Coding variants influencing FG levels

Through single-variant analysis, we identified 12 coding variants (all of which were non-synonymous changes) associated with FG levels at exome-wide significance, two low-frequency and ten common (S3 Table). These variants mapped to seven loci, six of them previously implicated in FG regulation. The signals at known loci included previously-reported common coding variants driving GWAS signals at GCKR (p.Pro446Leu, MAF = 36.9%, P = 5.3×10⁻¹⁸) and SLC30A8 (p.Arg325Trp, MAF = 35.7%, P = 2.5×10⁻¹⁰) [2, 11]. Three additional common coding variants associated with FG (in C2orf16, GPN1, and SLC5A6) were in moderate linkage disequilibrium (LD; r² = 0.2–0.4) with the p.Pro446Leu GCKR variant. Their associations were eliminated after conditioning on p.Pro446Leu, the known functional GWAS variant in this region [12, 13], indicating no causal role for these additional variants on FG regulation (S3 Table).

The sixth variant influencing FG levels at exome-wide significance was a low-frequency non-synonymous change which did not map to any previously known FG-influencing locus: p.Ala316Thr at GLP1R (MAF = 1.5%, P = 4.6×10⁻⁷; Table 1 and S1B Fig.). This variant showed modest association (P = 1.3×10⁻⁴) in a previous GWAS meta-analysis of FG [3], which partially overlaps with the present study, but now achieves exome-wide significance. The alanine residue at p.Ala316Thr is conserved across vertebrates, and the threonine substitution is predicted to be “possibly damaging” by in silico mutation analysis (S4 Table). GLP-1R (glucagon-like peptide 1 receptor) is the receptor for the incretin hormone glucagon-like peptide 1 (GLP1), which is released from enteroendocrine cells after food ingestion and potentiates insulin secretion. GLP1 receptor agonists are an established treatment for T2D [14].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Coding variants associated with FG and FI levels at exome-wide significance.

https://doi.org/10.1371/journal.pgen.1004876.t001

The six remaining coding variants all map to known FG-associated GWAS loci, but have not previously been implicated as playing a causal role. These include two variants in the islet-specific glucose-6-phosphatase catalytic subunit (G6PC2) gene at the G6PC2/ABCB11 locus: p.Val219Leu, a common variant (P = 6.0×10⁻⁹, MAF = 48.1%), and p.His177Tyr, a low-frequency variant (P = 3.1×10⁻⁸, MAF = 0.8%; S2A Fig.). These variants remained significantly associated (p.Val219Leu, P_conditional = 7.1×10⁻¹⁰ and p.His177Tyr, P_conditional = 1.3×10⁻¹¹) with FG after conditioning on the intronic lead GWAS SNP, rs560887 (Table 1 and S2B Fig.). Conversely, conditioning on the coding variants did not completely abolish the association signal at the lead GWAS SNP (P_{unconditional} = 6.4×10⁻⁷⁸; P_{conditional onHis177Tyr} = 3.1×10⁻⁵⁵, P_{conditional onVal219Leu} = 1.2×10⁻⁵⁸, and P_{conditional onHis177Tyr and Val219Leu} = 2.1×10⁻⁸³), confirming that the effect of the coding variants were largely independent of the lead GWAS SNP. Furthermore, the coding variants each remained associated with FG at exome-wide significance even after conditioning on both the lead GWAS SNP and the other coding variant, providing clear evidence of at least three association signals at this locus (Table 1 and S2C-S2D Fig.). These results are consistent with a recent study in Finnish individuals that reported a FG association signal at p.His177Tyr [15], but we extend that finding by demonstrating exome-wide significant association of multiple coding variants after conditioning on other associated variants in the region.

G6PC2 was also the only one of the 14,465 genes with multiple exome-array variants to demonstrate evidence of significant association with FG levels with any mask in gene-based tests (Table 2). In fact, for this gene we observed significant association with FG levels for all the masks encompassing multiple variants. These gene-based associations remained significant even after Bonferroni correction for testing of four different variant masks. Step-wise conditional analyses, adjusting for each variant included in the respective masks, revealed that the gene-based signals were primarily driven by two variants: p.His177Tyr and p.Tyr207Ser (S5 Table). The protein-truncating variant (PTV) p.Arg283*, which introduces a stop mutation in the last exon of the gene, showed no association with FG levels in single-variant or gene-based analyses. The most likely explanation is that this variant evades nonsense mediated decay (as is usual for PTV in the last exon) and that the truncated protein (missing only the terminal 72 amino acid residues) retains normal functional activity [16]. Due to its high allele frequency, and annotation as benign across multiple annotation algorithms, p.Val219Leu was not included in any of the gene-based variant masks (S6 Table). However, based on the single-variant it also has independent effects, with the result that we have three coding variants in G6PC2 (p.His177Tyr, p.Tyr207Ser, and p.Val219Leu) influencing FG levels. These three variants explain an additional 0.2% of the phenotypic variance in FG beyond that explained by the GWAS variant rs560887, bringing the total variance explained by this locus to 1.1%. However, we recognize that this estimate has been obtained in the discovery cohort and consequently might be inflated.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. G6PC2 gene-based association with FG levels using SKAT and BURDEN test

https://doi.org/10.1371/journal.pgen.1004876.t002

Both G6PC2 and ABCB11 have been considered strong biological candidate genes for glucose regulation, and advocated as potential effector transcripts at this GWAS locus [17, 18]. However, none of the 21 coding variants in ABCB11 that passed quality control (twelve rare, eight low-frequency, and one common) were significantly associated with FG levels, nor was there an aggregate association signal (P>0.05). Together, our genetic data provide compelling evidence that G6PC2 is an effector gene for FG regulation at this GWAS locus, with p.His177Tyr, p.Tyr207Ser, and p.Val219Leu as likely causal coding variants.

