EZH2/H3K27Me3 and phosphorylated EZH2 predict chemotherapy response and prognosis in ovarian cancer

Ethics statement

The study protocol was approved by the Ethical Committee of Union Hospital, Tongji Medical College, Huazhong University of Science and Technology (IOGR No: IORG0003571).

Clinical-pathological features

Non-stratified, non-matched clinical and pathological data were retrospectively collected. Basic characteristics included patient age, histology, International Federation of Gynecology and Obstetrics (FIGO) stage, tumor type, treatment regimen, chemotherapy response and follow-up data. Chemotherapy resistance was defined by relapse within six months after completing chemotherapy or progression during the primary chemotherapy. Relapses were diagnosed on clinical symptoms, radiological evidence and biochemical abnormalities such as elevated CA125. Overall survival (OS) was defined as the time from date of diagnosis to death or last follow-up date, progression-free survival (PFS) was defined as the time from surgery to relapse or last follow-up date. Diagnoses of all patients were confirmed pathologically.

Patient and tumor characteristics

Tissue samples from a total of 65 patients were obtained from ovarian cancer patients admitted to Union Hospital, Tongji Medical College, Huazhong University of Science and Technology between August 2008 and October 2015. All patients underwent radical surgery followed by standard platinum based chemotherapy. The median age of the patients was 52 (32–72) years.

Tissue micro-array construction

65 non-consecutive, unselected primary ovarian cancer specimens were included in the tissue microarray. The tumor samples were collected within one hour after resection from the primary site. Formalin-fixed paraffin-embedded (FFPE) tissue blocks were prepared according to the standard procedure. Tissue cylinders of 2 mm in diameter were punched from representative areas of each block with regard to the matching H&E staining control by a MiniCore Control Station (Alphelys Sarl, France). The Selected tissue cylinders were re-arranged and brought into three paraffin blocks by a semi-automated tissue arrayer (Beecher Instruments, Sun Prairie, WI, USA). 4 µm section slides were prepared for further use.

Immunohistochemistry

Immunohistochemistry (IHC) was performed as previously described. Briefly, the slides were dewaxed in xylene and went through a serial of descending ethanol to rehydrate. Antigen retrieval was performed through microwave irradiation. Blocking and staining were performed using Histostain Kits (SP9001 and SP9002, ZSGB-Bio, Beijing, China). Primary antibodies for EZH2 (1:100, Cell Signaling Technology #5246, Danvers, USA), pS21EZH2 (1:100, Bethyal #00388, Montgomery, USA), p S473Akt1 (1:100, Cell Signaling Technology #4060, Danvers, USA) and H3K27Me3 (1:400, Abclonal #A2363, Boston, USA) and were applied as recommended by the manufacturers. Replacements of primary antibodies by IgG were provided as negative and isotype controls. DAB color development and hematoxylin counterstaining were performed as appropriate.

IHC evaluation

IHC evaluation was performed by two trained researchers in a blinded manner, a third pathologist was referred to when disagreement occurred. IHC scores (0–12) were assessed using a semi-quantitative scale multiplying the percentage of positive tumor cells (0, 0%; 1, <25%; 2, 25%–50%; 3, 50%–75%; 4, >75%) by stain intensity (0, negative; 1, weak; 2, moderate; 3, strong). Only nuclear staining was considered valid when measuring the scores for EZH2, H3K27Me3, p-EZH2 and p-Akt1. All results were based on five high power (400X) fields.

Statistical analysis

All statistical analyses were performed using SPSS 20.0 software. The association between the IHC scores of EZH2, H3K27Me3, p-Akt1 and p-EZH2 and ovarian cancer clinical pathological characteristics were assessed using chi square, Fisher’s exact and Kruskal-Wallis tests. The correlation of EZH2/H3K27Me3 and p-EZH2/p-Akt1 were analyzed using Spearman test. Mann Whitney test was used to compare the difference between two groups as appropriate. Receiver operating characteristic (ROC) curve were plotted to examine the value of EZH2, p-EZH2 and EZH2/H3K27Me3 combination as predictive factor for chemotherapy response. Logistic and cox regression were used to analyze risks for chemotherapy response and survival. For measurement of chemotherapy response, odds ratio (OR) was used. When OR >1, the patients were prone to chemo-resistant; when OR <1, the patients were prone to chemo-sensitive. For measurement of OS and PFS, hazard ratio (HR) was used. When HR >1, the patients were prone to worse outcome; When HR <1, the patients were prone to better outcome. Kaplan–Meier method and log-rank test were used to plot and analyze survival curves. P values <0.05 were considered statistically significant.

Association between EZH2 associated pathways and clinical pathological features

To investigate the association of canonical and non-canonical EZH2 pathways to clinical pathological features, we first investigated each pathway component independently. The components of canonical and non-canonical EZH2 pathways were universally expressed in ovarian cancer specimens with a positive expression rate of 81.54% (53/65) for EZH2, 88.89% (48/54) for H3K27Me3, 63.07% (41/65) for pAkt1, and 75.38% (49/65) for pS21EZH2. EZH2-pS21EZH2 (Spearman r = 0.580, P < 0.0001) and pS21EZH2-pAkt1 (Spearman r = 0.546, P < 0.0001) were closely correlated while EZH2-H3K27Me3 were less closely correlated (Spearman r = 0.307, P = 0.002) (Fig. 1). Samples were classified as EZH2 High/Low, H3K27Me3 High/Low, pEZH2 High/Low and pAkt1 High/Low groups by the median IHC scores (6 for EZH2, 8.5 for H3K27Me3, 6 for pS21EZH2 and 2 for p-Akt1). High EZH2 (P = 0.053) and pEZH2 (P = 0.011) expression were closely related to chemotherapy resistance (Fig. 2). Specimens from FIGO stage I-II patients were prone to higher pAkt1 level (P = 0.097). H3K27Me3 IHC score tended to be lower in tissue samples from elderly aged patients (P = 0.057). No significant association was found between EZH2, H3K27Me3, pS21EZH2 or pAkt1 and histology or tumor type (Table 1).

