Association between 8q24 rs6983267 polymorphism and cancer susceptibility: a meta-analysis involving 170,737 subjects

Published data on the association between 8q24 rs6983267 polymorphism and cancer risk are inconsistent. Thus, we conducted a meta-analysis to evaluate the relationship between rs6983267 polymorphism and cancer risk. We searched on PubMed, EMBASE, Web of Science and China National Knowledge Infrastructure (CNKI) up to November 1, 2016 for relevant studies. Odds ratios (ORs) and 95% confidence intervals (CIs) were used to estimate the strength of this association. We included 78 case-control studies with a total of 73,996 cases and 96,741 controls in this meta-analysis. The pooled results showed that rs6983267 polymorphism was significantly associated with increased risk of overall cancer in all genetic models (dominant model: OR = 1.19, 95% CI = 1.13–1.26; recessive model: OR = 1.19, 95% CI = 1.14–1.25; homozygous model: OR= 1.31, 95% CI = 1.23–1.40; heterozygous model: OR = 1.14, 95% CI = 1.10–1.19; allelic model: OR = 1.14, 95% CI = 1.11–1.18). Stratified analyses indicated that rs6983267 significantly increased the risk of colorectal cancer in Caucasians, prostate cancer in Caucasians and Asians, thyroid cancer in Caucasians and lung cancer in Asians. When studies were stratified by study quality, source of controls and genotyping method, significant associations were especially found in the high quality studies, the publication-based studies, the hospital-based studies, and the PCR-RFLP studies. Additional well-designed studies with large samples should be performed to validate our results.


INTRODUCTION
Cancer has become a major public health problem. According to the GLOBOCAN 2012, approximately 14.1 million new cancer cases and 8.2 million cancer deaths were reported worldwide [1]. Epidemiological and biological evidence demonstrate that carcinogenesis is a complex process involving multiple environmental and genetic factors, although the etiology of carcinogenesis has not been fully elucidated.
Chromosomal 8q24 has emerged recently as a risk locus for various types of cancer among different ethnicities (Caucasians, Asians, and Africans) [2]. 8q24 has been described as a "gene desert", since the 600-kbp gene-poor region appears to have little transcriptional activity. Despite this, accumulating evidence suggested that 8q24 may play an active role in carcinogenesis. For example, POU5F1P1, which was originally considered as a pseudogene, has been identified on 8q24. It is now supposed that POU5F1P1 can encode a functional protein which contributes to carcinogenesis by acting as a weak transcriptional activator [3]; MYC, which acts as a transcriptional activator involving cell growth, differentiation, apoptosis, and other intracellular responses, is significantly associated with 8q24 [4,5]. It has been reported that the 8q24 region

Meta-Analysis
Oncotarget 57422 www.impactjournals.com/oncotarget contains multiple enhancer elements which can activate transcription of the nearby oncogene MYC; 8q24 also contains several other genes that functionally participate in cancer development, including ectonucleotide pyrophosphatase/phosphodiesterase 2 gene (ENPP2) and nephroblastoma overexpressed gene (NOV). ENPP2 encodes a phospholipase, which stimulates tumor cell proliferation [6]. NOV encodes regulatory protein CCN3, which involves in cancer development [6]. Furthermore, MetaCore regulatory network analysis found that both NOV and ENPP2 were indirectly linked by MYC [7].
Genome-wide association studies (GWASs) have identified the polymorphism rs6983267 as a new susceptibility locus for several cancer types [8][9][10][11]. Polymorphism rs6983267 which locates on 8q24 is a G/T single nucleotide polymorphism (SNP). There are three genotypes of rs6983267: homozygous risk alleles (GG), homozygous non-risk alleles (TT), and heterozygous alleles (G/T). Previous studies by Tuupanen et al. [12] and Sur et al. [13] demonstrated that the risk allele (G allele) serves as a binding element for the enhancer protein TCF4/LEF1 which can accelerate the transcription of MYC in vivo. Since then, a great number of studies have been performed on this polymorphism with the risk of many cancers in different populations but have generated equivocal results. Up to now, a number of meta-analyses have been published and implied a possible association between rs6983267 polymorphism and cancer risk. Unfortunately, some meta-analyses have presented contradictory results. For instance, Troutman et al. [14] indicated that rs6983267 was significantly associated with a high risk for prostate cancer among Caucasians, Asians and Africans. However, Zhu et al. [15] demonstrated that rs6983267 was associated to prostate cancer among Caucasians and Asians. Intriguingly, in 2016, Yang et al. [16] observed this association among Caucasians only. Besides, such contradictions also existed in some other meta-analyses [17][18][19]. Of note, lack of further evaluation in different stratified analyses prevented comprehensive understanding in some recent meta-analyses [20,21]. To better understand the precise relationship, we performed a comprehensive meta-analysis with increased statistical power.

