Association of Adiponectin Polymorphism with Metabolic Syndrome Risk and Adiponectin Level with Stroke Risk: A Meta-Analysis

Many previous studies have provided evidence that the ADIPOQ +45T>G polymorphism (rs2241766) might cause metabolic syndrome (MS). As a cardiovascular manifestation of MS, the incidence of stroke is associated with adiponectin; however, the results remain controversial and inconsistent. Systematic searches of relevant studies published up to Dec 2014 and Jan 2016 on the ADIPOQ +45T>G polymorphism and the risk of MS and adiponectin levels and the risk of stroke, respectively, were conducted in MEDLINE and EMBASE. The odds ratio (OR) or risk ratio (RR) and their 95% confidence interval (95% CI) were extracted. Sixteen studies containing 4,113 MS cases and 3,637 healthy controls indicated a weak positive association between ADIPOQ +45 T>G and MS in the dominant genetic model (OR = 1.30, 95% CI = 1.03–1.65), which was also validated by stratified subgroup analyses. Twelve studies including 26,213 participants and 4,246 stroke cases indicated that 5 μg/ml increments in adiponectin level were not relevant to stroke risk (RR = 1.05, 95% CI = 1.00–1.10, P = 0.069). This study suggested a weak positive association of ADIPOQ +45T>G with MS and a strong association with metabolic-related disease. Additionally, adiponectin level was not a causal factor of increasing stroke risk.

Literature Searches. This study was conducted according to the meta-analysis guidelines defined in the PRISMA statement 13,14 . The publication search was carried out in multiple electronic databases: PubMed, EMBASE, CNKI (China National Knowledge Infrastructure), and Wanfang databases.
For the primary analysis of the ADIPOQ + 45 T> G polymorphism and MS risk, the following subject terms were used to conduct the search: "adiponectin" or "ADIPOQ" or "adiponectin gene polymorphism", "AMP1", and "metabolic syndrome" or "MS" or "metabolic syndrome X" or "syndrome X". The relevant articles in English and Chinese published before Dec 31, 2014, were included.
For the secondary subject of adiponectin and stroke risk, the following subject terms were used for the search: "adiponectin" and "stroke". The relevant articles in English published before Jan 26 th , 2016, were included.
We also conducted a manual search to find additional relevant data based on the references identified in the retrieved articles.
Inclusion and Exclusion Criteria. The inclusion criteria of the primary meta-analysis were as follows: the studies (1) evaluated the relationship between ADIPOQ + 45T> G (rs2241766) and MS risk; (2) were case-control studies; (3) provided detailed genotype data to estimate odds ratios (ORs) and 95% confidence intervals (CIs).
The inclusion criteria of the secondary meta-analysis were as follows: the studies (1) used a prospective cohort to investigate the relationship of adiponectin and stroke risk; (2) included stroke as an endpoint; (3) had a follow-up duration of over 1 year; (4) used clear diagnostic criteria for stroke; and (5) provided data on the relative risk (RR) with 95% CI. In all identified studies, all subjects were free of stroke at baseline according to the stroke diagnostic criteria used and participated in the entire follow-up survey.
All the articles that met the above requirements were included, regardless of the sample size and the population of the studies. Unpublished conference papers were not included. For overlapping studies, the most recent publication or the largest sample size publication providing more information was selected. Data Extraction. Two investigators (H.-P.Y. and L.S.) systematically extracted and reviewed all studies according to a standardized form and then independently entered this information into an electronic database. The following items were reviewed and extracted from each eligible study for the primary meta-analysis when possible: name of the first author, year of publication, ethnicity, country where the study conducted, mean age, total number of cases and controls, allele frequencies, genotyping method, diagnostic criteria of MS, and language of the article. Different ethnicity descents were categorized as East Asian, South Asian, and Southeast Asians.
The extracted information for the articles for the secondary meta-analysis included the name of the first author, publication year, name of the study, country, mean age, duration of follow-up, proportion of men, number of stroke cases and total participants, and adjusted variables. In addition, all of the multivariate adjusted RRs and 95% CIs based on a comparison of the highest adiponectin level category with the lowest adiponectin level category and the ORs/RRs with their 95% CIs were extracted. We abstracted the values for each category of adiponectin level, the median value of the adiponectin level for each category, the number of cases and participants/ person-time for each category, and the most completely adjusted RRs with 95% CIs (odds, hazard or risk ratios were combined into RRs) for each category.