Loss of G6PC2 function leads to a reduction in FG levels

The phenotypic impact of these coding variants might, we reasoned, be influenced by the action of the non-coding GWAS variants on G6PC2 transcript regulation. Haplotype analysis of the four variants (rs560887, p.His177Tyr, p.Tyr207Ser, and p.Val219Leu) in a subset of 4,442 individuals revealed that the C allele encoding Leu219 was carried exclusively in cis with the glucose-raising allele at the GWAS SNP (Fig. 1). The estimated effect size of p.Val219Leu was considerably smaller ( =0.020 ± 0.004 mmol/L per allele) than rs560887 ( =0.070 ± 0.004 mmol/L per allele). These observations together explain the reversal in direction of effect of the Leu219 allele between conditional and unconditional analyses and indicate that, whilst Leu219 appears to have a glucose-raising effect in single-variant analyses, the molecular consequence of Leu219 is likely to be to lower glucose (Table 1). The minor alleles of the other two coding variants (p.His177Tyr, p.Tyr207Ser) also displayed glucose-lowering effects in conditional analyses. Given the role of G6PC2 in beta-cells we predicted that these would also be associated with reduced G6PC2 function, a hypothesis that we set out to test with a series of in vitro studies.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Haplotypes of the lead non-coding GWAS SNP rs560887 and the three coding variants.

rs138726309 (p.His177Tyr), rs2232323 (p.Tyr207Ser), and rs492594 (p.Val219Leu), obtained from 4,442 unrelated individuals from the Oxford Biobank. (A) Percentage minor allele frequency (MAF) and effect size estimates () of the four variants reported for the minor allele in mmol/L of FG after adjustment for age, sex, and BMI. (B) Haplotypes of the four associated variants in G6PC2 revealed that the glucose-lowering Leu219 allele was carried exclusively in cis with the glucose-raising allele at the GWAS SNP. Wild-type, glucose-raising alleles are circled in blue and the mutant, glucose-lowering alleles are circled in red. Diameter of the circle is proportional to the effect size estimates. Haplotype association was performed with FG derived residuals (after adjustment for age, sex, and BMI) using the most frequent haplotype as baseline.

https://doi.org/10.1371/journal.pgen.1004876.g001

First, we generated recombinant FLAG-tagged G6PC2 constructs to investigate the impact of these variants on protein expression and subsequent cellular localization. Transient transfection of all three constructs showed a marked reduction in G6PC2 protein levels. In HEK293 cells, protein levels were decreased by 99% (p.His177Tyr), 100% (p.Tyr207Ser), and 49% (p.Val219Leu), when compared to wild type (defined for this study as the major allele for each of these variants) G6PC2 (Fig. 2A). Reduced protein expression was also confirmed in the rat pancreatic β-cell line INS-1E (76%, 97%, and 49% reduction, respectively; Fig. 2B). The functional impact of the variant proteins mirrored our genetic observations; the variant protein with the p.Val219Leu substitution, which has only a modest effect on FG, was present at higher levels in both HEK293 and INS-1E cells than the p.His177Tyr and p.Tyr207Ser proteins, both of which have larger impacts on FG levels.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Functional characterization of wild type and variant G6PC2 proteins.

(A) Expression levels in HEK293 and (B) INS-1E cells were determined by western blot and densitometry analysis. The multiple bands on the western blot are likely to represent glycosylated G6PC2 protein products. Data are presented as mean ± standard error of the mean for at least three independent experiments. Significant differences are indicated as ** P<0.01; *** P<0.001; **** P<0.0001. EV, empty vector; WT, wild type. (C) Expression levels in HEK293 and INS-1E cells in the presence of proteasomal inhibitor MG-132 or lysosomal inhibitor chloroquine were determined by western blot. (D) Cellular localization in HEK293 cells was assessed by immunofluorescence microscopy. Cells were double immunostained for FLAG tag (green) and calnexin (red), and merged images with a DNA stain (blue) are shown. Images were taken with laser settings that were optimized separately for each sample. Scale bar, 10µm.

https://doi.org/10.1371/journal.pgen.1004876.g002

Our functional results for p.His177Tyr and p.Tyr207Ser are concordant with data from the in silico mutation assessment tools SIFT and PolyPhen-2, which predicted these variants as deleterious (S4 Table). In contrast, the common p.Val219Leu variant was predicted to be benign, whereas in vitro characterization clearly shows a significant reduction in protein expression, demonstrating the importance of experimentally evaluating coding variants for functional consequences. The observed decrease in G6PC2 protein levels caused by the FG-associated variants is also directionally consistent with the evidence indicating that the non-coding GWAS SNP rs560887 may have effects on pre-mRNA splicing [19].

Next, we used specific inhibitors of the proteasomal and lysosomal pathways (MG-132 and chloroquine, respectively) to demonstrate that the three G6PC2 variant proteins with p.His177Tyr, p.Tyr207Ser, and p.Val219Leu substitutions were predominantly degraded through the ubiquitin-proteasome pathway, an important cellular mechanism for clearing misfolded proteins. Protein expression could be rescued in the presence of MG-132 but not chloroquine (Fig. 2C). We also evaluated the impact of the variants on cellular localization of G6PC2 to the endoplasmic reticulum (ER) using calnexin as an ER marker. The variant proteins displayed similar localization patterns to wild type G6PC2 (Fig. 2D). Our finding that loss of G6PC2 function leads to a reduction in FG levels in humans is consistent with rodent data, which show that G6pc2 knockout mice have a ~15% decrease in FG levels [20]. Hence, our data, linking genetic associations with reduced protein function, indicate that normal G6PC2 function is critical for glucose homeostasis and that these variants most likely impact FG levels through altered intracellular catalysis of glucose-6-phosphate to glucose and inorganic phosphate in pancreatic β-cells.

