Random glucose GWAS in 493,036 individuals provides insights into diabetes pathophysiology, complications and treatment stratification

154 155 Homeostatic control of blood glucose requires different physiological responses in the fasting 156 and post-prandial states. We reasoned that glucose measurements under non -standardised 157 conditions (random glu cose; RG) may capture diverse glucoregulatory processes more 158 effectively than previous genome -wide association studies (GWAS) of fasting glycaemia or 159 after standardised glucose loads. Through GWAS meta -analysis of RG in 493,036 individuals 160 without diabetes of diverse ethnicities we identified 128 associated loci represented by 162 161 distinct signals, including 14 with sex -dimorphic effects, 9 discovered through trans -ethnic 162 analysis, and 70 novel signals for glycaemic traits. Novel RG loci were particularly enriched in 163 expression in the ileum and colon, indicating a prominent role for the gastrointestinal tract in 164 the control of blood glucose. Functional studies and molecular dynamics simulations of coding 165 variants of GLP1R, a well-established type 2 diabetes treatment target, provided a genetic 166 framework for optimal selection of GLP-1R agonist therapy. We also provided new evidence 167 from Mendelian randomisation that lung function is modulated by blood glucose and that 168 pulmonary dysfunction is a diabetes complic ation. Thus, our approach based on RG GWAS 169 provided wide-ranging insights into the biology of glucose regulation, diabetes complications 170 and the potential for treatment stratification. 171 172 Main text 173 174 Genetic factors are important determinants of glucose homeostasis and type 2 diabetes (T2D) 175 susceptibility. Heritability of both fasting glucose (FG) and T2D is high, at 35 -40%1 and 30-176 60%2, respectively . To date, more than 400 genetic loci have been described for T2D 3,4. 177 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 8 Genome-wide association studies (GWAS) for glycaemic traits in individuals without diabetes 178 have identified genetic predictors of blood glucose, insulin and other metabolic responses 179 during fasting or after oral or intravenous glucose challenge tests 5-8. However, physiological 180 glucose regulation involves responses to diverse nutritional and other stimuli that were, by 181 design, omitted from such studies. Blood glucose is frequently measured at different times 182 throughout the day in clinical practice and research studies (random glucose; RG ). Whilst RG 183 is inherently more variable than standardised measures, we reasoned that, across a very large 184 number of individuals, it may more comprehensively represent complex glucoregulatory 185 processes occurring in different organ systems. Therefore, to id entify and functionally 186 validate genetic effects influencing RG, explore its relationships with other traits and diseases, 187 and utilise these data to inform approaches to T2D treatment stratification, we performed 188 the first large-scale trans-ethnic GWAS meta-analysis for RG in individuals without diabetes. 189 190 RG GWAS significantly expands the catalogue of glycaemia-related genetic associations 191 192 We undertook RG GWAS in 493,036 individuals without diabetes of European (n=479,482) 193 and other ethnic (n=16,554) desc ent with adjustment for age, sex and time since last meal 194 (where available), along with exclusion of extreme hyperglycaemia (RG>20 mmol/L) and 195 individuals with diabetes (Supplementary Table 1) . The covariate selection was done upon 196 extensive phenotype mode lling (Methods, Supplementary Table 2, Supplementary Figure 197 1a). We identified 162 distinct signals (P<10-5) within 128 genetic loci reaching genome-wide 198 significance ( P<5x10-8) ( Figure 1a , Supplementary Table 3 ). Seventy RG signals had not 199 previously been reported for glycaemic traits (Table 1, Supplementary Table 3). In Europeans, 200 while the UK Biobank (UKBB) study provided 83.8% of the total study size, 128 detected 201 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 9 signals out of the 143 were directionally consistent in UKBB and other contributing studies 202 grouped together (Supplementary Table 3). Adjustment for last meal timing (Supplementary 203 Figure 1b) reduced effect sizes for several loci, including ITPR3, RREB1, RGS17, RFX6/VGLL2 204 and SYNGAP1, suggesting that these may be more related to the post-prandial state. RREB1, 205 RFX6 are transcription factors implicated in the development and function of pancreatic beta 206 cells9,10, and ITPR3 is a calcium channel involved in islet calcium dynamics in response to 207 glucose and G protein-coupled receptor (GPCR) activation11. Neither adjustment for body -208 mass index (BMI), nor a more stringent hyperglycaemia cut-off (RG>11.1 mmol/L or 209 HbA1c≥6.5%) ( Supplementary Figure 1 c-e) materially changed the magnitude and 210 significance of the RG effect estimates, although when all covariate models were individually 211 applied, nine additional signals at genome -wide significance were identified in UKBB (Table 212 1, Supplementary Table 4). 213 214 Several of the 162 signals identified in Europeans showed nominal significance ( P<0.05) in 215 specific UKBB ethnic groups, with GCK (rs2908286, r21000GenomesAllEthnicities=0.83 with rs2971670 216 lead in Europeans) reaching genome-wide significance in the African descent individuals alone 217 (Supplementary Table 3). Among the novel RG signals, USP47 was nominally significant in the 218 individuals of African, FAM46 and ACVR1C in the Indian and TRIM59/KPNA4 and ZC3H13 in 219 Chinese UKBB ancestry. Trans-ethnic meta-analyses combining Europeans and the other four 220 UKBB ancestral groups revealed seven novel RG signals, includi ng those at FOXN3, EPS8 and 221 ISG20L2 (Table 1). Overall, while being only 16,554 individuals larger in sample size than the 222 European meta-analysis, the trans-ethnic analysis expanded the novel locus discovery for RG 223 by one tenth (Supplementary Table 5). 