Aberrant Methylation of RASSF1A Closely Associated with HNSCC, a Meta-Analysis

The RAS association domain family protein 1a (RASSF1A), a tumor suppressor gene at 3p21.3, plays a very important role in various cancers, including the head and neck squamous cell carcinoma (HNSCC). Hypermethylation of CpG islands in the RASSF1A promoter region contribute to epigenetic inactivation. However, the association between RASSF1A promoter methylation and HNSCC remains unclear and controversial. Therefore, a meta-analysis was performed in the study to identify the association. We identified the eligible studies through searching PubMed, EMBASE, Web of Science, and China National Knowledge Infrastructure (CNKI) databases with a systematic searching strategy. The information on characteristics of each study and prevalence of RASSF1A methylation were collected. Pooled odds ratios (ORs) with corresponding confidence intervals (CIs) were calculated. Meta-regression was performed to analyze heterogeneity and funnel plots were applied to evaluate publication bias. A total of 550 HNSCC patients and 404 controls from twelve eligible studies were included in the meta-analysis. Overall, a significant association was observed between RASSF1A methylation status and HNSCC risk under a random-effects model (OR = 2.93, 95% CI: 1.58–5.46). There was no significant publication bias observed. The meta-analysis suggested that there was a significant association between aberrant RASSF1A methylation and HNSCC.

is originally described in lung and breast cancer 14 . Since then, it has emerged that RASSF1A is one of the most frequently hypermethylated genes so far described and was reported as a prognostic indicator in renal cell carcinoma, non-small cell lung cancer, neuroblastoma, endometrial cancer and breast cancer [15][16][17][18][19][20][21] . Furthermore, hypermethylation of RASSF1A within promoter CpG islands is frequently observed in the HNSCC cell lines 22 . All of these findings indicate that RASSF1A might play an important role in the development of HNSCC.
To date, a number of studies have investigated the association between aberrant methylation of RASSF1A and HNSCC through a comparison of the methylation prevalence of RASSF1A between cancerous tissues and controls. However, the obtained results of these studies are inconclusive and inconsistent 23,24 . Therefore, we conducted a meta-analysis of 12 published studies to conclude the association.

Results
Study characteristics. In total, the electronic search strategy initially identified 112 potentially relevant studies. Firstly, these potentially relevant studies were screened for inclusion based on their titles and abstracts. As a result, 19 duplications and 76 studies (four thesis, one conference proceeding, eight reviews, two animal studies, five cell lines, 49 not about HNSCC, six without RASSF1A and one without full text) were excluded. The remaining 17 citations were retrieved for full-text assessment. Upon the assessment, two articles which were not case-control studies and three articles with inadequate RASSF1A methylation data were excluded. Figure 1 showed the whole process of study selection and exclusion, with specification of reasons. Lastly, 12 studies, published between 2002 and 2012 with 18 to 111 cases, met the inclusion criteria and were included in our meta-analysis. The individual characteristics of the 12 included studies are summarized in Table 1.
The meta-analysis consisted of 550 cases of HNSCC tissues and 404 controls, with a total sample size of 954. Among the 12 included studies, the study populations were Caucasians in eight articles 22,[25][26][27][28][29][30][31] and Asians in four articles 23,24,32,33 . A total of nine studies conducted methylation-specific polymerase chain reaction (MSP) to assess the gene methylation status. Three articles used quantitative methylation-specific polymerase chain reaction (Q-MSP), bisulfite sequencing PCR (BSP) and methylation sensitive restriction analysis (MSRA) respectively to evaluate the RASSF1A methylation in cases and controls. The genomic location of the analyzed regions of eight studies included was the promoter. The genomic location of the analyzed regions of the remaining four articles was the CpG islands of the promoter. Hogg 34 analyzed the methylation status of CpG islands in the promoter region of RASSF1A from LCTSGR1 at 3p21.3 in the HNSCC patients. 11 of the articles were published in English, and 1 was published in Chinese. The specimens were cancerous tissues of HNSCC cases and non-cancerous tissues of controls. The control group was comprised of HNSCC patients, benign disease patients and healthy volunteers.
Meta-analysis results. The pooled ORs and corresponding ORs 95% CIs for the association between RASSF1A promoter methylation and HNSCC were shown in Fig. 2. A random-effects model was employed because a significant heterogeneity was observed among 12 included studies by the χ 2 -based Cochran Q statistic test and I 2 statistics (I 2 = 46.7%, Q = 20.65, P = 0.0372). In the overall meta-analysis, the RASSF1A promoter

