Meta-analyses of the association of G6PC2 allele variants with elevated fasting glucose and type 2 diabetes

Objective To collectively evaluate the association of glucose-6-phosphatase catalytic unit 2 (G6PC2) allele variants with elevated fasting glucose (FG) and type 2 diabetes (T2D). Design Meta-analysis Data sources PubMed, Web of Knowledge and Embase databases. Study selection Full text articles of studies that identified an association of G6PC2 with T2D and elevated FG. Patient involvement There was no T2D patient involvement in the analyses on the association of FG with G6PC2, there were T2D patients and non-diabetes patient involvement in the analyses on the association of T2D with G6PC2. Statistical analysis Random-effects meta-analyses were used to calculate the pool effect sizes. I2 metric and H2 tests were used to calculate the heterogeneity. Begg's funnel plot and Egger’s linear regression test were done to assess publication bias. Results Of the 423 studies identified, 21 were eligible and included. Data on three loci (rs560887, rs16856187 and rs573225) were available. The G allele at rs560887 in three ethnicities, the C allele at rs16856187 and the A allele at rs573225 all had a positive association with elevated FG. Per increment of G allele at rs560887 and A allele at rs573225 resulted in a FG 0.070 mmol/l and 0.075 mmol/l higher (ß (95% CI) = 0.070 (0.060, 0.079), p = 4.635e-50 and 0.075 (0.065, 0.085), p = 5.856e-48, respectively). With regard to the relationship of rs16856187 and FG, an increase of 0.152 (95% CI: 0.034–0.270; p = 0.011) and 0.317 (95% CI: 0.193–0.442, p = 6.046e-07) was found in the standardized mean difference (SMD) of FG for the AC and CC genotypes, respectively, when compared with the AA reference genotype. However, the G-allele of rs560887 in Caucasians under the additive model and the C-allele of rs16856187 under the allele and dominant models were associated with a decreased risk of T2D (OR (95% CI) = 0.964 (0.947, 0.981), p = 0.570e-4; OR (95% CI) = 0.892 (0.832, 0.956), p = 0.001; and OR (95% CI) = 0.923(0.892, 0.955), p = 5.301e-6, respectively). Conclusions Our meta-analyses demonstrate that all three allele variants of G6PC2 (rs560887, rs16856187 and rs573225) are associated with elevated FG, with two variants (rs560887 in the Caucasians subgroup and rs16856187 under the allele and dominant model) being associated with T2D as well. Further studies utilizing larger sample sizes and different ethnic populations are needed to extend and confirm these findings.


Introduction
Fasting plasma glucose (FPG) levels are associated with a risk of type 2 diabetes (T2D) and cardiovascular disease [1]. There is strong evidence suggesting that hyperglycemia is a risk factor in a dose-dependent manner for both micro-and macro-vascular complications in both type 1 and type 2 diabetes [2].
Both genetic and environmental factors contribute to the pathophysiology of T2D [3,4]. However, the contribution of genetic factors to T2D risk is not well understood. Global knockout of theglucose-6-phosphatase catalytic unit 2 (G6PC2) gene in mice led to a significant decrease in blood glucose [5]. Previous studies have showed that higher FPG levels within the normal glucose range constitute an independent risk factor for T2D [1,6]. Considering the genetic risk that might result from G6PC2 alleles, a number of studies have explored the association of G6PC2 with fasting glucose (FG) and T2D in different ethnicities [7][8][9][10][11][12]. However, individual studies have yielded inconsistent or conflicting findings, possibly caused by limitations associated with an individual study, such as different genetic backgrounds and ethnicity, sample size and so on. Wang H et al [13] have previously performed a meta-analysis on the association of G6PC2 rs560887 with T2D. To expand and evaluate more precisely the relationship between G6PC2 and FG and T2D, we carried out meta-analyses of published studies.

Search strategy
We included all studies published prior to 4st April of 2017 that reported an association between G6PC2 and T2D and FG. Eligible studies were found by searching the PubMed, Web of Knowledge and Embase databases for relevant reports. We used the gene name "G6PC2" as search term limited in all fields to retrieve association studies between genetic variants in G6PC2 and FG or T2D. We also reviewed reference lists of the identified publications for additional relevant studies. A literature search was performed on these databases without restriction to regions or publication types. Two investigators (YY.S and YQ.L) independently searched the articles, and disagreements were resolved by discussion.

Patient involvement
There was no T2D patient involvement in the analyses on the association of FG with G6PC2; There were T2D patients and non-diabetes patient involvement in the analyses on the association of T2D with G6PC2.

Data extraction
Data were extracted and summarized independently by two of the authors. The adjudicating senior authors resolved any disagreement. If the data were unavailable, an attempt was made to contact the corresponding author to request missing data via E-mail. The following information was extracted: (1) study characteristics such as the first author, study name, year published, country; (2) subjects and methods characteristics including sex, mean age, sample size (n), Body Mass Index (BMI), genotyping method and blood samples measured for FG; (3) primary outcomes such as risk allele, risk allele frequency (RAF), OR with 95% CI and their adjustment factors, statistical methods and the p value of Hardy-Weinberg equilibrium (HWE) test in control group for data on T2D; Mean and SD of FG and sample size (n) in every genotype, and ß and their standard error (SE) or 95% CI, p value for linear regression and their adjustment factors as in previously published studies [44][45][46], and the p value of HWE for data on FG. All data were extracted independently by two investigators (YY.S and YQ.L), and discrepancies were resolved by discussion.