Impact of the functional G6PC2 variants on other related quantitative traits and disease outcomes

We evaluated the impact of these functional G6PC2 variants on other related quantitative traits and disease outcomes to gain further insights into the metabolic processes involved (S7 Table). None of the variants showed any evidence of association with FI levels (P>0.1) in 30,825 non-diabetic individuals analyzed in the present study. No association of the lead G6PC2 GWAS variant with T2D risk has previously been shown in Europeans [2, 21, 22]. However, in a meta-analysis of exome-array data from 28,344 T2D cases and 51,801 controls of European ancestry (including the 33,231 controls used in the present meta-analyses), the common coding variant, p.Val219Leu, showed modest association with T2D risk (P = 0.0011; odds ratio = 1.05, 95% confidence interval 1.02–1.06). The G allele encoding Val219, which displayed a glucose-raising effect, conferred an increased risk of disease. In contrast to the observations in FG single-variant analysis, the direction of effect on T2D at this variant was unchanged, even after conditioning on the GWAS variant. This is consistent with a small case-control study in 3,676 individuals of Chinese ancestry where a nominal association (P = 0.0062) with T2D risk was reported [23]. Our finding provides evidence that G6PC2 is critical not only for controlling FG levels in the physiological range, but that impairment of its function contributes to T2D pathogenesis. The effect on T2D risk is modest given the impact of the p.Val219Leu variant on FG levels but the effect size is similar to that reported for the common non-coding variant GWAS signal at GCK [2]. GCK encodes the glycolytic enzyme glucokinase which catalyzes the reverse reaction to G6PC2. These observations are consistent with defects in glucose sensing rather than β-cell function.

Likely effector transcripts at established FG GWAS loci

Of the 12 coding variants significantly influencing FG levels, we have discussed eight above. The four remaining signals mapped to three additional established FG-associated GWAS regions and support the candidacy of particular effector transcripts at these loci. Two map to PCSK1 (p.Gln665Glu, MAF = 27.9%, P = 3.0×10⁻⁸; p.Ser690Thr, MAF = 27.9%, P = 4.1×10⁻⁸), and have previously been proposed as likely causal variants at this GWAS locus because of strong LD with the lead non-coding GWAS SNP (rs4869272) in Europeans (r² = 0.81) [3]. These two variants are in complete LD with each other (r² = 1) and both have residual association signals after adjusting for the lead GWAS SNP (p.Gln665Glu, P_conditional = 8.1×10⁻³; p.Ser690Thr, P_conditional = 0.01) (Table 1 and S3 Table). Conversely, conditioning on the coding variants completely abolished the association signal at the GWAS SNP (P_{unconditional} = 7.2×10⁻⁷; P_{conditional onGln665Glu} = 0.24, P_{conditional onSer690Thr} = 0.25, and P_{conditional onGln665Glu and Ser690Thr} = 0.34). Although we were unable to statistically distinguish between these two coding variants, there is suggestive in vitro evidence that the p.Gln665Glu variant may decrease PCSK1 activity whilst p.Ser690Thr behaved as wild-type [24], supporting the former as the most likely causal variant at the PCSK1 locus. PCSK1 encodes prohormone convertase 1/3 (PC1/3), a calcium-dependent serine endoprotease, which is essential for the conversion of a variety of prohormones, including proinsulin and proglucagon, to their bioactive forms.

The penultimate variant was in ZHX3 (p.Asn310Ser, MAF = 23.8%, P = 3.9×10⁻⁷) which maps to the TOP1 GWAS locus influencing FG levels [4]. Adjustment for p.Asn310Ser in a conditional analysis eliminated the association signal at the non-coding GWAS lead SNP rs6072275 (P_{unconditional} = 2.5×10⁻⁵ and P_conditional = 0.33), supporting ZHX3 as the plausible causal gene at this locus (Table 1 and S3 Table). ZHX3 (zinc fingers and homeoboxes 3) encodes a transcriptional repressor which belongs to a protein family known to regulate gene expression in the kidney podocytes and plays roles in both lipoprotein metabolism and triglyceride regulation in mice [25, 26].

The final variant was in RREB1 (p.Ser1554Tyr, MAF = 21.1%, P = 8.4×10⁻⁹) which resides in the RREB1/SSR1 GWAS locus influencing FG levels [4] and T2D risk [27] (Table 1). We were unable to explore the relationship between this association signal and the lead FG (rs17762454) or T2D (rs9502570) GWAS SNPs at this locus as neither the variants themselves, nor a close proxy (r²>0.80), was present on the exome-array. Based on European haplotypes from the 1000 Genomes Project, the coding variant was more strongly correlated (r² = 0.59) with rs9502570 as compared to rs17762454 (r² = 0.08), which might indicate multiple FG association signals in this region. These data point to a likely functional role for RREB1 at this locus, although further studies are needed to confirm this hypothesis.

Coding variants influencing FI levels

Turning to the analysis of FI, we identified six non-synonymous variants associated at exome-wide significance (one rare and five common). These mapped to four loci, three of them previously implicated in FI regulation (S3 Table). The single novel FI- influencing locus was represented by a rare variant in URB2 (p.Glu594Val, MAF = 0.1%, P = 3.1×10⁻⁷; Table 1 and S1 Fig.). The variant allele was observed in individuals across Europe, including UK, Denmark, Finland, and Sweden, and genotype calls across all cohorts passed visual inspection for clustering accuracy. Heterozygous genotypes (n = 6) were 100% concordant for 3,999 individuals genotyped on the array that had also been exome sequenced (S8 Table). This variant has a relatively large effect, with each copy of the minor allele increasing FI level by 32%. URB2 encodes ribosome biogenesis 2 homolog, which interacts with nuclear lamins [28]. The precise mechanistic role of URB2 is still poorly understood, but conditions caused by defects in lamin genes (laminopathies), including familial partial lipodystrophy, can cause loss of adipose tissue, insulin resistance, and metabolic syndrome [29, 30].

The remaining five coding variants implicated in the FI analysis, included three previously-reported FI-associated common variants in GCKR (p.Pro446Leu, P = 8.1×10⁻¹¹), PPARG (p.Pro12Ala, MAF = 14.7%, P = 1.3×10⁻⁷), and COBLL1 (p.Asn939Asp, MAF = 11.3%, P = 6.7×10⁻⁸) [3, 4]. The final two variants (in C2orf16 and GPN1) also mapped within the GCKR FG/FI-associated GWAS locus, and their associations were eliminated after conditioning on the GCKR p.Pro446Leu variant as seen in FG analysis (S3 Table).

