Androgen receptor gene polymorphisms and risk of prostate cancer: a meta-analysis

Although the association between CAG and GGN repeats in the androgen receptor gene and prostate cancer risk has been widely studied, it remains controversial from previous meta-analyses and narrative reviews. Therefore, we performed this meta-analysis to provide more precise estimates with sufficient power. A total of 51 publications with 61 studies for CAG repeats and 14 publications with 16 studies for GGN repeats were identified in the meta-analysis. The results showed that short CAG repeats (<22 repeats) carriers presented an elevated risk of prostate cancer than long CAG repeats (≥22) carriers (OR = 1.31, 95% CI 1.16 to 1.47). Prostate cancer cases presented an average fewer CAG repeats (MD = −0.85, 95% CI −1.28 to −0.42) than controls. Short GGN repeats (≤16) carriers presented an increased risk of prostate cancer than long GGN repeats (>16) carriers (OR = 1.38, 95% CI 1.05 to 1.82). In subgroup analyses, the abovementioned significant association was predominantly observed in Caucasian populations. The meta-analysis showed that short CAG and GGN repeats in androgen receptor gene were associated with increased risk of prostate cancer, especially in Caucasians.

a more precise and comprehensive result for the relationship between CAG and GGN repeat polymorphisms of androgen receptor gene and prostate cancer susceptibility.

Methods
Eligible criteria. For inclusion in this meta-analysis, the publication had to meet the following eligible criteria: (1) the exposure was androgen receptor gene CAG and GGN repeat polymorphisms; (2) populations were men with prostate cancer (cases) without prostate cancer (controls); (3) the outcome was incident of prostate cancer; (4) the study design was retrospective or prospective (i.e. nested) case-control study; (5) study provided distribution of genotype, odds ratio (OR) and corresponding 95% confidence interval (CI), mean difference (MD) and corresponding standard error (SE), and mean repeats in case and control groups with related SE. For duplicated publication, we included the most recent or that providing the most information. If one publication provided different groups of ethnicity, we considered the each group as a separate study. We conducted the meta-analysis according to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement in reporting meta-analysis. The protocol (registration number: CRD42016036971) of the meta-analysis was published in the PROSPERO register (http://www.crd.york.ac.uk/PROSPERO/).

Search strategy. A comprehensively literature search was performed in PubMed, Embase, CBM, CNKI and
Wanfang databases up to March, 2016, without restriction to regions, publication types, or languages. The search strategy was as following: ("polymorphism" AND "prostate cancer" AND "androgen receptor"). In addition, references in the recent reviews or meta-analysis and included articles were identified for any further potential related studies. Data extraction. Data from the included studies were extracted and summarized independently by two authors (HW and XT-Z). Any disagreement was resolved by discussion of which data should be extracted. The following information was extracted: last name of first author, publication year, country of study, ethnicity, study design, control status, sample size, age of the cases and controls, percentage of advanced prostate cancer cases (T3-T4, M0; T0-T4, M1), the repeat cutpoint of polymorphisms, mean number of repeats in case and control groups with related SE, dichotomous data (short versus long repeats), and estimate with corresponding 95% CI (including OR for dichotomous data and continuous data). We defined the long CAG and GGN repeats as ≥ 22 and > 16 repeats as previously published 6 , respectively. Otherwise, < 22 and ≤ 16 were short CAG and GGC repeats, respectively. Statistical analysis. We calculated ORs and 95% CIs for short CAG repeats (< 22) versus long CAG repeats (≥ 22) and short GGN repeats (≤ 16) versus long GGN repeats (> 16) using dichotomous data 6 . We summarized ORs and corresponding 95% CIs for per decrement of CAG and GGN repeats. We also summarized the MDs in number of repeats between cases and controls. In this meta-analysis, all pooled analyses were performed with random-effects model using the method of DerSimonian-Laird, with the estimate of heterogeneity being taken from the from the Mantel-Haenszel model. Subgroup analyses were also performed according to ethnicity (Caucasian, Asian, Africa, or Hispanic), study design (prospective, i.e. nested or retrospective case-control study), control status, and histology grade of prostate cancer (localized and advanced). In addition, meta-regression analysis was also performed for interaction of between-group. Sensitivity analysis was performed by removing each study at a time. Publication bias was detected using contour-enhanced funnel plot and Egger's linear regression method. Statistical analysis was performed using Stata 12.0 software. A two-sided P value of 0.05 was used, except for heterogeneity test (0.1).

