Large-scale exome array summary statistics resources for glycemic traits to aid effector gene prioritization

Abstract

Background

Genome-wide association studies for glycemic traits have identified hundreds of loci associated with these biomarkers of glucose homeostasis. Despite this success, the challenge remains to link variant associations to genes, and underlying biological pathways.

Methods

To identify coding variant associations which may pinpoint effector genes at both novel and previously established genome-wide association loci, we performed meta-analyses of exome-array studies for four glycemic traits: glycated hemoglobin (HbA1c, up to 144,060 participants), fasting glucose (FG, up to 129,665 participants), fasting insulin (FI, up to 104,140) and 2hr glucose post-oral glucose challenge (2hGlu, up to 57,878). In addition, we performed network and pathway analyses.

Results

Single-variant and gene-based association analyses identified coding variant associations at more than 60 genes, which when combined with other datasets may be useful to nominate effector genes. Network and pathway analyses identified pathways related to insulin secretion, zinc transport and fatty acid metabolism. HbA1c associations were strongly enriched in pathways related to blood cell biology.

Conclusions

Our results provided novel glycemic trait associations and highlighted pathways implicated in glycemic regulation. Exome-array summary statistic results are being made available to the scientific community to enable further discoveries.

Introduction

Genome-wide association studies (GWAS) have identified hundreds of loci associated with glycemic traits and type 2 diabetes (T2D) risk ^{1– 3} . Despite this tremendous success, the challenge remains to link the often lead non-coding variants with effector genes and mechanism of action. To complement these approaches, exome array studies ^{4, 5} and more recently, whole-exome sequencing approaches have focused on coding variant associations ^{6– 9} . These can be helpful to pinpoint potential effector genes for downstream functional studies. Here, we provide exome-array GWAS meta-analysis results for glycated hemoglobin (HbA1c, up to 144,060 participants), fasting glucose (FG, up to 129,665 participants), fasting insulin (FI, up to 104,140) and 2hr glucose post-oral glucose challenge (2hGlu, up to 57,878). Most of the data are from self-reported and genetically clustered European ancestry individuals (85%), with the remaining participants being of African American (6%), South Asian (5%), East Asian (2%) and Hispanic ancestry (2%). We identify single coding variant and gene-based associations to prioritize likely effector genes, and additionally perform pathway analyses to highlight relevant gene sets regulating each glycemic trait. Summary statistics from these analyses are publicly available through our website ( www.magicinvestigators.org), as well as through the GWAS catalog ( https://www.ebi.ac.uk/gwas/summary-statistics, study accessions GCST90256400 - GCST90256420) ¹⁰ .

Methods

Study design, cohorts, phenotypes and genotypes

MAGIC (Meta-Analysis of Glucose and Insulin-related traits Consortium) was established to focus on the genetic analysis of glycemic traits in individuals without diabetes. In this MAGIC effort, individuals without diabetes of self-reported and genetically clustered European (85%), African American (6%), South Asian (5%), East Asian (2%) and Hispanic (2%) ancestry from up to 64 cohorts participated. Sample sizes were up to 144,060 for HbA1c, 129,665 for FG, 104,140 for FI and 57,878 for 2hGlu. Participating cohorts and their characteristics are detailed in Supplementary Table S1 ¹¹ . Each cohort obtained ethical approval and written informed consent.

Phenotypes

Studied outcomes were FG (mmol/L), Ln-transformed FI (pmol/L), 2hGlu (mmol/L) and HbA1c (% of hemoglobin). Glycemic measurements are described in detail for each contributing cohort in Supplementary Table S1 ¹¹ . Individuals with diagnosed or treated diabetes, or those with diabetes based on FG (≥7 mmol/L), 2hGlu (≥11.1 mmol/L) and/or HbA1c (≥6.5%) were excluded from analyses.

Genotyping and QC

The Illumina HumanExome BeadChip is a genotyping array containing variants that have been observed in sequencing data of ~12,000 individuals. Non-synonymous variants seen at least three times across at least two datasets were included on the exome chip. More lenient criteria were used for splice and nonsense variants. Besides the core content of protein-altering variants, the exome chip contains additional variants including common variants identified in GWAS, ancestry informative markers, mitochondrial variants, randomly selected synonymous variants, HLA tag variants and Y chromosome variants. In this study we analyzed association with glycemic traits of 247,470 autosomal and X chromosome variants present on the exome chip. Genotype calling and quality control were performed following protocols developed by the UK Exome Chip or CHARGE consortium ¹² . The exact genotyping array, calling algorithm and QC procedure used by each cohort are depicted in Supplementary Table S1 ¹¹ .

Annotation and functional prediction of variants

Annotation of the exome chip variants was performed using the Ensembl Variant Effect Predictor v78 with plugin dbNSFP v2.9 to add in silico functional prediction from Polyphen HumDiv, Polyphen HumVar, LRT, Mutation Taster and SIFT (ensembl66 version) ^{13, 14} .

Statistical analyses

Single variant analyses. Individual cohorts ran linear mixed models using the raremetalworker (v 4.13.2) or rvtests (v20140723) software (Supplementary Table S1 ¹¹ ). For each glycemic outcome, analyses were performed using an additive model for the raw and the inverse normal transformed trait. In the manuscript and in all tables and figures effect estimates and standard errors are for the raw trait, while the p-values are from the inverse normal transformed trait analyses. Analyses were adjusted for age, sex, BMI, study-specific number of PCs and other study-specific covariates (Supplementary Table S1 ¹¹ ). Raremetal (v4.13.7 or higher) was used to combine results within and across ancestries by fixed-effect meta-analyses. Variants with P <10 ^-4 for deviation from Hardy-Weinberg equilibrium or with call rate <0.99 in individual cohorts were excluded from meta-analyses. In single variant analyses, the threshold for significance was P <2.2×10 ^-7 for coding variants (stop-gained, stop lost, frameshift, splice donor, splice acceptor, initiator codon, missense, in-frame indel and splice region variants). This P-value threshold was based on a Bonferroni correction weighted by the enrichment for complex trait associations among the functional annotation categories ^{15, 16} . We performed so called distance-based clumping; significant association signals located more than 500 kb apart were considered to represent distinct loci. Significantly associated variants located more than 500 kb from any variant already found to be associated in published large-scale glycemic trait and T2D GWAS analyses ^{1, 3, 17, 18} were considered novel glycemic trait associations. Gene-based and single-variant analyses results presented in the paper are for the meta-analyses of all ancestries combined, unless mentioned otherwise.

Gene-based analyses. Raremetal (v4.13.7 or higher) was used to perform gene-based burden and sequence kernel association (SKAT) tests. For both burden and SKAT tests, two in silico masks for inclusion of variants in the test were used: NSstrict and NSbroad. The NSstrict mask includes predicted protein truncating variants (PTVs, splice donor, splice acceptor, stop gained, frameshift, stop lost or initiator codon variant) OR variants that are missense and predicted to be damaging by five prediction algorithms (SIFT, Polyphen HumDiv, Polyphen HumVar, LRT, MutationTaster). The NSbroad mask additionally includes missense variants predicted to be damaging by at least one of the five prediction algorithms AND that have a MAF <1% in each ancestry group. These MAFs were derived from our single variant HbA1c meta-analyses results (N up to 144,060). Gene-based analyses were performed on genes containing at least two variants fulfilling the mask criteria. The P-value threshold for significance in gene-based analyses was 2.5 x 10 ^-6 (Bonferroni correction for 20,000 genes).

GeneMANIA network analysis

For network analyses, we used GeneMANIA (v3.5.1), a network approach that searches many large, publicly available biological datasets to find related genes. These include protein-protein, protein-DNA and genetic interactions, pathways, reactions, gene and protein expression data, protein domains and phenotypic screening profiles. GeneMANIA uses a label propagation algorithm for predicting gene function given the composite functional association network (calculated from the databases selected). The weights needed for the label propagation method to work are selected at the beginning of the process. In our case, and according to the defaults, we weighted the network using linear regression, to make genes in the input list interact as much as possible with each other. We analyzed all loci that had at least one non-synonymous variant with P <1 x 10 ^-5 with any trait, and then mapped the most significant non-synonymous variant at each locus to the gene (input genes). We performed four network analyses: (1) HbA1c-associated variants only, (2) FI-associated variants only, (3) FG-associated variants only, and (4) 2hGlu-associated variants only ( Figure 1, Supplementary Figure S1 ¹¹ ). We selected the 50 default databases to create the composite network, and we allowed the method to find at most 50 genes that are related to our query input list. The resultant networks were investigated to find enriched Gene Ontology (GO) terms and Reactome Pathways. Gene Set Enrichment (GSE) of networks and sub-networks were assessed with ClueGO ¹⁹ using GO terms and Reactome gene sets ²⁰ . The enrichment results were grouped using a Cohen’s Kappa score of 0.4, and terms were considered significant with a Bonferroni-adjusted p-value <0.05, provided that there was an overlap of at least three network genes in the relevant GO gene set when calculating GO enrichment. For the pathway selection (Reactome), we set a threshold that the network genes should represent at least 4% of the pathway. These values were applied given the recommended defaults when running ClueGO ¹⁹ . Cohen’s Kappa statistic was used to measure the gene-set similarity of GO terms and Reactome pathways and allowed us to group enriched terms into functional groups to improve visualization of enriched pathways. We used all genes with GO annotations and at least one interaction in our network database as the background set.

Figure 1. Network and pathway analyses identify relevant gene sets regulating glycemia using two different methods for variant associations with P <1 × 10 ^-5.

