XPC Lys939Gln polymorphism contributes to colorectal cancer susceptibility: evidence from a meta-analysis

We searched Pubmed, Embase and Cochrane library databases for all articles on the association between XPC Lys939Gln polymorphism and CRC risk using the following combined keywords: ‘xeroderma pigmentosum group C’, ‘XPC’, ‘colon cancer’, ‘rectal cancer’ and ‘colorectal cancer’. The latest search was done in December 2013, without any language restriction. Additional articles were identified through the references cited in the first series of articles selected. Articles included in the meta-analysis were in any language, with human subjects, published in the primary literature and had no obvious overlap of subjects with other studies. Among overlapping reports, only the studies with more information on origin of cases/controls were retained. The study was performed according to the proposal of Meta-analysis of Observational Studies in Epidemiology group (MOOSE) [19].

Two authors (Qiliu Peng and Xianjun Lao) independently reviewed and extracted data from all eligible studies. Data extracted included the first author, year of publication, country of origin, ethnicity, genotyping method, matching criteria, source of control, CRC ascertainment, total numbers of cases and controls and genotype frequencies of cases and controls. Ethnic backgrounds were categorized as Caucasian, and Asian. Smoking status (smoker or nonsmoker) was additionally recorded for stratified analysis. Smokers included current smokers and former smokers. Nonsmokers had never smoked. Cancer location was divided into colon cancer and rectum cancer and was also additionally recorded for the stratified analysis. To ensure the accuracy of the extracted information, the two authors checked the data extraction results and reached consensus on all of the data extracted. If different results were generated, they would check the data again and have a discussion to come to an agreement. A third reviewer (Weizhong Tang) was invited to the discussion if disagreement still existed.

The quality of eligible studies was evaluated independently by two authors (Qiliu Peng and Xue Qin) according to a set of predefined criteria (Table 1) based on the scale of Thakkinstian et al. [20]. The revised criteria cover the representativeness of cases, source of controls, ascertainment of CRC, total sample size, quality control of genotyping methods, and Hardy-Weinberg equilibrium (HWE) in the control population. Disagreements were resolved by consensus. Scores ranged from 0 (lowest) to 10 (highest). Articles with scores equal to or less than 6 were considered “low-quality” studies, whereas those with scores higher than 6 were considered “high-quality” studies.

Crude odds ratios (ORs) with 95% confidence intervals (CIs) were used to assess the association between the XPC Lys939Gln polymorphism and CRC risk. We evaluated the XPC Lys939Gln polymorphism and CRC risk using co-dominants (Gln/Gln vs. Lys/Lys and Gln/lys vs. Lys/Lys), recessive model (Gln/Gln vs. Gln/lys + Lys/Lys), and dominant model (Gln/Gln + Gln/lys vs. Lys/Lys). The Chi-square-based Q statistic test [21, 22] was used to evaluate the between-study heterogeneity. If the result of the heterogeneity test was P _Q  < 0.10, the pooled ORs were analyzed using the random-effects model [23]. Otherwise, the fixed-effects model [24] was selected. Subgroup analyses were performed by ethnicity, source of control, cancer location, smoking, and study quality. Sensitivity analysis was conducted by sequential omission of individual study to assess the robustness of the results. Publication bias was assessed using a Begg’s funnel plot and Egger’s regression asymmetry test [25]. If publication bias existed, the Duval and Tweedie non-parametric “trim and fill” method was used to adjust for it [26]. The distribution of the genotypes in the control population was tested for HWE using a goodness-of-fit Chi-square test. All analyses were performed using Stata software, version 12.0 (Stata Corp., College Station, TX). All P values were two-sided. To ensure the reliability and the accuracy of the results, two authors entered the data into the statistical software programs independently with the same results.