Study characteristics.
A total of 717 relevant publications were identified from the electronic literature search. The PRISMA flow diagram was presented in Fig. 1, which shows the detail of inclusion and exclusion of studies. Ultimately, 51 publications  were included in the meta-analysis, in which 51 publications 16 Association between CAG repeats polymorphism and prostate cancer risk. Fifty-one case-control studies conveyed data on the short versus long CAG repeats. The pooled analysis showed that men with short CAG repeats carried higher risk of prostate cancer than long CAG repeats (OR = 1.31, 95% CI 1.16 to 1.47; I 2 = 74.9%, P for heterogeneity < 0.01; Fig. 4). Thirty-three case-control studies presented the data for per one CAG decrement and the summarized OR was 1.04 (95% CI 1.02 to 1.07; I 2 = 83.4%, P for heterogeneity < 0.01; Fig. 5) for men with per one CAG decrement. The aggregated analysis suggested that prostate cancer cases seemed to have on average 0.85 fewer CAG repeat length than controls (MD = − 0.85, 95% CI − 1.28 to − 0.42; I 2 = 88.7%, P for heterogeneity < 0.01; Fig. 6) with 23 case-control studies.
Association between GGN repeats polymorphism and prostate cancer risk. Sixteen case-control studies provided data on the short versus long GGN repeats. The pooled results showed that men with short GGN repeats carried higher risk of prostate cancer than long GGN repeats (OR = 1.38, 95% CI 1.05 to 1.82; I 2 = 69.1%, P for heterogeneity < 0.01; Fig. 7). Six case-control studies presented the data for per one GGN decrement and the summarized OR was 0.99 (95% CI 0.95 to 1.03; I 2 = 0.0%, P for heterogeneity = 0.93; Fig. 8) per GGN. The summarized MD of GGN repeats showed no significant difference between the prostate cancer cases and controls (MD = 0.05, 95% CI − 0.09 to 0.18; I 2 = 0.0%, P for heterogeneity = 0.95; Fig. 9) with six case-control studies.

Haplotype analysis of CAG and GGN repeat polymorphisms.
Six case-control studies provided data for haplotype analysis. The estimated ORs were 2.06 (95% CI 1.29 to 3.29; I 2 = 69.3%, P for heterogeneity = 0.006), 1.79 (95% CI 1.08 to 2.96; I 2 = 75.8%, P for heterogeneity = 0.001), and 1.21 (95% CI 0.94 to 1.56; I 2 = 0, P for heterogeneity Subgroup, meta-regression and sensitivity analysis. Subgroup analyses were conducted according to ethnicity, study design, control status, and histology grade of prostate cancer. The results of subgroup analyses showed that the elevated risk of prostate cancer in both CAG and GGN repeat polymorphisms were more predominant among Caucasian populations (Tables 2, 3 and 4) and the increased risk of prostate cancer of long GGN repeats were more predominant in advanced prostate cancer cases (Fig. 11). Meta-regression analysis did not    3 and 4). Subgroup analysis showed that the result of CAG repeat length and risk of prostate cancer was robust (Fig. 12).
Publication bias. Publication bias was detected using contour-enhanced funnel plot and Egger's linear regression method. Contour-enhanced funnel plots showed that publication bias might exist for the short versus long CAG repeat polymorphism (Fig. 13) and no publication bias existed in the short versus long GGN repeat polymorphism (Fig. 14). Egger's linear regression method supported the aforementioned conclusion (P = 0.004 for CAG repeats; P = 0.07 for GGN repeats).