Characteristics of eligible studies
A total of 504 articles were preliminarily identified at first based on our selection strategy. We also identified three papers through the references. After scanning all of the abstracts, there were 71 articles that conformed to inclusion criteria. We excluded 12 articles that did not have completely extractable data [22][23][24][25][26][27][28][29][30][31][32][33], 3 articles were excluded because they did not conform to HWE [34][35][36], 2 studies were excluded because they were duplicated with previous publications [37,38]. Thus, we included 54 independent records [8,10,12,36,. One study [52] was treated as 9 independent case groups because nine cancer types were studied. Moreover, we retrieved 25 separated investigations from 9 articles [8,10,39,42,54,61,74,83,88]. Finally, a total of 78 separate studies involving 73,996 cases and 96,741 controls were included in our meta-analysis. Among them, there were 32 studies on colorectal cancer, 25 on prostate cancer, 6 on thyroid cancer, 4 on gastric cancer, 3 on breast cancer, 3 on lung cancer and 5 on other cancers. Figure 1 describes the process for the study. The baseline characteristics of all eligible studies are summarized in Supplementary Table 1.

Meta-analysis of the overall population
The main meta-analysis results of the association between rs6983267 polymorphism and cancer risk are shown in Table 1. We found that rs6983267 polymorphism significantly increased cancer risk in all five genetic models: dominant (GG+GT vs. TT, OR = 1.19, 95% CI = 1.13-1.26), recessive (GG vs. GT+TT, OR = 1.19, 95% CI = 1.14-1.25), homozygote (GG vs. TT, OR = 1.31, 95% CI = 1.23-1.40), heterozygous (GG vs. GT, OR = 1.14, 95% CI = 1.10-1.19) and allele (G vs. T, OR = 1.14, 95% CI = 1.11-1.18) models ( Figure 2). False-positive report probability (FPRP) values for all significant findings at different prior probability levels are summarized in Supplementary Table  2. FPRP values at pre-specified prior probability of 0.01 were all lower than 0.2, indicating that the association between rs6983267 polymorphism and cancer risk was noteworthy. Outcomes of trial sequential analysis (TSA) were concordant with our results and revealed that rs6983267 polymorphism was significantly associated with cancer risk. Moreover, it also revealed that enough number of samples were included in this meta-analysis to reach a concrete conclusion as the cumulative Z-curve surpassed the O'Brien-Fleming boundary ( Figure 3).

Subgroup analyses
When studies were stratified under source of controls and genotyping method, significant results were detected in all subgroups (Table 1). Moreover, when studies were stratified by quality score, an increased cancer risk was observed in high quality subgroup ( Figure 4). With the assumption of prior probability of 0.01, these statistically significant associations were noteworthy (FPRP value < 0.2) for population-based, hospital-based and high quality subgroups under all five models (Supplementary Table 2), and for PCR-RFLP subgroup under recessive, homozygote, heterozygous and allele models.
When studies were stratified in ethnicity, significant associations were found in Caucasians and Asians, but not in Africans (Table 1). Moreover, when studies were stratified in to cancer type, significant associations were www.impactjournals.com/oncotarget found in colorectal cancer, prostate cancer, thyroid cancer, lung cancer and other cancers subgroups, but not in gastric cancer and breast cancer. FPRP analyses suggested that these statistically significant associations were noteworthy for Caucasians, colorectal cancer, prostate cancer, thyroid cancer and lung cancer (Supplementary Table 2). Furthermore, stratified analyses revealed a significant association between rs6983267 polymorphism and risk of colorectal cancer among Caucasians (Table 2). FPRP analyses suggested that this positive association was noteworthy under all five models (FPRP range: 0.000-0.007, Table 3). For prostate cancer, increased risks were observed among Caucasians, Asians and Africans (Table 2). FPRP analyses indicated that these associations were noteworthy for Caucasians (FPRP < 0.001) under all five models and Asians (FPRP range: 0.026-0.194) under recessive model, homozygote model and allele model (Table 3). For thyroid cancer, increased risks were revealed among Caucasians and Asians ( Table  2). FPRP analyses suggested that these associations were noteworthy for Caucasians (FPRP range: 0.000-0.182) under dominant model, homozygote model and allele model (Table 3). For lung cancer, increased risk was found among Asians (Table 2). FPRP analyses suggested that this positive association was noteworthy under dominant model, homozygote model and allele model (FPRP range: 0.015-0.079, Table 3).