Quality assessment
The strengthening report of genetic association studies (STREGA) quality score system was used to assess the qualities of all included studies [47]. The STREGA system includes twentytwo quality assessment items with scores ranging from 0 to 22 (S1 Supplement). Studies are classified into three levels based on their scores: low quality (0-12), moderate-high quality (13)(14)(15)(16)(17), and high quality (18)(19)(20)(21)(22). Two authors (XJ.L and DD.Z) independently assessed the quality of included studies. Discrepancies over quality scores were resolved by discussing with all authors and subsequent consensus.

Statistical analysis
In our meta-analyses the included studies on rs560887 and rs573225 used an additive model to assess the genetic effect of G6PC2 polymorphisms [48]. For the studies on rs16856187, an additive model (AA versus AC versus CC) was used for the association of FG, and an allele model (A versus C), dominant model and recessive model were used for the association of T2D. The ß value and SEs were used to identify the association of rs560887 and rs573225 [14,32,44] with FG. SE of the ß value was calculated by 95% CI or ß value and p value when SE was not extracted directly from the original literature [49,50]. The SMD was used to analyze the association between rs16856187 and FG. ORs with 95% CIs were assessed to determine the relationship between T2D and rs560887 and rs16856187.
The aggregated results OR and 95% CI, ß and 95% CI and SMD were calculated using random-effects meta-analysis. A statistical test for heterogeneity was conducted using the I 2 metric and H 2 tests [51]. A I 2 greater than 50% or H 2 greater than 1 was suggestive of substantial between-study heterogeneity [52][53]. Sensitivity analyses were performed by omitting each study to identify possible study contributions to the heterogeneity. To evaluate the reliability and stability of our results, Begg's funnel plot and Egger's linear regression test were done to assess publication bias [54,55]. We divided the study populations into three ethnic subgroups, including Caucasians, Asians and African-Americans for the relationship between FG and rs560887, and two ethnic subgroups (Caucasians and Asians) for the relationship between T2D and rs560887. All analyses were performed using Stata 12.1 (Stata Corp, College Station, TX, USA).
The HWE for all the subjects of each study was evaluated using χ 2 test. For the studies which didn't include the distributions of genotypes but contained the information on the RAF in both cases and controls, we calculated the frequency of the different genotypes according to the HWE Law, which can be used to calculate the crude ORs and 95% CIs under an additive genetic model. All reported probabilities (p values) were two-sided, with p < 0.05 considered statistically significant.
Finally, for a better presentation of the public health relevance, we explored the PAR by taking into account both the pooled per-allele ORs and the pooled RAF (T2D risk allele frequency). PAR was calculated as PAR = (X − 1)/X. Assuming a multiplicative model, X = (1 − f) 2 + 2f(1 − f)γ + f 2 γ 2 , where f is RAF and γ is their estimated ORs. We calculated the pooled prevalence of each risk allele in various groups using the inverse variance method described previously [56].

Power calculations
Power to detect a genetic association was estimated using the QUANTO program version 1.2.4. For the association study with FG, we had an estimated power of more than 99.99% to detect a minimal per-allele effect at β of 0.070 mmol/l for rs560887 and 0.075 mmol/l for rs573225 under an additive model, depending on an allele frequency of 0.76 and 0.67. For the association study with T2D, we had an estimated power of 97.58% to detect an OR of 0.967 for rs560887 under the prevalence of 8.8% [57] under an additive model, and 98.78%, 64.23% and 10.01% power to detect genetic effects at an OR of 0.960, 0.892 and 0.923 for rs16856187 under allele, dominant and recessive model, respectively. A p value <0.05 was considered statistically significant (two-tailed).

Literature search results
Through literature searches a total of 423 articles from PubMed (National Center for Biotechnology Information), Web of Knowledge and Embase databases were identified up to 4st Aprilof 2017. A flow chart of study selection in the meta-analyses is shown in Fig 1. There were 52 articles included after duplicates were removed and following the screening of the titles and abstracts. As shown in S2 Supplement, 31 full-text articles were excluded. Overall, 21 articles were eligible and included (see study inclusion flowchart in Fig 1[7-12, [58][59][60][61][62][63][64][65][66][67][68][69][70][71][72]). Of these, 5 articles covered the loci rs560887 for FG and T2D, 3 articles covered the loci rs16856187 for FG and T2D, 2 articles covered the loci rs560887 and rs573225 for FG, 8 articles covered the loci rs560887 for FG, and 3 articles covered the loci rs560887 for T2D, respectively (Tables 1 and 2). In total, 18 studies comprising 69120 cases (62492 for rs560887, 6628 for rs16856187), 126483 non-diabetic controls (119627 for rs560887, 6856 for rs16856187), and 35 studies   containing 187968 non-diabetic participants (13752 for rs573225 and rs560887, 170772 for rs560887 and 3444 for rs16856187) were included. The meta-analyses were carried out according to the "Meta-analysis on Genetic Association Studies" statement (S3 Supplement).