224 225 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 10 Among established glycaemic trait signals, the well-known FG loci G6PC2 (P<5.86x10-754) and 226 GCK (P<6.93x10-301), with key roles in gluconeogenesis 12 and glucose sensing13, respectively, 227 showed the strongest associations with RG ( Supplementary Table 3). We also observed two 228 thirds of RG signals overlapping with T2D-risk loci ( Supplementary Figure 1e ), including 229 SLC30A8, DGKB, TCF7L2, GRB10 and THADA. The direction of effects at these loci between 230 RG, T2D and homeostasis model assessment of beta-cell function/insulin resistance (HOMA-231 B/-IR)6 (Supplementary Figure s 1e-f and 2, Supplementary Table 6 ) were consistent with 232 their epidemiological correlation. Notably, 14 established14,15, such as DGKB, THADA, RSPO3, 233 G6PC2, and novel, including TRIM59, POP7, SLC43A2, and SGIP1, loci showed sex-dimorphic 234 effects (Methods, Table 1, Figure 1a, Supplementary Table 3). Fine-mapping the associations 235 at RG loci through conditional analysis ( Table 1 ) we found three independent coding 236 nonsynonymous rare (minor allele frequency, MAF<1%) variants at G6PC2 with predicted 237 (rs2232326) and established (rs138726309, rs2232323)16 deleterious effects (Supplementary 238 Table 7). Within GCK, we observed five rare independent (r21000GenomesAllEthnicities<0.001) non-239 deleterious variants associated with RG at genome-wide significance, including a novel 3’UTR 240 rs2908276 for T2D, glycaemic traits or obesity (Supplementary Table 7). 241 242 Next, we sought to pinpoint the most plausibl e set of causal variants by calculating 99% 243 credible sets for each of RG loci. In the Europeans only analysis, 19 RG signals were explained 244 by one variant with posterior probability of ≥99% of being causal . For another 20 signals, a 245 lead variant had a posterior probability >80% (Figure 1b , Supplementary Table 8 ). The 246 credible sets were narrowed down in trans-ethnic RG meta-analysis (median credible set size 247 12.5 in the Europeans only , and 11.0 in the trans -ethnic analysis) (Supplementary Tables 9 248 and 10). This analysis helped to prioritise GLP1R for functional studies , in addition to the 249 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 11 already deeply characterised G6PC2 and CCND217, all three with lead SNPs of low frequency 250 (1%≤MAF99% of being causal. 251 252 The lead RG-associated SNPs at GLP1R, NEUROD1, and EDEM3 loci in our analysis were low-253 frequency coding variants (Supplementary Figure 3). NEUROD1 (Neuronal Differentiation 1) 254 and EDEM3 (ER Degradation Enhancing Al pha-Mannosidase Like Protein 3) are plausible 255 candidates for glucose homeostasis with the former reported for glucosuria 18 and the latter 256 linked to renal function 19,20. Additionally, lead variants at three previously reported for FG 257 (GCKR, TET2 and RREB1) and two novel RG (NMT1, WIPI1) loci were all common (MAF≥5%) 258 coding variants (Supplementary Figure 3). 259 260 Functional and structural characterisation of RG-associated GLP1R coding variants provides 261 a possible framework for T2D treatment stratification 262 263 The GLP1R gene, identified in our analysis and in previous T2D21 and glycaemic trait22 GWAS, 264 encodes a class B G protein-coupled receptor (glucagon-like peptide-1 receptor; GLP-1R) that 265 is an established target for glucose -lowering and weight loss in T2D using drugs such as 266 exenatide (exendin-4) and semaglutide 23. Within GLP1R, the lead missense variant at 267 rs10305492 (A316T) had a strong (0.058 mmol/l per allele) RG-lowering effect, second by size 268 only to G6PC2 locus variants. Previous attempts to functionally characterise A316T and 269 further GLP1R variants experimentally have been inconclusive 24, so we adopted a strategy 270 based on measuring ligand-induced coupling to mini-Gαs25, representing the most proximal 271 part of the Gα s-adenylate cyclase -cyclic adenosine monophosphate (cAMP) pathway that 272 links GLP-1R activation to insulin secretion. Mini-Gαs coupling efficiency was predictive of RG 273 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 12 effect for 16 GLP1R coding variants detected in the UKBB dataset with effect allele frequency 274 ranging from common (G168S, rs6923761 , P=4.40x10-5) to rare (R421W, rs146868158, 275 P=0.054) (Figure 2a, Supplementary Table 11), thereby linking differences in experimentally 276 measured GLP-1R function to blood glucose homeostasis. 277 278 To probe whether GLP1R coding variation could be therapeutically as well as physiologically 279 relevant, we also measured responses to several endogenous and pharma cological GLP-1R 280 agonists. Focussing on the two directly genotyped GLP1R missense variants in UKBB, we 281 observed that A316T (rs10305492 -A) showed increased responses , and R421W 282 (rs146868158-T) showed reduced responses , to all ligands except exendin -4 (both variants) 283 and semaglutide (A316T only), in line with their RG effects (Figure 2b). Agonist-induced GLP-284 1R endocytosis with R421W was normal despite its signalling deficit, suggestive of biased 285 agonism26. The imputed common G168S variant, with relatively small RG-lowering effect (=-286 0.0013 [SE=3.14x10-4]), also showed subtle increases in function. 287 288 To gain structural insights into GLP1R variant effects we performed molecular dynamics 289 simulations of the hum an GLP -1R bound to oxyntomodulin 27 (Extended Data Tables 1 -6). 290 A316T has a single amino acid substitution in the core of the receptor transmembrane domain 291 (Figure 2c) that leads to an alteration of the hydrogen bond network in close proximity (Video 292 S1). In A316T, residue T316 5.46 replaced Y242 3.45 in a persistent hydrogen bond with the 293 backbone of P312 5.42 one turn of the helix above T316 5.46 (Figures 2d -e, Video S1 ). This 294 triggers a local structural rearrangement that could transmit to the intracellular G protein 295 binding site through transmembrane helix 3 (TM3) and TM5. A structural water molecule was 296 found close to position 5.46 in both A316T and WT ( water cluster 5, Figure 2f). The same 297 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 13 water bridged the backbone of Y241 3.44 and A3165.46 in WT, or the backbone of Y2413.44 and 298 the side chain of T3165.46 in A316T. Given the importance of conserved water networks in the 299 process of activation of class A GPCRs 28,29, the presence of a stable hydrate d spot close to 300 position 5.4630 corroborates this site as important for tuning the intracellular conformational 301 landscape of GLP-1R. Also, a stabilising role for the water molecules at the binding site of the 302 G protein (water cluster apha5, Figure 2f) cannot be ruled out. Note that our results differ 303 from a previous analysis of A316T dynamics 22, which used an early model that does not fully 304 capture the full structural features of the current active GLP-1R models. 