Four hundred thirty-two formalin-fixed and paraffin-embedded surgical breast IDC specimens, 27 DCIS specimens, and 11 matched normal breast tissues obtained from the Soonchunhyang University Hospital from 2001 to 2013 were included in this retrospective study. Every included patient was diagnosed with breast cancer by pathologists based on pathological test results. Hematoxylin and eosin-stained slides were independently reviewed in each case to confirm the original diagnosis by two pathologists (S.H.J. and H.J.L.), based on the 2012 World Health Organization classification [14]. Data regarding patient age at initial diagnosis, tumor size, histological type, histological tumor grade, lymph node status, and surgery type were also collected. Pathologic TNM classification and staging were performed for the 432 cases using the current TNM international staging system (seventh edition of the American Joint Committee on Cancer criteria). Further data, including disease progression, patient survival, and cause of death, were obtained from medical records and/or interviews with patients' families. This study was approved by the Institutional Review Board at the Soonchunhyang University Hospital (SCHCA 2015-06-006).

For tissue microarray (TMA) construction, a sufficient viable tumor sample with no hemorrhage or necrosis was selected for each case. Representative 2-mm diameter core tissue sections were taken from paraffin blocks. They were arranged in new recipient TMA blocks using a manual TMA device (SuperBioChips Laboratories, Seoul, Korea). In cases with variable histological features, the most representative area was selected for TMA construction. In addition, 11 nonneoplastic breast tissues from IDC patients were included.

EZH2 expression was analyzed by IHC. Four-micrometer sections from the TMA blocks were deparaffinized in xylene and rehydrated with gradually decreasing concentrations of ethanol in distilled water. IHC staining of the TMA samples was performed with a Bond-Max Autostainer (Leica Microsystems, Bannockburn, USA). Primary antibody binding was detected with the Bond Polymer Refine Detection kit (Leica Microsystems), according to the manufacturer's instructions. The primary anti-EZH2 mouse monoclonal antibody (Novocastra, Newcastle Upon Tyne, UK) was used at a dilution of 1:100. Two independent observers (S.H.J. and H.J.L.) read the slides in a blinded manner, and consensus was made by repeated examination when results were discordant. Only epithelial cells were evaluated, and results for each core were recorded separately. At the time of review, neither of these investigators was aware of clinicopathologic data associated with each specimen because all slides had been coded. Nuclear EZH2 expression was scored as negative (score=0, no staining), weak (score=1, <25% of nuclei staining with any intensity), moderate (score=2, 25%–75% of nuclei staining with any intensity), or strong (score=3, >75% of nuclei staining with any intensity) [13]. Tumors with nuclear staining with moderate or strong were classified as having high EZH2 expression, whereas those with completely absent or weak nuclear staining were classified as having no or low EZH2 expression.

Immunohistochemical staining for estrogen receptor (ER; 1:50; Dako Co., Carpinteria, USA), progesterone receptor (PR; 1:50; Dako Co.), human epidermal growth factor receptor 2 (HER2; 1:200; Novocastra Laboratories Ltd., Newcastle, UK), Ki-67 (1:800; Dako Co.), cytokeratin 5/6 (CK5/6; 1:50; Dako Co.), epidermal growth factor receptor (EGFR; 1:100; Dako Co.), and p53 (1:1,200; Dako Co.) was done on 4-µm sections in the TMA blocks. ER and PR immunohistochemical staining was evaluated using the Allred method [15]. An Allred score of 3 or higher was classified as positive. HER2 was analyzed according to the general guidelines set by the American Society of Clinical Oncology/College of American Pathologists. When IHC yielded equivocal results, HER2 status was established with fluorescent in situ hybridization. Ki-67 and p53 expression levels were counted in 1,000 tumor cells, and positivity was defined as ≥14% and >10% staining, respectively. CK5/6 and EGFR expression levels were considered positive when the cytoplasmic and/or membranous reaction was ≥10%. The phenotypes were classified as follows: luminal A: ER- and/or PR-positive, HER2-negative, and Ki-67 index <14%; HER2-negative luminal B: ER- and/or PR-positive, HER2-negative, and Ki-67 index ≥14%; HER2-positive luminal B: ER- and/or PR-positive, HER2-positive, and any Ki-67 index; HER2 type: ER- and PR-negative and HER2-positive; basal triple-negative breast cancer (TNBC): ER-, PR-, and HER2-negative and CK5/6- and/or EGFR-positive; and non-basal TNBC: ER-, PR-, HER2-, CK5/6-, and EGFR-negative.