Meta-regression analysis and subgroup analysis.
A significant heterogeneity was found among the studies. Therefore we conducted a meta-regression to explore the source of heterogeneity with restricted maximum likelihood method (REML method 35 ). Based on previous studies, we assumed that the heterogeneity might arise from the ethnicity, control types, age of patients, RASSF1A methylation detection methods, case sample size, HPV infection status, gender proportion smoking status and histology types. However, only the data about ethnicity, control types, methods of RASSF1A methylation detection and case sample size were collected completely. Then, we conducted a multiple regression model with following four variables: races (Asians and Caucasians), control types (autologous control heterogeneous control), methods used to RASSF1A methylation detection (MSP, Q-MSP and BSP) and case sample size (≥40 and < 40). According to the result of the meta-regression analysis result, all 95% confidence intervals included 0 for the coefficients which indicated that none of the variables can explain the heterogeneity between-studies in Table 2. Furthermore, we performed a subgroup analysis of those variables in Table 3   Publication Bias. We performed a Begg's funnel plot 36 and Peter test 37 to assess the publication bias of the included studies. The shape of the Begg's funnel plot showed no obvious asymmetry and absence of symmetry indicated publication bias (Fig. 4). No publication bias was detected by Peter test (P = 0.73) and Begg's rank correlation test 36 (P = 0.87), respectively. Furthermore, the fail-safe number 38

Discussion
Previous studies have demonstrated that epigenetic alteration is an important event in the carcinogenic progression. Particularly, increased methylation in the promoter region of tumor suppressor gene can account for a progressive reduction of its expression, silencing and selective proliferative advantage in certain cells, which plays a vital role in the development of human cancer 39 . The aberrant methylation has been observed in the promoter region of RASSF1A in various cancers, including HNSCC 6 .  Our meta-analysis included 550 HNSCC tissues and 404 controls from 12 published studies. Overall, the pooled OR of RASSF1A methylation in cancer tissues and controls under the random-effects model was 2.93 (95% CI: 1.58-5.46), which suggested a significant association of the methylation of RASSF1A promoter with HNSCC. The overall heterogeneity between included studies was interpreted by the χ 2 -based Cochran Q statistic test and I 2 statistics and meta-regression was used to explore the sources of the heterogeneity. When I 2 > 50% and P < 0.1 for the Q statistic, the between-study heterogeneity was considered significant and the pooled ORs was calculated using a random-effects model (the DerSimonian-Laird estimate) 40 . Otherwise, a fixed-effects model (the Mantel-Haenszel test) was applied 40 . In the subgroups of races, the pooled ORs were 3.45 (95% CI: 1.85-6.44) in Caucasians subgroup under the fixed-effects model and 2.75 (95% CI: 0.96-7.89) in Asians subgroup under the random-effects model, respectively. This indicated that hypermethylation of RASSF1A had a stronger association with increased risk of HNSCC in Caucasians. Similarly, the methylation rates of the MGMT gene and GSTP1 gene in non-small cell lung cancer were also significantly higher in Caucasians than in Asians 41 , and this divergence might be due in large part to a combination of differences in allele frequencies and complex epistasis or interactions between the gene and environment 42 . The summary OR was 3.19 (95% CI: 1.30-7.83) in the autologous control subgroup under the random-effects model, and was 2.52 (95% CI: 1. 35-4.73) in the heterogeneous control subgroup under the fixed-effects model. Interestingly, this was consistent with a previous study of non-small cell lung cancer (NSCLC) 43 , which indicated an increased likelihood of RASSF1A methylation in heterogeneous controls compared to autologous controls. The reason for this might be because the benign lesions had a higher probability of RASSF1A methylation as an early stage of carcinoma. In the method subgroup, the ORs were 10.76 (95% CI: 0.56-211.78) in the BSP group, 1.06 (95% CI: 0.38-2.94) in the Q-MSP group, and 3.30  (95%CI: 2.17-5.01) in the MSP group under the fixed-effects model. The differences of ORs in these subgroups were potentially caused by the different sensitivities and specificities of the method used to the detection of gene methylation. Q-MSP is a sensitive quantitative assay with normalization of the amplifiable DNA content of samples. The cut-off point of Q-MSP was derived from the best distinguish point. However, the cut-off point of MSP is defined by visual detection of the presence or absence of PCR product compared to the intensity of controls 44 . Therefore, MSP (a nonquantitative and nonfluorometric method) would be hard to detect low levels of promoter methylation, while Q-MSP can detect up to 1/1000 methylated alleles 45 and this would have an impact on the results. BSP, a method of genomic sequencing, can provide a more direct and quantitative analysis for most CPG sites within a defined region than MSP and Q-MSP 46 . The 95% confidence intervals of ORs of BSP method subgroup and Q-MSP method subgroup included 1. This was potentially attributed to the relatively small sample size (less than 60) in these subgroups. The pooled OR also differed according to different sample size. In the < 40 cases subgroup, the OR was 2.30 (95% CI: 1. 25-4.21) under the fixed-effects model. In the ≥40 cases subgroup, the OR 3.87 (95% CI: 1. 46-10.25) under the random-effects model. However, there was no significant difference between different sample size subgroups.
Some potential limitations of the study should be taken into consideration when interpreting the results of meta-analysis. Firstly, due to the 12 included studies were retrospective, there might be a potential unidentified confounding bias, information bias and selection bias. Secondly, although we explored and evaluated the source of heterogeneity in four variables, we could not explore heterogeneity from other aspects because of the insufficient demographic and clinical data. Thirdly, previous studies demonstrated that time of sampling 47 and fixation techniques 48 potentially influenced methylation status in paraffin-embedded tumors. The 12 included studies varied in time of sampling and fixation techniques and these could result in heterogeneity. Additionally, since the number of included studies and samples were relatively small, further investigations with a large number of samples were required.