( A–B) The networks represent composite networks for ( A) HbA1c and ( B) FG, from the GeneMANIA analysis using genes with variant associations at P <1 × 10 ^-5 for each trait as input. Nodes outlined in red correspond to genes from the input list. Other nodes correspond to related genes based on 50 default databases. Based on the network, GO terms and Reactome pathways that were significantly enriched are depicted. To summarize these results, the most significant term of all calculated terms within the same group is represented. Barplots with the Bonferroni-adjusted -log10(p-values) of the most significant terms within each group are are shown. Each group was assigned a specific color; if a gene is present in more than one term, it is displayed in more than one color. ( C–D) Heatmaps showing EC-DEPICT results from analysis of ( C) all traits except HbA1c and ( D) FG. The columns represent the input genes for the analysis. In ( C), these are genes with variant associations of P <1 × 10 ^-5 for FG, FI, and/or 2hGlu, and in ( D) these are genes with variant associations of P <1 × 10 ^-5 for FG. Rows in the heatmap represent significant meta-gene sets (FDR <0.05). The color of each square indicates DEPICT’s z-score for membership of that gene in that gene set, where dark red means “very likely a member” and dark blue means “very unlikely a member.” The gene set annotations indicate whether that meta-gene set was significant at FDR <0.05 or not significant (n.s.) for each of the other EC-DEPICT analyses. For heatmap intensity and EC-DEPICT P-values, the meta-gene set values are taken from the most significantly enriched member gene set. The gene variant annotations are as follows: (1) the European minor allele frequency (MAF) of the input variant, where rare is MAF <1%, low-frequency is MAF 1–5%, and common is MAF >5%, 2) whether the gene has an Online Mendelian Inheritance in Man (OMIM) annotation as causal for a diabetes/glycemic-relevant syndrome or blood disorder, 3) to 6) whether each variant was significant ( P <2 × 10 ^-7), suggestively significant ( P <1 × 10 ^-5), or not significant in Europeans for each of the four traits, and 7) whether each variant was included in the analysis or excluded by filters (see Methods). AWS: array-wide significant.

Gene set enrichment analysis (GSEA)

An extension of the GWAS GSEA method DEPICT ²¹ , EC-DEPICT ^{22, 23} , was used for GSEA. The key feature of EC-DEPICT is the use of “reconstituted” gene sets, which are gene sets collected from many different databases (e.g. canonical pathways, protein-protein interaction networks, and mouse phenotypes) that have been extended based on large-scale microarray co-expression data ^{21, 24} .

Six groups of variants were analyzed: (1) HbA1c-associated variants only, (2) FI-associated variants only, (3) FG-associated variants only, (4) 2hGlu-associated variants only, (5) all trait-associated variants, and (6) all trait-associated variants except for HbA1c. For each trait, the associated variants based on the European summary statistics were identified and clumped using a +/- 500 kb window. Then, the most significant nonsynonymous variant for each locus was included in the analysis, with a cut-off of P <10 ^-5. Annotations from the CHARGE consortium were used to assign variants to genes (see URL ). After GSEA, highly correlated gene sets were grouped by affinity propagation clustering of all 14,462 gene sets ²⁵ into “meta-gene sets” using SciKitLearn.clustering.AffinityPropagation version 0.17 ²⁶ . For all visualizations, the gene set within a meta-gene set with the best enrichment P-value was used; heat maps were created with the ComplexHeatmap package in R ²⁷ .

URL: CHARGE Consortium ExomeChip annotation file (v6).

Method and choice of data for permutations: We performed the EC-DEPICT analysis as described elsewhere ^{22, 23} . All analyses are based on a group of 14,462 “reconstituted” gene sets, which contains a z-score for probability of gene set membership for each gene (for details, see ^{21, 24} ).

The basic EC-DEPICT method is as follows. We first obtain a list of significant input variants (the most significant nonsynonymous variant per locus) and then map variants to genes based on annotations from the CHARGE consortium (see URL ). For each gene set, we obtain the gene set membership z-scores for all trait-associated input genes and sum them to generate a test statistic. We then take 2,000 permuted ExomeChip association studies (described in more detail below) and calculate the average permuted test statistic for that gene set, as well as the permuted standard deviation. For each permutation, the number of top genes we take as “input genes” is matched to the actual observed number of input genes. We then calculate (observed test statistic – average permuted test statistic)/(permuted standard deviation) to generate a z-score, which is converted to a p-value via the normal distribution. False discovery rates were calculated by comparing the observed p-values to a permuted P-value distribution generated with an additional set of 50 permuted association studies.

The permuted ExomeChip association studies are conducted by (1) generating 2,200 sets of normally distributed phenotypes and (2) using these randomly generated phenotypes to conduct 2,200 association studies with real ExomeChip data. Using these permutations to adjust the observed test statistics corrects for any inherent structure in the data (e.g. that pathways made up of longer genes may be more likely to come up as significant by chance).

For these analyses, we first generated permutations based on ExomeChip data we had used previously for this purpose: 11,899 samples drawn from three cohorts (Malmö Diet and Cancer [MDC], All New Diabetics in Scania [ANDIS], and Scania Diabetes Registry [SDR]). For simplicity, we refer to these cohorts as the “Swedish permutations.”

As part of our GSEA pipeline, we remove input trait-associated variants that are not present in the permuted data to ensure that all variants are appropriately modeled. When using the Swedish permutations, this generally results in removing a substantial fraction of the variants, especially of the very rarest variants (due to the smaller sample size of the Swedish data relative to the data being analyzed). We have previously observed that this filtering can actually improve the GSEA signal, possibly due to more heterogeneous biology or a higher false-positive rate in these very rare variants ²³ . However, in this case, we observed that in performing this filtering, we excluded variants in several known monogenic disease genes, such as HNF1A and SLC2A2. Therefore, we wished to repeat the analysis with a set of permutations which would allow us to retain these variants. We thus repeated the analysis with a second set of permutations consisting of 152,249 samples from the UK Biobank (referred to as the “UKBB permutations”). The larger sample size in the UKBB permutations means more variants are present and can therefore be included in the analysis.

Concordance of results from two different sets of permuted distributions across phenotypes: For completeness, we report the results from the use of both sets of permutations. We note that the results are strongly concordant. The larger number of significant gene sets reported based on the UK Biobank permutations is generally a combination of 1) overall improved power (i.e. more variants are included) and 2) the inclusion of variants in key driver genes absent in the Swedish permutations, encompassing both the monogenic genes mentioned above (e.g. SLC2A2) and additional genes with clearly relevant biology (e.g. SLC30A8). The results from both sets of permutations are summarized below. For all analyses, “significance” refers to a false discovery rate of <0.05.

All-trait analysis: After filtering, 78 input genes were included for the analysis with the UKBB permutations and 60 for the analysis with the Swedish permutations. (Note that the difference in the number of input genes is due to the presence of a larger number of input variants in the UKBB permutations – see above). We found 234 significant gene sets in 86 meta-gene sets based on the UKBB permutations (Supplementary Figure S2 ¹¹ ) and 133 gene sets in 51 meta-gene sets based on the Swedish permutations (Supplementary Figure S3 ¹¹ ). The correlation between the UKBB and Swedish analyses was r = 0.902, P <10 ^-300.

All-traits-except-HbA1c analysis: After filtering, 45 input genes were included for the analysis with the UKBB permutations and 33 for the analysis with the Swedish permutations. We found 128 significant gene sets in 53 meta-gene sets based on the UKBB permutations (Supplementary Figure S2 ¹¹ ) and 45 significant gene sets in 18 meta-gene sets based on the Swedish permutations (Supplementary Figure S3 ¹¹ ). The correlation between the UKBB and Swedish analyses was r = 0.882, P <10 ^-300.

HbA1c-only analysis: After filtering, 41 input genes were included for the analysis with the UKBB permutations and 33 for the analysis with the Swedish permutations. We found 191 significant gene sets in 73 meta-gene sets based on the UKBB permutations (Supplementary Figure S2 ¹¹ ) and 120 gene sets in 41 meta-gene sets based on the Swedish permutations. (Supplementary Figure S3 ¹¹ ). The correlation between the UKBB and Swedish analyses was r = 0.936, P <10 ^-300.

FG-only analysis: After filtering, 26 input genes were included for the analysis with the UKBB permutations and 22 for the analysis with the Swedish permutations. We found 106 significant gene sets in 39 meta-gene sets based on the UKBB permutations (Supplementary Figure S2 ¹¹ ) and 48 significant gene sets in 15 meta-gene sets based on the Swedish permutations (Supplementary Figure S3 ¹¹ ). The correlation between the UKBB and Swedish analyses was r = 0.939, P <10 ^-300.

2hGlu-only analysis: After filtering, 12 input genes were included for the analysis with the UKBB permutations and seven for the analysis based on the Swedish permutations. We found 56 significant gene sets in 17 meta-gene sets based on the UKBB permutations (Supplementary Figure S2 ¹¹ ), with no significant gene sets based on the Swedish permutations. The correlation between the UKBB and Swedish analyses was r = 0.787, P <10 ^-300.

FI-only analysis: After filtering, 11 input genes were included for the analysis with the UKBB permutations and eight for the analysis with the Swedish permutations. There were no significant gene sets from either analysis. The correlation between the UKBB and Swedish analyses was r = 0.860, P <10 ^-300.