Data availability

Summary statistics of single-variant and gene-based analyses are available at http://www.diagram-consortium.org/Mahajan_2014_ExomeChip/

Ethics statement

All human research was approved by the relevant institutional review boards, and conducted according to the Declaration of Helsinki and all patients provided written informed consent. FIN-D2D 2007, DPS, DR’s EXTRA, FINRISK 2007, FUSION, and METSIM were approved by the University of Michigan Health Sciences and Behavioral Sciences Institutional Review Board (ID: H03-00001613-R2). The Danish studies (Health 2006, Inter99, and Vejle Biobank) were approved by the local Ethical Committees of Capital Region (approval # H-3-2012-155, KA 98155 and KA-20060011) and Region of Southern Denmark (approval # S-20080097). The GoDARTS study was approved by EoS REC 09/S1402/44. The Twins UK study was approved by EC04/015. The OBB study was a approved by South Central, Oxford C, 08/H0606/107+5, IRAS project 136602. The PIVUS study is approved by 00–419 and ULSAM study by 251/90 and 2007/338. The PPP study was approved by the Committee On the Use of Humans as Experimental Subjects at MIT (IRB 0912003615).

Phenotypes

Study participants. FG was measured in mmol/L and FI in pmol/L. Individuals were excluded from the analysis if they had a physician diagnosis of diabetes, were on diabetes treatment (oral or insulin), or had a fasting plasma glucose concentration ≥7 mmol/L or ≥11.1 mmol/L following a 2-hour oral glucose tolerance test. Individual studies applied further sample exclusions, including for pregnancy, non­fasting measurements, and type 1 diabetes (S1 Table). Measures of fasting glucose made in whole blood were corrected to approximate plasma level by multiplying by 1.13 [35].

Trait transformations and adjustment. To achieve approximate normality of the traits within each study, FG and natural logarithm–transformed FI levels were adjusted for age, sex, BMI, and study specific covariates followed by inverse normalization of the residuals. Inverse normalized residuals were used as the dependent quantitative trait in genetic association models to calibrate type 1 error. Effect estimates were obtained using untransformed FG and natural logarithm–transformed FI levels after adjusting for age, sex, BMI, and study specific covariates.

Genotyping and quality control

The Illumina HumanExome Beadchip array was genotyped by individual studies. This custom array was designed to facilitate large-scale genotyping of 247,870 mostly rare (MAF<0.5%) and low-frequency (MAF 0.5–5%) protein altering variants selected from sequenced exomes and genomes of ~12,000 individuals (see URLs). Details of genotype calling and quality control are presented in S1 Table. To confirm the genotyping quality of all the variants discussed in the manuscript, we compared the genotype calls in 2,312 to 4,000 individuals for which we have both exome-array genotypes and exome-sequence data (S8 Table). The results are highly concordant for all variants (99% heterozygous genotype concordance and 100% concordance observed for non-reference homozygotes). Call rate and Hardy-Weinberg equilibrium p-values for each variant are provided in S9 Table.

Association analysis

Single-variant analysis. We tested single variants for association with FG- and FI-derived inverse normalized residuals assuming an additive genetic model using a linear mixed model to account for relatedness with EMMAX [5]. Study-specific QQ plots and genomic lambdas are shown in the S3 Fig. We repeated single-variant association analyses with untransformed FG and natural logarithm–transformed FI residuals to obtain effect estimates. We then combined the association summary statistics across studies by using a fixed-effects meta-analysis (sample size Z-score weighted) using METAL [36]. Genotype cluster plots of all variants described here were inspected visually in all studies.

Single-variant conditional analysis. Conditional analysis was performed to identify additional association signals at known or novel loci. The analysis included the genotypes of the lead variant(s) at the conditioning loci as covariate(s) in the regression analysis in EMMAX. We then performed meta-analysis of the association summary statistics across studies by using a fixed-effects meta-analysis (sample size Z-score weighted).

Gene-based analysis. For gene-based testing, we first computed single-variant score statistics and their covariance matrices for all polymorphic markers within each study. We then combined the single-variant score statistics from all studies using the Cochran-Mantel-Haenszel method and computed corresponding variance-covariance matrices centrally [6]. These variance-covariance matrices were used to compute gene-level statistics. We applied a frequency-weighted burden test which assumes variants have similar effect sizes and SKAT, a dispersion test that performs well when both protective and deleterious variants are present [8, 9]. Test-specific asymptotic distributions were used to evaluate significance. For gene-based analyses, we used only unrelated individuals (n = 32,223 for FG and n = 29,848 for FI) and included principal components as covariates to adjust for population structure. We generated four variant lists using frequency data and functional annotation. Variants were mapped to transcripts in Ensembl 66 (GRCh37.66). Using annotations from CHAoS v0.6.3, SnpEff v3.1, and VEP v2.7, we identified variants predicted to be protein-truncating (PTV; e.g. nonsense, frameshift, essential splice site) or protein-altering (e.g. missense, in-frame indel, non-essential splice site) in at least one mapped transcript (by at least one of the three algorithms) [37, 38]. We additionally used the procedure described by Purcell et al. to identify subsets of missense variants meeting “strict” or “broad” criteria for being deleterious, using annotation predictions from Polyphen2-HumDiv, PolyPhen2-HumVar, LRT, MutationTaster, and SIFT [10]. Masks 1 (“PTV-only”) and 3 (“PTV + NS_strict”) are restricted to variants with predicted major effect on protein function, and, as a result, disproportionately favor inclusion of rare variants, whilst masks 2 (“PTV + missense”) and 4 (“PTV + NS_broad”) are more permissive. In total, we tested 1,028, 14,465, 4,603, and 13,093 genes having at least two such variants in the above four categories respectively. For gene-based tests reaching exome-wide significance, if the conditional analysis showed that the signal was driven by a single variant, we required the variant to achieve exome-wide significance in the single-variant test as well.