305 306 In analogy with A316T, molecular simulations with the G168S variant indicate the formation 307 of a sta ble new hydrogen bond between the side chain of residue S168 1.63 and A164 1.59, 308 located one turn above on the same helix (Video S2, Figure 2g). This moved the C -terminal 309 end of TM1 closer to TM2 and reduced the overall flexibility of ICL1 ( Figure 2h), which could 310 potentially alter the role of ICL1 in G protein activation. In contrast to A316T and G168S, the 311 site of mutation R421W is consistent with persistent interactions with the G protein . 312 Simulations predicted a propensity of R421W to interact with a different region of the G 313 protein -subunit to that engaged by WT (Figure 2i). 314 315 For a broader view of the impact of GLP1R coding variation, we screened an additional 178 316 missense variants id entified from exome sequencing 31 for exendin -4-induced mini -Gs 317 coupling and endocytosis ( Figures 2j-k, Supplementary Table 12 ). 110 variants showed a 318 reduced response in either or both pathways (“LoF1”), and 67 displayed a specific response 319 deficit that was not fully explained by differences in GLP-1R surface expression (“LoF2”), with 320 many of these defects being larger than in the analysis in Figure 2a. 321 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 14 322 Overall, these data suggest GLP1R variation influences blood glucose levels in health and is 323 likely to be a direct modifier of responses to drug treatment 32. As some patients fail to 324 respond adequately to GLP-1R agonist treatment, and others are particularly sensitive to side 325 effects33, this approach may feed into optimised treatment selection in T2D. 326 327 Functional annotation of RG associations and intestinal health 328 329 Previous T2D and glycaemic trait GWAS have primarily implicated pancreatic, adipose and 330 liver tissues3. To leverage our RG GWA results to identify additional cell and tissue types with 331 aetiological roles in glucose metabolism, we performed a range of complementary functional 332 annotation analyses in relation to RG GWAS. DEPICT34, which predicts enriched tissu e types 333 from prioritised gene sets (Methods), highlighted intestinal tissues including ileum and colon, 334 as well as pancreas, adrenal glands, adrenal cortex and cartilage (False Discovery Rate<0.20) 335 (Figures 3a-b, Supplementary Tables 13a -c). Similarly, CELLECT35, which facilitates cell -type 336 prioritisation based on single cell RNAseq datasets (Methods), identified large intestinal 337 tissue as the second ranked only to pancreatic cell types (Figure 3c, Supplementary Table 14); 338 interestingly, RG variants were related particularly to enriched expression in pancreatic 339 polypeptide (PP) cells, exceeding even the more conventionally implicated insulin -secreting 340 beta cells. Supporting evidence was obtained from transcriptome -wide association study 341 (TWAS) analysis (Methods), where w e identified a total of 216 (119 unique) significant 342 genetically driven associations across the ten tested tissues; 52 (26 unique) of highlighted 343 genes are located at genome-wide significant RG loci ( Supplementary Tables 15a) . TWAS 344 signals in skeletal muscle showed the largest overlap with RG signals , such as GPSM136 and 345 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 15 WARS; with combined results from ileum and colon also highly enriched , including the novel 346 NMT1 and the established FADS1/3 and MADD genes (Figure 1a, Supplementary Tables 15a-347 b). Moreover, e pigenetic annotations using the GARFIELD tool highlighted significant 348 (P<2.5×10-5, Methods) enrichment of RG-associated variants in foetal large intestine, as well 349 as blood, liver and other tissues (Supplementary Figure 4, Supplementary Table 16 ). Adult 350 intestinal tissues are not available in GARFIELD except for colon. Prompted by multiple 351 analyses highlighting a potential role for the digestive tract in glucose regulation, we assessed 352 the overlap between our signals and those from the latest microbiome GWAS 37 (Methods) 353 and identified three genera sharing signals with RG at two loci: Collinsella and 354 LachnospiraceaeFCS020 at ABO-FUT2 and Slackia at G6PC2 (Figure 1a, Supplementary Table 355 17). The ABO-FUT2 locus effects on RG could be mediated by abundance of bacteria 356 producing glucose from lactose and galactose38. 357 358 eQTL colocalization analyses, using eQTLgen blood expression data from 31,684 individuals39 359 and the COLOC2 approach (Methods), identified 14 loci with strong links (posterior 360 probability >50%) to gene expression data, including SMC4, TRIM59, EIF5A2, TET2, COG5, 361 CHMP5, NFX1, FNBP4, MADD, RAPSN, WARS1, HBM, NUFIP2, and PPDPF (Supplementary 362 Table 18). This further supported elucidation of biological candidates at novel and established 363 glycaemic loci. 364 365 Finally, we observed associations at two RG loci ( GCKR, HNF1A) with nine total plasma N -366 glycome traits 40 at a Bonferroni corrected threshold ( Methods, Figure 1a, Supplementary 367 Table 19). These traits represent highly branched galactosylated sialylated glycans (attached 368 to alpha1 -acid protein - an acute -phase protein 41), known to lead to chronic low -grade 369 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 16 inflammation42,43 and an increased risk of T2D 44-46 that might be explained by the role of N -370 glycan branching o f the glucagon receptor in the glucose homeostasis 47. In additio n, ten 371 glycans showed association with five RG loci ( GCKR, HNF1A, BAG1, PLUT, ACVR1C) loci at a 372 suggestive level of significance ( Figure 1a ). Among them, three are attached to 373 immunoglobulin G molecules41 and their increased relative abundances are associated with a 374 lower risk of T2D48 and diminished inflammation status49. 