In order to assess the clinical relevance of different EZH2 related pathways, we next analyzed clinical pathological features in tissues with different EZH2/H3K27Me3 and pAkt1/pS21EZH2 levels. For EZH2-H3K27Me3, samples were classified as EZH2^high/H3K27Me3^high, EZH2^high/H3K27Me3^low, EZH2^low/H3K27Me3^high and EZH2^low/H3K27Me3^low. For pAkt1-pS21EZH2, samples were classified as pAkt1^high/pEZH2^high, pAkt1^high/pEZH2^low, pAkt1^low/pEZH2^high and pAkt1^low/pEZH2^low. The distribution of number of cases in each group were compared in relation to clinical features including age, histology, FIGO stage, tumor type and chemo-response. The results suggested that EZH2/H3K27Me3 level (P = 0.053) and pAkt1/pEZH2 level (P = 0.025) were closely correlated with chemotherapy response (Tables 2 and 3).

EZH2/H3K27Me3 and pS21EZH2 predicted platinum-based chemotherapy response

As both EZH2/H3K27Me3 and pAkt1/pS21EZH2 pathways correlated with chemotherapy response, subsequent assessment of the prognostic value of these two pathways were performed. First, the prognostic value of independent EZH2, H3K27Me3, pS21EZH2 and pAkt1 for chemotherapy response was assessed by logistic regression and ROC analysis. Logistic regression revealed that patients with lower pS21EZH2 (OR = 0.184; 95%CI [0.052–0.647], P = 0.008) and lower EZH2 (OR = 0.095; 95%CI [0.948–10.107], P = 0.061) IHC scores had better chemotherapy response (Table 4). The area under the curve (AUC) for EZH2 and pS21EHZ2 were 0.688 (specificity 74.47%, sensitivity 63.16%) and 0.789 (specificity 89.36%, sensitivity 68.42%) (Fig. 2). Although H3K27Me3 and pAkt1 did not have significant prognostic value as single factors, combined with EZH2 and pS21EZH2, EZH2^low/H3K27Me3^low (OR = 0.110; 95%CI [0.013–0.906].107, P = 0.040) and pAkt1^low/pEZH2^low (OR = 0.260; 95%CI [0.066–1.023], P = 0.054) status predicted better chemotherapy response (Table 4). For patients with serous ovarian cancer, pS21EZH2 remained prognostic for chemotherapy response (OR = 0.188; 95%CI [0.050–0.705], P = 0.013). Patients with EZH2^low/H3K27Me3^low (OR = 0.133; 95%CI [0.016–1.133], P = 0.065) and pAkt1^low/pEZH2^low (OR = 0.288; 95%CI [0.070–1.190], P = 0.086) were prone to better chemotherapy response (Table 5).

EZH2/H3K27Me3 and pS21EZH2 predicted PFS in ovarian cancer patients

Since chemotherapy response was an independent factor influencing prognosis, we next evaluate whether EZH2 associated pathways representing varied chemotherapy response status were of prognostic significance to survival of ovarian cancer patients. In all the factors included for analysis, Advanced FIGO stage and poor chemotherapy response were closely related with poorer OS and PFS through both uni- and multi-variate analyses (Tables 6 and 7). Kaplan–Meier survival analysis and univariate cox regression analysis for OS revealed that lower EZH2 (HR = 0.464; 95%CI [0.205–1.048], P = 0.065) and pS21EZH2 (HR = 0.464; 95%CI [0.206–1.046], P = 0.064) staining tended to correlate with better OS. A combination of EZH2^low/H3K27Me3^low staining was significantly associated with improved OS (HR = 0.257; 95%CI [0.076–0.863], P = 0.028). Kaplan–Meier survival analysis and univariate cox regression analysis for PFS revealed that lower pS21EZH2 staining was significantly associated with better PFS (HR = 0.477; 95%CI [0.226–0.882], P = 0.020). Similarly, a combination of EZH2^low/H3K27Me3^low staining was significantly associated with increased PFS (HR = 0.344; 95%CI [0.145–0.813], P = 0.015) (Fig. 3). When adjusted for FIGO stage, multivariate Cox regression analysis for PFS revealed that pS21EZH2 staining (HR = 0.453; 95%CI [0.229–0.895], P = 0.023) and EZH2^low/H3K27Me3^low status (HR = 0.388; 95%CI [0.164–0.917], P = 0.031) remained to be independent prognostic factors for PFS. For patients with serous ovarian cancer, chemotherapy response was closely related with OS and PFS through univariate analyses. Lower pS21EZH2 staining (HR = 0.431; 95%CI [0.207–0.896], P = 0.024) and pAkt1^low/pEZH2^lowstatus (HR = 0.479; 95%CI [0.219–1.046], P = 0.065) were associated with improved PFS (Table 8). We did not found significant relation of age, histology, tumor type, H3K27Me3 expression and pAkt1 expression to OS or PFS through uni- or multi-variate analyses.