Gene-based conditional analysis. We performed conditional gene-level analysis to evaluate whether rare or low-frequency variants associated with the trait in single-variant analysis could account for or were due to a gene-based test association signal [6]. The genotypes of the variant(s) at the conditioning locus were included as covariate(s) in this analysis.

Estimating phenotypic variance explained by SNPs. We used a subset of Finnish samples (FIN-D2D 2007, DPS, DR’s EXTRA, FINRISK 2007, and METSIM; n = 10,266) to calculate variance explained by G6PC2. We ran a model regressing BMI adjusted FG on the four G6PC2 variants: intronic GWAS lead SNP rs560887, p.His177Tyr (rs138726309), p.Tyr207Ser (rs2232323), and p.Val219Leu (rs492594), assuming an additive model (and adjusting for sex, age, age², and study origin). A separate model was run excluding the GWAS SNP to determine the additional variance captured with the three coding variants.

Power calculations.We had >99.9% power to identify variants that explain >0.3% of the phenotypic variance and 80% power to detect coding variants that explain >0.1% of the phenotypic variance. To achieve >80% power for variants with MAF<0.05% would require effect sizes of at least 1 SD unit of residuals of mmol/L for FG and pmol/L for FI.

When we estimate power to detect association for aggregation tests, we make many assumptions: i) proportion of variants contributing to trait variation, ii) direction of effects, iii) number of variants aggregated, and iv) allele frequency distribution of the variants [34]. In this study, assuming 100% of the variants contribute to trait variation, we had >99.9% power to detect association for genes that explain >0.25% of the phenotypic variance and 80% to detect genes that explain >0.1% of the phenotypic variance using a burden test and >0.5% of the phenotypic variance using SKAT. In a less favorable scenario, for example assuming a gene explains 0.25% of the phenotypic variance, 25% of the variants contribute to trait variation, different directions of effect, 20 variants tested, and variants sampled from the reported allele frequency distribution in the exome-array design, power to detect association in this study may be near 0% for a burden test and 63% for SKAT [34].

Haplotype analysis. In 4,442 individuals from the Oxford Biobank, we used an expectation-maximization (EM) algorithm to obtain the posterior distribution of haplotypes consistent with the observed genotypes at four G6PC2 variants: intronic GWAS lead SNP rs560887, p.His177Tyr (rs138726309), p.Tyr207Ser (rs2232323), and p.Val219Leu (rs492594). Haplotype association with FG- and FI-derived residuals (after adjustment for age, sex, and BMI) was tested in a linear regression framework, as a function of haplotype dosage from posterior distribution, and including principal components as covariates to account for population structure using the most frequent haplotype as baseline.

In-silico mutation analysis. SIFT [39], PolyPhen-2 [40], and Condel [41] algorithms were used to predict the functional effects of the associated non-synonymous variants on protein function. Genomic Evolutionary Rate Profiling (GERP) [42] scores were calculated to indicate the degree of evolutionary conservation at a given human nucleotide based on multiple genomic sequence alignments and were measured as site-specific ‘rejected substitutions’: higher scores indicate greater conservation.

Functional studies

Site directed mutagenesis. Human G6PC2 cDNA (NM_021176.2) within a pCMV6-Entry vector (with a C-terminal FLAG-tag) was purchased from OriGene (RC211146). We generated non-synonymous variants in the G6PC2 coding sequence of the clone using Quikchange II Site-Directed Mutagenesis (Agilent). All mutations were verified by Sanger sequencing; in each case, only the desired nucleotide substitutions was introduced.

Western blot analyses. HEK293 and INS-1E 832–13 cells were transfected with each FLAG-tagged wild type or mutant G6PC2 construct using Lipofectamine 2000 (Invitrogen). In protein degradation assays, cells were treated with 10μM MG-132 (Calchembio) or 100μM chloroquine (Sigma) for 15h. At 48h after transfection, cells were collected and homogenized by sonication in lysis buffer. Total cellular protein was separated by 4–12% SDS-PAGE (Invitrogen) and transferred to nitrocellulose membranes. We determined G6PC2 expression by immunoblotting using a mouse monoclonal FLAG M2 antibody (Sigma, F1804). A rabbit polyclonal antibody against β tubulin (Santa Cruz, sc-9104) was used as a loading control. Secondary antibodies specific to mouse or rabbit IgG were obtained from Thermo Fisher Scientific. Protein bands were detected using the ECL reagent (Pierce Thermo Fisher Scientific). Western blots were quantified by densitometry analysis using ImageJ and paired t tests of densitometric data were carried out in GraphPad Prism.

Immunofluorescence. HEK293 cells were transfected with each FLAG-tagged G6PC2 construct using FuGene 6 transfection reagent (Promega) in 4-well chamber slides (BD Biosciences). After 48h, cells were fixed with 4% paraformaldehyde in PBS for 15 min. Cells were permeabilized with 0.05% Triton X-100 in PBS for 15 min, and blocked for 1 h with 10% BSA in PBS-Tween 20. We carried out double immunostaining of cells using FLAG M2 (Sigma, F1804) and calnexin (Santa Cruz, sc-11397) primary antibodies followed by anti-mouse fluorescein (Vector Labs) and anti-rabbit TRITC (Dako). DRAQ5 fluorescent probe (Thermo Fisher Scientific) was applied at 20μM as a far-red DNA stain. Slides were mounted with Vectashield mounting medium (Vector Labs) and visualized on a BioRad Radiance 2100 confocal microscope with a 60× 1.0 N.A. objective. Images were acquired with different laser settings that were optimized for each sample and therefore fluorescent intensities are not comparable across samples.