Publication bias, sensitivity test
Visual inspection of funnel plots for the primary outcomes did not show distinct asymmetry, and based on Begg's funnel plots (S1 Fig) and Egger's linear regression, no publication bias was observed (all p > 0.1). In the sensitivity test (S2 Fig), the leave-one-out influential analyses did not show any major change in the primary outcome, indicative of a good stability of results.

Discussion
To the best of our knowledge, this is the most comprehensive meta-analyses on the evaluation of the associations between G6PC2 SNPs and FG and T2D. Our meta-analyses include the most SNPs in G6PC2 on the associations of these SNPs with FG and T2D studied to date. In these meta-analyses, we analyzed three SNPs (rs560887, rs16856187 and rs573225) in the G6PC2 gene for an effect on FG and two SNPs (rs560887 and rs16856187) for an association with T2D. We found that all three SNPs were associated with elevated FG level in participants with normal glucose regulation, and rs560887 in the Caucasians subgroup and rs16856187 under allele and dominant model were all associated with T2D.
However, no associations with T2D risk were found for G at rs560887 in overall and Asians populations, which is consistent with a meta-analysis published in 2013 [13]. Compared with this previous study, our study contained greater sample sizes in the Caucasians subgroup (number of case/control were 47673/97909 and 54586/111122 for previous and current studies, Meta-analyses of the association of G6PC2 with FG and T2D respectively). Association of G at rs560887 with T2D and FG in Caucasians is similar to individuals of European descent in the MAGIC study [OR (95% CI): 0.97 (0.95-0.99), ß (SE):0.075 (0.003) mmol/l, respectively] [70].
There may be some reasonable explanations for these differences between ethnic groups. First, the gene-gene interactions and different environmental factors may affect susceptibility to the genetic variant and diabetes [73,74]. Second, the sample size for Asian populations may be too small. Third, a previous study has reported that Asian subgroups have unique risk-factor profiles for developing diabetes, which differ from other populations [75]. Thus, further investigations on Asian populations are needed to replicate the observed association with type 2 diabetes. We will explore the association of G6PC2 with T2D in the Chinese population in the near future.
G6PC2 belongs to the G6PC family of proteins, which catalyze the dephosphorylation of glucose-6-phosphate to glucose [76]. Thus, glucose-6-phosphatase activity could control glucose metabolism and insulin secretion [76]. However, carriers of the G allele at rs560887 in Caucasians subgroup and A allele at rs16856187 in allele and dominant model all displayed a lower risk of type 2 diabetes and a higher risk of elevated FPG level, which is inconsistent with these previous studies [1,6]. The mechanism linking the SNP rs560887-A to reduced G6PC2 activity might be connected to the relative expression of the full-length active protein [23,77]. Studies have shown that heightened beta-cell sensitivity to glucose and a lowered glucose setpoint for insulin secretion are early steps toward ß cell apoptosis [78]. Recent reports in individuals of European descent also demonstrated a strong association between G6PC2 variants and insulin secretion [32]. The allele that decreased FPG was also found to lower beta cell function [65]. However, carriers of G at G6PC2 rs560887 displayed a higher risk of type 2 diabetes and a higher FPG level in Asians. It is unclear whether ethnic differences in beta cell function [79,80] contributed to the different results. Moreover, our relatively small sample size of Asians may also limit our ability to reach a reliable conclusion.
In addition, we also found that the presence of T at rs13387347, A at rs2232316, G at rs492594, A at rs483234, T at rs3755157 and C at rs478333 in G6PC2 among Asians were correlated with a higher risk of T2D [10, 65,69]. Meanwhile, C at rs478333 in adolescents, T at rs3755157 and A at rs483234 among Asians displayed a higher FG level. However, T at rs13387347 displayed a lower FG level [10, 65,69]. Due to a lack of data, a meta-analysis was not completed for these SNPs, yet they still provide evidence for the association of G6PC2 with FG and T2D.
This study shows that ß value (linear regression coefficient) rather than SMD for the association of FG with G6PC2 SNPs (rs560887 and rs573225) when pooled, which was not seen in the previous studies. This is currently the most comprehensive meta-analyses on G6PC2.
Some limitations in our meta-analyses should be mentioned. First, our results on FG and T2D were based on slightly different adjusted estimates. Second, the studies included in the analyses may be insufficient to allow firm conclusions. Thus, potential publication bias is likely to exist, in spite of the lack of evidence for this obtained from our statistical tests. The power to detect bias is limited, particularly for moderate amounts of bias or meta-analyses based on a small number of small studies [81]. Third, heterogeneity is also a potential problem, with estimates of zero or even just low heterogeneity being a concern since heterogeneity is very likely present but undetected [82]. Finally, the sample size for rs16856187 is small, and the estimate of the effect of rs16856187 on FG may be imprecise. Therefore, further study is necessary to confirm this finding.