A meta-analysis of associations of LEPR Q223R and K109R polymorphisms with Type 2 diabetes risk

Background Leptin receptor (LEPR) plays a pivotal role in the control of body weight, energy metabolism, and insulin sensitivity. Various genetic association studies were performed to evaluate associations of LEPR genetic variants with type 2 diabetes (T2D) susceptibility. Methods A comprehensive search was conducted to identify all eligible case-control studies for examining the associations of LEPR single nucleotide polymorphisms (SNPs) Q223R (rs1137101) and K109R (rs1137100) with T2D risk. Odds ratios (OR) and corresponding 95% confidence intervals (CIs) were used to measure the magnitudes of association. Results For Q223R, 13 studies (11 articles) consisting of a total of 4030 cases and 2844 controls, and for K109R 7 studies (7 articles) consisting of 3319 cases and 2465 controls were available. Under an allele model, Q223R was not significantly associated with T2D risk (OR = 1.09, 95% CI: 0.80–1.48, P-value = 0.5989), which was consistent with results obtained under four genotypic models (ranges: ORs 1.08–1.20, 95% CIs: 0.58–2.02 to 0.64–2.26; P-values, 0.3650–0.8177, which all exceeded multiplicity-adjusted α = 0.05/5 = 0.01). In addition, no significant association was found between K109R and T2D risk based on either an allele model (OR = 0.93, 95% CI: 0.85–1.03, P-value = 0.1868) or four genotypic models (ranges: ORs 0.81–0.99, 95% CIs: 0.67–0.86 to 0.97–1.26, P-values, 0.0207–0.8804 which all exceeded multiplicity-adjusted α of 0.01). The magnitudes of association for these two SNPs were not dramatically changed in subgroup analyses by ethnicity or sensitivity analyses. Funnel plot inspections as well as Begg and Mazumdar adjusted rank correlation test and Egger linear regression test did not reveal significant publication biases in main and subgroup analyses. Bioinformatics analysis predicted that both missense SNPs were functionally neutral and benign. Conclusions The present meta-analysis did not detect significant genetic associations between LEPR Q223R and K109R polymorphisms and T2D risk.


Methods
A comprehensive search was conducted to identify all eligible case-control studies for examining the associations of LEPR single nucleotide polymorphisms (SNPs) Q223R (rs1137101) and K109R (rs1137100) with T2D risk. Odds ratios (OR) and corresponding 95% confidence intervals (CIs) were used to measure the magnitudes of association.

Results
For Q223R, 13 studies (11 articles) consisting of a total of 4030 cases and 2844 controls, and for K109R 7 studies (7 articles) consisting of 3319 cases and 2465 controls were available. Under an allele model, Q223R was not significantly associated with T2D risk (OR = 1.09, 95% CI: 0.80-1.48, P-value = 0.5989), which was consistent with results obtained under four genotypic models (ranges: ORs 1.08-1.20, 95% CIs: 0.58-2.02 to 0.64-2.26; Pvalues, 0.3650-0.8177, which all exceeded multiplicity-adjusted α = 0.05/5 = 0.01). In addition, no significant association was found between K109R and T2D risk based on either an allele model (OR = 0.93, 95% CI: 0.85-1.03, P-value = 0.1868) or four genotypic models (ranges: ORs 0.81-0.99, 95% CIs: 0.67-0.86 to 0.97-1.26, P-values, 0.0207-0.8804 which all exceeded multiplicity-adjusted α of 0.01). The magnitudes of association for these two SNPs were not dramatically changed in subgroup analyses by ethnicity or sensitivity analyses. Funnel plot inspections as well as Begg and Mazumdar adjusted rank correlation test and Egger linear regression test did not reveal significant publication biases in main and subgroup analyses. Bioinformatics analysis predicted that both missense SNPs were functionally neutral and benign. PLOS  Introduction their roles in T2D risk, for which sufficient numbers of single studies (i.e., > 5) were obtained for each SNP. We therefore conducted a comprehensive meta-analysis focusing exclusively on these two missense SNPs aiming at elucidating their associations with T2D susceptibility.

Search strategy
We searched relevant studies from the following electronic databases: PubMed, Excerpta Medica Database (EMBASE), Cochrane Library, and Google Scholar up to February 1, 2016. The following search terms were used in the electronic searches: "leptin receptor", "gene", "lepr", "T2D", "T2D and Type 2 Diabetes" with language restrictions to either English or Chinese. This study was performed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement checklist (S1 PRISMA Checklist) and the Meta-analysis of Genetic Association Studies checklist (S2 Checklist).