Heterogeneity analyses
Q test and I 2 statistics were applied to evaluate the heterogeneity during our study. There was significant heterogeneity observed in the overall analysis. Therefore, we conducted meta-regression to explore the source of heterogeneity by ethnicity, cancer type, genotyping method, study quality and source of controls. As shown in Figure 5 and Supplementary Table 3, cancer type (heterozygote model: P = 0.043, recessive model: P = 0.020) and study quality (allelic model: P < 0.001, homozygote model: P < 0.001, heterozygote model: P = 0.003, recessive model: P < 0.001, dominant model: P < 0.001) were factors that contributed to the observed heterogeneity across all studies. However, combining with these two factors could explain only 33.89% (heterozygote model) or 49.57% (recessive model) of the τ 2 value, indicating that cancer type and study quality could explain one part of the heterogeneity. Otherwise, ethnicity, genotyping method and source of controls did

Publication bias
Begg's and Egger's tests were performed to assess the publication bias. The shape of the Begg's funnel plots seemed symmetrical ( Figure 6). Meanwhile, Egger's test suggested that there is no evidence of significant publication bias (P Egger =0.100 for dominant model, P Egger = 0.944 for recessive model, P Egger = 0.233 for homozygote model, P Egger =0.692 for heterozygote model, and P Egger = 0.484 for allele model) in this meta-analysis.

Sensitivity analysis
To evaluate the influence of individual study on the pooled ORs and 95% CIs, we excluded one study at each time. Results indicated that none of single study substantially changed the corresponding pooled ORs and 95% CIs (Figure 7), and demonstrated that our metaanalysis was relatively stable and credible.

