Ps. These results suggest that EZH2 suppresses miR-31 in colorectal cancer and may correlate with differentiation and evolution of serrated pathway.

A recent study has reported that EZH2 suppresses miR-31 expression by inducing H3K27me3 on the miR-31 promoter and that the inhibition of EZH2 increased miR-31 expression in prostate cancer [2]. Furthermore, EZH2-mediated histone methylation has been shown to suppress miR-31 expression in adult T-cell leukemia [16]. With regard to melanoma, genetic and epigenetic loss of miR-31 produced a feed-forward EZH2 expression [6]. Thus, accumulating evidence indicates that EZH2 may downregulate miR-31 expression in human cancers; however, no study has reported the relationship between EZH2 and miR-31 in colorectal cancer.
We conducted this study to clarify the association of EZH2 expression with miR-31 and epigenetic alterations using a database comprising more than 500 colorectal tumors. Furthermore, we performed functional analyses to identify whether EZH2 suppressed miR-31 expression in colorectal cancers.

EZH2 expression and patient survival
The influence of EZH2 expression on clinical outcomes was assessed in 299 patients with colorectal cancer (stages I-IV). During follow-up among eligible patients, 100 patients died, of which 81 deaths were attributed to colorectal cancer. The median follow-up time for censored patients was 4.4 years. Kaplan-Meier analysis was performed using categorical variables (score 0, score 1, score 2, or score 3). In terms of cancer-specific survival, significantly lower mortality was observed in patients with high EZH2 expression (log-rank test: P = 0.010) than in those with low EZH2 expression ( Figure 3A).
We made a dichotomous expression variable for EZH2, defining a score of 3 as high expression and scores of 0-2 as low expression. In terms of cancer-specific   A. In terms of cancer-specific survival, significantly lower mortality was observed in patients with high EZH2 expression than in those with low EZH2 expression (log-rank test: P = 0.010). B. In terms of cancer-specific survival, significantly lower mortality was observed in patients with high expression (score 3) than in those with low expression (score 0-2) (log-rank test: P = 0.0022). www.impactjournals.com/oncotarget survival, significantly lower mortality was observed in patients with high expression (log-rank test: P = 0.0022) than in those with low expression ( Figure 3B). However, no significant differences were observed between the high and low expression groups when we defined scores of 1-3 as high expression and score of 0 as low expression (log-rank test: P = 0.69) or when we defined scores of 2-3 as high expression and scores of 0-1 as low expression (P = 0.075).

Knockdown of EZH2 increases miR-31 expression in colon cancer cell lines
To examine whether EZH2 suppressed miR-31 expression, we knocked down EZH2 mRNA by siRNAs in colon cancer cell lines and measured the resulting miR-31 expression. Figure 4A shows the expression level of EZH2 in 7 colon cancer cell lines using quantitative reverse transcription-PCR (qRT-PCR) ( Figure 4A). Our data demonstrated that in RKO cells, EZH2 expression was successfully downregulated by approximately 30% and 56% when transfecting two different EZH2 siRNAs (siEZH2_7644 and siEZH2_7882, respectively) ( Figure  4B). Moreover, we found that there was a considerable increase in miR-31 expression in RKO cells transfected with EZH2 siRNAs ( Figure 4C). Similar results were observed in HT29 and SW480 cells (Supplementary Figure 4).

Promoter region of miR-31 is marked by histone H3 lysine 27 trimethylation (H3K27me3)
To clarify the functional link between EZH2 and miR-31 expression, we examined H3K27me3 levels around the promoter region of miR-31 by performing chromatin immunoprecipitation (ChIP) assay. We observed that H3K27me3 was steadily enriched at the promoter region of miR-31 in RKO cells, and the H3K27me3 levels were decreased after knockdown of   Figure 5). Taken together with increased expression of miR-31 after EZH2 knockdown ( Figure 4C), these results suggest the expression of miR-31 is suppressed by EZH2 through H3K27me3 in RKO cells.