Study selection
The inclusion criteria were: (1) an original human-based case-control study using either a hospital-based or a population-based design; (2) a clear definition of T2D; (3) the relationship  [24]. Only Q223R was meta-analyzed by Liu et al. (2015) [69], and only Q223R, K109R, K656N, and P1019P were meta-analyzed by Su et al. (2016) [70]. between either Q223R (rs1137101) or K109R (rs1137100) and T2D risk was evaluated; and (4) providing sufficient data for calculating genotype and allele odds ratios (ORs) with their respective corresponding 95% confidence intervals (CIs). The exclusion criteria were: (1) reviews, conference abstracts, editorials and letters, (2) animal and in vitro studies, and (3) data about genotype frequencies could not be obtained. In case of overlapping or repeated studies, the one with most completed information was chosen. In addition, if more than one study shared the same subjects, the one with smaller sample size is excluded. All assessments were performed independently by two reviewers (YY and TN).

Data extraction
Data extraction was performed independently by two investigators (YY and TN) based on a pre-defined standard protocols. Any disagreements were solved by discussion. From each qualified study, the following information was collected: year of publication, first author's name, study location, ethnicity, source of controls (population-based or hospital-based), diagnosis criteria of T2D (i.e., how T2D is defined), sample sizes and respective genotypic frequencies in case and control groups, mean±standard deviation (SD) of age, distribution of gender, genotyping methods, and Hardy-Weinberg equilibrium (HWE) in controls (To present study characteristics more succinctly, T2D diagnosis criteria, genotyping methods, and HWE in controls were omitted from Tables 1 and 2). For each variable, corresponding measurements were shown using the same unit.

Quality assessment
Two authors (YY and TN) evaluated each individual study's quality independently according to the Newcastle-Ottawa scale (NOS) [25], which assesses the quality of each individual study in three sections: (1) selection of study subjects: 0-4; (2) comparability of study subjects: 0-2; and (3) clinical outcome: 0-3. The NOS score has a range of 0-9; and a score ! 7 is indicative of a good quality, e.g., [26,27]. Studies with a NOS score ! 6 are considered to be of sufficient quality for inclusion in a meta-analysis (e.g., [24,28]).

Statistical analysis
The ORs with 95% CIs were computed to evaluate respective associations of LEPR Q223R and K109R SNPs with T2D risk. For each polymorphism, 5 genetic models were employed, i.e., (1) an allele model (G vs. A), (2) a homozygote model (GG vs. AA), (3) a heterozygote model (AG vs. AA), (4) a dominant model (GG+AG vs. AA), and (5) a recessive model (GG vs. AG+AA). Heterogeneity among studies was assessed by Cochrane's Q-test [29], which follows a chisquare distribution. I 2 statistic, which is on a scale of 0-100% (0-25%, no heterogeneity; 25-50%, moderate heterogeneity; 50-75%, large heterogeneity; 75-100%, extreme heterogeneity) [30], is also computed. A Cochrane's Q test P-value < 0.10 [30] or an I 2 > 50% [31] was considered indicative of a statistically significant heterogeneity. A random effects model (the DerSimonian and Laird method) [32] was employed when a significant heterogeneity was detected among studies. Otherwise, a fixed effects model (the Mantel-Haenszel method) [33] was applied. Subgroup analyses stratified by ethnicity (Chinese populations vs. non-Chinese populations) were performed. The stability of the results was assessed using sensitivity analysis by removing each single study involved in the meta-analysis one at a time to reflect the influence of the individual study to the pooled ORs. The potential presence of publication bias was assessed by means of funnel plot inspection, and both Begg and Mazumdar adjusted rank correlation test [34] and Egger's linear regression test [35] were applied to test for funnel plot asymmetry. All statistical analyses were conducted using R version 3.2.3 software meta package

Characteristics of included studies
A flow diagram depicting the study selection process is shown in Fig 2