Visualization: As in previous work ^{22, 23} , we have included all trait-associated variants in the heat maps, even if they were excluded from the analysis (e.g. because they were absent in the permutations or did not have a nonsynonymous annotation in the CHARGE annotation file). This is because we assume that if the genes harboring those variants have strong predicted membership in significantly trait-associated gene sets, they are still good candidates for prioritization. In fact, this may be even stronger evidence in favor of these genes because they did not contribute to the enrichment analysis and therefore their prioritization is independently derived (and provides even more support to the implicated biology).

Results

Study design overview

We performed single-variant and gene-based association analyses with FG, FI, HbA1c, and 2hGlu levels on exome-array coding variants in up to 144,060 individuals without diabetes (to exclude any consequence of diabetes treatments or related interventions on these quantitative traits) of European (85%), African-American (6%), South Asian (5%), East Asian (2%), and Hispanic (2%) ancestry from up to 64 cohorts (Supplementary Table S1 ¹¹ , Methods). We used a linear mixed model to test single-variant associations in each individual cohort and combined results by fixed-effect meta-analyses within and across ancestries. As body mass index (BMI) is a major risk factor for T2D and is correlated with glycemic traits, all analyses were adjusted for BMI to identify loci influencing glycemia independently from their effects on overall adiposity. We have previously demonstrated that collider bias did not significantly affect results with BMI adjustment ¹ . We used distance-based clumping to define distinct loci and considered signals to be novel if they were located more than 500 kb from a variant with an established association with any of the glycemic traits or T2D in large published GWAS (Methods). We considered a coding variant to meet exome-wide significance for association if P <2.2 × 10 ^{-7 15, 16} ( Table 1, Methods). To increase power to detect rare variant associations, we additionally performed gene-burden and sequence kernel association (SKAT) tests for gene-level analyses to identify genes with significant evidence of association ( P <2.5 × 10 ^-6) ( Table 2, Methods). Finally, to identify relevant biological pathways enriched in associations with glycemic traits we conducted pathway and network analyses.

Table 1. Single-point coding variant associations meeting the significance threshold for coding variants of P <2.2 × 10 ^-7.

This table includes all coding variants meeting this threshold, irrespective of whether they fall in completely new loci or in previously-established loci, provided that the association at the established locus was not shown to be due to a non-coding variant (Table S2) or another coding variant at the same locus. Novel loci are highlighted in bold. HbA1c: glycated haemoglobin; FG: fasting glucose; FI: fasting insulin; 2hGlu: 2h glucose; Alleles E/O: effect allele/other allele; EAF: effect allele frequency; Effect (SE): effect size (standard error); P: p-value; N: number of samples in the analysis; Novel/previous glycemic trait association: Novel corresponds to a new association result in this study; Locus name of previous association – name used for previously reported locus. ¹Significant in the European-only analysis in our study. Genes in this table are listed in order of chromosomal position.

Trait	SNP	Gene	Protein Consequence	Alleles E/O	EAF	Effect (SE)	P	N	Previous glycemic trait association (if any)	Locus name of previous association
FG	rs1886686	WDR78	p.G12A	G/C	0.739	0.014 (0.002)	2.24×10 -11	123558	Novel
HbA1c	rs267738	CERS2	p.E106A	G/T	0.186	-0.01 (0.002)	6.96×10 -10	144043	HbA1c	CERS2
HbA1c	rs863362	OR10X1	p.W66X	T/C	0.465	0.011 (0.001)	6.76×10 -15	114945	HbA1c	SPTA1
HbA1c	rs857725	SPTA1	p.K1693Q	G/T	0.262	0.022 (0.001)	1.56×10 -50	143956	HbA1c	SPTA1
HbA1c	rs11887523	MFSD2B	p.A60T	A/G	0.007	-0.072 (0.01)	1.44×10 -12	122060	HbA1c	ATAD2B
FG	rs1260326	GCKR	p.L446P	C/T	0.631	0.029 (0.002)	6.36×10 -48	129588	FG, FI, 2hGlu	GCKR
FI	rs1260326	GCKR	p.L446P	C/T	0.626	0.024 (0.002)	5.55×10 -32	104076	FG, FI, 2hGlu	GCKR
2hGlu	rs1260326	GCKR	p.L446P	C/T	0.618	-0.069 (0.009)	4.48×10 -15	57813	FG, FI, 2hGlu	GCKR
FG	rs35720761	THADA	p.C845Y	T/C	0.108	-0.018 (0.003)	4.35×10 -9	129622	T2D, FG	THADA
HbA1c	rs35720761	THADA	p.C845Y	C/T	0.113	0.014 (0.002)	2.58×10 -12	144001	T2D, FG	THADA
FG	rs7578597	THADA	p.T897A	C/T	0.106	-0.019 (0.003)	1.99×10 -8	113162	T2D, FG	THADA
FI	rs7607980	COBLL1	p.N901D	C/T	0.128	-0.032 (0.003)	1.30×10 -24	97817	FI	COBLL1
FG	rs2232323	G6PC2	p.Y207S	C/A	0.006	-0.129 (0.012)	1.05×10 -28	123981	FG, HbA1c	G6PC2
HbA1c	rs2232323	G6PC2	p.Y207S	C/A	0.007	-0.053 (0.007)	3.25×10 -13	144038	FG, HbA1c	G6PC2
FG	rs146779637	G6PC2	p.R283X	T/C	0.002	-0.138 (0.02)	1.78×10 -12	127278	FG, HbA1c	G6PC2
HbA1c	rs146779637	G6PC2	p.R283X	T/C	0.002	-0.074 (0.012)	4.58×10 -10	141728	FG, HbA1c	G6PC2
FI	rs1983210	OBSL1	p.E1365D	G/C	0.729	0.016 (0.003)	8.48×10 -10	79767	Novel
FI	rs3183099	OBSL1	splice region variant	A/G	0.226	-0.013 (0.002)	4.70×10 -8	100713	Novel
FI	rs1801282	PPARG	p.P12A	G/C	0.117	-0.031 (0.003)	3.50×10 -23	98631	FI	PPARG
HbA1c	rs35726701	RNF123	p.K596E	G/A	0.019	0.025 (0.005)	4.19×10 -8	131203	HbA1c	USP4
FG	rs5400	SLC2A2	p.T110I	A/G	0.161	-0.022 (0.003)	2.14×10 -17	129591	FG, HbA1c	SLC2A2
HbA1c	rs5400	SLC2A2	p.T110I	A/G	0.153	-0.013 (0.002)	2.27×10 -13	144012	FG, HbA1c	SLC2A2
HbA1c 1	rs223705	EGF	p.M708I	A/G	0.374	-0.007 (0.001)	2.11×10 -7	121204	HbA1c	EGF
HbA1c	rs7683365	GYPB	p.T48M	A/G	0.312	0.012 (0.002)	1.61×10 -8	45191	HbA1c	FREM3
FG	rs146886108	ANKH	p.R187Q	T/C	0.004	-0.088 (0.014)	5.67×10 -10	129647	T2D	ANKH
HbA1c	rs31244	SV2C	p.D543N	A/G	0.083	0.012 (0.002)	6.05×10 -8	144000	Novel
FG	rs6235	PCSK1	p.S690T	G/C	0.264	-0.022 (0.002)	9.22×10 -24	123560	FG	PCSK1
2hGlu	rs2549782	ERAP2	p.K392N	T/G	0.519	-0.055 (0.009)	6.81×10 -10	57836	2hGlu	ERAP2
HbA1c	rs35742417	RREB1	p.S1499Y	A/C	0.173	-0.01 (0.002)	3.76×10 -9	143967	FG, T2D	RREB1
FG	rs35742417	RREB1	p.S1499Y	A/C	0.183	-0.019 (0.002)	1.27×10 -16	129577	FG, T2D	RREB1
HbA1c	rs1799945	HFE	p.H63D	G/C	0.129	-0.023 (0.002)	1.20×10 -30	128354	HbA1c	HFE, HIST1H4A
HbA1c	rs1800562	HFE	p.C279Y	A/G	0.051	-0.042 (0.003)	3.30×10 -47	138093	HbA1c	HFE, HIST1H4A
FG	rs10305492	GLP1R	p.A316T	A/G	0.014	-0.08 (0.008)	2.37×10 -25	129601	FG	GLP1R
HbA1c	rs35332062	MLXIPL	p.A358V	A/G	0.117	0.011 (0.002)	6.18×10 -9	144042	HbA1c	MLXIPL
HbA1c	rs3812316	MLXIPL	p.Q241H	G/C	0.112	0.012 (0.002)	2.15×10 -8	108605	HbA1c	MLXIPL
FG	rs194524	STEAP2	p.R456Q	A/G	0.523	0.01 (0.002)	7.65×10 -8	129629	FG, T2D, RG	STEAP2-AS1
HbA1c	rs34664882	ANK1	p.A1503V	A/G	0.026	-0.049 (0.004)	2.43×10 -39	144034	HbA1c	ANK1
FG	rs13266634	SLC30A8	p.R276W	T/C	0.305	-0.029 (0.002)	1.63×10 -46	129614	FG, HbA1c, T2D	SLC30A8
HbA1c	rs13266634	SLC30A8	p.R276W	T/C	0.300	-0.015 (0.001)	8.50×10 -28	143982	FG, HbA1c, T2D	SLC30A8
HbA1c	rs11557154	DCAF12	p.R113Q	T/C	0.138	-0.009 (0.002)	1.70×10 -7	144045	T2D, HbA1c	Mahajan 2022 from CMD KP
FG	rs17853166	IKBKAP	p.S251G	C/T	0.026	-0.037 (0.006)	4.82×10 -11	129640	FG	IKBKAP
HbA1c	rs60980157	GPSM1	p.S391L	T/C	0.246	-0.013 (0.002)	6.71×10 -17	118824	FG, T2D	GPSM1
FG	rs60980157	GPSM1	p.S391L	T/C	0.254	-0.014 (0.002)	2.35×10 -9	110915	FG, T2D	GPSM1
HbA1c	rs906220	HK1	p.H7R	G/A	0.916	0.025 (0.003)	2.16×10 -21	94970	HbA1c	HK1
FG	rs701865	PDE6C	p.S270T	A/T	0.366	-0.01 (0.002)	1.14×10 -7	118580	FG, RG	PDE6C
HbA1c	rs61732434	OR51V1	p.S161N	T/C	0.008	-0.052 (0.009)	1.75×10 -8	127507	HbA1c	HBB
HbA1c	rs415895	SWAP70	p.Q447E	G/C	0.641	-0.013 (0.001)	1.15×10 -21	138028	HbA1c	SWAP70
HbA1c	rs117706710	AMPD3	p.V311L	T/G	0.009	0.037 (0.006)	2.32×10 -10	144048	HbA1c	AMPD3
FG	rs2167079	ACP2	p.R29Q	T/C	0.340	0.016 (0.002)	7.99×10 -15	129580	FG	MADD
HbA1c	rs35233100	MADD	p.R766X	T/C	0.055	-0.015 (0.003)	1.13×10 -8	144034	FG	MADD
FG	rs35233100	MADD	p.R766X	T/C	0.054	-0.029 (0.004)	1.46×10 -12	126231	FG	MADD
FG	rs56200889	ARAP1	p.Q802E	C/G	0.270	-0.016 (0.002)	1.79×10 -14	122674	FG	ARAP1
HbA1c	rs643788	DPAGT1	p.I393V	C/T	0.425	-0.006 (0.001)	1.77×10 -7	144009	HbA1c	C2CD2L
FI 1	rs145878042	RAPGEF3	p.L300P	G/A	0.011	-0.054 (0.01)	1.15×10 -7	91485	FI/HbA1c	HDAC7/ PFKM
HbA1c	rs2732481	ZNF641	p.Q363P	G/T	0.315	-0.009 (0.001)	2.07×10 -11	142280	HbA1c	SENP1
HbA1c	rs3184504	SH2B3	p.W262R	C/T	0.567	0.007 (0.001)	5.98×10 -8	138551	HbA1c	ATXN2
2hGlu	rs1169288	HNF1A	p.I75L	C/A	0.345	0.06 (0.011)	7.90×10 -9	44278	T2D, 2hGlu	HNF1A
HbA1c	COSM147717	ATP11A	p.M317V	G/A	0.748	0.009 (0.001)	3.77×10 -12	144022	HbA1c	ATP11A,TUBGCP3
HbA1c	rs229587	SPTB	p.S439N	T/C	0.357	0.007 (0.001)	2.60×10 -8	134780	HbA1c	SPTB
HbA1c	rs35097172	SLC25A47	splice region variant, 5’ UTR variant	T/C	0.216	-0.008 (0.002)	5.67×10 -8	144028	FG	SLC25A47
2hGlu	rs3784634	VPS13C	p.R974K	T/C	0.540	-0.069 (0.011)	6.40×10 -10	37217	2hGlu	VPS13C/ C2CD4A/ C2CD4B
HbA1c 1	rs3747481	PRR14	p.P359L	T/C	0.261	0.009 (0.002)	3.30×10 -8	103338	HbA1c	ITGAD
HbA1c	rs201226914	PIEZO1	p.L939M	T/G	0.002	-0.159 (0.015)	4.42×10 -26	144024	HbA1c	CDT1,CYBA
2hGlu	rs72839768	DVL2	p.T529I	A/G	0.020	0.197 (0.03)	4.10×10 -11	57866	T2D, 2hGlu	SLC16A13
HbA1c	rs2748427	TMC6	p.W125R	G/A	0.233	0.027 (0.002)	8.56×10 -70	132326	HbA1c	TMC6
HbA1c	rs7225887	B3GNTL1	p.A163T	T/C	0.211	-0.015 (0.002)	5.73×10 -22	125749	HbA1c	FN3KRP, FN3K
HbA1c	rs35413309	RGS9BP	p.A223V	T/C	0.030	-0.02 (0.004)	1.42×10 -8	141598	HbA1c	PDCD5
2hGlu	rs1800437	GIPR	p.E318Q	C/G	0.217	0.103 (0.011)	2.59×10 -23	56252	2hGlu	GIPR
FG	rs17265513	ZHX3	p.N310S	C/T	0.188	0.016 (0.002)	2.59×10 -10	126253	FG	ZHX3
HbA1c	rs855791	TMPRSS6	V727A	G/A	0.577	-0.019 (0.001)	9.46×10 -51	143907	HbA1c	TMPRSS6
FG	rs15943	MAP3K15	p.Q1083E	C/G	0.005	-0.084 (0.014)	2.83×10 -9	67004	glucose	PDHA1/MAP3K15
FG	rs56381411	MAP3K15	p.G670S	T/C	0.005	-0.085 (0.013)	1.51×10 -11	62319	glucose	PDHA1/MAP3K15
HbA1c	rs2229241	RENBP	splice acceptor variant	C/T	0.012	-0.123 (0.007)	1.14×10 -62	95622	HbA1c	G6PD
HbA1c	rs1050828	G6PD	p.V68M	T/C	0.007	-0.334 (0.008)	7.41×10 -322	112209	HbA1c	G6PD