DISCUSSION
We performed this study to identify the association of EZH2 expression with molecular alterations in colorectal tumors. In a database comprising 301 patients with colorectal cancer, high EZH2 expression was inversely associated with miR-31 expression, independent of clinicopathological and molecular features. Our data also showed that high EZH2 expression was a favorable prognostic factor in colorectal cancer. Regarding premalignant lesions, negative EZH2 expression was higher in SSA/Ps than in HPs, TSAs or non-serrated adenomas. In functional analysis, we showed that the siRNA knockdown of EZH2 increased miR-31 expression in colon cancer cell lines.
EZH2 reportedly downregulates miR-31 expression in human cancers such as prostate cancer [2] and adult T-cell leukemia [16]. However, no study has reported the potential role of EZH2 in the regulation of miR-31 expression in colorectal cancer. The present multivariate analysis showed that high EZH2 expression was inversely associated with miR-31 expression in colorectal cancer, whereas functional analysis showed that EZH2 knockdown increased miR-31 expression in colon cancer cell lines. Recent studies that used colon cancer cell lines have reported that EZH2 expression was inversely associated with microRNA-506 [38], microRNA-26a, and let-7b [39]; these microRNAs downregulated EZH2 expression by directly targeting 3′-UTR. Conversely, EZH2 has been reported to suppress miR-31 expression by inducing H3K27me3 on the miR-31 promoter in prostate cancer. Therefore, to determine if EZH2 is involved in the regulation of miR-31 expression through histone H3K27me3, we examined H3K27me3 levels around the transcription start site on the miR-31 promoter by performing ChIP assay and found that EZH2mediated histone methylation downregulates miR-31 expression in colon cancer. To the best of our knowledge, this is the first report describing the downregulation of miR-31 by EZH2 in colorectal cancer. Recently, we also reported that high miR-31 expression was associated with shorter progression-free survival in patients with colorectal cancer treated using anti-epidermal growth factor receptor (EGFR) therapy [27]. Therefore, because of the relationship with miR-31, EZH2 may represent a new prognostic biomarker for molecular targeted therapies and can provide a promising therapeutic target in patients with colorectal cancer.
With regard to the association between EZH2 expression and outcomes in patients with colorectal cancer, previous studies have reported that EZH2 overexpression was associated with a favorable prognosis [11][12][13]. Our data also showed that high EZH2 expression was associated with favorable survival using multivariate stage-stratified Cox analysis. These results are reasonable because we recently reported that high miR-31 expression, which is inversely correlated with EZH2 expression, is an unfavorable prognostic factor in patients with colorectal cancer [28]. We summarized the association of EZH2 expression with miR-31 and cancer-specific survival in Supplementary Figure 6. Our current study had some important limitations, particularly because of its crosssectional nature and unknown bias (i.e. selection bias). Nevertheless, our multivariate regression analysis was adjusted for potential confounders, including clinical and molecular features, and we were able to show that high EZH2 expression was inversely associated with miR-31 expression and associated with better prognosis.
CIMP is a distinct form of epigenomic instability that causes most cases of sporadic MSI-high colorectal cancers through the epigenetic inactivation of MLH1 [30,31,[40][41][42][43][44]. RUNX3 belongs to the RUNX family of genes, which has been reported to be the single best marker for diagnosing CIMP-high status [41]. Previous studies have reported that EZH2 downregulates RUNX3 by inducing histone H3 methylation in various human cancers such as gastric [10], bile duct [4], and pancreatic cancers [5]. With regard to colorectal cancer, Kodach et al. reported, that despite higher levels of EZH2 and lower levels of RUNX3, no inverse correlation was present between EZH2 and RUNX3 [9]. Consistent with those results, when using 4 CIMP-specific promoters (CACNA1G, IGF2, MLH1, and RUNX3), we showed that neither RUNX3 methylation nor CIMP-high status was associated with EZH2 expression. These results suggest that EZH2 overexpression may not correlate with the epigenetic silencing of CIMP-specific promoters by histone H3 methylation in colorectal cancer.
Premalignant colorectal neoplasms appear to be important precursor lesions in the pathogenesis of colorectal cancer. In particular, the serrated neoplasia pathway has attracted considerable attention as an alternative pathway of colorectal cancer development, and serrated lesions exhibit unique clinicopathological and molecular features [45][46][47][48]. Both SSA/Ps and TSAs are recognized premalignant lesions, but SSA/Ps are the principle serrated precursors of colorectal cancer [47][48][49][50]. Because there are many clinicopathological and molecular similarities between SSA/Ps and CIMP-high colorectal cancers, including proximal tumor location, BRAF mutation, and MLH1 methylation, SSA/Ps are hypothesized to be precursor lesions that develop into CIMP-high colorectal cancers [31,42,46,50,51].
We recently reported that high miR-31 expression was more pronounced in SSA/Ps with cytological dysplasia than in other SSA/Ps, but we did not find a significant difference between TSAs with high-grade dysplasia and other TSAs [31]. These data imply that miR-31 correlate with the progression of SSA/Ps. However, our current data revealed that EZH2 expression (moderate or strong) was not detected in any SSA/Ps and that negative EZH2 expression was higher in SSA/Ps than in HPs, TSAs, and non-serrated adenomas. In addition, we demonstrated that the knockdown of EZH2 by siRNAs increased miR-31 expression in colon cancer cell lines, suggesting that negative EZH2 expression causes miR-31 upregulation in the progression of SSA/Ps. Therefore, EZH2 may be a key molecule in the differentiation and evolution of serrated lesions, and microvesicular HPs without EZH2 expression may progress to SSA/Ps. Future independent studies are required to clarify the role of EZH2 expression in the serrated pathway.
In conclusion, we identified an inverse association between the expressions of EZH2 and miR-31 in colorectal cancer and that the upregulation of EZH2 expression may be a rare event in SSA/Ps. Hence, we suggest that EZH2 suppresses miR-31 expression in colorectal cancer and may correlate with differentiation and evolution of the serrated pathway. These findings improve our understanding of the mechanism of colorectal tumorigenesis and have the potential to significantly affect clinical and translational research on colorectal cancer.

