Meta-analysis of association between rs1447295 polymorphism and prostate cancer susceptibility

Aims A number of studies have found that the single nucleotide polymorphisms (SNPs) within the 8q24 region of genome were associated with the susceptibility of prostate cancer. Association between 8q24 SNP variant rs1447295 and higher risk of prostate cancer had been investigated, but those studies were incomplete and the conclusions were obscure. Methods To better elucidate the relationship between rs1447295 polymorphism and the susceptibility of prostate cancer, we performed a more comprehensive meta-analysis about the association between rs1447295 polymorphism and prostate cancer susceptibility by collecting relevant articles published up to November, 2016 and excluding many replicated cohort data existing in previous reports, which made the conclusion more reliant and objective. Results The results showed that there was a significant prostate cancer risk associated with rs1447295 polymorphism not only in the total groups, but also in American, European and Asian descent subgroups. Meanwhile, a comprehensive analysis about the association between rs1447295 polymorphism and prostate cancer risk were conducted by using different clinical characteristic stratifications including Gleason score, tumor stage and PSA level. The result showed that rs1447295 polymorphism was correlated with different stages of prostate cancer Conclusions There are strong association between rs1447295 polymorphism and prostate cancer susceptibility in different ethnic groups and different prostate cancer stage, suggesting that rs1447295 might serve as a reliable biomarker for prostate cancer diagnosis.


INTRODUCTION
Prostate cancer (PCa) is the most common cancer and the third leading cause of cancer-related deaths among males in developed countries [1]. Approximately 233,000 new cases were expected to be diagnosed with 29,480 estimated deaths in the USA in 2014 [2]. However, the underlying etiology of PCa is poorly understood. The most recognized factors associated with PCa risk include age, ethnicity, cigarette smoking, and alcohol consumption and so on. In addition, it was suggested that several genetic polymorphisms could influence an individual's susceptibility to PCa [3,4]  Distinguishing the genetic variants that increase the risk of more advanced disease is important for improving regimens for screening, diagnosis, and the treatment of prostate cancer [5,6]. A region on chromosome -8q24 was first shown to confer a PCa risk in a genome-wide linkage scan of 871 Icelandic men [6,7]. Subsequently multiple independent studies with compelling evidence demonstrated that the risk of PCa was influenced by the genetic variations in the region of 8q24 independently [6,7].
One polymorphism in region of 8q24, rs1447295 (A/C, A was considered as risk allele), has been reported a genome-wide association studies (GWAS) association with PCa risk [7,8]. It was reported that rs1447295 polymorphism was associated with greater tumor aggressivenessin African Americans [9] and American Whites [10], advanced stage diseases in Eastern Whites [11] and African American [9,12]. Although there were numerous studies investigating the association between rs1447295 and the risk of PCa, the results were incomplete and cursory. Amundadottir et al. [7] showed that the rs1447295 riskallele tended to be more strongly associated with Gleason score 7-10 tumors than with Gleason score 2-6 tumors in African Americans and American Whites. However, this trend did not appear to translate to Eastern White populations [7,13]. The relationship between PCa risk and this SNP did not appear to be affected by PSA level in Eastern Whites [11,14]. Besides, Zheng et al. [11] showed that this allele was associated with a younger age at onset in Eastern Whites, but Schumacher et al. [15] and Zheng et al. [16] didn't show such evidence. Those contradictory results make it imperative to do a comprehensive analysis to clear up the confusion.
To meet this demand, we did a pilot study to analyze the association of rs1447295 polymorphism and different prostate cancer clinical characteristics, including Gleason score, tumor stage and PSA level. Meanwhile, we also performed a meta-analysis to offer a more comprehensive estimation of the association between rs1447295 and PCa susceptibility in three different ethic subgroups -American, Asian and European descents.

Characteristics of studies
A total of 208 articles were retrieved after the first search, among which 81 articles contained case-control studies targeting at prostate cancer. We removed the articles without the exact quantity information about the genotypes of rs1447295 and those studies with overlapping samples. To simplify the analysis, we restricted our study subjects only as American descent, Asian descent and European descent. Finally, 27 case-control studies from 20 articles were suitable for our meta-analysis as shown in Table 1 [Among these, 12 studies were about American descent, 7 studies were about European descent and 8 studies were about Asian descent]. All the data in these studies were related to association between 8q24 rs1447295 A/C polymorphism and human PCa susceptibility. The flow chart of selecting studies and the reasons for exclusion were presented in Figure 1. Table 1 presented the following characteristics collected from each study: year of publication and first author, race or ethnicity of samples, exact quantity of each genotype for cases and controls, HWE (Hardy-Weinberg equilibrium) p value and genotyping method. We also collected the clinical characteristic of cases, including Gleason score, tumor stage and PSA (prostate level).