DISCUSSION
It has been well established that genetics determine the risk of cancer in the last decades [90]. Since SNP is the main cause of human genetic variation, the connection between SNP and individual risk of cancer has drawn considerable attention. Recently, epidemiological studies evaluated the relationship between rs6983267 polymorphism and risk of multiple cancer types, while the results were inconsistent. In previous meta-analyses [15-18, 20, 91, 92] , the limited number of studies that included or not exclusion of studies that were not in HWE, then the validity of conclusions may decrease. Moreover, many relevant case-control studies, including more cancer types, were published [40,89], while these Oncotarget 57425 www.impactjournals.com/oncotarget articles have not been discussed in previous metaanalyses. Hence, to provide a good comprehensive conclusion, a meta-analysis of all available studies was conducted.
We performed a meta-analysis of 78 case-control studies from 54 articles (73,996 cases and 96,741 controls) to clarify the relationship between rs6983267 polymorphism and cancer susceptibility. Significant associations between rs6983267 polymorphism and cancer susceptibility were found under most of assumed comparisons, either in overall or in stratified analyses by ethnicity, cancer type, study quality, source of controls and genotyping method. When studies were stratified by study quality and source of controls, significant associations were observed under the assumed comparisons in the high quality subgroup, the publication-based subgroup and the hospital-based subgroup, but not in the low quality subgroup. Lack of association in the low quality subgroup was probably due to the fact that this group could not represent the general population sufficiently.
In addition, stratified analyses by ethnicity revealed a significant association between rs6983267 and colorectal cancer in Caucasians, but not in Asians and Africans. The rs6983267 was identified as a common susceptibility variant for colorectal cancer by three previous GWASs [8,9,93] in Caucasians, which was consistent with our study. However, our results were different than two previous meta-analyses [17,18], which reported that there was a significant association between rs6983267 and colorectal cancer in Asians. Possible reasons for this difference could be explained as follows: 1) one study by Hutter et al. [61] was carried out in Caucasians, while it was incorporated into the Asian subgroup in a previous meta-analysis [17]; 2) we included 7 case-control studies in Asians, instead of only 4 studies in a meta-analysis by Haerian et al. [18], and therefore, results of our meta-analysis were more credible; 3) lack of further research in different stratified analyses prevented comprehensive understanding in a recent meta-analysis [21]; 4) our results were based on sufficient evidence, which was proved by FPRP for the first time.
Moreover, we also found that rs6983267 polymorphism was a risk factor for the susceptibility to prostate cancer in Caucasians and Asians, which was consistent with two independent GWASs [10,11]. However, our outcomes were different to the results shown by Ho et al. [73], which demonstrated that rs6983267 polymorphism was not associated with prostate cancer. This discrepancy may be caused by the limited sample size. Ho et al. [73] included only 521 subjects (247 cases and 274 controls), which may lack sufficient power to support or deny an association. Previous metaanalyses also focused on the relationship between the rs6983267 and prostate cancer. However, our outcomes were different to previous meta-analyses [16,91], which indicated that rs6983267 polymorphism was associated with prostate cancer risk in Caucasians but not in Asians. Possible reasons for this difference could be explained as follows: 1) this discrepancy may be come from the limited sample size. For example, Yang et al. [16] included only 3 studies (805 cases and 703 controls) in Asians, whereas we included 5 studies (2200 cases and 1864 controls); 2) we excluded the studies that do not follow HWE, however, Li et al. [91] did not. In addition, when compared with the meta-analysis by Zhu et al. [15], although we reached the same conclusion, our analysis has some advantages: 1) the sample size of Zhu et al. was relatively small (16,753 cases and 14,802 controls); 2) to avoid false positive findings, FPRP analyses were performed for all significant findings observed in our study.  Oncotarget 57427 www.impactjournals.com/oncotarget Furthermore, we found that rs6983267 conferred a higher thyroid cancer risk among Caucasians than Asians. It was partially consistent with the consequence of the meta-analysis by Li et al. [20], while the sample size in our study was much more times elevated than theirs. Similarly, we also found that rs6983267 conferred a higher lung cancer risk among Asians, and FPRP analyses suggested that this positive association was noteworthy.
Moderate heterogeneity between eligible studies was identified for all genetic models in the overall comparisons. Common reasons of heterogeneity may include differences in sample selection (e.g., source of controls, HWE) or studied populations (e.g., geographic location), or methods (e.g., genotyping method), or other factors (e.g., study quality and cancer type). Metaregression analyses indicated that the potential sources of heterogeneity were cancer type and study quality. The results in false-positive report probability analysis were in bold, if the prior probability < 0.2. OR: odds ratio; CI: confidence interval; a P value for significant test; b Statistical power was calculated using the number of observations in the meta-analysis and the OR and P value in this table.
Oncotarget 57428 www.impactjournals.com/oncotarget   Nevertheless, when studies were stratified by cancer type and study quality, heterogeneity was still high in the colorectal cancer subgroup, the high quality subgroup and the low quality subgroup. These two analyses provided evidence that heterogeneity might also be explained by other confounding factors. In general, more uniform and rigorous studies were required.
Since quite many false positive results were found in those association studies between genetic variants and complex diseases due to the widely use of significance threshold (P < 0.05) [94]. To avoid false positive findings, our meta-analysis adopted FPRP analysis, which is based not only on the observed P value but also on the prior probability of hypothesis, making our results more reliable [95].
In the current study, there existed several advantages: 1) more studies were included in our metaanalysis; 2) more comprehensive subgroup analyses were conducted, and significant associations were found when we restricted to the high quality subgroup and populationbased controls subgroup; 3) our results were based on sufficient evidence, which were proved by TSA for the first time; 4) to avoid false positive findings, FPRP analyses were used for all significant findings observed in our study. However, some limitations should also be emphasized. First, in the subgroup analysis, we found that our analysis was limited on Caucasians, Asians and Africans, so we do not know whether these conclusions can also be adopted in other populations. Second, the number of included studies in some subgroups was relatively small, which might create significant or insignificant associations by chance due to insufficient statistical power. Third, this study is a summary of multiple data sources. Due to the Oncotarget 57433 www.impactjournals.com/oncotarget lack of original data, we could not evaluate the cancer susceptibility stratified by drinking status, smoking, carcinogen, radiation exposure, and other risk factors. Thus, more studies by standardized unbiased methods are required to offer more detailed individual data.
In summary, this systematical meta-analysis indicated that rs6983267 polymorphism significantly increased the risk of colorectal cancer in Caucasians, prostate cancer in Caucasians and Asians, thyroid cancer in Caucasians and lung cancer in Asians. In addition, significant association between rs6983267 polymorphism and cancer risk was observed in the high quality subgroup, the publication-based subgroup, the hospitalbased subgroup, and the PCR-RFLP subgroup. Further multi-center, large-cohort and well-designed studies are necessary to validate our findings.