Table 2. Gene-based results from broad (NSbroad mask) and strict (NSstrict mask) analyses.

Genes in bold are newly discovered from this effort. N var: total number of variants in that gene-based analysis; P _burden: p-value from burden test which assumes all variants have the same direction of effect; P _SKAT: p-value from SKAT test which allows for different directions of effect between variants. The lowest p-value is highlighted in bold.

Trait	Gene	NSbroad mask			NSstrict mask
		N var	P burden	P SKAT	N var	P burden	P SKAT
FG	G6PC	9	1.41×10 -6	1.32×10 -5	3	1.41×10 -3	7.43×10 -4
FI	G6PC	8	1.62×10 -6	8.58×10 -6	3	1.85×10 -3	7.80×10 -3
HbA1c	TF	10	2.15×10 -6	5.98×10 -3	3	5.48×10 -2	5.48×10 -2
FG	MAP3K15	18	1.86×10 -25	1.07×10 -18	7	1.34×10 -14	4.01×10 -11
HbA1c	MAP3K15	18	1.27×10 -7	1.53×10 -04	7	2.65×10 -4	9.46×10 -3
FG	G6PC2	18	4.09×10 -67	5.38×10 -58	7	7.8×10 -69	3.83×10 -56
HbA1c	G6PC2	18	6.18×10 -30	4.65×10 -27	7	1.04×10 -31	1.92×10 -26
FG	SLC30A8	13	5.69×10 -4	6.42×10 -11	7	6.55×10 -11	3.74×10 -10
HbA1c	SLC30A8	12	7.20×10 -8	2.18×10 -5	6	5.66×10 -8	3.22×10 -6
FG	VPS13C	52	9.66×10 -6	3.73×10 -7	26	1.27×10 -5	1.44×10 -5

Identification of single-variant associations

Our single variant analyses identified 62 distinct coding variant associations at 58 genes associated with at least one of the glycemic traits at exome-wide significance ( P <2.2 × 10 ^-7) ( Table 1). Of these, four variants at three genes represented novel associations. These included a missense (rs1983210, p.E1365D) and a splice region variant (rs3183099) in OBSL1 associated with FI, another missense variant (rs1886686, p.G12A) in WDR78 associated with FG, and a missense variant (rs31244, p.D543N) in SV2C associated with HbA1c ( Table 1). In addition, the missense variant (rs146886108, p.R187Q) in ANKH which was previously associated with T2D was associated for the first time with FG.

Identification of gene-based associations

Our gene-based analyses identified six genes associated with glycemic traits, including G6PC and TF that had not been associated with glycemic traits before ( Table 2 and Supplementary Table S2 ¹¹ ). These findings provide new hypotheses for downstream follow-up studies in the context of glycemic trait biology. G6PC, encoding glucose-6-phosphatase, is associated with FG and FI and is a homolog of G6PC2. G6PC2 is an established effector gene at a GWAS locus which contains multiple coding variants known to influence FG and HbA1c but not FI levels ^{4, 5, 28– 30} . Loss-of-function variants at SLC30A8 have been previously associated with reduced risk of T2D ^{31– 33} , while VPS13C maps to the VPS13C/ C2CD4A/ C2CD4B T2D risk locus. Follow-up studies at this locus have with varying levels of evidence suggested C2CD4A, encoding a calcium-dependent nuclear protein, as the causal gene for T2D through its potential role in the pancreatic islets ^{34– 37} . We found evidence of association at MAP3K15 with reduced levels of FG and HbA1c ( Table 2 and Supplementary Table S2 ¹¹ ), which is consistent with recent reports of the gene’s association with reduced levels of HbA1c and glucose, and reduced T2D risk ^{6, 38} . Our analyses also detected TF (encoding transferrin) as a novel gene-based association signal associated with HbA1c but not any of the other glycemic traits, consistent with the role of the protein as the main iron carrier in the blood ( Table 2 and Supplementary Table S2 ¹¹ ).

