Association between XRCC1polymorphism 399 G->A and glioma among Caucasians: a systematic review and meta-analysis

A systematic literature search of the PubMed, EMBASE and Scopus databases was performed on October 25^th, 2011 using search terminology [(XRCC1 OR x-ray cross complementing group 1) AND (glioma OR brain tumor)]. This was supplemented by a query of the HuGE Literature Finder based on a search of (glioma OR brain tumor) for gene XRCC1. Reference lists of identified studies were reviewed in order to identify additional relevant articles.

Included studies were required to meet all of the following conditions: 1) study employed a case–control design of human subjects; 2) outcome was microscopically-confirmed malignant glioma; and 3) data were presented on the genotype counts of cases and controls for the 399 G → A polymorphism in XRCC1. No language restriction was implemented, and corresponding authors were contacted in an attempt to obtain unreported genotype counts if studies were otherwise eligible. Genotype counts for cases and controls for the study by Liu et al. [22] were provided.

The following data were extracted from each eligible article: last name of first author, date of publication, country in which study was performed, ethnicity of subjects, genotype counts of cases and controls, genotyping method and quality control, glioma subtype (grade III and/or grade IV), and age and gender composition of subjects. There was no blinding to the names of authors, journals, institutions, or funding sources.

Deviation from Hardy-Weinberg equilibrium in control subjects was tested using a chi-squared goodness-of-fit test (P < 0.05 was considered significant). Assuming a fixed effects model, the inverse variance method was used to estimate separate genotype-specific summary odds ratios (OR) and 95% confidence intervals for the association between 399 G → A and risk of glioma. A fixed effects model was used due to the exhaustive nature of the literature review and a detailed exploration of possible sources of heterogeneity, as opposed to a random effects model which is not a solution for heterogeneity and assumes that identified studies are a subset of a larger pool of studies. Three pooled ORs were calculated: AA versus GG (OR₁), AG versus GG (OR₂), and AA versus AG (OR₃). The resulting ORs were used to determine the most appropriate genetic model using a previously described approach [23]:

After determination of the most appropriate genetic model, heterogeneity across studies was assessed using the I ² statistic [24]. To explore potential sources of heterogeneity, subgroup analyses were conducted according to gender, glioma subtype (Grade III vs. Grade IV), country (US vs. other), study size (<500 vs. ≥500), and genotyping method (PCR-RFLP vs. other). Influence analyses were also conducted, in which pooled estimates were calculated after omission of one study at a time to identify studies excessively influencing the summary estimate. Finally, publication bias was qualitatively assessed through use of a funnel plot of the log(standard error of the effect estimates) versus the effect estimates. Analyses were conducted using Review Manager Version 5.1.4 (Cochrane Collaboration, Oxford, UK) and R Version 2.13.1.

The initial database search yielded 25 articles, and four additional articles were retrieved from a search of the reference lists of the originally identified articles. Screening of the titles and abstracts of these 29 articles for study eligibility led to the initial exclusion of 18 articles which did not meet eligibility requirements, leaving 11 full articles for review [22, 25‐34]. Of these articles, four were excluded due to not reporting on a case control study [28], providing duplicate data [31, 33], and reporting the incorrect phenotype [29]. This systematic review resulted in the identification of seven eligible studies (Figure 1).

Of the seven studies retained for review, three were conducted in the United States [22, 27, 32], two in Turkey [25, 34], one in Finland [30], and one in Brazil [26]. All studies reported on a mix of Grade III and Grade IV gliomas among exclusively Caucasian populations, based on DNA extracted from collected blood samples. Genotyping was conducted using restriction fragment length polymorphism PCR (PCR-RFLP) for all studies except Rajaraman et al. [32] which employed real-time TaqMan PCR, and Liu et al. [22] which used an array based approach. All control groups were in Hardy Weinberg equilibrium with the exception of that assembled by Custódio et al. [26] which deviated considerably (P < 0.001) and was consequently excluded from the meta-analysis. Thus, six studies [22, 25, 27, 30, 32, 34] were included in the final meta-analysis with data on 2,362 unique cases and 3,085 controls (Table 1).

For the association between variants in 399 G → A and glioma, genotype-specific odds ratios OR₁, OR₂, and OR₃ were 1.33 (95% CI 1.12-1.58, P = 0.001), 1.14 (95% CI 1.01-1.28, P = 0.03), and 1.16 (95% CI 0.98-1.37, P = 0.09), respectively. As these genotype-specific odds ratios are most suggestive of a dominant model, genotypes AG and AA were collapsed and the studies were meta-analyzed using a dominant model (AA + AG versus GG).

Results of the overall meta-analysis are presented in Figure 2. The pooled odds ratio among Caucasians for the association between 399 G → A AG + AA genotypes, as compared to GG genotype, and risk of glioma was statistically significant (OR = 1.17, 95% CI 1.05-1.30, P = 0.006), with indication of high among-study heterogeneity (I ²  = 87%). Data were available to explore potential sources of this heterogeneity through stratification by gender, glioma subtype (Grade III vs. Grade IV), study country (US vs. other), genotyping method (PCR-RFLP vs. other), and study size (n < 500 vs. n ≥ 500), and results of these analyses are presented in Table 2. A statistically significant association was observed among females (OR = 2.27, 95% CI 1.45-3.57, P < 0.001), but this stratification was only feasible for one study [22] based on only 158 female cases. Furthermore, although significant associations were detected for studies with small sample size (OR = 3.09, 95% CI 2.06-4.65, P < 0.001), studies conducted outside of the US (OR = 1.22, 95% CI 1.05-1.42, P = 0.009), and studies employing PCR-RFLP genotyping (OR = 1.16, 95% CI 1.01-1.32, p = 0.03), it was clear that these associations were being driven by results of the study by Yosunkaya et al. [34]. An influence analysis demonstrated that there was no longer a statistically significant association between the XRCC1 399 G → A polymorphism and glioma after removing this study from the meta-analysis (OR = 1.10, 95% CI 0.98-1.23, P = 0.10), and there were no significant associations detected upon repeating previous subgroup analyses except for that among females as described above (Table 3). A funnel plot of the remaining studies did not indicate any evidence of publication bias (available upon request).