Analyzing 74,248 Samples Confirms the Association Between CLU rs11136000 Polymorphism and Alzheimer’s Disease in Caucasian But Not Chinese population

Clusterin (CLU) is considered one of the most important roles for pathogenesis of Alzheimer’s Disease (AD). The early genome-wide association studies (GWAS) identified the CLU rs11136000 polymorphism is significantly associated with AD in Caucasian. However, the subsequent studies are unable to replicate these findings in different populations. Although two independent meta-analyses show evidence to support significant association in Asian and Caucasian populations by integrating the data from 18 and 25 related GWAS studies, respectively, many of the following 18 studies also reported the inconsistent results. Moreover, there are six missed and a misclassified GWAS studies in the two meta-analyses. Therefore, we suspected that the small-scale and incompletion or heterogeneity of the samples maybe lead to different results of these studies. In this study, large-scale samples from 50 related GWAS studies (28,464 AD cases and 45,784 controls) were selected afresh from seven authoritative sources to reevaluate the effect of rs11136000 polymorphism to AD risk. Similarly, we identified that the minor allele variant of rs11136000 significantly decrease AD risk in Caucasian ethnicity using the allele, dominant and recessive model. Different from the results of the previous studies, however, the results showed a negligible or no association in Asian and Chinese populations. Collectively, our analysis suggests that, for Asian and Chinese populations, the variant of rs11136000 may be irrelevant to AD risk. We believe that these findings can help to improve the understanding of the AD’s pathogenesis.

participants carrying 2 copies of minor allele in rs11136000 are associated with a decreased risk of AD 17 . The consistent result in North American Caucasian population is also identified by Carrasquillo et al. 18 . While in Canadian and Korean populations, the rs11136000 is found not associated with AD according to the studies of Bettens et al. 24 and Chung et al. 27 , respectively. Then, two independent meta-analysis studies re-assess the results of these GWAS studies published before June 20, 2013 (18 studies) and August 31, 2014 (25 studies), respectively, and both of them found this SNP is significantly associated with AD in populations of Asian and Caucasian 29,30 . But among the subsequent 18 GWAS studies published after August 31, 2014, many of them report inconsistent results in the corresponding populations [31][32][33][34][35][36][37][38][39][40][41][42][43][44][45][46][47] . Moreover, by comparing the selected GWAS articles published before June 20, 2013 in the two meta-analysis studies, we find the selection is incomplete for both of them. In particular, Liu et al. 29 miss two GWAS articles about Caucasian populations 16,24 , and Du et al. 30 miss a GWAS article about Asian population 27 . In fact, through our further investigation, a total five related GWAS articles published before August 31, 2014 are not collected in the two meta-analysis studies [48][49][50][51][52] . In addition, a GWAS study about American and German populations is misclassified to the Asian ethnicity subgroup in Du et al. 's study 22 .
We suspected that the small-scale and incompletion or heterogeneity of the samples maybe lead to different results of these studies. In this study, we selected 50 related GWAS studies with large-scale samples from 40 articles (28,464 cases and 45,784 controls, about 40.3% increase over the total number of the previous two meta-analysis studies 29,30 ) by searching the PubMed, ClinicalKey, AlzGene, Google Scholar, CNKI, Wanfang and VIP databases, and reevaluated the association between AD and rs11136000 polymorphism in Caucasian, Asian and Chinese population using the method of meta-analysis as previously described [53][54][55][56][57][58][59][60][61][62][63] . The use of more complete and larger scale samples would make the results more reliable.
After that, the appropriate studies were identified by the following criteria: (1) The study is a GWAS to analysis the association of rs11136000 polymorphism and AD. (2) It is a case-control design study. (3) The study provides Extraction of the related data. We extracted the related data for subsequent analysis from these identified studies: (1) each study's publication date. (2) The first the author's name in each of these studies. (3) The numbers of AD patients and controls of each study. (4) The sample's ethnicity of each study. (5) The detailed genotype data of rs11136000 polymorphism both in AD patients and controls. (6) The types of genotyping platforms. (7) The key results of each study (i.e. the OR value and its 95% CI, as well as the corresponding P value). Moreover, if these results are not provided in the study directly, we would calculate them by the genotype data using the R program (http://www.r-project.org/).

Genetic model choice.
The rs11136000 polymorphism contains two types of variants (T and C). T is the minor allele and C is the major allele. We assumed that they are the lower and high risk factor for AD, respectively. Then, the dominant model (TT + TC allele versus CC allele), allele model (T versus C) and recessive model (TT versus TC + CC) were used in this study. According to Table 1, all these studies were meta-analyzed using allele model, while only the studies offering CC, CT and TT genotypes data were analyzed using dominant or recessive model.