Meta-analysis results
For assessing the relationship between LEPR Q223R polymorphism and T2D risk, a total of 13 studies (11 articles) were included (Table 3) and a random effects model was employed because of the presence of significant heterogeneity. Under an allelic model, a comparison of G vs. A produced an OR of 1.09 (95% CI: 0.80-1.48), which was not statistically significant (Pvalue = 0.5989) ( Table 3 and Fig 3). Under genotypic models, comparisons of GG vs. AA, AG vs. AA, GG/AG vs. AA, and GG vs. AG/AA gave rise to ORs of 1.20, 1.08, 1.13, and 1.13 with P-values of 0.5741, 0.8177, 0.6871, and 0.3650, respectively, which also did not attain statistical significance. For assessing the relationship between LEPR K109R polymorphism and T2D risk, a total of 7 studies (7 articles) were included (Table 4) and a fixed effects model was employed because of a lack of significant heterogeneity. Under an allelic model, a comparison of G vs. A produced an OR of 0.93 (95% CI: 0.85-1.03), which did not reach statistical significance (Pvalue = 0.1868) (  P-values of 0.8087, 0.0207, 0.0384, and 0.8804 respectively, which all exceeded multiplicityadjusted α = 0.05/5 = 0.01 with control for 5 genetic models.

Test of heterogeneity
In the pooled analysis, for LEPR Q223R, a significant heterogeneity was detected for comparisons under 5 different genetic models, i.e., G vs. A, GG vs. AA, AG vs. AA, GG/AG vs. AA, and GG vs. AG/AA, such that I 2 was 90.20%, 86.10%, 82.90%, 88.00%, and 75.40%, respectively (P-value for heterogeneity < multiplicity-corrected α = 0.05/5 = 0.01 for considering 5 genetic models), as shown in Table 3. Therefore, a random effects model was chosen to estimate this SNP's pooled OR. For LEPR K109R, no statistically significant heterogeneity was detected for comparisons under 5 different genetic models, i.e., G vs. A, GG vs. AA, AG vs. AA, GG/AG vs. AA, and GG vs. AG/AA, such that I 2 s ranged from 0.00% to 13.60%, and P-values for heterogeneity ranged from 0.3274 to 0.8044, which exceeded multiplicity-corrected α = 0.05/ 5 = 0.01, as shown in Table 4. Because I 2 was under 50% and P-values for heterogeneity were not significant for all these genetic models, a fixed effects model was applied in estimating this SNP's pooled effect.

Subgroup analysis
To explore sources of heterogeneity across studies, subgroup analyses by ethnicity (i.e., Chinese populations vs. non-Chinese populations) were conducted. For LEPR Q223R, 7 studies   (Table 7). Under each of 5 genetic models, no significant heterogeneity was detected (I 2 ranged from 0% to 55.10%, and P-value for heterogeneity ranged from 0.1078 to 0.4121). For this SNP, 4 studies were performed in non-Chinese populations (Table 8). Under each of 5 genetic models, no significant heterogeneity was detected (I 2 was consistently 0.00% for each comparison, and P-value for heterogeneity ranged from 0.5877 to 0.7808). Therefore, a fixed effects model was employed under each of these 5 genetic models in Chinese and non-Chinese populations, respectively.

Sensitivity analysis
In order to assess the influence of each individual study on the pooled OR, we performed a sensitivity analysis by excluding each single study involved in the meta-analysis one at a time.  (Table 4). These findings show that our results were statistically robust for both of these two polymorphisms.