Statistical analyses were performed using the software package SPSS version 19.0, for Windows (IBM Corp., Armonk, USA). Associations between EZH2 expression and the clinicopathologic characteristics were analyzed using Pearson chi-square test, Fisher exact test, or an independent t-test, according to test conditions. Survival rates were plotted using the Kaplan-Meier method and log-rank test. Disease-free survival (DFS) periods were defined as the intervals between primary surgeries and final follow-up visits without disease or evidence of recurrence or metastasis of breast cancers (locoregional relapse and distant metastasis). Overall survival (OS) periods were defined as the intervals between primary surgeries and the last follow-up visits or deaths from any cause. Cox proportional hazard model was used to evaluate the associations between clinicopathological factors and survival rates. Hazard ratio (HR) and associated 95% confidence interval (CI) were calculated for each variable. A p-value <0.05 was considered statistically significant.

This study included 426 women (98.6%) and six men (1.4%). The mean age was 52.4±12.4 years (range, 24–81 years), and the mean tumor size was 2.4±1.4 cm (range, 0.3–12 cm). Of the 432 patients, 222 (51.4%) underwent breast-conserving surgeries, and 210 (48.6%) underwent mastectomies. Of the 432 samples, 54 were grade 1 (12.5%), 213 were grade 2 (49.3%), and 165 were grade 3 (38.2%). Primary tumor size data were available for the 432 patients. From this, 212 (49.1%), 197 (45.6%), 19 (4.4%), and 4 (0.9%) tumors were categorized as pT1, pT2, pT3, and pT4, respectively. Of the 432 patients, 163 patients (37.7%) had lymph node positivity at the time of surgery. The 432 patient tumors were classified using the TNM classification system as stage I (n=154, 35.6%), stage II (n=198, 45.8%), and stage III (n=80, 18.5%). The proportions of patient tumors positive for ER and PR expression were 66.0% and 35.2%, respectively. Upon HER2 expression analysis, 18.1% of all patient tumors were positive. High Ki-67 expression was noted in 45.6% of tumors. Positive CK5/6 and EGFR expression levels were found in 9.3% and 20.6% of tumors, respectively. For p53 expression, 18.5% of patient tumors were positive.

EZH2 protein in IDC was expressed mainly in tumor cell nuclei (Figure 1). We evaluated EZH2 expression in 11, 27, and 432 normal breast tissues, DCIS specimens, and IDC specimens, respectively. EZH2 expression was positive in 0 of 11 (0%) normal breast tissue, 8 of 27 (29.6%) DCIS specimens, and 215 of 432 (49.8%) IDC specimens, in increasing order, and the differences in positivity between groups were statistically significant (p=0.001) (Table 1). The mean EZH2 expression score was 0.18±0.41 in normal breast tissue, 0.89±1.09 in DCIS, and 1.49±1.12 in IDC (Figure 2). EZH2 expression was significantly higher in DCIS and IDC specimens than in normal breast tissue (p<0.001). After evaluation of the 432 immunostained IDC specimens, 109 (25.2%) were negative; 108 (25.0%) were scored 1, indicating weak positivity; 109 (25.2%) were scored 2, indicating moderate positivity; and 106 (24.5%) were scored 3, indicating strong positivity. For statistical analyses, the cases were subdivided into an EZH2-high expression group (scores 2 and 3; n=215, 49.8%) and an EZH2-low expression group (scores 0 and 1; n=217, 50.2%).

