Association between aberrant APC promoter methylation and breast cancer pathogenesis: a meta-analysis of 35 observational studies

Search strategy

Eligible studies were identified by searching the following databases until February 2016: the Cochrane Library, PubMed, Web of Science and Embase. No language restrictions or lower data limits were imposed; only abstracts, unpublished and incomplete studies were excluded. Titles, abstracts of potential references and reference lists from relevant studies were carefully checked. We performed the search strategy using the following search terms and their various combinations: “APC,” “Adenomatous polyposis coli,” “methylation,” “breast cancer,” “breast neoplasm” and “mammary carcinoma.”

Selection criteria

The studies included in the present meta-analysis addressed the association between APC methylation and increased BC risk. Our inclusion criteria were as follows: (1) provided sufficient data on the frequency of APC methylation in BC patients and controls; (2) original observational studies in full-text form; and (3) when several studies overlapped, the most recent or large-scale article was selected. The following were exclusion criteria: (1) data based on reviews, animal models, case reports or cell line studies; (2) studies lacking key information necessary for calculations; (3) duplicated studies; and (4) studies including BC patients or controls who underwent radiotherapy and chemotherapy which may influence APC promoter methylation levels.

Data extraction

The relevant data were extracted from the eligible studies independently by two authors (D Zhou and WW Tang). Differing opinions, if any, were resolved by discussion in accordance with the original literature. The following information was extracted in a predefined table: the name of the first author, the year of publication, the country of origin, the sample type, the experimental methods used to detect APC methylation, sample size, tumor stage, tumor grade and APC methylation frequencies. Additionally, we classified stage 0, I and II as early-stage BC and stage III and IV as late-stage BC, as confirmed by the AJCC staging system. Furthermore, grades I and II were combined as low-grade BC; grade III was regarded as high-grade BC. This meta-analysis was performed following the statement of preferred reporting items set by the PRISMA Group (File S1) (Moher et al., 2009).

Statistical analysis

All analyses were carried out using Review Manager 5.3 (Cochrane Collaboration) and Stata 12.0 (Stata Corporation) software. Forest plots were designed to estimate relative study-specific effects according to the 95% confidence interval (CI). The association between APC promoter methylation and BC risk was evaluated by calculating the odds ratio (OR) with corresponding 95% CI. For individual studies the OR was represented by a square and the 95% CI by a horizontal line in the centre of the forest plot. The OR and associated 95% CIs in the centre of the forest plot were plotted on a logarithmic scale. When a CI did not include 1.0, a correlation was deemed statistically significant. Heterogeneity between the included studies was quantified through Q-tests based on the chi-square test and I² value. An I² value >50% and a p < 0.10 denoted strong heterogeneity, an I² value = 25–50% denoted a moderate degree of heterogeneity and an I² value <25% or a p > 0.10 denoted mild heterogeneity (Higgins et al., 2003). A random effect model was used when statistical heterogeneity existed among studies (p < 0.1). Otherwise, the fixed effect model was employed (Li et al., 2014). Moreover, the subgroup meta-analyses were also performed according to region, experimental methods for detecting APC methylation, and sample types in order to explore the potential origin of inter-study heterogeneity. In addition, we conducted a sensitivity analysis by removing a single study to examine the stability of the results. The funnel plot, Begg’s test and Egger’s test were investigated in order to determine the degree of publication bias. The treatment effect was plotted against a measure of study size in the funnel plot. When publication bias was present, the shape of the funnel plot was asymmetric. Trim and fill analysis was used to estimate the number of potential missing studies resulting from the asymmetry of the funnel plot.

Study selection and characteristics

The selection process is displayed as a flow chart in Fig. 1 based on the search strategies as previously described. After a careful initial search of the abstracts, 74 potentially relevant articles were identified excluding 1 duplicate and 93 irrelevant studies. Then, we reviewed the full text articles. Among these studies, 43 were excluded (21 articles did not design a control group; 9 articles focused on BC cell lines; 8 articles lacked available data; and 5 articles were reviews). Finally, 31 studies published from 2001 to 2016 involving 35 studies were included in this systematic meta-analysis (PubMed 19, Web of Science 10, Embase 2).