Bioinformatics analysis
Based on 7 different in silico tools, both LEPR Q223Rand K109R are predicted to exert a low impact on protein function (by Mutation Assessor), to be evolutionarily more acceptable (by BLOSUM62) neutral (by PROVEAN, PANTHER, SNPs&GO, and SNPs3D) and benign (by PolyPhen-2) ( Table 11). Discussion LEP, a pleiotropic hormone produced primarily by adipose tissue, plays an essential role in signaling energy status to the central nervous system (CNS), which has helped to redefine adipose tissue as an endocrine organ [60]. By binding to LEPRs expressed by neurons in CNS [61], leptin exerts its physiological effects on food intake, body weight, glucose and lipid metabolism, and regulation of immune function [15]. Although several independent studies identified significant associations between genetic variants of LEPR and obesity (e.g., [62,63]), others did not (e.g., [58,64]). Three meta-analysis studies (i.e., [65][66][67]) did not find significant relationships of LEPR polymorphisms with either obesity or obesity-related outcomes. In current study, 13 studies (11 articles; 4030 cases and 2844 controls) for Q223R, and 7 studies (7 articles;  3319 cases and 2465 controls) for K109R were included, which far exceed the sample size of any individual study. By employing 5 different genetic models to meta-analyze potential effects of these two missense SNPs on T2D risk, we did not detect statistically significant associations of either Q223R or K109R with T2D risk in either main analyses or subgroup analyses. Further, based on 7 software tools, both missense SNPs were predicted to be functionally neutral and benign. The VAFs for Chinese and non-Chinese populations for LEPR Q223R and K109R are not uniform across different ethnic groups. For Q223R, higher VAFs were observed in Chinese T2D cases (0.82) and controls (0.79) than in non-Chinese T2D cases (0.64) and controls (0.63), respectively (S1 Fig). Further, for K109R, higher VAFs were observed in Chinese T2D cases (0.83) and controls (0.82) than in non-Chinese T2D cases (0.40) and controls (0.42), respectively (S2 Fig). VAFs for both missense SNPs in Chinese populations of current study were similar to those reported in other studies, e,g., [61] and [68], which appear to be higher than in non-Chinese populations. As shown in Fan and Say (2014) [61], even among Asians, the respective allele frequencies of variant alleles R223 and R109 were notably higher in Chinese than Indians and Malays.
A comparison between the current meta-analysis and three other meta-analysis studies, i.e., Yang et al. (2016) [24], Liu et al. (2015) [69], Su et al. (2016) [70], is shown in Table 12. For  [70] such that no statistically significant associations were found. However, significant association was found by Yang et al. (2016) [24]. With respect to K109R, our results were concordant with those of Yang et al. (2016) [24] and Su et al. (2016) [70], such that no significant relationship was found between this missense SNP and T2D risk. With respect to another LEPR missense SNP, i.e., K656N, which was meta-analyzed by Yang et al. (2016) [24] and Su et al. (2016) [70], 5 and 4 studies were included in each of these two meta-analysis studies, respectively, which limited their abilities to draw robust conclusions on them. Therefore, to ensure that there are sufficiently large numbers of individual studies (i.e., > 5) amenable to subgroup analyses, only Q223R and K109R were assessed in the current study. We found that neither of these two missense SNPs is significantly associated with T2D risk. Taken together, based on our careful assessments, for Yang et al.  [70], there are errors (i.e., the genotype count data were incorrectly assigned to at least one included study) in data extraction from individual studies (affecting all these three studies) (affecting all of [24], [69], and [69), and errors (i.e., included studies contain overlapping data) in the number of individual studies included for meta-analysis a SNP (affecting [24] and [69]) (Table 12).
Caution should be taken when interpreting our results on the associations of gene polymorphisms with T2D. A significant heterogeneity was detected for Q223R (P-values for heterogeneity < multiplicity-corrected α = 0.05/5 = 0.01 for considering 5 genetic models (Table 3)], but not for K109R [range of P-values, 0.0205-0.6487, which were > multiplicity-corrected α = 0.05/5 = 0.01 (Table 4)] and subgroup analyses were conducted to explore reasons of heterogeneity. When stratified by ethnicity (i.e., Chinese vs non-Chinese populations), for Q223R, heterogeneity remained significant in each subgroup [P-values for heterogeneity < 0.0001 in Chinese populations (Table  5) and 0.002 non-Chinese populations (Table 6), respectively, which were all < multiplicitycorrected α = 0.05/5 = 0.01], and therefore, ethnicity did not appear to explain heterogeneity for Q223R. No heterogeneity was detected for K109R in either Chinese populations (Table 7) or non-Chinese populations (Table 8), because P-value for heterogeneity for each model was > multiplicity-corrected α = 0.05/5 = 0.01. In order to evaluate the influence of single studies on the overall estimate, a sensitivity analysis was performed by deleting each single study one at a time for allele model. The omission of any single study did not significantly alter pooled effect estimates for either Q223R (Table 9) or K109R ( Table 10), suggesting that our meta-analysis results were both reliable and credible. For assessments of publication bias, funnel plots were generated and their symmetries were tested using Begg and Mazumdar rank correlation and Egger's linear regression tests. Both tests revealed that no significant biases existed (P-values > 0.05 for all 5 genetic models for each SNP), and inspections of funnel plots also indicated no evidence of publication bias for the entire study sample [Fig 5 (Q223R) and Fig 6 (K109R)], and for either Chinese populations  Tables 2 and 4, and S10-S13 Figs of the study, i.e., Su et al., (2016)].
All data extraction problems of three previously published meta-analysis studies were addressed.