Quality assessment. A quality assessment of the primary meta-analysis was performed by two authors independently according to the Newcastle-Ottawa scale 15 for non-randomized studies. The scale contains nine points: selection (4 points), comparability (2 points), and estimation of outcomes or exposures (3 points).
The qualities of the identified studies for the secondary meta-analysis were assessed by two authors independently according to a "methodological quality assessment scale" (Supplementary Table S3), which was modified from a previous publication 16,17 . Five items including the representativeness of the cases, source of controls, sample size, quality control of the genotyping methods, and Hardy-Weinberg equilibrium (HWE) were assessed in this scale. The quality score ranged from 0 to 10, with a high score indicating a good quality study. Any disagreements between the two investigators were resolved through discussion.
Scientific RepoRts | 6:31945 | DOI: 10.1038/srep31945 Statistical Analysis. The primary meta-analysis of the relationship between the ADIPOQ + 45 T> G polymorphism and MS risk was conducted as follows. ORs and 95% CIs were calculated as estimates of the relative strength of the associations between the ADIPOQ + 45 T> G polymorphism and MS risk. The pooled ORs were calculated from a weighted average of OR from each study. Due to the potential functional role of the ADIPOQ + 45 T> G polymorphism, carriers with at least one risk allele will present the corresponding phenotype or traits. Moreover, to obtain all of the data included, the dominant model (GG + GT vs. TT) was adopted. Allele contrast (G vs. T) and recessive models (GG vs. GT + TT) were also performed in the meantime. After adjusting for the potential confounding factors that might have contributed to the heterogeneity, the dominant model was adopted for further investigation. Subgroup analyses were also conducted to explore the effects of the following confounding factors: language of the article, diagnostic criteria, and ethnicities.
The secondary meta-analysis of the relationship between adiponectin and risk of stroke was conducted as follows. Multivariable-adjusted RRs and 95% CIs were used to obtain the overall RR for the highest adiponectin level category compared with the lowest category using random-effects models. RRs and 95% CIs for each SD increase in log-transformed adiponectin were used to obtain the overall RR per SD in log μ g/ml. The RR of each 5 μ g/ml adiponectin level increment was calculated for studies that provided enough raw data to quantify the association between adiponectin level and risk of stroke using GLST (generalized least-squares trend estimation) and VWLS (variance-weighted least square regression model) for the dose-response analysis, as proposed by Greenland, Longnecker, Orsini et al. The detailed methods of the dose-response meta-analysis have been described before 18 .
The between-study heterogeneity was calculated using Cochran's Q test (Mantel-Haenszel chi-squared test) and was considered significant if the P value was less than 0.1 19,20 . Additionally, the I 2 statistic was used to test for heterogeneity, with values of I 2 less than 25%, 25-50%, and greater than 50% representing low, moderate and high degrees of inconsistency, respectively. When the P value was ≥ 0.1 or I 2 ≤ 25%, a fixed effects model (Mantel-Haenszel method) was selected; otherwise, a random effects model (DerSimonian and Laird method) was selected. We then explored the sources of heterogeneity using meta-regression and stratified analysis. Sensitivity analyses were performed to evaluate the stability of the results. A leave-one-out method was used to evaluate each study, and a pooled estimate was calculated for the remaining studies. Begg's funnel plots were conducted to evaluate the publication bias qualitatively; Begg's test and Egger's test were performed to quantitatively assess the publication bias. All P values were two-sided, with the statistically significant level defined as lower than 0.05. The statistical analyses were conducted using Stata (version 12.0; StataCorp LP, TX, USA).