Pathway analyses identify relevant gene sets regulating glycemia

Next, we used our coding variant association results to identify pathways enriched for glycemic trait associations, and to subsequently determine the extent to which different associations within the same trait implicate the same or similar pathways (as indicated by the functional connectivity of the network). To do this we used GeneMANIA network analysis ³⁹ , which takes a query list of genes and finds functionally similar genes based on large, publicly available biological datasets, that include protein-protein, protein-DNA and genetic interactions, pathways, protein domains, protein and gene expression data. GeneMANIA taps on updated versions of these databases for its core and network analyses, to identify related genes of known functions based on our input list of genes. To increase power to connect genes in a network, we considered all genes harboring non-synonymous variants that reached P <1 × 10 ^-5 (Supplementary Table S3 ¹¹ ) for any of the four glycemic traits in our study and mapped the most significant non-synonymous variant at each locus to the respective gene (totaling 121 associations across all traits) (Methods). A high degree of connectivity was observed within the HbA1c network, with enrichment of processes related to blood cell biology such as porphyrin metabolism, erythrocyte homeostasis and iron transport ( Figure 1A and Supplementary Table S4 ¹¹ ). In comparison, the network generated from FG-associated genes captured several processes known to contribute to glucose regulation and islet function, including insulin secretion, zinc transport and fatty acid metabolism ( Figure 1B and Supplementary Table S4 ¹¹ ). Given that there were fewer genes associated with FI and 2hGlu, we were less powered to draw meaningful insights from the enriched pathways in those traits (Supplementary Figure S1 and Supplementary Table S4 ¹¹ ).

We also performed gene set enrichment analysis (GSEA) using EC-DEPICT ^{22, 23} (Methods). The primary innovation of EC-DEPICT is the use of 14,462 gene sets extended based on large-scale co-expression data ^{21, 24} . These gene sets take the form of z-scores, where higher z-scores indicate a stronger prediction that a given gene is a member of a gene set. To reduce some of the redundancy in the gene sets (many of which are strongly correlated with one another), we clustered them into 1,396 “meta-gene sets” using affinity propagation clustering ²⁵ . These meta-gene sets are used to simplify visualizations and aid interpretation of results. As before, we considered all loci with variants that reached P <1 × 10 ^-5 (Supplementary Table S3 ¹¹ ) for any of the four glycemic traits for defining input genes (Methods). When looking across all traits combined, we found 234 significant gene sets in 86 meta-gene sets with false discovery rate (FDR) of <0.05 (Supplementary Table S5A, Supplementary Figure S2A ¹¹ ). As expected, we observed a strong enrichment of insulin- and glucose-related gene sets, as well as hormone secretion and cytoplasmic vesicle gene sets (in keeping with pancreatic beta cell insulin vesicle release). In agreement with the GeneMANIA network analyses, we also noted a particularly strong enrichment for blood-related pathways represented by gene sets such as erythrocyte differentiation and heme metabolic process, which was primarily driven by HbA1c-associated variants. This was likely because HbA1c levels are influenced not only by glycation but also by blood cell turnover rate ^{1, 40, 41} . To disentangle blood cell turnover from effects due to glycation, we repeated the analysis excluding variants that were significantly associated with HbA1c only and found 128 significant gene sets in 53 meta-gene sets (FDR <0.05) ( Figure 1C, Supplementary Table S5B, Supplementary Figure S2B ¹¹ ). Indeed, we noted that majority of the gene sets now implicated pathways relevant to the pancreatic islets and metabolic tissues, such as “abnormal glucose homeostasis”, “peptide hormone secretion”, “Maturity Onset Diabetes of the Young”, and multiple pathways involved in the regulation of glycogen, incretins, and carbohydrate metabolism, that were also seen in the FG only analysis ( Figure 1D, Supplementary Table S5D, Supplementary Figure S2D ¹¹ ).

We also analyzed each of the four traits separately, to reveal trait-specific enriched gene sets (Supplementary Table S5, Supplementary Figure S2C-E, Supplementary Figure S3C-D ¹¹ , Methods). Overall, our network and pathway enrichment analyses provide insight into the biology underlying each glycemic trait and may facilitate the prioritization of specific genes or pathways across multiple different phenotypes.

Discussion

Here we have described large scale meta-analyses results for coding variant and gene-based associations for four glycemic traits, FG, FI, HbA1c and 2Glu, and the downstream pathways and networks that are regulated by the associated genes. Our results identified three genes with novel single-variant associations with glycemic traits OBSL1 (FI), WDR78 (FG) and SVC2 (HbA1c). OBSL1 encodes a cytoskeletal protein related to obscurin, mutations in which have been shown to lead to an autosomal recessive primordial growth disorder (OMIM: 612921). Loss of OBSL1 leads to downregulation of CUL7, a protein known to interact with IRS-1, downstream of the insulin receptor signaling pathway ⁴² . WDR78 encodes a WD repeat-containing protein 78, the same variant rs1886686-C has been previously associated with a decrease in systolic blood pressure ⁴³ . However, none of the OBSL1 (rs1983210, b = -0.018, p = 1.20 x 10 ^-4, N = 144,114; rs3183099, b = -0.019, p = 1.36 x 10 ^-4, N = 125,397) or WDR78 (rs1886686, b = -0.017, p = 3.83 x 10 ^-5, N = 164,878) variants we detected here reached exome-wide significance in our recent large multi-ancestry study ¹ . This, despite larger sample sizes and good genotype quality (info >0.8 for each of the variants for the majority of cohorts), suggesting caution in the interpretation of these findings, and the need for additional datasets testing these associations. The final variant, p.D543N in SV2C, was associated with HbA1c with p = 5.5 x 10 ^-5 in the European meta-analysis ¹ , and with p = 1.37 x 10 ^-12 in UK biobank ⁴⁴ . A second missense variant at this gene, p.T482S, is also strongly associated with HbA1c (p = 1.9 x 10 ^-16) and with red blood cell distribution width in UK biobank (p = 3.3 x 10 ^-11) ⁴⁴ , and with mean corpuscular volume (p = 3 x 10 ^-11) ⁴⁵ . Given that variation in red blood cell traits can influence HbA1c levels ^{1, 41} , associations between these missense variants suggest SV2C as the likely effector gene at this locus. Also, the absence of evidence for association between this gene and other glycemic traits suggests its effect on HbA1c is independent of glycemia.

The novel gene-based association of G6PC with FG and FI was notable. Homozygous inactivating alleles in G6PC, including both p.R83C and p.Q347X which are contained in our gene-based association (Table S2), are known to give rise to glycogen storage disease type 1a (GSD1a). GSD1a is a rare autosomal recessive metabolic disorder ^{46, 47} , but this is the first time that rare coding variants in G6PC have been shown to influence FG and FI levels in normoglycemic individuals. The other novel gene-based association was between TF and HbA1c. TF encodes transferrin, an iron-binding transport protein that circulates at high levels in blood plasma as an important biological carrier of iron. Dysregulation of iron concentrations due to reduced transferrin levels or function could affect the measurement of HbA1c independently of glycemia ⁴⁸ . The presence of multiple coding variants within TF associated with red blood cell traits in UK biobank ⁴⁴ lends additional support to this hypothesis.

Overall, our network and pathway analyses were highly concordant with each other and with other published data identifying processes related to glucose regulation and islet function, including insulin secretion and zinc transport associated with FG loci, and red blood cell biology processes amongst HbA1c associated loci ¹ . The FG network revealed linking nodes (that are not among the association signals) with known links to glucose homeostasis and diabetes, such as GCK (encoding the beta cell glucose sensor glucokinase), GCG (encoding the peptide hormone glucagon secreted by the alpha cells of the pancreas) and GIP (encoding the incretin hormone gastric inhibitory polypeptide). Notably, lipid related pathways associated with fasting glucose. One gene within the FG cluster for lipid-related pathways is CERS2, which encodes ceramide synthase 2, an enzyme known to be associated with the sphingolipid biosynthetic process ( Figure 1B, Supplementary Table S3 ¹¹ ). Although CERS2 is only nominally associated with FG and is significantly associated with HbA1c (rs267738: P _FG = 3.54 × 10 ^-7; P _HbA1c = 6.96 × 10 ^-10), it does not cluster together with any HbA1c-enriched pathway, suggesting that CERS2 is regulating FG and HbA1c indirectly through its role in lipid metabolism.

Conclusions

In conclusion, our results provided novel glycemic trait associations and highlighted pathways implicated in glycemic regulation. The summary statistics results are being made publicly available through various platforms so they can be harnessed with other data to aid effector gene identification.