Results for each SNP
[The results shown were those reported by the study, which included incorrectly extracted data as indicated in above "Data accrual" section]: Q223R Pvalues ( [The results shown were those reported by the study, which included incorrectly extracted data as indicated in above "Data accrual" section]: Q223R P-value (   [70] their mistakes in data extraction were corrected and their weaknesses in considering 4 genetic models were well-addressed. (2) Both subgroup and sensitivity analyses were performed in the current study whereas only one of these two important types of analyses was employed by each of the three previously published meta-analysis studies (Table 12), which demonstrated that our results were statistically stable.
(3) The current study applied both Begg and Mazumdar adjusted rank correlation test and Egger's linear regression test whereas [of the three previously performed meta-analyses, only one study, i.e., Su et al. (2016) [70] employed both, but only for 4 genetic models], we did not detect any publication biases by funnel plot inspections in either main analyses or subgroup analyses, indicating that our results were unbiased. (4) In the current study, all included studies were of sufficiently high quality (i.e., NOS score ! 7), which all met our inclusion criteria. (5) To assess functional impacts of these two common missense SNPs, 7 in silico tools were applied, and their results were consistent with each other.
There are several limitations in the current study: (1) Our meta-analysis was based on unadjusted OR estimates due to a lack of individual participants' data. There is an important potential source of type II error β in the inference that LEPR genetic variants does not contribute to diabetes-susceptibility in our meta-analysis. Some of the individual studies, e.g., Liao et al. (2012) [23] and Roszkowska-Gancarz et al. (2014) [52], which were included for meta-analysis of both Q223R and K109R, did not match body weight and age between cases and controls, or adjust computationally for these important covariates which are critical to penetrance of genes predisposing to T2D. Since T2DM is highly correlated with body weight and age, using thinner and younger control subjects compared to T2D cases (e.g., Etemad et al. (2013) [49]), could confound the estimate of a non-weight dependent T2DM effect of LEPR genetic variants. (2) The study examined two most widely studied missense SNPs of LEPR in T2D, i.e., Q223R (rs1137101) and K109R (rs1137100) which were in a moderate level of linkage disequilibrium (LD) (e.g., r 2 = 0.3647 in Caucasians [71]), and haplotype-based association analysis could provide more statistical power than single SNP analysis [72][73][74]. (3) We applied a Bonferroni procedure to correct for the 5 genetic models tested, as in Wong et al. (2015) [75], and this procedure could be conservative. (4) The number of studies included in our meta-analysis, particularly the subgroup analyses according to ethnicity, was limited. (5) For Q223R, because individual studies had diverse population characteristics, significant between-study heterogeneity was observed, which could affect the precision of results, although the random effects model was applied in the presence of significant heterogeneity to pool ORs for this SNP. (6) T2D is polygenic and multifactorial, and there are a variety of possible genetic (> 80 genetic susceptibility loci have been identified [76], e.g., TCF7L2, PPARG), environmental (e.g., air pollution by nitrogen dioxide, PM 2.5 , and PM 10 [77]), nutritional (e.g., dietary fiber, fat intake [78]), lifestyle (e.g., physical inactivity [79]) and sociodemographic (e.g., age, ethnicity, education [80]) risk factors involved in the etiology of this disease. Because the definition of T2D varies among the individual studies [The World Health Organization (WHO) and American Diabetes Association (ADA) represent the two most widely used criteria (Tables 1 and 2)], over-(i.e., too many) or under-(i.e., too few) inclusion of subjects could be a possibility for each study. (7) Potential gene-gene and gene-environment interactions may influence the associations of LEPR gene Q223R and K109R polymorphisms and T2D risk. (8) This metaanalysis focused only on articles published in the English and Chinese languages, and there may be other eligible studies that were published in other languages.
In conclusion, to the best of our knowledge, the current study is most up-to-date, robust, and unbiased, when compared to previously published meta-analysis studies (i.e., Yang et al.  [70]) in this field. Neither Q223R nor K109R was significantly associated with T2D risk in the current meta-analysis, and bioinformatics analysis predicted that both SNPs are functionally neutral and benign. Additional welldesigned independent studies with sufficiently large sample sizes in various ethnicities could be conducted to confirm our findings.