Conclusions
Our meta-analysis identified an association between aberrant rmethylation of RASSF1A promoter with HNSCC, which indicated that hypermethylation of RASSF1A promoter might be a potential biomarker in the process of HNSCC. Prospective studies with larger sample size are needed to confirm these results in the future.

Methods
The meta-analysis was performed according to the latest meta-analysis guidelines (PRISMA).

Studies identification.
Studies were identified via an electronic search of a range of computerized databases, including PubMed, Embase, Web of Science and CNKI using the following key words: 'squamous cell carcinoma or cancer' , 'oropharyngeal or oropharynx or head and neck or tonsil' , 'RAS association domain family protein 1A' , 'RASSF1A' , 'methylation' and 'hypermethylation' . Articles were searched in the databases form Jan 1, 2000 to May 8, 2015 without language limitation. Two independent reviewers screened the titles and abstracts identified by the electronic search to identify relevant studies. The inclusion criteria of the meta-analysis were as follows: (1) case-control study design; (2) presentation of data necessary for calculating odds ratios (ORs); (3) studies primarily evaluating the incidence of RASSF1A methylation in HNSCCs and corresponding autologous/heterogeneous control, including non-tumor tissue, plasma and sputum of HNSCC patients. The excluded studies were as follows: duplication, review, animal study, experimental study and adequate specific data.
Data extraction. Data retrieved from the eligible studies including first author's name, year of publication, published journal and country, patient ethnicity, population size, methods used to determine methylation status, histology, control type, and methylation status of RASSF1A promoter in extracted cancer tissues and controls. Data extraction was conducted by two reviewers independently using a standard data extraction form. If there were disagreement between them, a third reviewer was used to reach a consensus. Statistical analysis. All statistical analyses were conducted by using the Meta package (version 2.2-1) in R (version 3.0.2; http://www.r-project.org/). The pooled odds ratios (ORs) of different studies and corresponding 95% confidence intervals (CIs) were calculated to evaluate the strength of the association between RASSF1A methylation and HNSCC risk. In order to assess the percentage of variability across studies attributable to heterogeneity beyond by sampling error, the χ 2 -based Cochran Q statistic test and I 2 statistics were employed. When I 2 > 50% and P < 0.1 for the Q statistic, the between-study heterogeneity was considered significant and the pooled ORs was calculated using a random-effects model (the DerSimonian-Laird estimate) 40 . Otherwise, a fixed-effects model (the Mantel-Haenszel test) was applied 40 . If the heterogeneity was significant, to explore and assess the source of heterogeneity, a meta-regression (restricted maximum-likelihood estimator method, REML 35 ) was initially performed and a subgroup analysis was followed according to the results of meta-regression. Sensitivity analysis was employed to assess the effects of single study on pooled ORs after omitting one study. Publication bias was assessed by a funnel plot for Egger's test. When the individual studies had cells with zero counts, the default was to add 0.5 to all zero counts in the Meta package. Statistical significance was defined as a two-tailed P value of 0.05 in our study.