Funding Statement

This work was supported by Wellcome [grant numbers 095101 and 200837 to Anna Gloyn]; Inês Barroso is funded by Wellcome (WT206194) and this work was partly supported by “Expanding excellence in England” award from Research England; The Fenland Study is funded by Wellcome Trust and the Medical Research Council (MC_U106179471); Genotyping of the Generation Scotland GS:SFHS samples was funded by the Wellcome Trust (Wellcome Trust Strategic Award “STratifying Resilience and Depression Longitudinally” (STRADL) Reference 104036/Z/14/Z) and the Medical Research Council UK. Generation Scotland received core funding from the Chief Scientist Office of the Scottish Government Health Directorate CZD/16/6 and the Scottish Funding Council HR03006; GoDARTS study was funded by The Wellcome Trust Study Cohort Wellcome Trust Functional Genomics Grant (2004-2008) (Grant No: 072960/2/03/2) and The Wellcome Trust Scottish Health Informatics Programme (SHIP) (2009-2012). (Grant No: 086113/Z/08/Z); For HELIC MANOLIS and HELIC Pomak this work was funded by the Wellcome Trust (098051) and the European Research Council (ERC-2011-StG 280559-SEPI); The LOLIPOP study is supported by the Wellcome Trust (084723/Z/08/Z), the National Institute for Health Research (NIHR) Comprehensive Biomedical Research Centre Imperial College Healthcare NHS Trust, the British Heart Foundation (SP/04/002), the Medical Research Council (G0601966,G0700931), the NIHR (RP-PG-0407-10371), European Union FP7 (EpiMigrant, 279143) and Action on Hearing Loss (G51). Mark I McCarthy was funded by Wellcome (grant numbers 090532, 098381, 106130, 203141 and 212259) and NIH U01-DK105535; Genotyping and analysis of PIVUS and ULSAM were funded by the Wellcome Trust under awards WT064890, WT090532 and WT098017. PIVUS and ULSAM are supported by the Swedish Research Council, Swedish Heart-Lung Foundation, Swedish Diabetes Foundation and Uppsala University; TwinsUK is funded by the Wellcome Trust, Medical Research Council, European Union, Chronic Disease Research Foundation (CDRF), and the National Institute for Health Research (NIHR)-funded BioResource, Clinical Research Facility and Biomedical Research Centre based at Guy’s and St Thomas’ NHS Foundation Trust in partnership with King’s College London; 1958 British Birth cohort: analysis was supported by BHF programme grant (Deloukas) RG/14/5/30893; AGES has been funded by NIH contracts N01-AG-1-2100 and 271201200022C, the NIA Intramural Research Program, Hjartavernd (the Icelandic Heart Association), and the Althingi (the Icelandic Parliament); The Atherosclerosis Risk in Communities (ARIC) study is carried out as a collaborative study supported by the National Heart, Lung, and Blood Institute (NHLBI) contracts (HHSN268201100005C, HHSN268201100006C, HHSN268201100007C, HHSN268201100008C, HHSN268201100009C, HHSN268201100010C, HHSN268201100011C, and HHSN268201100012C). Funding support for “Building on GWAS for NHLBI-diseases: the U.S. CHARGE consortium” was provided by the NIH through the American Recovery and Reinvestment Act of 2009 (ARRA) (5RC2HL102419). ML was supported by a National Heart, Lung, and Blood Institute T32-HL0072024 Cardiovascular Epidemiology Training Grant; ASCOT: this work was supported by Pfizer, New York, NY, USA, for the ASCOT study and the collection of the ASCOT DNA repository, by Servier Research Group, Paris, France and by Leo Laboratories, Copenhagen, Denmark; Athero-Express Biobank Study: Dr. Sander W. van der Laan is funded through EU H2020 TO_AITION (grant number: 848146). We are thankful for the support of the Netherlands CardioVascular Research Initiative of the Netherlands Heart Foundation (CVON 2011/B019 and CVON 2017-20: Generating the best evidence-based pharmaceutical targets for atherosclerosis [GENIUS I&II]), the ERA-CVD program ‘druggable-MI-targets’ (grant number: 01KL1802), and the Leducq Fondation ‘PlaqOmics’; BioMe: The Mount Sinai BioMe Biobank is supported by The Andrea and Charles Bronfman Philanthropies; Caroline Hayward is supported by an MRC University Unit Programme Grant “QTL in Health and Disease” (U. MC_UU_00007/10); CHS: this CHS research was supported by NHLBI contracts HHSN268201200036C, HHSN268200800007C, HHSN268201800001C, N01HC55222, N01HC85079, N01HC85080, N01HC85081, N01HC85082, N01HC85083, N01HC85086 and 75N92021D00006; and NHLBI grants U01HL080295, R01HL068986, R01HL087652, R01HL105756, R01HL103612, R01HL120393, and U01HL130114 with additional contribution from the National Institute of Neurological Disorders and Stroke (NINDS). Additional support was provided through R01AG023629 from the National Institute on Aging (NIA). The provision of genotyping data was supported in part by the National Center for Advancing Translational Sciences, CTSI grant UL1TR001881, and the National Institute of Diabetes and Digestive and Kidney Disease Diabetes Research Center (DRC) grant DK063491 to the Southern California Diabetes Endocrinology Research Center; COPSAC2000: We greatly acknowledge the private and public research funding allocated to COPSAC and listed on www.copsac.com, with special thanks to The Lundbeck Foundation (Grant nr. R16-A1694), Ministry of Health (Grant nr. 903516), Danish Council for Strategic Research (Grant nr.: 0603-00280B), The Danish Council for Independent Research and The Capital Region Research Foundation as core supporters; CROATIA_Korcula: Exome array genotyping was funded by UK’s Medical Research Council; DIABNORD: the current study was funded by Novo Nordisk, the Swedish Research Council, Påhlssons Foundation, the Swedish Heart Lung Foundation, and the Skåne Regional Health Authority (all to PWF); The DPS has been financially supported by grants from the Academy of Finland (117844 and 40758, 211497, and 118590 (MU); The EVO funding of the Kuopio University Hospital from Ministry of Health and Social Affairs (5254), Finnish Funding Agency for Technology and Innovation (40058/07), Nordic Centre of Excellence on ‘Systems biology in controlled dietary interventions and cohort studies, SYSDIET (070014), The Finnish Diabetes Research Foundation, Yrjö Jahnsson Foundation (56358), Sigrid Juselius Foundation and TEKES grants 70103/06 and 40058/07; The DR's EXTRA Study was supported by the Ministry of Education and Culture of Finland (722 and 627;2004-2011), Academy of Finland (102318; 104943;123885; 211119), Kuopio University Hospital, Finnish Diabetes Association, Finnish Foundations for Cardiovascular Research, Päivikki and Sakari Sohlberg Foundation, by European Commission FP6 Integrated Project (EXGENESIS); LSHM-CT-2004-005272, City of Kuopio and Social Insurance Institution of Finland (4/26/2010); EFSOCH: this paper presents independent research supported by the National Institute for Health Research (NIHR) Exeter Clinical Research Facility; The EPIC-Norfolk study (DOI 10.22025/2019.10.105.00004) has received funding from the Medical Research Council (MR/N003284/1 and MC-UU_12015/1) and Cancer Research UK (C864/A14136). The genetics work in the EPIC-Norfolk study was funded by the Medical Research Council (MC_PC_13048); EPIC-Potsdam: the study was supported in part by a grant from the German Federal Ministry of Education and Research (BMBF) and the State of Brandenburg to the German Center for Diabetes Research (DZD e.V.) (82DZD00302). The recruitment phase of the EPIC-Potsdam study was supported by the Federal Ministry of Science, Germany (01 EA 9401) and the European Union (SOC 95201408 05 F02). The follow-up of the EPIC-Potsdam study was supported by German Cancer Aid (70-2488-Ha I) and the European Community (SOC 98200769 05 F02); EpiHealth was supported by the Swedish Research Council strategic research network Epidemiology for Health, Uppsala University and Lund University. Genotyping in EpiHealth was supported by Swedish Heart-Lung Foundation (grant no. 20120197 and 20140422), Knut och Alice Wallenberg Foundation (grant no. 2013.0126), and Swedish Research Council (grant no. 2012-1397); Erasmus Rucphen Family (ERF) was supported by the Consortium for Systems Biology (NCSB), both within the framework of the Netherlands Genomics Initiative (NGI)/Netherlands Organisation for Scientific Research (NWO). ERF study as a part of EUROSPAN (European Special Populations Research Network) was supported by European Commission FP6 STRP grant number 018947 (LSHG-CT-2006-01947) and also received funding from the European Community's Seventh Framework Programme (FP7/2007-2013)/grant agreement HEALTH-F4-2007-201413 by the European Commission under the programme “Quality of Life and Management of the Living Resources” of 5th Framework Programme (no. QLG2-CT-2002-01254) as well as FP7 project EUROHEADPAIN (nr 602633). High-throughput analysis of the ERF data was supported by joint grant from Netherlands Organisation for Scientific Research and the Russian Foundation for Basic Research (NWO-RFBR 047.017.043).The exome-chip measurements have been funded by the Netherlands Organization for Scientific Research (NWO; project number 184021007) and by the Rainbow Project (RP10; Netherlands Exome Chip Project) of the Biobanking and Biomolecular Research Infrastructure Netherlands (BBMRI-NL; www.bbmri.nl (http://www.bbmri.nl) ). Ayse Demirkan is supported by a Veni grant (2015) from ZonMw. Ayse Demirkan, Jun Liu and Cornelia van Duijn have used exchange grants from PRECEDI; The Family Heart Study (FamHS) was supported by NIH grants R01-HL-087700 and R01-HL-088215 from NHLBI, and R01-DK-8925601 and R01-DK-075681 from NIDDK; The FIA3 study was supported in part by a grant from the Swedish Heart-Lund Foundation (to PWF); The FIN-D2D 2007 study was supported by funds from the hospital districts of Pirkanmaa; Southern Ostrobothnia; North Ostrobothnia; Central Finland and Northern Savo; the Finnish National Public Health Institute; the Finnish Diabetes Association; the Ministry of Social Affairs and Health in Finland; Finland’s Slottery Machine Association; the Academy of Finland [grant number 129293] and Commission of the European Communities, Directorate C-Public Health [grant agreement no. 2004310]; FINRISK 2007: VS was supported by the Finnish Foundation for Cardiovascular Research. PJ was supported by the Academy of Finland #118065; Folkert Asselbergs is supported by UCL Hospitals NIHR Biomedical Research Centre; Framingham Heart Study: Genotyping, quality control and calling of the Illumina HumanExome BeadChip in the Framingham Heart Study was supported by funding from the National Heart, Lung and Blood Institute Division of Intramural Research (Daniel Levy and Christopher J. O’Donnell, Principle Investigators). Also supported by National Institute for Diabetes and Digestive and Kidney Diseases (NIDDK) U01 DK078616, UM1 DK078616, NIDDK K24 DK080140 and American Diabetes Association Mentor-Based Postdoctoral Fellowship Award #7-09-MN-32, all to Dr. Meigs, and NIDDK Research Career Award K23 DK65978, a Massachusetts General Hospital Physician Scientist Development Award and a Doris Duke Charitable Foundation Clinical Scientist Development Award to Dr. Florez; The FUSION study was supported by DK093757, DK072193, DK062370, and ZIA-HG000024. HAK has received funding from Academy of Finland (support for clinical research careers, grant no 258753); Support for GENOA was provided by the National Heart, Lung and Blood Institute (HL054464; HL054481; HL087660; HL119443; HL086694) of the National Institutes of Health; GIANT: Anne E Justice (AEJ) is funded under NIH 5K99HL130580-02; GLACIER: the current study was funded by Novo Nordisk, the Swedish Research Council, Påhlssons