Patients and tissue specimens
Formalin-fixed paraffin-embedded (FFPE) tissues of 310 colorectal cancers (stages I-IV), 158 serrated lesions and 57 non-serrated adenomas (i.e. tubular or tubulovillous adenomas) of patients who underwent endoscopic resection or other surgical treatment at Sapporo Medical University Hospital and Keiyukai Sapporo Hospital between 1999 and 2014 were collected. To avoid selection bias as much as possible, we consecutively collected FFPE tissue specimens of colorectal cancers, serrated lesions, and non-serrated adenomas.
The criterion for diagnosis of colorectal cancer was invasion of malignant cells beyond the muscularis mucosa. Intramucosal carcinoma and carcinoma in situ were classified as adenoma. Colorectal tumors were classified by location as follows: the proximal colon (cecum, ascending, and transverse colon), distal colon (splenic flexure, descending, and sigmoid colon) and rectum. The patients were followed until death or December 2014, whichever came first.
Serrated lesions [hyperplastic polyps (HPs) (N = 54), sessile serrated adenomas/polyps (SSA/Ps) (N = 53) and traditional serrated adenomas (TSAs) (N = 51)] were classified on the basis of the current World Health Organization (WHO) criteria [52]. All HPs were found to be microvesicular. Informed consent was obtained from all the patients before specimen collection. This study was www.impactjournals.com/oncotarget approved by the respective institutional review boards of the participating institutions.

Immunohistochemistry for EZH2 expression
Immunohistochemistry was performed on 5 μm FFPE sections. Sections were autoclave-pretreated in target retrieval solution (Dako Cytomation, Carpinteria, CA, USA). Endogenous peroxidase activity was blocked using 3% hydrogen peroxide, and the sections were incubated overnight at 4°C with anti-EZH2 antibody (#612667, BD Biosciences, San Jose, CA, USA) at a dilution of 1:100. A subsequent reaction was performed using a horseradish peroxidase enzyme-labeled polymer of the EnVision™ Plus detection system (Dako). When a positive reaction was visualized using a diaminobenzidine (DAB) solution, counterstaining was performed using Mayer's hematoxylin. Five random high-power fields were evaluated per lesion to determine the mean nuclear positivity, which was categorized as follows: score 0 (negative, <5%), score 1 (weak, 5%-39%), score 2 (moderate, 40%-79%), or score 3 (strong, ≥80%). EZH2 expression was visually interpreted by H.K., who was unaware of the other data. For the agreement study on EZH2 expression, 147 randomly selected cases were examined by a second pathologist (by K.N.), who was also unaware of the other data. The concordance between the two pathologists (P < 0.0001) was 0.87 (κ = 0.74), indicating substantial agreement.