375 376 Analysis of genetic relationships between RG and other metabolic or non-metabolic traits 377 378 To quantify the shared genetic contribution between RG and other phenotypes, we estimated 379 their genetic correlations using linkage -disequilibrium score regression analyses. We 380 detected positive genetic correlations between RG , squamous cell lung cancer ( rg=0.28, 381 P=0.0015), and lung cancer (rg=0.12, P=0.037, Figure 4, Supplementary Table 20); as well as 382 inverse genetic correlations with lung function related traits, such as forced vital capacity 383 (FVC, rg=-0.090, P=0.0059) and forced expiratory volume in 1 second (FEV1 , rg=-0.054, 384 P=0.017) ( Figures 3a and 4, Supplementary Table 20 ). To investigate this further, we 385 conducted a bi-directional Mendelian Randomisation (MR) analysis, which suggested a causal 386 effect of RG and T2D on lung function, including FEV1 (βMR-RG=-0.60, P=0.0015; βMR-T2D=-0.049, 387 P=1.27x10-13) and FVC (β MR-RG=-0.61, P=3.5x10-4; βMR-T2D=-0.062, P=1.42x10-21), but not vice 388 versa (Methods, Supplementary Table 21). Previous observational studies have highlighted 389 worsening lung function, as defined by FVC, in T2D patients50,51. More recently, it was shown 390 that patients with diabetes are at an increased risk of death from the viral infection COVID -391 1952, with pulmonary dysfunction contributing to mortality53. Our data therefore support the 392 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 17 causal effect of glycaemic dysregulation on a decline in lung function as a novel complication 393 of diabetes. 394 395 Genome-wide genetic correlation analyses also showed strong positive genetic correlation of 396 RG with FG (rg=0.88, P=6.93×10-61, Figure 4, Supplementary Table 20). We meta-analysed RG 397 studies other than UKBB with FG GWAS summary statistics 54, observing 77 signals reaching 398 nominal significance that were directionally consistent in both UKBB and RG+FG 399 (Supplementary Table 3), providing an additional support to our RG findings. Given the large 400 genetic overlap between RG, other glycaemic traits and T2D, we evaluated the ability of a 401 trait-specific polygenic risk score (PRS) to predict RG, T2D and glycated haemoglobin (HbA1c) 402 levels using UKBB effect estimates and the Vanderbilt cohort (Methods). The RG PRS 403 explained 0.58% of the variance in RG levels when individuals with T2D were included, 404 (Supplementary Table 22 ) and 0.71% of the variance after excluding those who developed 405 T2D within one year of their last RG measurement. The RG PRS performance was comparable 406 to that of the FG loci PRS (0.38% vs. 0.42% for T2D; 0.40% vs. 0.44% for HbA1c) indicating 407 wide similarities with the latter. 408 409 We previously highlighted diverse effects of FG and T2D loci on pathophysiological processes 410 related to T2D development by grouping associated loci in relation to their effects on multiple 411 phenotypes6. Cluster analysis of the RG signals with 45 related phenotypes identified three 412 separate clusters that give insights into the aetiology of glucose regulation and associated 413 disease states (Methods, Figure 1a, Supplementary Table 23, Supplementary Figures 5a-d). 414 Cluster 1 (“metabolic syndrome” cluster) clearly separated 33 loci with effects on higher 415 waist-to-hip ratio, blood pressure, plasma triglycerides, insulin resistance (HOMA -IR) and 416 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 18 coronary artery disease risk, as well as lower testosterone and sex hormone binding globulin 417 levels in men. Cluster 3 was characterised in particular by insulin secretory defects6. Cluster 2 418 was less clearly defined by a primary effect on insulin release versus insulin a ction3, but 419 interestingly included a sub-cluster of 21 loci which exert protective effects on inflammatory 420 bowel disease. Moreover, cluster 2 was notable for generally reduced impact on T2D risk in 421 comparison to clusters 1 and 3, underscoring the partial overlap between genetic 422 determinants of glycaemia and T2D that is known to exist55. 423 424

425 426 Taking advantage of data from 493,036 individuals, we have expanded by 58 the number of 427 loci associated with glycaemic traits. By using RG, our analysis integrates genetic contributions 428 to a wider range of physiological stages than possible with FG or other standardised 429 measures. Moreover, the greater statistical power obtained from large trans -ethnic meta-430 analysis improves confidence in i dentification of potentially causal variants, thereby helping 431 to prioritise loci for more detailed functional analyses in the future. Our observation of ligand-432 specific responses to the A316T, G168S and R421W GLP1R variants provides a mechanism 433 that can explain why some individuals respond better or worse to particular GLP-1R-targeting 434 drugs. We note that other class B GPCRs identified in our current analysis and other glycaemic 435 or T2D GWAS include GIPR, GLP2R3 and SCTR21, all of which are investigational targets for T2D 436 treatment. Our functional annotation analyses point to underexplored tissue mediators of 437 glycaemic regulation, with several sources of evidence highlighting a likely role of the 438 intestine. This observation is compatible with the well -described and profound effects of 439 gastric bypass surgery on T2D resolution56, as well as links between the intestinal microbiome 440 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 19 and responses to several diabetes drugs 57. Finally, through Mendelian randomisation we 441 were able to identify a causal effect of glucose levels and T2D on lung function, demonstrating 442 the utility of this approach for the corroboration of findings from observational studies an d 443 elevating lung dysfunction as a new complication of diabetes. 444 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 20

445 446 Phenotype definition and model selection for RG GWAS 447 We used RG (mmol/l) measured in plasma or in whole blood (corrected to plasma level using 448 the correction factor of 1.13). Individuals were excluded from the analysis , if they had a 449 diagnosis of T2D or were on diabetes treatment (oral or insulin). Individual studies applied 450 further sample exclusions, including pregnancy, fasting plasma glu cose equal to or greater 451 than 7 mmol/l in a separate visit, when available, and having type 1 diabetes . Detailed 452 descriptions of study -specific RG measurements are given in Supplementary Table 1 . All 453 studies were approved by local ethics committees and all participants gave informed consent. 454 We examined the distributions of untransformed and natural logarithmic transfor med RG in 455 the first set of six available cohorts. We observed that RG was approximately normally 456 distributed after natural log transformation. We then determined the variables that could 457 have a significant effect on RG by fitting several regression models using naturally log -458 transformed RG as the outcome with age, sex, BMI and time since last meal as predictors. 