Results
Characteristics of the Included Studies. The characteristics of each study included in these two meta-analyses are summarized in Tables 1 and 2. Supplementary Fig. S1A shows the detailed procedure by which articles are identified and assessed for inclusion in the primary meta-analysis. the genotype frequencies of the controls were all consistent with the HWE (all P > 0.05). The serum adiponectin levels presented in Table 1 shows that the concentration of adiponectin is lower in MS patients and higher in controls (P = 0.0016). Supplementary Fig. S1B shows the detailed procedure by which articles are identified and reviewed for inclusion in the secondary meta-analysis. The 12 studies 11,[36][37][38][39][40][41][42][43][44][45][46] in this analysis contained 26,213 participants and 4,246 stroke cases. Seven studies 11,36,37,40,41,43,44 (8 data points due to separate data for men and women) were included to compare the highest adiponectin level category with the lowest category. Two studies 37,45 (3 data points due to separate data for men and women) were included to estimate the RR per SD increase in log-transformed adiponectin level. Finally, 11 studies 11,36-44,46 (13 data points due to separate data for men and women) were included to estimate the dose-response relationship between adiponectin level and stroke. Quality assessment. Quality assessments were performed for the included studies both in the primary and the secondary meta-analyses. All studies had acceptable qualities according to the methodological quality assessment (Table 1 and Supplementary Table S3) and the Newcastle-Ottawa scale (Supplementary Table S4).  Fig. S2). Two studies (3 points) that provided the RRs of stroke per SD in log-transferred adiponectin level demonstrated that the overall RR was 1.20 (95% CI = 0.99-1.44, P = 0.059). Additionally, 11 studies with 13 data points that provided sufficient raw data for the dose-response analysis indicated that the RR of stroke for each 5 μ g/ml adiponectin level increment was 1.05 (95% CI = 1.00-1.10, P = 0.069) ( Table 3 and Fig. 1B). After excluding two papers by Prugger C et al. and Gardener H et al., the RR of stroke for each 5 μ g/ml adiponectin level increment was 1.02 (95% CI = 0.99-1.05, P = 0.261; I 2 = 27.8%). Therefore, there was no association between adiponectin levels and stroke risk.

ADIPOQ +45 T>G polymorphism and MS risk.
Stratified Analysis and Meta-Regression. Considering the significant heterogeneity in both meta-analyses, a stratified analysis and meta-regression were performed successively to investigate the potential sources of heterogeneity.
For the primary meta-analysis, first of all, we conducted a stratified analysis based on language and diagnostic criteria of the overall data and found that they were not the sources of heterogeneity (I 2 was largely unchanged, data were shown in Supplementary Fig. S3A,B). In addition, the results of the sensitivity analysis and publication bias analysis showed that the data from SHEN J et al.   Fig. S3A,B). Based on these findings, as well as similar results observed in the meta-regression ( Tau-square Fig. S3C).
For the secondary meta-analysis, the stratified analysis was performed based on study design, country, mean age, sex, follow-up duration, and measure of association (data were shown in Supplementary Table S1). There was no relationship between the risk of stroke and elevated adiponectin levels and it remained for all factors we proposed (i.e., all RR values were around 1). We found that these factors were not the sources of heterogeneity (I 2 was largely unchanged). The meta-regression showed that the association did not substantially differ due to the effect of these potential confounding factors (i.e., all P were > 1). Supplementary Fig. S4A, one-way sensitivity analyses for the primary analysis were performed by excluding one study at a time in each genetic model. The results of the dominant genetic models indicated that the overall results were dependable. As we can see, the two papers by SHEN J et al.  Supplementary Fig. S4C). After removing these studies, the tau-squared decreased for the primary and secondary analysis (from 0.1354 to 0.0194 for the primary analysis and from 0.0038 to 0.0011 for the secondary analysis) and explained 85.67% and 71.1% of the sources of heterogeneity, respectively.

Summary of Meta-Analysis Related to ADIPOQ +45T>G. Considering the close relationship
between ADIPOQ + 45 T > G and the typical components of MS as well as the many resulting potential disease risks such as obesity, cancer, and hypertension, we summarized data from the published meta-analyses as well as the primary meta-analysis in our study (Fig. 2, Supplementary Table S2). The results showed that ADIPOQ + 45 T> G appeared to be a strong risk factor for MS components, obesity, non-alcoholic fatty liver disease (NAFLD),  Test of Heterogeneity. For the primary analysis, we performed a heterogeneity test based on the dominant model in our meta-analysis and found strong heterogeneity among the included studies (I 2 = 82.5%, P < 0.001) (Fig. 1A). The allele contrast and recessive genetic models also showed substantial heterogeneity (data not shown). It was thus very important to identify the sources of heterogeneity and weaken their influence. For the secondary analysis, clear heterogeneity was observed for both the RR of the highest adiponectin level category compared with the lowest category and the RR of each 5μ g/ml increment in adiponectin level (I 2 = 54.6%, P = 0.009) (Fig. 1B and Supplementary Fig. S2).