Identification of the eligible studies
We systematically searched on PubMed, EMBASE, Web of Science and China National Knowledge Infrastructure (CNKI) electronic databases for relevant studies published before November 1, 2016. A detailed search strategy is presented in Supplementary Table 4. If studies were performed with overlapping data, only the largest or the latest studies would be included. Two independent authors conducted the search. Finally, we also searched the reference lists of all retrieved articles for potential studies manually.

Inclusion criteria
Enrolled studies should meet the following eligibility criteria: (1) case-control design; (2) investigating the association between rs6983267 polymorphism and cancer risk; (3) describing the genotype distributions in detail to calculate the OR and 95% CI in cases and controls; (4) observed genotype frequencies in controls must be consistent with Hardy-Weinberg Equilibrium (HWE).

Data extraction
Two investigators (M.Z. and X.W.) independently screened and extracted data from all eligible studies, with any disagreement resolved by consensus. The following information was collected: first author's surname, year of publication, ethnicity, country of origin, cancer type, source of controls, genotyping method, numbers of cases and controls, P-value of HWE in controls.

Quality score assessment
The quality of each study was independently assessed by two investigators (X.W. and C.L.) who used quality scoring criteria modified from previous studies (Supplementary Table 5) [96,97]. The evaluation items were as follows: ascertainment of cancer case, representativeness of case, representativeness of control, control selection, genotyping examination, and total sample size. Quality scores ranged from 0 (worst) to 12 (best). Studies scoring higher than 9 points were classified as high quality.

Statistical analysis
All statistical analyses were performed by STATA version 12.0 (STATA Corporation, College Station, Texas, USA). P < 0.05 was considered statistically significant. The strength of association between rs6983267 and cancer risk was estimated by OR with 95% CI. Z test was applied to confirm whether an association was statistically significant. We measured the association based on five different genetic models: dominant model (GG+GT vs. TT), recessive model (GG vs. GT+TT), homozygote model (GG vs. TT), heterozygote model (GG vs. GT), and allele model (G vs. T). Cochran Q-test and I 2 statistics were used to assess the between-study heterogeneity.
A random-effect model was used to assess pooled ORs when I2 (%) > 50% or P (Q) < 0.10, otherwise, a fixedeffect model was selected. Additionally, meta-regression analyses were used to detect the main sources of heterogeneity in our meta-analysis. Stratification analyses were performed by ethnicity, cancer type, study quality, genotyping method and source of controls. Sensitivity analyses were performed to assess the stability of the results. Furthermore, publication bias was assessed by Begg's and Egger's funnel plots, with potential publication bias if P < 0.05 and the plot was asymmetrical [98]. For each statistically significant association, false positive report probability (FPRP) analysis was performed using the method reported by Wacholder et al. [95]. We calculated FPRP assuming a prior probability of 0.01 as previously proposed [99]. An FPRP cutoff value of 0.2 was used and only result with FPRP value less than 0.2 was referred as noteworthy [95].

Trial sequential analysis (TSA)
A meta-analysis is prone to systematic errors (bias) or random errors (play of chance) due to dispersed data and repeated significance testing [100]. To obtain more comprehensive assessment, trial sequential analysis (TSA) (version 0.9; Copenhagen Trial Unit, Copenhagen, Oncotarget 57434 www.impactjournals.com/oncotarget Denmark, 2011) was used to calculate required information size (number of samples) and to confirm statistical reliability of meta-analysis.
In our study, we calculated the required information size by setting an overall type-I error of 5% and type-II error of 10% (a power of 90%). TSA plotted a twosided graph where blue line indicates cumulative Z-score, red straight lines show significance boundaries of the conventional meta-analysis, and red lines sloping inwards represent trial sequential monitoring boundaries with adjusted P-values.

Author contributions
M. Z., X.W. and J.T. performed the research design and data collection; X.W. and C. L. assessed the studies quality and conducted the stratified analysis; X.L. and Y. W. did the sensitive analysis and publication bias test. M. Z., X. W., C. L. and J.T. wrote the paper. All authors confirmed the final edition.