459 Modelling of RG revealed significant effects (P<0.05) of age, sex, BMI and time since last meal 460 (accounted for as T, T2 and T3) in these cohorts (Supplementary Table 2 ). Compared to RG 461 models without T, inclusion of T, T2 and T3 increased the proportion of variance explained in 462 the range of 1-6%. Thus, inclusion of this covariate is potentially equivalent to 1 -6% increase 463 in study sample size. For the GWAS , we included individuals based on two RG cut -offs: <20 464 mmol/l (20) to account for the effect of extreme RG values and <11.1 mmol/l (11), which is 465 an established threshold for T2D diagnosis. We then evaluated six different models in GWAS 466 according to covariates included and cut -offs used : 1) age (A) and sex (S), RG<20 mmol/L 467 (AS20), 2) age, sex and BMI (B), RG<20 mmol/L (ASB20), 3) age and sex, RG<11.1 mmol/L 468 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 21 (AS11), 4) age, sex and BMI, RG<11.1 mmol/L (ASB11), 5) age, sex, T, T2 and T3, RG<20 mmol/L 469 (AST20) and 6) age, sex, T, T2 and T3 and BMI, RG<20 mmol/L (ASTB20). Apart from above, 470 additional adjustments for study site and geographical covariates were also applied. 471 472 Genotyping and quality control 473 Commercial genome-wide arrays and the Metabochip58 were used by individual studies for 474 genotyping. Studies with genome -wide arrays undertook imputation of miss ing genotypes 475 using at least the HapMap II CEU reference panel via MACH 59, IMPUTE60 or MINIMAC 61 476 software ( Supplementary Table 1 ). For each study, samples reflecting duplicates, low call 477 rate, gende r mismatch, or p opulation outliers were removed . Low -quality SNPs were 478 excluded by the following criteria: call rate <0.95, minor allele frequency (MAF) <0.01, minor 479 allele count <10, Hardy-Weinberg P-value <10−4. GWAS were performed with PLINK, SNPTEST, 480 EMMAX, R package LMEKIN, Merlin, STATA, and ProbABEL (Supplementary Table 1). 481 482 GWAS in the UKBB 483 For the GWAS of the UKBB data we excluded non-white non-European individuals and those 484 with discrepancies in genotyped and reported sex. For the RG definition , we used the same 485 criteria as in the other studies described above. To control for population structure, we 486 adjusted the analyses for six first principal components. The GWAS was performed using the 487 BOLT-LMM v2.3 software 62,63 restricting the analyses to variants with MAF>1% and 488 imputation quality>0.4. 489 490 RG meta-analyses 491 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 22 The GWAS meta-analysis of RG consisted of four components: (i) 37,239 individuals from 10 492 European GWAS imputed up to the HapMap 2 reference panel, (ii ) 3,156 individuals from 493 three European GWAS with Metabochip coverage, (iii) 21,083 individuals from two European 494 GWAS imputed up to 1000 genomes reference panel and iv) 401,810 individuals of white 495 European o rigin from the UKBB and (iv) 16,983 individuals from the Va nderbilt cohort 496 imputed to the HRC panel. We imputed the GWAS meta -analysis summary statistics of each 497 component to all -ancestries 1000 Genomes reference panel64 using summary statistics 498 imputation method implemented in the SS -Imp v0.5.5 software 65. SNPs with imputation 499 quality score < 0.7 were excluded. We then conducted inverse variance meta -analyses to 500 combine the association summary statistics from all components using METAL (version from 501 2011-03-25)66. We focused our meta-analyses on models AS20 (17 cohorts, N max=481,150) 502 and AST20 (when time from last meal was available in the cohort) (12 cohorts, Nmax=438,678). 503 For FHS cohort, where no information was available for individuals with RG>11.1 ( an 504 established threshold for 2hGlu concentration, which is a criterion for T2D diagnosis), AS11 505 model results were used. In order to maximise the association power while taking into 506 account T , we also performed meta -analysis using AST20 (when time from last meal was 507 available in the cohort) combined with AS20 (otherwise) and we termed this analysis as 508 AS20+AST20 in the following text (17 cohorts, Nmax=480,250). 509 A signal was considered to be associated with RG if it had reached genome-wide significance 510 (P<5x10-8) in the meta -analysis of UKBB and other cohorts in either of our two models of 511 interest (AS20) or (AST20) or in their combination (AS20+AST20). We report the P-value from 512 the combined model, unless otherwise stated. Full results from all models are provided in the 513 Supplementary Table 3 . All the follow -up analyses were conducted using the combined 514 AS20+AST20 model. We checked for nominal significance (P<0.05) and directional consistency 515 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 23 of the effect sizes for the selected leads in the combined model in UKBB results vs other 516 cohort results. We further extended the check between UKBB results and meta -analysis of 517 other cohorts including FG GWAS meta-analysis54 excluding overlapping cohorts. This meta-518 analysis conducted in METAL was sample size and P-value based due to the measures being 519 at different scale (natural logarithm transformed RG and untransformed FG). 520 521 Trans-ethnic analyses and meta-analysis 522 We performed GWAS in those non-European populations within UKBB that had a sample size 523 of at least 1,500 individuals. These were Black (N=7,644), Indian (N=5,660), Pakistani 524 (N=1,747) and Chinese (N=1,503). We further meta -analysed our European cohorts with the 525 trans-ethnic UKBB cohorts. The analyses were performed with BOLT-LMM and METAL. 526 527 Sex-dimorphic analysis 528 To evaluate sex -dimorphism in our results, we meta -analysed the UKBB and the V anderbilt 529 cohort with the GMAMA software 67, which provides a 2 degrees of freedom (df) test of 530 association assuming different effect sizes between the sexes. We considered a signal to show 531 evidence of sex-dimorphism if the 2 df test P-value was <5x10-8 and if the sex heterogeneity 532 P-value (1 df) was <0.05. 533 534 Clumping and GCTA analysis 535 We performe d a standard cl umping analysis [ PLINK 1.9 ( v1.90b6.4)68 criteria: P≤510-8, 536 r2=0.01, window -size=1Mb, 1000 Genomes Phase 3 data as linkage disequilibrium ( LD) 537