Due to the high heterogeneity in the two meta-analyses, stratified analyses, meta-regression, and sensitivity analyses were conducted. Based on the results above, we found that although some studies were far from the central line in the sensitivity analyses and might have been the sources of the heterogeneity, the conclusions regarding the associations in the two meta-analyses did not change when these studies were excluded. Therefore, the results of the two meta-analyses were reliable and stable. Supplementary Fig. S4B, we found no obvious asymmetry in the dominant genetic model in the primary analysis, and this result was statistically supported by the Begg's and Egger's test (P Begg's = 0.325, P Egger's = 0.118). Additionally, the studies by SHEN J et al. and CHEN F-2 et al. were found to have deviated far from the others. For the secondary analysis, there was no obvious publication bias based on the funnel plot, Begg's test, and Egger's test (P Begg's = 0.631, P Egger's = 0.378) (Supplementary Fig. S4D).

Discussion
Adiponectin is a plasma glycoprotein of adipose tissue origin 47 . Adiponectin levels were reported to be lower in subjects with the MS phenotype, including obesity, T2DM, dyslipidemia, and hypertension 48 . Studies also confirmed that the ADIPOQ gene (+ 45T> G polymorphism) is associated with adiponectin levels and resulting MS 9,49,50 . However, the findings are still conflicting. Therefore, we conducted an updated meta-analysis to derive a more precise evaluation of the association of the ADIPOQ + 45 T> G polymorphism and MS risk. Furthermore, genetic factors like the ADIPOQ + 45T> G polymorphism is related to serum adiponectin levels 51,52 . Thus, in addition to the ADIPOQ + 45T> G contribution to MS, adiponectin levels might also contribute to stroke risk, which is considered a cardiovascular manifestation of MS 12,53 . Due to the inconsistent conclusions regarding the association between adiponectin levels and stroke risk, 12 studies on this topic were also included in our secondary meta-analysis.
In the primary analysis, due to the strong heterogeneity, a sensitivity analysis, stratified analysis and meta-regression were conducted in the dominant genetic model. Diversity in ethnicity, diagnostic criteria, and distinction between published English and Chinese language studies might be the common source of heterogeneity 54-56 , thus we tried to clarify the sources of heterogeneity through subgroup analysis by the above three factors. The stratified analysis and meta-regression indicated that language and diagnostic criteria were not sources of heterogeneity (Tau-squared value was not decreased). Meta-analysis is often affected by the potential "outlier points" among the involved studies, which might be correlated to the data reliability, small sample size, publication bias and clinical confounding. In the current meta-analysis, both of sensitivity analysis and publication bias analysis indicated 2 articles from SHEN J et al. and CHEN F-2 et al. deviated potentially from other studies ( Supplementary Fig. S4A,B). The articles from SHEN J et al. and CHEN F-2 et al. contributed inappropriate data to the overall assessment and were thus removed. After their exclusion, the heterogeneity greatly improved (I2 reduced to 40.1%) and the Tau-squared value reduced accordingly whereas this positive association still existed. Therefore, we excluded the 2 studies for the further stratified analysis and meta-regression. Taken together, our study indicated that a weak positive association was visible in the relationship between ADIPOQ + 45T> G and MS in the dominant genetic model (OR = 1.30, 95% CI = 1.03-1.65).
The first meta-analysis focused on the relationship between the ADIPOQ + 45T> G polymorphism and the risk of MS in the whole population and had several limitations. First, the number of subjects in the included studies was still small, and only two studies used data from outside the Chinese population; thus, more large-scale studies are needed to assess the association between ADIPOQ + 45T> G and MS. Second, environmental factors such as smoking, drinking, physical activity and family income might have a substantial influence on the progression of MS, but the lack of raw data makes it impossible for us to account for gene-environment interactions. Third, our analysis only focuses on ADIPOQ + 45T> G; however, the development of MS is not due to a single gene, and there should be cross-talk between ADIPOQ + 45T> G and other gene polymorphisms that should be considered 27 .