RNA isolation and quantitative RT-PCR
Total RNA was extracted from cell pellets using the TRIzol ® reagent (Invitrogen by Life Technologies) and reverse transcribed to cDNA with the PrimeScript™ RT Reagent Kit (Takara Bio Inc., Kusatsu, Japan). Quantitative RT-PCR was performed using the TaqMan ® universal master mix with specific primers and a probe set for EZH2 (Applied Biosystems; TaqMan Gene Expression Assay). Actin-beta (ACTB) expression was used to normalize for variance. All genes were tested in triplicates.

Chromatin immunoprecipitation (ChIP)-PCR
ChIP was performed as described previously [35] with minor modifications. Briefly, 1 × 10 6 cells were fixed in 1% formaldehyde for 10 min at room temperature, rinsed in glycine, then washed in cold phosphate buffered saline (PBS) twice. The cells were re-suspended in SDS lysis buffer and were sonicated using a Covaris S2 device (Covaris Inc., Woburn, MA, USA), following the manufacturersʼ instructions. Sheared chromatin was immunoprecipitated for more than eight hours at 4°C using 2 μg of anti-H3K27me3 antibody (#9733; Cell Signaling Technology, Danvers, MA, USA). Before adding antibodies, 10 μl of sheared chromatin was saved as input DNA sample. After washing, elution and reversal of the cross-links, DNA was purified using Agencourt AMPure XP ® (Agencourt Biosciences, Beverly, MA, USA), according to the product manual. The purified DNA was amplified by real-time quantitative PCR with SYBR ® Select Master Mix (Life Technologies) and 7500 fast realtime PCR system (Life Technologies) and was analysed for enrichment. The sequences of the ChIP primers are provided in Supplementary Table 2.

Statistical analysis
JMP (version 10) and SAS (version 9) software programs were used for statistical analyses (SAS Institute, Cary, NC, USA). All P-values were two-sided. Univariate analyses were performed to investigate clinicopathological and molecular characteristics according to the EZH2 expression level; a chi-square test or Fisher's exact test was used for categorical data, while analysis of variance was used to compare the mean patient age and tumor size. To account for multiple hypothesis testing in associations between EZH2 expression and other 13 covariates, the P-value for significance was adjusted by Bonferroni correction to P = 0.0038 (= 0.05/13).
In survival analysis, the Kaplan-Meier method and log-rank test were used to assess the survival time distribution. Cox proportional hazards regression models were used to compute mortality HRs according to the EZH2 expression status. Stratification by the tumor-nodemetastasis (TNM) disease stage (I, IIA, IIB, IIIA, IIIB, IIIC and IV) was performed using the "strata" option in the SAS "proc phreg" command. The multivariate, stage-stratified Cox model included the EZH2 expression variable stratified by gender (male vs. female), age at diagnosis (continuous), tumor size (continuous), year of diagnosis (continuous), tumor location (proximal colon vs. distal colon and rectum), tumor differentiation (well to moderate vs. poor), MSI status (MSI-high vs. MSS/ MSI-low), CIMP status (CIMP-high vs. CIMP-low/zero), mutations of BRAF, KRAS and PIK3CA (present vs. absent), and miR-31 (high expression vs. low expression). A backward elimination was performed with a threshold of P = 0.10, to avoid overfitting.
A multivariate logistic regression analysis assessing the relationships with EZH2 expression status initially included gender, age, tumor size, year of diagnosis, tumor location, tumor differentiation, disease stage, MSI, CIMP, mutations of BRAF, KRAS and PIK3CA, and miR-31, considering potential confounding and causal relationships. For multiple hypothesis testing, the P-value for significance was adjusted by Bonferroni correction to 0.0038 (= 0.05/13).