panel] to select a list of near -independent signals. We then performed a stepwise 538 model selection analysis (GCTA conditional analysis) to replicate the analysis using GCTA 539 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 24 v1.93.069 with the following parameters: P≤510-8 and window -size=1Mb. We further 540 checked for additional distinct signals by using a region -wide threshold of P≤110-5 for 541 statistical significance. 542 543 GLP-1R pharmacological and structural analysis 544 Reagents 545 Custom peptides were purchased from Wuxi Apptec and were at least 9 5% pure. SNAP -546 Surface probes were purchased from New England Biolabs. BG -S-S-64970 was provided by 547 New England Biolabs on a collaborative basis. Furimazine was obtained from Promega. 548 549 Plasmids and cell line generation 550 Wild-type and variant GLP-1R expression plasmid s, termed pcDNA5-SNAPf-GLP-1R-SmBiT, 551 were generated by Genewiz, as previously described71, to the following design: a fast-labelling 552 SNAPf tag and upstream signal peptide based on that of the 5 -HT3A receptor 553 (MDSYLLMWGLLTFIMVPGCQA), plus C -terminal SmBiT tag , were appended to the c odon-554 optimised wild-type or variant human GLP-1R sequence (without the endogenous N-terminal 555 signal peptide, which would lead to cleavage of the N-terminal SNAP-tag; accordingly, known 556 missense variants in the signal peptide region were not included), and inserted into the 557 pcDNA5/FRT/TO expression vector . These constructs allow bio -orthogonal labelling of 558 expressed GLP -1R using SNAP -labelling probes and monitoring of cytosolic protein 559 interactions made to GLP -1R. Constructs were used either for transient transfection or to 560 generate stable cell lines. To obtain cell populations with inducible expression of SNAP-GLP-561 1R-SmBiT from a single genomic locus, Flp-In™ T-REx™ 293 cells72 (Thermo Fisher) were co-562 transfected with pOG44 (Thermo Fisher) and wild-type or variant pcDNA5-SNAPf-GLP-1R-563 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 25 SmBiT in a 9:1 ratio, followed by selection with 100 µg/ml hygromycin. The resulting cell lines 564 were maintained in DMEM supplemented with 10% foetal bovine serum (FBS) and 1% 565 penicillin/streptomycin. 566 567 Mini-Gs recruitment assay 568 Assays were performed as previously described 71. Where stable cell lines were used (i.e. 569 Figures 2a and 2b), wild-type or variant T -REx-SNAP-GLP-1R-SmBiT cells were seeded in 12 -570 well plates and transfected with 1 µg/well LgBiT -mini-Gs25 (a gift from Prof Nevin Lambert, 571 Medical College of Georgia). The following day GLP-1R expression was induced by addition of 572 tetracycline (0.2 µg/ml) to the culture medium for 24 hours. For transient transfection assays 573 (i.e. Figure 2j), HEK293T cells in poly-D-lysine-coated white 96-well plates were transfected 574 using Lipofectamine 2000 with 0.05 µg/well wild -type or variant SNAP -GLP-1R-SmBiT plus 575 0.05 µg/well LgBiT -mini-Gs and the assay performed 24 hours later. Cells were then 576 resuspended in Hank’s balanced salt solution (HBSS) + furimazine (Promega) diluted 1:50 and 577 seeded in 96 -well half area white plates , or the same reagent added to adherent cells for 578 transient transfection assays . Baseline luminescence was measured over 5 min using a 579 Flexstation 3 plate reader at 37°C before addition of ligand or vehicle. Agonists were applied 580 at a series of concentrations spanning the response range. After agonist addition, luminescent 581 signal was serially recorded over 30 min, and ligand-induced effects were quantified by 582 subtracting individual well baseline. Signals were cor rected for differences in cell number as 583 determined by BCA assay. 584 585 High content imaging-based GLP-1R internalisation assay 586 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 26 The assay was performed as previously described 71. Where stable cell lines were used (i.e. 587 Figures 2a and 2b), wild -type or variant T -REx-SNAP-GLP-1R-SmBiT cells were seeded 588 (10,000/well) in poly-D-lysine-coated black, clear-bottom 96-well plates, in complete medium 589 supplemented with tetracycline (0.2 µg/ml) for 24 hours before the assay . Medium was 590 removed and cells labelled with 0.5 µM BG -S-S-649 (a gift from New England Biolabs) in 591 complete medium for 20 min at 37°C. Agonists were then applied in serum -free medium at 592 the indicated dose for a 30-min stimulation period to induce GLP-1R internalisation. A series 593 of concentrations spanning the response range were used. Cells were then washed with HBSS, 594 followed by a 5 -min treatment ± 100 mM sodium 2 -mercaptoethanesulfonate (Mesna) in 595 alkaline TNE buffer (pH 8.6) to cleave residual surface B G-S-S-649 w ithout affecting that 596 internalised whilst bound to SNAP -GLP-1R. After re -washing, the plate was imaged using a 597 0.75 numerical aperture 20 x phase contrast objective , with 9 fields-of-view (FOVs) per well 598 acquired for both transmitted phase contra st and epifluorescence. Flat-field correction of 599 epifluorescence images was performed using BaSiC 73 and cell segmentation was performed 600 using PHANTAST74 for the phase contrast image. To determine specific GLP-1R labelling, cell-601 free background per image was determined from the segmented epifluorescence image and 602 subtracted from the mean fluorescence intensity from the cell -containing regions . Ligand 603 induced effects were determined by subtracting the signal from vehicle-treated cells exposed 604 to Mesna . Responses were normalised to signal from labelled, untreated cells (i.e. total 605 surface labelling) within the same assay. GLP-1R surface expression levels were also obtained 606 from these assays from wells not treated w ith GLP-1RA or Mesna. For transient transfection 607 assays (i.e. Figure 2j), the assay was performed similarly but with the following changes: 1) 608 HEK293T cells in poly-D-lysine-coated black clear-bottom 96-well plates were transfected 609 using Lipofectamine 2000 with 0.1 µg/well wild -type or variant SNAP-GLP-1R-SmBiT and the 610 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 27 assay performed 24 hours later; 2) the plate was imaged as above both prior to and after 611 ligand treatment (+subsequent Mesna cleavage); 3) surface labelling quantification was 612 obtained from the pre-treatment read, and total internalised receptor was obtained from the 613 post-treatment read. 614 615 Analysis of pharmacological data 616 Technical replicates within the same assay were averaged to give one biological replicate. For 617 concentration-response assays (Figures 2a and 2b), ligand-induced responses were analysed 618 by 3-parameter fitting in Prism 8.0 (Graph Pad Software). As a composite measure of 619 agonism75, log 10-transformed E max/EC50 values were obtained for each ligand/variant 620 response. The wild -type response was subt racted from the variant response to give 621 ∆log(max/EC50), a measure of gain - or loss-of-function for the variant relative to wild -type. 622 Log10-transformed surface expression levels were obtained for each variant relative to wild -623 type; these were then used to correct mini-Gs ∆log(max/EC50) values for differences in variant 624 GLP-1R surface expression levels, by subtraction with error propagation. GLP -1R 625 internalisation responses were already normalised to surface expression within each assay. 