Evaluation of association between prostate cancer and rs1447295 polymorphism
Although many studies found the association between rs1447295 polymorphism and prostate cancer risk, the correlation of prostate cancer susceptibility and rs1447295 polymorphism is inconsistent. Therefore, we collected all the relevant studies with clear information of rs1447295 genotypes to address this question.  Figure  2(A-D). The results indicated that there's a strong association between prostate cancer susceptibility and rs1447295 polymorphism and this was consistently observed under different genetic models.
Meanwhile, we performed the similar analysis for the ethic subgroup database (    Corresponding forest plots of four models were shown in Figure 2(A-D). The result indicated that rs1447295 had strong association with prostate cancer susceptibility regardless of American, European or Asian descent.

Evaluation of association between rs1447295 polymorphismand prostate cancer clinical characteristics
Prostate cancer clinical characteristics, such as Gleason score, tumor stage and PSA, are important indexes for evaluating the progress of this disease. Correlation between rs1447295 polymorphismand prostate cancer clinical characteristics has been touched before but not in systematic study. Clarification of the correlation is essential to assess the diagnosis value of rs1447295 polymorphism in different prostate cancer status. Therefore, in addition to overall and subgroup     (Figure 3 and 4). As for the PCa cases with tumor stage 1 to 2, the risk allelic OR was 1.287 (95% CI 1.182 -1.402; p value < 0.001); for cases with tumor stage 3 to 4, the risk allelic OR was 1.554 (95% CI 1.362 -1.774; p value < 0.001). As for cases with PSA level not more than 10 ng/mL, the risk allelic OR was 1.274 (95% CI 1.034 -1.568; p value = 0.023), for cases with PSA level more than 10 ng/mL, the risk allelic OR was 1.243 (95% CI 1.019 -1.515; p value = 0.032). Due to the limited sample size, we were not able to perform detailed ethic subgroup analyses for tumor stage and PSA level. Corresponding forest plots were shown as Figure 5 and 6. These results suggested that rs1447295 polymorphism was associated with prostate cancer risk by different clinical characteristic stratifications and might serve as a solid and sensitive marker for prostate cancer diagnosis even in the early stage.

Publication bias
Funnel plots were inspected to evaluate the possibility of reporting a publication bias for skewness. For the most part these did not suggest the presence of significant reporting bias (Figure 7). To test the publication bias of the literature, both Begg's test and Egger's test were performed. Results of publication bias were shown in Table 2. No publication bias was observed under any model (all p values of Egger's test and Begg's test were over 0.01).
Similar evaluation was also conducted in analysis of clinical characteristics, and no publication bias was observed too. Corresponding funnel plots were shown as Figure 8, 9, 5(C, D), and 6(C, D).

DISCUSSION
Even though there had been meta-analysis related to association between rs1447295 and prostate cancer before, we performed a more detailed analysis. We       removed replicated cohort data in previous study and enlarged the data pool with the mostly recent research (2006 to 2016) in our analysis which made the conclusion more scientific and reliable. Most importantly, we found significant association between rs1447295 polymorphism and the risk of prostate cancer in total groups and also in both European descent and Asian descent. Meanwhile, our results also showed that rs1447295 polymorphism was associated with different prostate cancer clinical characteristics, providing an important evidence to use rs1447295 polymorphism as a useful marker for prostate cancer diagnosis even in the early stage.
The most recent meta-analysis for rs1447295 and PCa risk was done in 2012 [17], our analysis updated the data collected from 2006 to 2016 and focus on European descent, American descent and Asian descent. Compared with the previous meta-analysis [17], we incorporated three new articles in our analysis of Asian group. Distinct from previous analysis which only included the case and control number as the dataset but did not analyze the correlation according to rs1447295 genotypes, our study analyzed the correlation of PCa risk with rs1447295 in four different genotypes. In the four different gene models of rs1447295, the frequency of PCa significantly increased in European descent, American descent and Asian descent.
Analysis of the association between risk allele at rs1447295 and PCa clinical features showed rs1447295 polymorphism tends to correlate with advanced prostate cancer risk. For example, rs1447295 polymorphism may be associated with greater tumor aggressiveness in African Americans [9] and American Whites [10], advanced stage diseases in Eastern Whites [11] and African American [9,12]. The rs1447295 risk allele tended to be more strongly associated with Gleason score 7-10 tumors than with Gleason score 2-6 tumors in African Americans and American Whites [7]. However, this trend was not observed in Eastern White populations [7,13]. In addition, the correlation between rs1447295 polymorphism and PCa risk did not appear to be affected by PSA level in Eastern Whites [11,14]. Zheng et al. [11] showed that this allele was associated with a younger age at onset in Eastern Whites, but Schumacher et al. [15] and Zheng et al. [16] didn't show such evidence. It looks like that the risk allele is only associated with a younger age at onset in African Americans populations. To clear up the confusion, we evaluated the association between rs1447295 and PCa risk for cases with different clinical characteristics including Gleason score, tumor stage and PSA (prostate specific antigen) level. Our result turned out that rs1447295 polymorphism was associated with different prostate cancer clinical characteristics in American, descent, European descent and Asian descent. The interesting results suggested that risk allele at rs1447295 might be a sensitive marker for prostate cancer diagnosis. However, a more detailed study with larger sample size and more diverse population is needed to further confirm this observation in the future.