URLs

1. Exome-array design,
   http://genome.sph.umich.edu/wiki/Exome_Chip_Design
2. EPACTS,
   http://genome.sph.umich.edu/wiki/EPACTS
3. METAL,
   http://www.sph.umich.edu/csg/abecasis/Metal/download/
4. RareMETALS,
   http://genome.sph.umich.edu/wiki/RareMETALS
5. The 1000 Genomes Project,
   www.1000genomes.org

Members of T2D-GENES consortium and GoT2D consortium

Hanna E Abboud¹, Uzma Afzal², David Aguilar³, Rector Arya¹, Gil Atzmon⁴, Tin Aung^{5, 6}, Eric Banks⁷, Inês Barroso^{8, 9}, Nir Barzilai⁴, Jennifer E Below¹⁰, Dwaipayan Bharadwaj¹¹, Thomas W Blackwell¹², Lori L Bonnycastle¹³, Don Bowden^14–16, Jason Carey⁷, Mauricio O Carneiro⁷, John C Chambers^{2, 17, 18}, Edmund Chan¹⁹, Juliana Chan^20–22, Giriraj R Chandak²³, Peng Chen²⁴, Yuhui Chen²⁵, Han Chen^{26, 27}, Ching-Yu Cheng^{5, 6, 24, 28}, Kee Seng Chia²⁴, Yoon Shin Cho²⁹, Adolfo Correa³⁰, Joanne E Curran³¹, Mark J Daly³², Aaron G Day-Williams⁸, Ralph A DeFronzo¹, Mark DePristo⁷, Peter J Donnelly^{25, 33}, Shah B Ebrahim³⁴, Paul Elliott^{2, 35}, Tõnu Esko⁷, ^36–38, João Fadista³⁹, Yossi Farjoun⁷, Andrew J Farmer⁴⁰, Vidya S Farook³¹, Timothy Fennell⁷, Teresa Ferreira²⁵, Tasha Fingerlin⁴¹, Tom Forsén^{42, 43}, Sharon P Fowler¹, Paul W Franks^44–46, Timothy M Frayling⁴⁷, Barry I Freedman⁴⁸, Philippe Froguel⁴⁹, Eric R Gamazon⁵⁰, Christian Gieger⁵¹, Benjamin Glaser⁵², Min Jin Go⁵³, Jacqueline I Goldstein^{7, 32}, Harald Grallert^54–56, George Grant⁷, Todd Green⁷, Michael Griswold⁵⁷, Daniel Esten Hale⁵⁸, Bok-Ghee Han⁵³, Christopher Hartl⁷, Andrew T Hattersley⁵⁹, Pamela J Hicks^14–16, Dylan Hodgkiss⁶⁰, Momoko Horikoshi^{25, 61}, Martin Hrabé de Angelis⁶², Cheng Hu⁶³, Frank B Hu^{44, 64}, Iksoo Huh⁶⁵, Mohammad Kamran Ikram^{5, 6}, Thomas Illig^{55, 66}, Kathleen A Jablonski⁶⁷, Christopher P Jenkinson^{1, 68}, Weiping Jia⁶³, Hyun Min Kang¹², Chiea-Chuen Khor^{5, 6, 24, 69, 70}, Yongkang Kim⁶⁵, Young Jin Kim⁵³, Bong-Jo Kim⁵³, Leena Kinnunen⁷¹, Jaspal Singh Kooner⁷², Jasmina Kravic³⁹, Jennifer Kriebel^{55, 56}, Ashish Kumar^{25, 73}, Satish Kumar³¹, Teemu Kuulasmaa⁷⁴, Min-Seok Kwon⁷⁵, Claudia Langenberg⁷⁶, Torsten Lauritzen⁷⁷, Selyeong Lee⁶⁵, Jaehoon Lee⁶⁵, Juyoung Lee⁵³, Jong-Young Lee⁷⁸, Donna M Lehman¹, Benjamin Lehne², Jonathan C Levy⁶¹, Jiang Li¹⁴, Liming Liang^{27, 64}, Wei Yen Lim²⁴, Keng-Han Lin¹², Jianjun Liu^{24, 70}, Marie Loh^{2, 79}, Ronald C W Ma^20–22, Clement Ma¹², Reedik Mägi³⁷, Jared Maguire⁷, Taylor J Maxwell¹⁰, Gilean McVean²⁵, Christa Meisinger⁵⁴, Thomas Meitinger^80–82, Olle Melander⁸³, Andres Metspalu³⁷, Evelin Mihailov³⁷, Lili Milani³⁷, Loukas Moutsianas²⁵, Martina Müller-Nurasyid^{51, 80, 84, 85}, Solomon K Musani⁸⁶, Yoshihiko Nagai^87–89, Narisu Narisu¹³, Benjamin M Neale^{7, 32}, Maggie C Y Ng^{14, 15}, Peter Nilsson⁹⁰, Stephen P O'Rahilly⁹, Marju Orho-Melander⁹¹, Katharine R Owen^{61, 92}, Nicholette D Palmer^14–16, Taesung Park^{65, 75}, Dorota Pasko⁴⁷, Richard D Pearson²⁵, John R B Perry^{25, 47, 76}, Annette Peters^{54, 56, 80}, Toni I Pollin⁹³, Ryan Poplin⁷, Dorairaj Prabhakaran³⁴, Sobha Puppala³¹, Shaun Purcell^{7, 94, 95}, Lu Qi^{44, 96}, Qibin Qi^{44, 97}, Michael Roden⁹⁸, Olov Rolandsson⁴⁵, Anders H Rosengren³⁹, Manjinder Sandhu^{8, 99}, Thomas Schwarzmayr⁸², Laura J Scott¹², Robert A Scott⁷⁶, James Scott⁷², William R Scott², Jobanpreet Sehmi^{17, 72}, Khalid Shakir⁷, Rob Sladek^{87, 88, 100}, Joshua D Smith¹⁰¹, Alena Stančáková⁷⁴, Konstantin Strauch^{51, 85}, Tim M Strom^{81, 82}, Amy Swift¹³, E Shyong Tai^{19, 24, 102}, Juan Fernandez Tajes²⁵, Sian-Tsung Tan^{17, 72}, Nikhil Tandon¹⁰³, Herman A Taylor Jr³⁰, Yik Ying Teo^{24, 104, 105}, Farook Thameem¹, Barbara Thorand^{54, 56}, Martijn van de Bunt^{25, 61}, Tibor V Varga⁴⁶, Mark Walker¹⁰⁶, Nicholas J Wareham⁷⁶, Ryan P Welch¹², Thomas Wieland⁸², Gregory Wilson Sr¹⁰⁷, Tien Yin Wong^{5, 6}, Andrew R Wood⁴⁷, Joon Yoon⁷⁵, Eleftheria Zeggini⁸, Weihua Zhang^{2, 17}