Foundation, the Swedish Heart Lung Foundation, and the Skåne Regional Health Authority (all to PWF); The Health, Aging, and Body Composition (HABC) Study is supported by NIA contracts N01AG62101, N01AG62103, and N01AG62106. The genome-wide association study was funded by NIA grant 1R01AG032098-01A1 to Wake Forest University Health Sciences; The HANDLS study was supported by the Intramural Research Program of the NIH, National Institute on Aging and the National Center on Minority Health and Health Disparities (project # Z01-AG000513 and human subjects protocol number 09-AG-N248); HRS is supported by the National Institute on Aging (NIA U01AG009740). The genotyping was funded separately by the National Institute on Aging (RC2 AG036495, RC4 AG039029); The Health2006 was financially supported by grants from the Velux Foundation; The Danish Medical Research Council, Danish Agency for Science, Technology and Innovation; The Aase and Ejner Danielsens Foundation; ALK-Abello A/S, Hørsholm, Denmark, and Research Centre for Prevention and Health, the Capital Region of Denmark; The Health2008 was supported by the Timber Merchant Vilhelm Bang’s Foundation, the Danish Heart Foundation (Grant number 07-10-R61-A1754-B838-22392F), and the Health Insurance Foundation (Helsefonden) (Grant number 2012B233); Heather M. Highland is supported by funding from NHLBI training grant T32 HL007055; Hidetoshi Kitajima was funded by Manpei Suzuki Diabetes Foundation Grant-in-Aid for the young scientists working abroad; Inter99: the study was financially supported by research grants from the Danish Research Council, the Danish Centre for Health Technology Assessment, Novo Nordisk Inc., Research Foundation of Copenhagen County, Ministry of Internal Affairs and Health, the Danish Heart Foundation, the Danish Pharmaceutical Association, the Augustinus Foundation, the Ib Henriksen Foundation, the Becket Foundation, and the Danish Diabetes Association; Funding for the InterAct project was provided by the EU FP6 programme (grant number LSHM_CT_2006_037197); Jennifer L. Asimit: Medical Research Council Methodology Research Fellowship (MR/K021486/1); The JHS is supported by contracts HHSN268201300046C, HHSN268201300047C, HHSN268201300048C, HHSN268201300049C, HHSN268201300050C from the National Heart, Lung and Blood Institute and the National Institute on Minority Health and Health Disparities. ExomeChip genotyping was supported by the NHLBI of the National Institutes of Health under award number R01HL107816 to S. Kathiresan; Joel N. Hirschhorn: NIH R01DK075787; The KORA study was initiated and financed by the Helmholtz Zentrum München – German Research Center for Environmental Health, which is funded by the German Federal Ministry of Education and Research (BMBF) and by the State of Bavaria. Furthermore, KORA research was financed by a grant from the BMBF to the German Center for Diabetes Research (DZD) and a grant from the Ministry of Innovation, Science, Research and Technology of the state North Rhine-Westphalia (Düsseldorf, Germany). It was also supported within the Munich Center of Health Sciences (MC-Health), Ludwig-Maximilians-Universität, as part of LMUinnovativ; Leipzig Adults: this work was supported by grants from the German Research Council (SFB- 1052 “Obesity mechanisms”; B01; B03), from the German Diabetes Association and from the DHFD (Diabetes Hilfs- und Forschungsfonds Deutschland). IFB AdiposityDiseases is supported by the Federal Ministry of Education and Research (BMBF), Germany, FKZ: 01EO1501 (AD2-060E, AD2-06E99). This work was further supported by the Kompetenznetz Adipositas (Competence network for Obesity) funded by the Federal Ministry of Education and Research (German Obesity Biomaterial Bank; FKZ 01GI1128); Leipzig-Childhood-IFB: this work was supported by grants from Integrated Research and Treatment Centre (IFB) Adiposity Diseases, from the German Research Foundation for the Clinical Research Group “Atherobesity” KFO 152 (KO3512/1 to AK), and by the European Commission (Beta-JUDO) and by EFRE (LIFE Child Obesity); Lothian Birth Cohort 1921 and Lothian Birth Cohort 1936: phenotype collection in the Lothian Birth Cohort 1921 was supported by the UK’s Biotechnology and Biological Sciences Research Council (BBSRC), The Royal Society and The Chief Scientist Office of the Scottish Government. Phenotype collection in the Lothian Birth Cohort 1936 was supported by Age UK (The Disconnected Mind project). Genotyping was supported by Centre for Cognitive Ageing and Cognitive Epidemiology (Pilot Fund award), Age UK, and the Royal Society of Edinburgh. The work was undertaken by The University of Edinburgh Centre for Cognitive Ageing and Cognitive Epidemiology, part of the cross council Lifelong Health and Wellbeing Initiative (MR/K026992/1). Funding from the BBSRC and Medical Research Council (MRC) is gratefully acknowledged; MESA and the MESA SHARe projects are conducted and supported by the National Heart, Lung, and Blood Institute (NHLBI) in collaboration with MESA investigators. Support for MESA is provided by contracts 75N92020D00001, HHSN268201500003I, N01-HC-95159, 75N92020D00005, N01-HC-95160, 75N92020D00002, N01-HC-95161, 75N92020D00003, N01-HC-95162, 75N92020D00006, N01-HC-95163, 75N92020D00004, N01-HC-95164, 75N92020D00007, N01-HC-95165, N01-HC-95166, N01-HC-95167, N01-HC-95168, N01-HC-95169, UL1-TR-000040, UL1-TR-001079, and UL1-TR-001420, UL1TR001881, DK063491, and R01HL105756. Funding for SHARe genotyping was provided by NHLBI Contract N02-HL-64278; The METSIM study was supported by the Academy of Finland (contract 124243), the Finnish Heart Foundation, the Finnish Diabetes Foundation, Tekes (contract 1510/31/06), and the Commission of the European Community (HEALTH-F2-2007 201681), and the US National Institutes of Health grants DK093757, DK072193, DK062370, and ZIA- HG000024; Natasha H J Ng (NHJN) is supported by the National Science Scholarship from the Agency for Science, Technology and Research (A*STAR) in Singapore; The genotyping in the NEO study was supported by the Centre National de Génotypage (Paris, France), headed by Jean-Francois Deleuze. The NEO study is supported by the participating Departments, the Division and the Board of Directors of the Leiden University Medical Center, and by the Leiden University, Research Profile Area Vascular and Regenerative Medicine. Dennis Mook-Kanamori is supported by Dutch Science Organization (ZonMW-VENI Grant 916.14.023); NFBC1966 and 1986 received financial support from the Academy of Finland (project grants 104781, 120315, 129269, 1114194, 24300796, Center of Excellence in Complex Disease Genetics and SALVE), University Hospital Oulu, Biocenter, University of Oulu, Finland (75617), NIHM (MH063706, Smalley and Jarvelin), Juselius Foundation, NHLBI grant 5R01HL087679-02 through the STAMPEED program (1RL1MH083268-01), NIH/NIMH (5R01MH63706:02), the European Commission (EURO-BLCS, Framework 5 award QLG1-CT-2000-01643), ENGAGE project and grant agreement HEALTH-F4-2007-201413, EU FP7 EurHEALTHAgeing -277849, the Medical Research Council, UK (G0500539, G0600705, G1002319, PrevMetSyn/SALVE) and the MRC, Centenary Early Career Award. The program is currently being funded by the H2020-633595 DynaHEALTH action and academy of Finland EGEA-project (285547). The DNA extractions, sample quality controls, biobank up-keeping and aliquotting was performed in the National Public Health Institute, Biomedicum Helsinki, Finland and supported financially by the Academy of Finland and Biocentrum Helsinki; The Botnia and The PPP-Botnia studies have been financially supported by grants from Folkhalsan Research Foundation, the Sigrid Juselius Foundation, The Academy of Finland and university of Helsinki (grants no. 263401, 267882, 312063, 312072 and 336826), the European Research Council under the European Union's Seventh Framework Programme (FP7/2007-2013) / ERC grant agreement n° 269045, Nordic Center of Excellence in Disease Genetics, EU (EXGENESIS, MOSAIC FP7-600914), Ollqvist Foundation, Swedish Cultural Foundation in Finland, Finnish Diabetes Research Foundation, Foundation for Life and Health in Finland, Signe and Ane Gyllenberg Foundation, Finnish Medical Society, Paavo Nurmi Foundation, Helsinki University Central Hospital Research Foundation, Perklén Foundation, Närpes Health Care Foundation and Ahokas Foundation. The study has also been supported by the Ministry of Education in Finland, Municipal Heath Care Center and Hospital in Jakobstad and Health Care Centers in Vasa, Närpes and Korsholm; PROMIS: Dr. Saleheen has received grants from the National Heart, Lung and Blood Institute, Fogarty International, Pfizer, Regeneron, Eli Lilly, and Genentech; Robert Sladek is the recipient of a Chercheur Boursier award from the Fonds de la Recherche en Santé du Québec and a New Investigator Award from the Canadian Institutes of Health Research and is supported by operating funds from the Canadian Institutes of Health Research; Rebecca Fine is supported by NHGRI F31 HG009850; The RISC study was supported by European Union grant QLG1-CT-2001-01252 and AstraZeneca. The initial genotyping of the RISC samples was funded by Merck & Co Inc.; Rona J Strawbridge is supported by a UKRI Innovation at HDR and University of Glasgow LKAS Fellowship; The Rotterdam Study is funded by Erasmus Medical Center and Erasmus University, Rotterdam, Netherlands Organization for the Health Research and Development (ZonMw), the Research Institute for Diseases in the Elderly (RIDE), the Ministry of Education, Culture and Science, the Ministry for Health, Welfare and Sports, the European Commission (DG XII), and the Municipality of Rotterdam. The Exome chip array data set was funded by the Genetic Laboratory of the Department of Internal Medicine, Erasmus MC, from the Netherlands Genomics Initiative (NGI)/Netherlands Organisation for Scientific Research (NWO)-sponsored Netherlands Consortium for Healthy Aging (NCHA; project nr. 050-060-810), the Netherlands Organization for Scientific Research (NWO; project number 184021007) and by the Rainbow Project (RP10; Netherlands Exome Chip Project) of the Biobanking and Biomolecular Research Infrastructure Netherlands (BBMRI-NL; www.bbmri.nl (http://www.bbmri.nl)); SardiNIA: the study is supported by National Human Genome Research Institute grants HG005581, HG005552, HG006513, HG007022 and HG007089; by National Heart, Lung, and Blood Institute grant HL117626; by the Intramural Research Program of the US National Institutes of Health, National Institute on Aging, contracts N01-AG-1-2109 and HHSN271201100005C; by Sardinian Autonomous Region (L.R. 7/2009) grant cRP3-154; The Singapore Chinese Eye Study was funded by the Agency for Science, Technology and Research - Biomedical Research Council (A*STAR BMRC) Grant, Singapore [08/1/35/19/550] and Singapore Ministry of Health’s National Medical Research Council [NMRC/CIRG/1417/2015]; Sorbs: this work was supported by grants from the German Research Council (DFG - SFB 1052 “Obesity mechanisms”; A01, C01, B03 and SPP 1629 TO 718/2-1), from the German Diabetes Association and from the DHFD (Diabetes Hilfs- und Forschungsfonds Deutschland); The UK Household Longitudinal Study is funded by the Economic and Social Research Council; The Vejle Diabetes Biobank was supported by The Danish Research Council for Independent Research; The WGHS is funded by the National Heart, Lung, and Blood Institute (HL043851 and HL080467) and National Cancer Institute (CA047988 and UM1CA182913). Funding for genotyping on the Exome Array was funded by Amgen.