Literature search and data extraction
We searched related articles through six databases, including "PubMed", "Science Direct", "Karger", "Web of Science", "Wiley Online Library" and "Springer". To avoid omitting related literatures, "8q24" or "rs1447295" and "prostate cancer" were set as key words. The search coverage was those published in English before November, 2016. We exclude the books and other literatures that were not related with case-control study or not aimed at prostate cancer research. Then full texts of the left articles were carefully checked and we removed articles which didn't contain the exact quantity information about the genotypes of rs1447295. Finally, we noticed there're some studies with overlapping samples so that we picked out adequate ones, and here we restricted our study subjects as American descents, Asian descents and European descents.
All data were extracted independently by two reviewers. Preliminary evaluation was conducted based on the titles and abstracts, and then full texts of potentially relevant with our studies were obtained and re-evaluated for the inclusion. The following characteristics were collected from each study: year of publication and first author, race or ethnicity of samples, exact quantity of each genotype for cases and controls, HWE (Hardy-Weinberg equilibrium) p value and genotyping method. In addition, clinical characteristic of cases such as Gleason score, tumor stage and PSA (prostate specific antigen) level were also collected, especially those with detailed genotype.

Statistical methods
The statistical analysis was conducted with STATA 12 (Stata Corp LP, College Station, Texas, United States). In the whole process, p value less than 0.05 was considered to be statistically significant. Hardy-Weinberg equilibrium in cases or controls was evaluated by the Chi-Square test, and p value over 0.05 was considered as significant equilibrium. HWE was also taken as a dataextraction factor and those studies with HWE p value over 0.05 were chosen for further analysis.
To get a more reasonable result, four genetic models of inheritance were adopted in our analysis: dominant model (AA + AC vs. CC; A was considered as risk allele), recessive model (AA vs. AC + CC), homozygote model (AA vs. CC) and additive model (A vs. C). In dominant model, we investigated the distribution of genotype AA and AC comparing to genotype CC; as for recessive model, the distribution of genotype AA comparing to genotype AC and CC was analyzed; in homozygote model we used CC as reference genotype and investigated the distribution of AA; as for additive model the distribution of allele A comparing to C was analyzed.
For each study, the quantities of four genotypes in cases and control groups were used as pooled data. As for pooling analysis, Mantel-Haenszel (M-H) fixed-effect model was applied to analyze datasets with insignificant heterogeneity, and DerSimonian and Laird (D-L) randomeffect model was suitable for datasets with obvious heterogeneity. In our analysis, the heterogeneity among studies was evaluated using I 2 index. The higher I 2 was, the more significant the heterogeneity was. To be specific, when I 2 was less than 50%, we could believe there was no significant heterogeneity among pooled data, and then M-H model is applied; for I 2 more than 75%, an obvious heterogeneity existed and D-L model should be adopted; otherwise both models could be applied. For each analysis, M-H model was used firstly to test the heterogeneity, and then an adequate model was chosen based on the test result of I 2 value. Pooled odds ratio (OR) and 95% confidence interval were calculated with correspondent model, and corresponding forest plot was generated to summarize the result.
As for the elaborated evaluation, factors, like Gleason score, tumor stage and PSA level, were chosen and corresponding genotypes were collected, and risk allelic OR of comparing cases with controls was calculated to see whether there were differences between cases with distinct clinical characteristics.
To evaluate publication bias, Begg's funnel plot was generated based on the analysis result and database size. The more asymmetry the funnel plot looked, the more publication bias was introduced. Meanwhile, Egger's test was also performed for further investigation. For the Egger's test, the significance level was set as p value <0.01.