The general characteristics of eligible studies were summarized and displayed in Table 1. A total of 2,483 BC patients and 1,218 controls were employed in multiple countries or regions including Asia (n = 10) (Jin et al., 2001; Jing et al., 2010; Jung et al., 2013; Lee et al., 2004; Liu et al., 2007; Park et al., 2011b; Prasad et al., 2008; Zhang et al., 2007), Europe (n = 13) (Fridrichova et al., 2015; Hoque et al., 2009; Jeronimo et al., 2008; Martins et al., 2011; Matuschek et al., 2010; Muller et al., 2003; Parrella et al., 2004; Rykova et al., 2004; Van der Auwera et al., 2009a; Van der Auwera et al., 2009b; Van der Auwera et al., 2008; Wojdacz et al., 2011b), Africa (n = 2) (Hoque et al., 2006; Swellam et al., 2015), North America (n = 9) (Brooks et al., 2010; Chen et al., 2011; Cho et al., 2010; Dulaimi et al., 2004; Lewis et al., 2005; Shinozaki et al., 2005; Taback et al., 2006; Virmani et al., 2001) and Oceania (n = 1) (Pang et al., 2014). Furthermore, the methylated APC levels in BC patients and controls were examined with 6 methods. Of these methods, methylation specific PCR (MSP) was adopted in 17 studies, quantitative real-time MSP (QMSP) was used in 9 studies, methylation specific-multiplex ligation-dependent probe amplification (MethyLight) was used in 4 studies, methylation specific-multiplex ligation-dependent probe amplification (MS-MLPA) was employed in 2 studies, methylation-sensitive high-resolution melting analysis (MS-HRM) was used in 2 studies and pyrosequencing was used in only 1 study. Furthermore, BC tissues (i.e., fresh frozen tissues, formalin fixed paraffin-embedded tissues and tissues from surgery), samples derived from blood (i.e., blood cells and serum) and needle aspirated fluid (NAF) were enrolled to assess the methylation levels of the APC promoter.

Meta-analysis

The pooled results of this meta-analysis reflected the association between APC promoter methylation and BC pathogenesis (Fig. 2). Due to the existence of significant heterogeneity among the included studies (p < 0.00001, I² = 65%), the random effect model was adopted to evaluate the combined effects of APC promoter methylation. The overall analysis indicated that the frequency of APC promoter methylation was remarkably higher in BC patients than in cancer-free controls. The combined OR for 35 included relevant studies showed that APC methylation was significantly correlated with increased BC risk and the absence of APC expression played an important role in BC pathogenesis (OR = 8.92, 95% CI [5.12–15.52]).

Sensitivity analysis

A sensitivity analysis was conducted by omitting one individual study every time to evaluate the stability of the pooled OR and to choose the heterogeneous study. As shown in Fig. 3, the combined OR between APC methylation and increased BC risk was indeed reliable without heterogeneous studies.

Subgroup analysis

Due to the significant existence of inter-study heterogeneity (p < 0.00001, I² = 65%), subgroup analysis based on region, experimental methods for the detection of APC methylation and sample types were carried out to appraise the sources of the heterogeneity (Table 2). With regard to subgroup analysis based on region, heterogeneity in Asian subgroups disappeared completely (I² = 0%) and the pooled OR value was 24.48 [10.94, 54.74]. The I² value representing heterogeneity in the European and North American subgroups decreased by 50% and 42%, compared with the overall I² value. Furthermore, their OR values also decreased to 4.63 [2.44, 8.78] and 3.79 [1.70, 8.44]. In the African subgroup, the OR was 172.05 [1.76, 16792.96] with higher heterogeneity (I² = 80%) due to the small subset containing only 2 studies. These results indicated that the heterogeneity might result from different regions and APC methylation was remarkably related to increased BC risk without geographical restrictions. For the subgroup analyses based on sample types, the blood or serum group (OR = 9.44, 95% CI [2.56–34.83]) made the largest contribution to the heterogeneity (I² = 78%). In the tissue subgroup, the OR was 9.93 [5.10, 19.34] with lower heterogeneity (I² = 50%). Moreover, heterogeneity in the NAF subgroups could be ignored (I² = 6%). These results confirmed the stable association between APC methylation and BC risk in different sample types. For studies based on the methods used to detect the methylation of the APC promoter, the combined OR value was 18.18 for MSP (95% CI [7.96–41.52]), 3.93 for QMSP (95% CI [1.78–8.69]), 3.29 for MethyLight (95% CI [1.27–8.52]) and 31.81 for MS-MLPA (95% CI [5.30–191.06]). Heterogeneity in the QMSP (I² = 36%) and MS-MLPA (I² = 0%) subgroups was far lower than that of the MethyLight and MS-HRM subgroups (I² = 83%).

To assess the association between APC methylation and tumor stage, 11 studies comprising 681 BC patients were pooled to calculate the OR. The results showed that the frequency of APC promoter methylation was significantly lower in early-stage patients than in late-stage patients (OR = 0.62, 95% CI [0.42–0.93], I² = 34%). Meanwhile, the OR of 8 studies revealed that the association between APC methylation and tumor grade was not statistically significant (OR = 0.78, 95% CI [0.51–1.21], I² = 0%).

Publication bias

We used the funnel plot, Begg’s test and Egger’s test to evaluate the degree of publication bias. The shape of the funnel plot had no obvious asymmetry (Fig. 4A). Moreover, Begg’s test (Pr > |z| = 0.239 > 0.05) suggested no significant publication bias (Fig. 4B). Interestingly, Egger’s test revealed evident statistical proof for the existence of publication bias (p > |t| = 0.000 < 0.05). Therefore, we carried out trim and fill analysis to identify and revise the bias. As shown in Fig. 4C, 12 adjusted studies were added to the initial meta-analysis. The corrected OR was still highly significant for the association between APC methylation and BC risk (OR = 1.444, 95% CI [1.081–1.965]), further proving the stability of our meta-analysis.