1. Department of Medicine, University of Texas Health Science Center, San Antonio, Texas, USA.
2. Department of Epidemiology and Biostatistics, Imperial College London, London, UK.
3. Cardiovascular Division, Baylor College of Medicine, Houston, Texas, USA.
4. Departments of Medicine and Genetics, Albert Einstein College of Medicine, New York, USA.
5. Department of Ophthalmology, Yong Loo Lin School of Medicine, National University of Singapore, National University Health System, Singapore.
6. Singapore Eye Research Institute, Singapore National Eye Centre, Singapore.
7. Program in Medical and Population Genetics, Broad Institute, Cambridge, Massachusetts, USA.
8. Department of Human Genetics, Wellcome Trust Sanger Institute, Hinxton, Cambridgeshire, UK.
9. Metabolic Research Laboratories, Institute of Metabolic Science, University of Cambridge, Cambridge, UK.
10. Human Genetics Center, School of Public Health, The University of Texas Health Science Center at Houston, Houston, Texas, USA.
11. Functional Genomics Unit, CSIR-Institute of Genomics & Integrative Biology (CSIR-IGIB), New Delhi, India.
12. Department of Biostatistics and Center for Statistical Genetics, University of Michigan, Ann Arbor, Michigan, USA.
13. Medical Genomics and Metabolic Genetics Branch, National Human Genome Research Institute, National Institutes of Health, Bethesda, Maryland, USA.
14. Center for Genomics and Personalized Medicine Research, Wake Forest School of Medicine, Winston-Salem, North Carolina, USA.
15. Center for Diabetes Research, Wake Forest School of Medicine, Winston-Salem, North Carolina, USA.
16. Department of Biochemistry, Wake Forest School of Medicine, Winston-Salem, North Carolina, USA.
17. Department of Cardiology, Ealing Hospital NHS Trust, Southall, Middlesex, UK.
18. Imperial College Healthcare NHS Trust, Imperial College London, London, UK.
19. Department of Medicine, Yong Loo Lin School of Medicine, National University of Singapore, National University Health System, Singapore.
20. Hong Kong Institute of Diabetes and Obesity, The Chinese University of Hong Kong, Hong Kong, China.
21. Department of Medicine and Therapeutics, The Chinese University of Hong Kong, Hong Kong, China.
22. Li Ka Shing Institute of Health Sciences, The Chinese University of Hong Kong, Hong Kong, China.
23. CSIR-Centre for Cellular and Molecular Biology, Hyderabad, Andhra Pradesh, India.
24. Saw Swee Hock School of Public Health, National University of Singapore, National University Health System, Singapore.
25. Wellcome Trust Centre for Human Genetics, Nuffield Department of Medicine, University of Oxford, Oxford, UK.
26. Department of Biostatistics, Boston University School of Public Health, Boston, Massachusetts, USA.
27. Department of Biostatistics, Harvard School of Public Health, Boston, Massachusetts, USA.
28. Centre for Quantitative Medicine, Office of Clinical Sciences, Duke-NUS Graduate Medical School Singapore, Singapore.
29. Department of Biomedical Science, Hallym University, Chuncheon, Republic of Korea.
30. Department of Medicine, University of Mississippi Medical Center, Jackson, Mississippi, USA.
31. Department of Genetics, Texas Biomedical Research Institute, San Antonio, Texas, USA.
32. Analytic and Translational Genetics Unit, Department of Medicine, Massachusetts General Hospital, Boston, Massachusetts, USA.
33. Department of Statistics, University of Oxford, Oxford, UK.
34. Centre for Chronic Disease Control, New Delhi, India.
35. MRC-PHE Centre for Environment and Health, Imperial College London, London, UK.
36. Department of Genetics, Harvard Medical School, Boston, Massachusetts, USA.
37. Estonian Genome Centre, University of Tartu, Tartu, Estonia.
38. Division of Endocrinology, Boston Children’s Hospital, Boston, Massachusetts, USA.
39. Department of Clinical Sciences, Diabetes and Endocrinology, Lund University Diabetes Centre, Malmö, Sweden.
40. Department of Primary Care Health Sciences, University of Oxford, Oxford, UK.
41. Department of Epidemiology, Colorado School of Public Health, University of Colorado, Aurora, Colorado, USA.
42. Department of General Practice and Primary Health Care, University of Helsinki, Helsinki, Finland.
43. Vasa Health Care Center, Vaasa, Finland.
44. Department of Nutrition, Harvard School of Public Health, Boston, Massachusetts, USA.
45. Department of Public Health and Clinical Medicine, Umeå University, Umeå, Sweden.
46. Department of Clinical Sciences, Lund University Diabetes Centre, Genetic and Molecular Epidemiology Unit, Lund University, Malmö, Sweden.
47. Genetics of Complex Traits, University of Exeter Medical School, University of Exeter, Exeter, UK.
48. Department of Internal Medicine, Section on Nephrology, Wake Forest School of Medicine, Winston-Salem, North Carolina, USA.
49. Genomics and Molecular Physiology, CNRS (Institut de Biologie de Lille), Lille, France.
50. Department of Medicine, Section of Genetic Medicine, The University of Chicago, Chicago, Illinois, USA.
51. Institute of Genetic Epidemiology, Helmholtz Zentrum München, German Research Center for Environmental Health, Neuherberg, Germany.
52. Endocrinology and Metabolism Service, Hadassah-Hebrew University Medical Center, Jerusalem, Israel.
53. Center for Genome Science, Korea National Institute of Health, Chungcheongbuk-do, Republic of Korea.
54. Institute of Epidemiology II, Helmholtz Zentrum München, German Research Center for Environmental Health, Neuherberg, Germany.