Based on the search criteria, ten studies investigating the XPC Lys939Gln polymorphism and CRC susceptibility were identified. Two of these articles were excluded because they did not present sufficient data for calculating OR and 95% CI [27, 28]. Manual search of references cited in the eligible studies did not reveal any additional articles. As a result, a total of 8 relevant studies containing 3,301 cases and 4,177 controls were included in the meta-analysis [29‐36] (Additional file 1: Figure S1). The main characteristics of these studies were listed in Table 2. Among these publications, four studies were conducted in Caucasian descent [29, 30, 33, 36], and four were conducted in Asian descent [31, 32, 34, 35]. Two were population–based studies [34, 36] and six were hospital–based studies [29‐33, 35]. Three of these studies presented XPC Lys939Gln polymorphism genotype distributions according to smoking status (smokers and nonsmokers). The cases were histologically or pathologically confirmed as CRC in five studies [29, 31, 32, 34, 35]. Controls were mainly healthy or hospital-based populations and matched with age and gender. The genotype distributions of the controls in all of the included studies were consistent with HWE.

As shown in Table 3, we found that the XPC Lys939Gln polymorphism was significantly correlated with increased CRC risk when all studies were pooled into the meta-analysis (Gln/lys vs. Lys/Lys: OR = 1.293, 95% CI 1.169–1.430, P = 0.000; Gln/Gln + Gln/lys vs. Lys/Lys: OR = 1.260, 95% CI 1.145–1.388, P = 0.000). In subgroup analysis by ethnicity, significant increased CRC risk was found in Asian populations (Gln/lys vs. Lys/Lys: OR = 1.345, 95% CI 1.187–1.523, P = 0.000, Figure 1; Gln/Gln + Gln/lys vs. Lys/Lys: OR = 1.317, 95% CI 1.170–1.484, P = 0.000, Figure 2), but not in Caucasian populations. In stratified analysis according to study quality, significant increased CRC risk was found in high quality studies (Gln/lys vs. Lys/Lys: OR = 1.290, 95% CI 1.148–1.450, P = 0.000; Gln/Gln + Gln/lys vs. Lys/Lys: OR = 1.248, 95% CI 1.117–1.395, P = 0.000), but not in low quality studies. In subgroup analysis by smoking status, significant increased CRC risk was observed in nonsmokers (Gln/Gln + Gln/lys vs. Lys/Lys: OR = 1.286, 95% CI 1.020–1.622, P = 0.033), but not in smokers. In subgroup analysis according to source of control, significant increased CRC risk was found in both hospital-based studies (Gln/lys vs. Lys/Lys: OR = 1.335, 95% CI 1.182–1.507, P = 0.000; Gln/Gln + Gln/lys vs. Lys/Lys: OR = 1.282, 95% CI 1.142–1.439, P = 0.000) and population-based studies (Gln/lys vs. Lys/Lys: OR = 1.204, 95% CI 1.003–1.444, P = 0.046; Gln/Gln + Gln/lys vs. Lys/Lys: OR = 1.213, 95% CI 1.020–1.443, P = 0.029). However, in subgroup analysis according to cancer location, statistical significant association was not detected in both colon cancer patients and rectum cancer subjects.

Heterogeneity between studies was estimated using the Chi-square-based Q test and the significance of which was set at P _Q  < 0.10. There was no statistical significant heterogeneity among studies when all eligible studies were pooled into the meta-analysis. In subgroup analyses according to ethnicity, source of control, cancer location, smoking, and study quality, statistical significant heterogeneity was not observed in all subgroups (Table 3).

Sensitivity analysis was performed by sequential omission of individual studies. For analyses of pooling more than three individual studies, the significance of the pooled ORs was not influenced excessively by omitting any single study (Figure 3), indicating that our results were statistical robust.

Begg’s funnel plot and Egger’s test were performed to assess the publication bias of literatures in all comparison models. The shape of the funnel plot did not reveal any evidence of obvious asymmetry (Figure 4). Then, the Egger’s test was used to provide statistical evidence of funnel plot symmetry. All the p values of Egger’s tests were more than 0.05 (P = 0.660 for GlnGln vs. LysLys; P = 0.584 for Glnlys vs. LysLys; P = 0.670 for dominant model GlnGln + Glnlys vs. LysLys; and P = 0.627 for recessive model GlnGln vs. Glnlys + LysLys), providing statistical evidence for the funnel plots’ symmetry. The results suggested that publication bias was not evident in this meta-analysis.