Tracking the Features of CIMP-positive Glioma: A Systematic Review and Meta-Analysis

Backgrounds Controversy surrounding CpG island methylator phenotype (CIMP) in glioma exists with regard to its prognostic value. Methods PubMed, EMBASE, and Cochrane Library databases were searched for studies describing molecular and clinicopathological features, and overall survival of gliomas stratified by CIMP status. Associations of CIMP with outcome parameters were estimated using odds ratio (OR) or hazard ratios (HRs) with a 95% confidence interval (CI) using a fixed or random effects model.

in specimen for a single-aim study of glioma, discussion of all stage of tumors, and no exclusion based on molecular marker.

Statistical analysis
All statistical tests were performed with Stata Version 13.0 (Stata Corporation, College Station, TX, USA). ORs and 95% CI were used to assess the relationship between CIMP status and clinicopathological and molecular parameters. For quantitative aggregation of survival results, HRs and their 95% CIs were combined as the effective value. The HRs were calculated directly from the reported data by number of events or calculated from Kaplan-Meier survival curve using Engauge Digitizer software (freely downloaded from http://sourceforge.net). Heterogeneity between studies was analyzed using a X 2 -based Qtest (P > 0.1 was considered a lack of heterogeneity) and an I 2 test (I 2 < = 50% indicated low heterogeneity, and I 2 > 50% indicated substantial heterogeneity). For the low heterogeneity group, each study was analyzed using the fixed effects model. Otherwise, the random effects model was used. Significance of the pooled OR or HR was determined by a Z-test, with P < 0.05 considered statistically significant. An estimate of potential publication bias was carried out using a funnel plot, with an asymmetric plot suggesting possible publication bias. Funnel plot asymmetry was assessed using Egger's linear regression test, a linear regression approach measuring funnel plot asymmetry on the natural logarithm scale of the OR. As suggested by Egger, significance of the intercept was determined by a t-test (with P < 0.05 considered representative of statistically significant publication bias).

Study characteristics
The electronic search from three databases yielded 121 potential articles. After title screening, abstract screening and full-text evaluation, 109 articles were excluded (Fig. 1).  (12,13,14,17), two from the Erasmus medical cancer brain tumor tissue bank (13,16), two from the Spanish National Tumor Bank Network (18,19), two from the NOA trial in Germany (15,21), and one from the Chinese Glioma Genome Atlas (CGGA) (20). In addition, one study used data from a publicly available dataset (23) and another used mixed samples from two publicly available datasets and one newly generated dataset from MD Anderson (22). For each included study, Fig. 2 lists the risk of bias from selection, exposure assessment, other variable assessment, outcome assessment, and confounding factors. Based on a strict exclusion and inclusion criteria, studies with high risk in selection bias were not included.

Clinicopathological features
Extractable data related to clinicopathological factors were gender and histopathology.

Overall survival and publication bias
In total, five studies compared the overall survival of individuals with CIMP + and CIMPgliomas. Our pooled analysis showed that CIMP + glioma was significantly associated with longer overall survival (HR -0.57; 95% CI -0.97--0.16; P = 0.003; P heterogeneity 0.000; Begg's funnel plot was performed to assess publication bias. Heterogeneity tests for comparing the 12 combined studies showed heterogeneity in certain analyses such as those of IDH1 mutation, 1p19q LOH, EGFR mutation, AOA, OD, GBM and overall survival. However, no single study influenced the pooled OR qualitatively as indicated by the sensitivity analyses (data not shown).

Discussion
Epigenetic alterations have been reported to be involved in human cancerization through various mechanisms such as DNA methylation, histone modifications, small and long noncoding RNA, and chromatin architecture remodeling (27). With unclear significant influence of aberrant DNA sequence changes in human cancers and the irreversible and hereditary nature of epigenetic alterations (28), the presence of epigenetic alterations in noncancerous tissue suggests epigenetic alterations are involved in the field for cancerization. CpG island methylator phenotype (CIMP) is one of the most reported epigenetic alterations and is recognized as a major event in the origin of many cancers (27).
Prognostic value of CIMP has been reported for a variety of tumors. For example, CIMP + phenotype is significantly associated with a worse outcome in colorectal cancer (24) and is a predictor of poor prognosis to esophageal adenocarcinoma (25). CIMP is also a promising biomarker for the management of patients with gastric cancer (26). However, the prognostic role of CIMP status in glioma is unclear. In this study, we expanded upon previous tumor-associated studies of the prognostic value of CIMP to examine CpG islands associated with glioma.
We identified 12 published studies that included 2386 glioma patients in order to estimate the association between CIMP and other molecular and clinicopathological features in glioma. We found a trend toward more IDH1 mutations and 1p19q LOH and OD, and less GBM in CIMP-positive glioma than in CIMP-negative glioma. Moreover, we also demonstrated that CIMP did not show a correlation with MGMT promoter methylation, EGFR mutation, AOA, OA or gender, but CIMP-positive was significantly associated with longer overall survival. Taken together, these results suggest that CIMP may be used as an independent prognostic marker.
Heterogeneity in the relationship between CIMP status and certain clinicopathological features was significant in this study. One of the major confounding factors in the significant heterogeneity may be the lack of a standardized definition of CIMP, with the number, type and identity of genes employed in the selection panel different in every study. Until 2010, Noushmeh and colleagues (12) (29). Further studies are needed to validate a consistent CIMP definition.
So far, the value of CIMP as a predictive biomarker to guide prescription of neoadjuvant or adjuvant chemotherapy in glioma is uncertain. However, considering the influence of CIMP in therapeutic and clinical trial strategy may be necessary. It is clear that there is heterogeneity, even within other molecule biomarker combinations, which is likely to lead to potential prognostic value for individualized therapy. Aldap and colleagues (29) reported that glioma were divided into 3 distinct survival groups based on CIMP markers (CIMP-positive, CIMP-intermediate and CIMP-negative). Multivariate analysis showed that both IDH1 mutation and CIMP were independent predictors of outcome, suggesting CIMP and IDH1 mutation are potential prognostic biomarkers in glioma. Furthermore, G-CIMP tumor-related genes exhibited a demethylated pattern, and reversing the methylated pattern of G-CIMP tumor-related genes may be a potential solution for glioma. Further work should be conducted to verify and confirm the clinical value of CIMP in glioma patients.
The main limitation of our study the spectrum of gene panel markers used for CIMP.
Unfortunately, this is a common finding in CIMP studies, and other systematic reviews and meta-analyses in colorectal cancer (30) and gastric cancer (31,32) have accepted this relative limitation when performing pooled analyses. This study has significant strength in that it is a systematic review and meta-analysis of the currently available literature regarding the prognostic value of CIMPs in glioma.

Conclusions
In conclusion, this meta-analysis highlights that CIMP-positive glioma have their own molecular (such as IDH1 mutations) and clinicopathological features and a better prognosis for CIMP-positive glioma, suggesting CIMP could be used as an independent prognostic marker for glioma.    Risk of bias of each included study. Red cycle: study with high risk of bias; green cycle: study with low risk of bias; yellow cycle: study with insufficient information for assessing risk of bias.