626 Statistical signifi cance between wild -type and variant responses was inferred if the 95% 627 confidence intervals for ∆log(max/EC 50) did not cross zero 75. Changes to the profile of 628 receptor response between mini -Gs recruitment and GLP-1R internalisation were inferred if 629 p<0.05 with unpaired t -test analysis, with Holm -Sidak correction for multiple comparisons. 630 For transient transfection assays ( Figure 2j ), responses were normalised to wild -type 631 response and log10 transformed to give Log ∆ response. Additionally, the impact of differences 632 in surface expression on functional responses was determined by subtracting log -633 transformed normalised expression level from log-transformed normalised response. 634 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 28 635 Variance explained in RG effects by mini-Gs recruitment at coding GLP1R variants 636 RG (AST20) effects estimated in the UKBB study at 18 independent (r2<0.02) coding GLP1R 637 variants (Supplementary Table 10) were regressed on mini-Gs coupling in response to GLP-1 638 stimulation (corrected for surface expression) giving more weight to variants with higher 639 minor allele frequency. Adjusted R2 is reported as variance explained in RG effects by mini-640 Gs coupling. 641 642 Computational methods including molecular dynamics simulations 643 The active state structure of GLP-1R in complex with OXM27 and Gs protein was modelled as 644 previously described30 and used to simulate the WT GLP -1R and G168S, A316T and R421W. 645 The systems were prepared for molecular dynamics (MD) simulations an d equ ilibrated as 646 reported in30. AceMD376 was employed for production runs (four MD replicas of 500 ns each). 647 AquaMMapS analysis77 was performed as previously described30. 648 649 Credible set analysis 650 After selecting the signals with each region based on different M-A results from AS20, AST20 651 and AS20+AST20 models, we further performed a credible set analysis to obtain a list of 652 potential causal variants for each of the 143 selected signals. Based on the method adopted 653 from78 under the assumption that there is one causal variant within each region, we created 654 99% credible sets. We also calculated credible sets for the trans -ethnic meta-analysis and 655 compared the results between the European only and trans-ethnic meta-analyses. 656 657 DEPICT analysis 658 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 29 DEPICT uses GWAS summary statistics and computes a prioritization of genes in associated 659 loci, which are used to prioritise tissues via enrichment analysis. DEPICT v1 (rel 194) was used 660 with default settings and RG GWAS summary statistics as input against a genetic background 661 of SNPsnap data 79 derived from the 1000 Genomes Project Phase 3 80 in order to prioritis e 662 genes. Tissue and c ell types enriched for prioritis ed genes were computed on normalis ed 663 expression data comprised of 209 tissues and cell types from 37,427 Affymetrix U133 Plus 2.0 664 Array, as previously described 34. We used 500 permutations for bias adjustment and 50 665 replications for false discovery rate estimation in our analysis in order to calculate empirical P-666 values and false discovery rate cutoffs for prioritised tissues. 667 668 CELLECT analysis 669 CELL type Expression-specific integration for Complex Traits (CELLECT)35 v1.0.0 and Cell type 670 EXpression-specificity (CELLEX) 35 v1.0.0 are two toolkits for genetic identification of likely 671 etiologic cell types using GWAS summary statistics and single -cell RNA-sequencing (scRNA-672 seq) data. Tabula muris gene expression data81, a scRNA-seq dataset derived from 20 organs 673 from adult male and female mice, was pre -processed as described previously 82. Briefly, 674 expression values were normalised by using a scaling factor of 10k transcripts. The normalised 675 values were transformed by taking log(x+1), followed by filtering out infrequently expressed 676 genes, and keeping only those mouse transcripts with 1 -1 mapping to human genes in 677 Ensembl v.91. This data was supplied to CELLEX to compute a cumulative expression 678 specificity metric (ESμ) of every gene for each Tabula muris cell type by combining four 679 different expression specificity measures82. ESμ values were converted to stratified LD-score 680 regression (S-LDSC) annotations using the 1000 Genomes Project SNPs and mapping each SNP 681 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 30 to the strongest ESμ value within 100kb. Cell types were prioritised by S-LDSC on the basis of 682 ESμ-derived annotations and GWAS summary statistics from the current RG meta-analysis. 683 684 Genetically regulated gene expression analysis 685 We used MetaXcan (S-PrediXcan) v0.6.10 83 to identify genes whose genetically predicted 686 gene expression levels are associated with RG in a number of tissues. The tested tissues were 687 chosen based on their involvement i n glucose metabolism. Those were adipose v isceral 688 omentum, adipose subcutaneous, skeletal muscle, liver, pancreas and whole blood. 689 Additionally, we tested ileum, transverse colon, sigmoid colon and adrenal g land, because 690 they were highlighted by DEPICT analysis. The models for the tissues of interest were trained 691 with GTEx Version 7 transcriptome data from European individuals 84. The tissue 692 transcriptome models and 1000 Genomes 85 based covariance matrices of the SNPs used 693 within each model were downloaded from PredictDB Data Repository. The association 694 statistics between predicted gene expression and RG were estimated from the effect s and 695 their standard errors coming from the AS20+AST20 model. Only statistically significant 696 associations after Bonferroni correction for the number of genes tested across all tissues (P  697 8.996x10-7) were included into the table. Genes, where less than 80% of the SNPs used in the 698 model were found in the GWAS summary statistics , were excluded due to low reliability of 699 association result. 