The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Data availability

Underlying data

Open Science Framework (OSF): Underlying data for ‘Large-scale exome array summary statistics resources for glycemic traits to aid effector gene prioritization’, https://doi.org/10.17605/OSF.IO/K6W3B ¹¹

This project contains the following underlying data:

• Table S1: Supplementary Table S1 – Cohort characteristics, genotyping and quality control (QC), glucose, insulin, 2hGlu and HbA1c analyses and covariates.

• Table S2: Supplementary Table S2 - Full gene-based results including all variants included in the masks, for both novel and previously-established genes

• Table S3: Supplementary Table S3 - All variants associated with FG, FI, HbA1c and/or 2hGlu in our analyses with P<10-5

• Table S4: Supplementary Table S4 - Gene Set Enrichment Analysis by GeneMANIA network analysis showing enriched GO terms and Reactome pathways in the network for (A) HbA1c; (B) FG; (C) FI; (D) 2hGlu

• Table S5: Supplementary Table S5 - EC-DEPICT results

• Figure S1: Supplementary Figure S1 – GeneMANIA network analysis results

• Figure S2: Supplementary Figure S2 – EC-DEPICT results (UKBB permutations)

• Figure S3: Supplementary Figure S3 - EC-DEPICT results (Swedish permutations)

Data are available under the terms of the Creative Commons Attribution 4.0 International license (CC-BY 4.0)

Accession numbers

GWAS Catalog: meta-analysis summary statistics of 2-hour glucose in African American ancestry. MAGICExome_2hGlu_AFR.tsv.gz, study accession number GCST90256400. https://identifiers.org/gcst:GCST90256400

GWAS Catalog: meta-analysis summary statistics of 2-hour glucose in European ancestry. MAGICExome_2hGlu_EUR.tsv.gz, study accession number GCST90256401. https://identifiers.org/gcst:GCST90256401

GWAS Catalog: multi-ancestry meta-analysis summary statistics of 2 hour glucose. MAGICExome_2hGlu_ALL.tsv.gz, study accession number GCST90256402. https://identifiers.org/gcst:GCST90256402

GWAS Catalog: meta-analysis summary statistics of fasting glucose in African American ancestry. MAGICExome_FG_AFR.tsv.gz, study accession number GCST90256403. https://identifiers.org/gcst:GCST90256403

GWAS Catalog: meta-analysis summary statistics of fasting glucose in East Asian ancestry. MAGICExome_FG_EAS.tsv.gz, study accession number GCST90256404. https://identifiers.org/gcst:GCST90256404

GWAS Catalog: meta-analysis summary statistics of fasting glucose in European ancestry. MAGICExome_FG_EUR.tsv.gz, study accession number GCST90256405. https://identifiers.org/gcst:GCST90256405

GWAS Catalog: meta-analysis summary statistics of fasting glucose in Hispanic ancestry. MAGICExome_FG_HISP.tsv.gz, study accession number GCST90256406. https://identifiers.org/gcst:GCST90256406

GWAS Catalog: meta-analysis summary statistics of fasting glucose in South Asian ancestry. MAGICExome_FG_SAS.tsv.gz, study accession number GCST90256407. https://identifiers.org/gcst:GCST90256407

GWAS Catalog: multi-ancestry meta-analysis summary statistics of fasting glucose. MAGICExome_FG_ALL.tsv.gz, study accession number GCST90256408. https://identifiers.org/gcst:GCST90256408

GWAS Catalog: meta-analysis summary statistics of fasting insulin in African American ancestry. MAGICExome_FI_AFR.tsv.gz, study accession number GCST90256409. https://identifiers.org/gcst:GCST90256409

GWAS Catalog: meta-analysis summary statistics of fasting insulin in East Asian ancestry. MAGICExome_FI_EAS.tsv.gz, study accession number GCST90256410. https://identifiers.org/gcst:GCST90256410

GWAS Catalog: meta-analysis summary statistics of fasting insulin in European ancestry. MAGICExome_FI_EUR.tsv.gz, study accession number GCST90256411. https://identifiers.org/gcst:GCST90256411

GWAS Catalog: meta-analysis summary statistics of fasting insulin in Hispanic ancestry. MAGICExome_FI_HISP.tsv.gz, study accession number GCST90256412. https://identifiers.org/gcst:GCST90256412

GWAS Catalog: meta-analysis summary statistics of fasting insulin in South Asian ancestry. MAGICExome_FI_SAS.tsv.gz, study accession number GCST90256413. https://identifiers.org/gcst:GCST90256413

GWAS Catalog: multi-ancestry meta-analysis summary statistics of fasting insulin. MAGICExome_FI_ALL.tsv.gz, study accession number GCST90256414. https://identifiers.org/gcst:GCST90256414

GWAS Catalog: meta-analysis summary statistics of HbA1c in African American ancestry. MAGICExome_HbA1c_AFR.tsv.gz, study accession number GCST90256415. https://identifiers.org/gcst:GCST90256415

GWAS Catalog: meta-analysis summary statistics of HbA1c in East Asian ancestry. MAGICExome_HbA1c_EAS.tsv.gz, study accession number GCST90256416. https://identifiers.org/gcst:GCST90256416

GWAS Catalog: meta-analysis summary statistics of HbA1c in European ancestry. MAGICExome_HbA1c_EUR.tsv.gz, study accession number GCST90256417. https://identifiers.org/gcst:GCST90256417

GWAS Catalog: meta-analysis summary statistics of HbA1c in Hispanic ancestry. MAGICExome_HbA1c_HISP.tsv.gz, study accession number GCST90256418. https://identifiers.org/gcst:GCST90256418

GWAS Catalog: meta-analysis summary statistics of HbA1c in South Asian ancestry. MAGICExome_HbA1c_SAS.tsv.gz, study accession number GCST90256419. https://identifiers.org/gcst:GCST90256419

GWAS Catalog: multi-ancestry meta-analysis summary statistics of HbA1c. MAGICExome_HbA1c_ALL.tsv.gz, study accession number GCST90256420. https://identifiers.org/gcst:GCST90256420

These data are also available from https://magicinvestigators.org/downloads/