Discussion
The present meta-analysis summarizes the evidence to date regarding the association between CAG and GGN repeat polymorphisms of androgen receptor and the risk of prostate cancer. The results suggested that short CAG and GGN repeats in the androgen receptor gene were associated with increased risk of prostate cancer, especially in Caucasians.   The short CAG repeats (< 22) and short GGN repeats (≤ 16) carry a roughly 1.31-and 1.38-fold higher risk of developing prostate cancer compared with subjects with long CAG (≥ 22) repeats and long GGN repeats (> 16), respectively. Each decrement in CAG repeat presented 1.04-fold higher risk of developing prostate cancer. Prostate cancer cases presented an average 0.85 fewer CAG repeats than controls. In Caucasians, the aforementioned elevated risk was increased. This could be due to that more studies conducted in Caucasians, which provided greater statistical power for detecting small gene effect. Specifically, the prostate cancer cases in Caucasian population carried an average 1.09 fewer CAG repeats than controls. This difference might yield certain measurable biological impact in prostate carcinogenesis, such as early diagnosis and gene therapy.   An interaction between CAG and GGN repeat polymorphisms in increasing the prostate cancer susceptibility was documented by our meta-analysis. Haplotype analysis showed that short CAG and short GGN repeats carriers presented 2.06-fold higher risk of developing prostate cancer compared with long CAG and long GGN      repeats carriers. Moreover, the short CAG repeats and long GGN repeats carriers presented 1.79-fold higher risk of developing prostate cancer compared with long CAG and long GGN repeats carriers.
In 2004, Zeegers et al. 6 published the first meta-analysis regarding the association between CAG and GGN repeat length polymorphisms in the androgen receptor gene and prostate cancer risk, in which included 23 articles with 19 retrospective case-control studies and 5 prospective case-control studies, comprising a total of 4274 cases and 5275 controls. They found that the presence of shorter repeats seemed to be modestly associated with prostate cancer risk. However, they did not found any significant difference in number of repeats between cases and controls. In 2012, Gu et al. 10 aggregated 27 articles to evaluate the relationship between CAG repeat polymorphism and prostate cancer risk. Their meta-analysis demonstrated that the CAG repeat polymorphism in androgen receptor gene with more than 20 repeats might confer a protective effect among the prostate cancer cases among men 45 years or older only. In 2013, Sun et al. 11 carried out another meta-analysis regarding the association between CAG repeat polymorphism and prostate cancer risk, which included 47 studies with 13 346 cases and 15 172 controls. They suggested that a short CAG repeat polymorphism might increase the risk of prostate cancer compared with the longer CAG repeat, especially in Caucasians and Asians. Compared with the previous meta-analysis 6,10,11 , our meta-analysis was more comprehensively searched and our meta-analysis included 51 case-control studies (14 803 cases and 18 888 controls) for CAG repeats and 16 case-control studies (2986 cases and 3705 controls) for GGN repeats. In addition, our meta-analysis performed haplotype analysis and suggested that there exists an interaction between CAG and GGN repeat polymorphisms in increasing the prostate cancer susceptibility. Moreover, we found a significant difference in number of CAG repeat length between cases and controls, and the absolute difference in more than 1 repeat in Caucasians.
The present retrospective analysis has some limitations. First, the evidence of between study heterogeneity was apparent, and the heterogeneity might distort the conclusion of the current meta-analysis [67][68][69] . Additionally, the meta-regression analysis failed to identify the source of heterogeneity. Second, the standard of cutpoint of repeat length polymorphisms varied in different studies. This might in part contribute to the between study heterogeneity. Third, the screening policy of prostate cancer also varies between countries. Especially in United States, the prostate-specific antigen screening of the general population is more commonly used than other countries 6 . These different screening policies might also be responsible for the between study heterogeneity. Fourth, the publication bias was detected in the present meta-analysis for the association between CAG repeat polymorphism and risk of prostate cancer. The existing publication bias indicated that certain studies with negative results for the association between CAG repeat polymorphism and prostate cancer risk are under-represented in the literature. The  publication bias also might distort the conclusion of the present meta-analysis. Ultimately, the meta-analysis is a secondary analysis; therefore, we could not handle the problem of between study heterogeneity.
In summary, our meta-analysis indicated that short CAG and GGN repeats in androgen receptor gene were associated with increased risk of prostate cancer, especially in Caucasians.