Hardy-Weinberg equilibrium (HWE) test.
The HWE test of the rs11136000 polymorphism in AD patient and control groups was performed using a non-continuity correction chi-squared method with the significance level P < 0.01 as previously described 65 . Briefly, for the SNP in each case and control group, the simulated Heterogeneity test. In this study, the heterogeneity among the kinds of populations was measured by the two parameters, I 2 value and Cochran's Q. I 2 value range from 0 to 100%, and it is calculated by Cochran's Q according to the formula = × . The Cochran's Q is based on a chi-squared distribution with k − 1 degrees of freedom, and k means the number of studies. Usually, the extreme, high, moderate and low heterogeneity was considered corresponding to the I 2 value of >75%, 50-75%, 25-50%, and <25%, respectively. In this study, the threshold of significant heterogeneity was set as I 2 > 50% and P < 0.01 according to previous studies 53-56 . Meta-analysis in entirety and subgroup. According to the results of heterogeneity test, the random and the fixed effect model were performed when the heterogeneity was significant or not, respectively 66 . We used the R package 'meta' to perform the meta-analysis, and determine the significance level of association between rs11136000 and AD through the pooled OR value and its 95% CI, as well as the corresponding P value (http://cran.r-project.org/web/packages/meta/index.html). And then, the original samples were further split into Caucasian, Asian, East Asian and Chinese populations, and the meta-analysis was performed in these subgroups. Publication bias analysis and sensitivity analysis. We first evaluated the publication bias of the studies used in dominant, allele and recessive model, respectively, by the two common checking methods, the Begg's test 67 and Egger's test 68 . The threshold of significant publication bias was set as P < 0.05. Then, we used the asymmetry of the funnel plots to describ the results of the publication bias analysis. Finally, for sensitivity analyses, we excluded each study in turn from the whole sample to measure the influence of each study. Data availability. All the datasets used in this are available from the corresponding author.

Results
Study acquisition and data extraction. By a keyword search in the publicly available databases and a screening according to the criteria, a total 46 studies from 36 articles were identified which mainly involved in Caucasian and Asian populations. Moreover, a study about Sweden population was selected from AlzGene database, and three studies involved in Asian populations were identified by the citation check using Google Scholar. Figure 1 showed the workflow of selection. Then, the related data of these 50 studies were extracted, and the main information was described in Table 1 (the detailed genotype data, the OR value and its 95% CI, as well as the corresponding P value were shown in Supplementary Table S1).

Hardy-Weinberg equilibrium test.
We calculated the P value of HWE to assess the genotype distribution of rs11136000 polymorphism in AD patients and controls separately. Using a significance level of P < 0.01, we observed that a few of the samples deviated from HWE, including the case samples from the study of Yu et al.  Heterogeneity Test and Meta-analysis. After the test, we found that there is no the significant genetic heterogeneity of rs11136000 polymorphism among all of the 50 selected studies using the dominant (I 2 = 0% and P = 0.60), allele (I 2 = 10% and P = 0.28) and recessive model (I 2 = 33% and P = 0.04). Therefore, the meta-analysis with fixed effect model was performed to assess the association between rs11136000 and the risk of AD, and we found significant results in all the three models. In particular, the significant association between the minor allele (T) of rs11136000 and a decreased risk of AD was identified in the allele (OR = 0.875, 95% CI = 0.854-0.896, P < 0.0001) (Fig. 2), dominant (OR = 0.848, 95% CI = 0.817-0.879, P < 0.0001) and recessive model (OR = 0.822, 95% CI = 0.779-0.868, P < 0.0001) (Supplementary Figs S1 and S2).
The Asian population in this study was composed of the Indian, Iranian, Korean and Japanese individuals separately from a GWAS study, and the Chinese individuals from 12 GWAS studies. Therefore, we then assessed the association between this SNP and risk of AD in East Asian and Chinese populations. Interestingly, the results of meta-analysis in East Asian population were similar to these in Asian population ( Supplementary Figs S6-S8). However, the association was not significant in Chinese population using the allele (OR = 0.939, 95% CI = 0.878-1.004, P = 0.0654) (Fig. 4a), dominant (OR = 0.988, 95% CI = 0.887-1.101, P = 0.8270) (Fig. 4b) and recessive model (OR = 0.615, 95% CI = 0.355-1.068, P = 0.0841) (Fig. 4c), which was different from the findings in the previous studies 29,30 .
Moreover, given that a few samples from four GWAS studies (three Asian populations and a mixed population) deviated from HWE, we further tested whether they affected the accuracy of the results by removing these studies from whole sample, Asian, East Asian and Chinese subgroups, respectively. The results were consistent  with what we had been observed previously in whole sample and the subgroups using allele, dominant and recessive model. Table 2 showed the detailed information of the results.
Publication bias analysis and sensitivity analysis. As the funnel plots show (Fig. 5), we did not identify the significant publication bias in the three genetic models.

Discussion
AD was characterized by accumulation and toxic effect of the Aβ deposits in brain 3 , and previous studies reported that the CLU could markedly influence the fibrillary Aβ formation and accumulation to mediate its toxicity in vivo, and likely as one of the most important roles for pathogenesis of AD 6,7 . Then, the subsequent GWAS studies found some variants in CLU were differently distributed between AD patients and controls [11][12][13][14][15][16][17][18] . Among these variants, a significant association was found between the minor allele (T) of rs11136000 and a decreased risk of AD by Harold et al. 11 , Lambert et al. 12 , Carrasquillo et al. 13 and Seshadri et al. 14 . However, these results could not be repeated in other populations by the following studies [19][20][21][22][23][24][25][26][27][28] .
As far as we know, our meta-analysis about the association of the CLU rs11136000 polymorphism with the risk of AD is by far the largest scale study. The results reveal a significant association between them in Caucasian ethnicity but not Chinese ethnicity, which is consistent with the findings of most of the corresponding GWAS studies. In summary, we believe that these findings can help to improve the understanding of the AD's pathogenesis.