55. Research Unit of Molecular Epidemiology, Helmholtz Zentrum München, German Research Center for Environmental Health, Neuherberg, Germany.
56. German Center for Diabetes Research (DZD), Neuherberg, Germany.
57. Center of Biostatistics and Bioinformatics, University of Mississippi Medical Center, Jackson, Mississippi, USA.
58. Department of Pediatrics, University of Texas Health Science Center, San Antonio, Texas, USA.
59. University of Exeter Medical School, University of Exeter, Exeter, UK.
60. Department of Twin Research and Genetic Epidemiology, King’s College London, London, UK.
61. Oxford Centre for Diabetes, Endocrinology and Metabolism, Radcliffe Department of Medicine, University of Oxford, Oxford, UK.
62. Institute of Experimental Genetics, Helmholtz Zentrum München, German Research Center for Environmental Health, Neuherberg, Germany.
63. Department of Endocrinology and Metabolism, Shanghai Diabetes Institute, Shanghai Jiao Tong University Affiliated Sixth People’s Hospital, Shanghai, China.
64. Department of Epidemiology, Harvard School of Public Health, Boston, Massachusetts, USA.
65. Department of Statistics, Seoul National University, Seoul, Republic of Korea.
66. Hannover Unified Biobank, Hannover Medical School, Hannover, Germany.
67. The Biostatistics Center, The George Washington University, Rockville, Maryland, USA.
68. Research, South Texas Veterans Health Care System, San Antonio, Texas, USA.
69. Department of Paediatrics, Yong Loo Lin School of Medicine, National University of Singapore, National University Health System, Singapore.
70. Division of Human Genetics, Genome Institute of Singapore, A*STAR, Singapore.
71. Diabetes Prevention Unit, National Institute for Health and Welfare, Helsinki, Finland.
72. National Heart and Lung Institute, Cardiovascular Sciences, Hammersmith Campus, Imperial College London, London, UK.
73. Chronic Disease Epidemiology, Swiss Tropical and Public Health Institute, University of Basel, Basel, Switzerland.
74. Faculty of Health Sciences, Institute of Clinical Medicine, Internal Medicine, University of Eastern Finland, Kuopio, Finland.
75. Interdisciplinary Program in Bioinformatics, Seoul National University, Seoul, Republic of Korea.
76. MRC Epidemiology Unit, Institute of Metabolic Science, University of Cambridge, Cambridge, UK.
77. Department of Public Health, Section of General Practice, Aarhus University, Aarhus, Denmark.
78. Ministry of Health and Welfare, Seoul, Republic of Korea.
79. Institute of Health Sciences, University of Oulu, Oulu, Finland.
80. Deutsches Forschungszentrum für Herz-Kreislauferkrankungen (DZHK), Partner Site Munich Heart Alliance, Munich, Germany.
81. Institute of Human Genetics, Technische Universität München, Munich, Germany.
82. Institute of Human Genetics, Helmholtz Zentrum München, German Research Center for Environmental Health, Neuherberg, Germany.
83. Department of Clinical Sciences, Hypertension and Cardiovascular Disease, Lund University, Malmö, Sweden.
84. Department of Medicine I, University Hospital Grosshadern, Ludwig-Maximilians-Universität, Munich, Germany.
85. Institute of Medical Informatics, Biometry and Epidemiology, Chair of Genetic Epidemiology, Ludwig-Maximilians-Universität, Neuherberg, Germany.
86. Jackson Heart Study, University of Mississippi Medical Center, Jackson, Mississippi, USA.
87. McGill University and Génome Québec Innovation Centre, Montreal, Quebec, Canada.
88. Department of Human Genetics, McGill University, Montreal, Quebec, Canada.
89. Research Institute of the McGill University Health Centre, Montreal, Quebec, Canada.
90. Department of Clinical Sciences, Medicine, Lund University, Malmö, Sweden.
91. Department of Clinical Sciences, Diabetes and Cardiovascular Disease, Genetic Epidemiology, Lund University, Malmö, Sweden.
92. Oxford NIHR Biomedical Research Centre, Oxford University Hospitals Trust, Oxford, UK.
93. Department of Medicine, Program in Personalized and Genomic Medicine, University of Maryland, Baltimore, Maryland, USA.
94. Center for Human Genetic Research, Department of Medicine, Massachusetts General Hospital, Boston, Massachusetts, USA.
95. Department of Psychiatry, Icahn Institute for Genomics and Multiscale Biology, Icahn School of Medicine at Mount Sinai, New York, USA.
96. Channing Division of Network Medicine, Department of Medicine, Brigham and Women’s Hospital and Harvard Medical School, Boston, Massachusetts, USA.
97. Department of Epidemiology and Population Health, Albert Einstein College of Medicine, New York, USA.
98. Institute of Clinical Diabetology, German Diabetes Center, Leibniz Center for Diabetes Research at Heinrich Heine University, Düsseldorf, Germany.
99. Department of Public Health and Primary Care, Institute of Public Health, University of Cambridge, Cambridge, UK.
100. Division of Endocrinology and Metabolism, Department of Medicine, McGill University, Montreal, Quebec, Canada.
101. Department of Genome Sciences, University of Washington School of Medicine, Seattle, Washington, USA.
102. Cardiovascular & Metabolic Disorders Program, Duke-NUS Graduate Medical School Singapore, Singapore.
103. Department of Endocrinology and Metabolism, All India Institute of Medical Sciences, New Delhi, India.
104. Department of Statistics and Applied Probability, National University of Singapore, Singapore.
105. Life Sciences Institute, National University of Singapore, Singapore.
106. The Medical School, Institute of Cellular Medicine, University of Newcastle, Newcastle, UK.
107. College of Public Services, Jackson State University, Jackson, Mississippi, USA.