Given the complexity of the MS and the lack of clarity surrounding its definition, most reports to date have been restricted to the examination of the relationship between variant polymorphisms and the individual criteria of MS, rather than with the syndrome as a whole. For example, individuals carrying ADIPOQ + 45T> G polymorphism were found to be associated with type 2 diabetes mellitus in Chinese, however, conflicting results have been reported in Asian and European subjects 57,58 . To our knowledge, this prospective study is the first to date to demonstrate the findings of studies about the association of ADIPOQ + 45T> G with MS in the whole population together with a stratified analysis. There was only one prior meta-analysis, conducted by GAO M et al., that showed a significant relationship between ADIPOQ + 45T> G and MS that differed from ours 9 . There might be several potential reasons for this difference. First, GAO M et al. performed their meta-analysis specifically in the Chinese population. In the present study, we focused on studies throughout the world that were up to date, and we presented a more comprehensive conclusion rather than studying a single ethnic group. Additionally, considering the distinct genetic backgrounds in the population, we further performed a subgroup analysis on East Asians. Second to detect associations. Third, we also conducted a stratified analysis for different languages and diagnostic criteria as well as a subgroup analysis in East Asians to derive an accurate evaluation and validated our conclusions. Finally, and most importantly, we summarized the previous meta-analysis with ours and evaluated the pivotal role of ADIPOQ + 45T> G in all potential metabolic-related disease risks including MS and metabolic-related disease. A clear risk of ADIPOQ + 45T> G for obesity, cancer, NAFLD, T2DM, DN, hypertension, CAD, CHD and POVS was found. These disorders usually co-exist with each other 59,60 . This noteworthy finding led to the possible hypothesis that ADIPOQ + 45T> G might be a common genetic factor behind these diseases and could explain the molecular mechanism of comorbidity in some patients.
Therefore, there exist some shared and distinct mechanisms that need to be clarified regarding MS, metabolic-related disease and their comorbidities. The answer might be somewhat as follows. First, the populations in the existing studies on MS and metabolic-related disease were so different that the confounding factors could not be controlled properly. Second, there must be a key individual characteristic on top of general factors (ADIPOQ + 45T> G) that is present in MS and metabolic-related disease. According to the forest plot of our summarized analysis, although the ADIPOQ gene might be the "common soil" of individual MS components and other potential metabolic-related diseases, its contribution to MS as a whole appeared quite differently 61 . A genome wide association study (GWAS) suggested plasma adiponectin did not relate to genetic loci associated with MS parameters, indicating that MS is not modulated directly by the genetically determined adiponectin 29,62 . In fact, previous clinical observations categorized MS into two major components: one cluster was glucose/obesity/lipids, and the other was blood pressure 29 . These two groups significantly contributed to the genetic background of MS, indicating the heterogeneous nature of MS and the possibly distinct pathogenesis between MS with and without each cluster. In addition, metabolic-related diseases and MS might partially overlap with each other as specific clinical sub-phenotypes with a common genetic background. This could explain the inconsistent relationship found between ADIPOQ and individual MS parameters. Third, a well-designed follow-up cohort study including patients with both MS and metabolic-related disease that considers confounding factors and assesses a larger sample size with a focus on ADIPOQ + 45T> G is required in the future.
In the secondary meta-analysis, various analytical methods demonstrated that there was no association between 5 μ g/ml increments in adiponectin level and stroke risk (RR = 1.05, 95% CI = 1.00-1.10, P = 0.069). The conflicting results of the studies published previously may have existed for several reasons. First, the standards of measuring adiponectin level increments were not uniform in the included prospective and nested case-control cohort studies, which suggest it would be inappropriate to combine these measures in the overall RR. In our analysis, we stratified the analysis by the different measurement standards and evaluated the RR of stroke in 3 ways, as follows: stroke risk for the highest adiponectin level category compared with the lowest category; the overall RR per SD in log μ g/ml of adiponectin; and the quantified RR of each 5 μ g/ml increment in adiponectin level. Therefore, the conclusions we drew from these three analyses would be more accurate and reasonable. Second, serum adiponectin levels are related to various lipoproteins 63 . A limitation of the secondary analysis might be the limited data on HMW-adiponectin and the lack of data on other adiponectin fractions, which could have resulted in the influence of specific molecular-weight fractions of adiponectin on risk of stroke; this influence remains unclear.
In conclusion, our analysis suggested a weak positive association of ADIPOQ + 45T> G with MS as well as a strong association with metabolic-related disease. As a result, the ADIPOQ gene could be treated as a promising physiological and pharmacological target in the prevention and treatment of disease with either a single component or comorbidity. Second, our analysis suggested that an elevated adiponectin level was not a causal factor contributing to an increase in the risk of stroke. Further large-scale studies in a rigorously defined cohort should be carried out to validate our findings.