High EZH2 expression significantly correlated with high histologic grade (p<0.001), large tumor size (p=0.014), advanced pathologic stage (p-stage, p=0.006), negative ER expression (p<0.001), positive HER2 expression (p<0.001), high Ki-67 staining index (p<0.001), positive CK5/6 expression (p=0.003), positive EGFR expression (p<0.001), positive p53 expression (p<0.001), endocrine therapy status (p=0.001), and chemotherapy (p=0.032) (Table 2). Other clinicopathologic variables, including age (p=0.700), sex (p=0.448), operation method (p=0.453), lymph node metastasis (p=0.372), negative PR expression (p=0.132), neoadjuvant chemotherapy (p=0.098), and radiotherapy (p=0.488), did not correlate with EZH2 expression.

For molecular subtypes, high EZH2 expression was significantly associated with HER2-negative luminal B (72.2%), HER2-positive luminal B (73.5%), and HER2 type (75.0%) as well as basal TNBC (83.6%) cancers, but high EZH2 was not associated with luminal A cancer (20.7%) or non-basal TNBC (23.3%, p<0.001).

The median follow-up period for all patients was 41.0 months, with a range of 1 to 158 months. At the time of analysis, the numbers of overall deaths and patients with disease progression were 48 and 82, respectively. The 4-year overall survival rates for IDC patients with high or low EZH2 expression were 86.0% and 91.7%, respectively. By univariate analysis, conventional prognostic parameters, including tumor size, lymph node metastasis, and p-stage were significant predictors of DFS and OS (Table 3). In addition, expression levels of ER and HER2 significantly predicted DFS or OS of breast cancer patients. Patients with high EZH2 expression had a significantly worse DFS (p=0.041) and OS (p=0.009) than did patients with low EZH2 expression (Figure 3A, B). Based on molecular subtypes, the DFS (p=0.016), and OS (p=0.032) rates were significantly different (Figure 3C, D). Luminal A patient had the best survival, closely followed by HER2-negative luminal B patients during the first 40 months of follow-up with respect to DFS and OS. Patients with HER2-positive cancer had the poorest prognoses with respect to DFS and OS. All TNBC patients (both basal and non-basal) had intermediate survival times, with deaths occurring earlier than deaths of luminal A patients.

To evaluate EZH2 positivity in IDC as an independent predictor of DFS and OS, multivariate analysis using the Cox proportional hazard model was performed and included age, tumor size, lymph node metastasis, and EZH2 expression. All variables with p<0.05 in univariate analysis were included in the multivariate Cox model. Three variables, age (p=0.002, only for OS), lymph node metastasis (p=0.028, only for DFS), and p-stage (p=0.042 and p=0.021 for DFS and OS, respectively) were significant prognostic factors for IDC patients (Table 3). Multivariate analysis showed no significant association between EZH2 expression and disease progression and overall death (HR, 1.24, 95% CI, 0.78–1.97, p=0.370; and HR, 1.25, 95% CI, 0.66–2.37, p=0.502).

The DFS and OS rates of the EZH2-high and -low expression groups, stratified according to molecular subtype, are shown in Figure 4 and Supplementary Figure 1 (available online). In patients with luminal A disease, high EZH2 expression was significantly associated with shorter OS compared to low EZH2 expression (p=0.045) (Figure 4B). In contrast, DFS was not significantly different between the EZH2-high and -low expression groups (p=0.142) (Figure 4A). In patients with HER2-negative luminal B disease, neither DFS nor OS was significantly different according to EZH2 expression (p=0.386 and p=0.147, respectively) (Supplementary Figure 1A, B). In patients with HER2-positive luminal B disease, neither DFS nor OS was significantly different according to EZH2 expression (p=0.634 and p=0.782, respectively) (Supplementary Figure 1C, D). In patients with HER2-positive disease, neither DFS nor OS was significantly different according to EZH2 expression (p=0.522 and p=0.690, respectively) (Supplementary Figure 1E, F). In patients with basal TNBC, neither DFS nor OS was significantly different according to EZH2 expression (p=0.367 and p=0.535, respectively) (Supplementary Figure 1G, H). In patients with non-basal TNBC, neither DFS nor OS was significantly different according to EZH2 expression (p=0.274 and p=0.374, respectively) (Supplementary Figure 1I, J). Multivariate analysis using the Cox proportional hazard model showed that high EZH2 expression in luminal A cancer was not a significant independent prognostic factor for OS (HR, 2.85; 95% CI, 0.80–10.16; p=0.106).