700 701 GARFIELD analysis 702 We applied the GARFIELD tool v2 86 on the RG AS20 +AST20 meta-analysis results to assess 703 enrichment of the RG -associated variants within functional and regulatory features. 704 GARFIELD integrates various types of data from a number of publicly available cell lines. Those 705 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 31 include gen etic annotations, chromatin states, DNaseI hypersensitive sites, t ranscription 706 factor binding sites, FAIRE -seq elements and histone modifications. We considered 707 enrichment to be statistically significant if the RG GWAS P-value reached P=1×10-8 and the 708 enrichment analysis P-value was <2.5×10-5 (Bonferroni corrected for 2040 annotations). 709 710 Genetic association with gut microbiome 711 We assessed the genetic overlap between RG GWAS results and those for gut microbiome. 712 GWAS of microbiome profiles were publicly available and downloaded from the 713 https://mibiogen.gcc.rug.nl/ [mibiogen.gcc.rug.nl] . For each of the 211 taxa, the 714 corresponding P-values for the 143 RG GWAS SNPs and their proxies were extracted. 715 716 Genetic association with GLP-1 and GIP 717 We assessed the genetic overlap between RG GWAS results and those for glucagon-like 718 peptide-1 (GLP-1) and gastric inhibitory polypeptide (GIP) measured at 0 and 120 minutes. 719 We extracted the results for the 143 RG signals from the GWAS summary statistics for GLP -1 720 and GIP87. 721 722 eQTL co-localization analysis 723 We further performed co-localization analysis using whole blood gene expression-QTL (eQTL) 724 data provided by eQTLGen39 and AS20+AST20 meta-analysis results. Only cis-eQTL data from 725 eQTLGen was incorporated to reduce the computational burden. The COLOC2 Bayesian -726 based method 88 was used to interrogate the potential co -localization between RG GWAS 727 signals and the genetic control of g ene expression. We first extracted the RG GWAS test 728 statistics of all the SNPs within +/ -1Mb region around the 143 RG signals. Then, for each RG 729 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 32 signal, we matched the eQTLGen results with the RG results and performed COLOC2 analysis 730 evaluating the posteri or probability (PP) of five hypotheses for each region: H 0, no 731 association; H1, GWAS association only; H 2, eQTL association only; H 3, both GWAS and eQTL 732 association, but not co-localised; and H4, both GWAS and eQTL association and co-localised. 733 Only GWAS signals with at least one nearby gene/probe reaching PP (H4) ≥ 0.5 were reported. 734 735 Genetic association with human blood plasma N-glycosylation 736 We assessed the genetic association between 143 RG signals and 113 human blood plasma 737 N-glycome traits using prev iously published genome -wide summary association statistics 89. 738 The description of the analysed traits and details of the association ana lysis can be found 739 elsewhere40. We considered association s to be significant when P-740 value<0.05/113/143=3.09e-6 ( after Bonferroni correction ). Association was considered as 741 suggestive when P-value<10-4. 742 743 Genetic correlation analysis 744 We i nvestigate the shared genetic component between RG and other traits , including 745 glycaemic ones, by performing genetic correlation analysis using the bivariate LD score 746 regression method (LDSC v1.0.0)90. To reduce multiple testing burden, only the GWAS results 747 of the UKBB model AS20 were used. We used GWAS summary statistics available in LDhub 91 748 and the Meta-Analysis of Glucose and Insulin -related Traits Consortium (MAGIC) website 749 (https://www.magicinvestigators.org) for several traits including FG/FI 54, HOMA -B/HOMA-750 IR92. In total, 228 different traits were included in the geneti c correlation analysis with RG. 751 We considered P≤0.05 as the nominal significant level. 752 753 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 33 MR analysis 754 We applied a bidirectional two-sample MR strategy to investigate causality between RG and 755 lung function, as well as T2D and lung function using independent genetic variants as 756 instruments. MR can provide estimates of the effect of modifiable exposures on an outcome 757 (e.g. disease) unaffected by classical confounding or reverse causation, whenever randomised 758 clinical trials are not feasible. We looked for evidence for the presence of a causal e ffect of 759 RG and T2D on two lung function phenotypes; FVC and FEV1 in a two -sample MR setting. 760 Genome-wide summary statistics for the lung function phenotypes were available93, involving 761 cohorts from the SpiroMeta consortium and the UKBB study. T2D susceptibility variants and 762 their effects were obtained from the largest-to-date T2D GWAS4. 763 To avoid conf ounding due to sample overlap, lung function summary statistics used as 764 outcome data were those estimated in the SpiroMeta consortium alone. Similarly, when 765 testing the effect of lung function on RG, RG genetic effects used as outcome data were 766 estimated in all cohorts except UK Biobank. There was no sample overlap between the lung 767 function- and the T2D GWAS, thus allowing the use of T2D effects estimated in all contributing 768 European studies. Genome-wide T2D summary statistics were availabl e from a previous 769 study3 to test for the causal effect of lung function on T2D. All analyses were conducted using 770 the R software package TwoSampleMR v0.5.494. 771 Instrument selection: Independent (established by conditional analyses for both RG and the 772 lung function phenotypes) genome -wide significant ( P<5x10-8) variants were selected as 773 genetic instruments. In total, 143 independent vari ants were defined for RG by the current 774 study, 424 T2D signals were reported for Europeans by Vujkovic et al. and 130/162 775 independent signals were reported by Shrine et al. for FVC and FEV1, respectively. We looked 776 for proxy variants with a minimum r 2 of 0.8 where the instrumental variant was not present 777 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 34 in the outcome data. Palindromic variants with minor allele frequency larger than 45% were 778 excluded to avoid uncertainty when harmonizing effects to the exposure -increasing allele. 779 After filtering, 136 variants were used to instrument RG and 413 variants were available as 780 T2D instruments. For FVC, 125 and 115 variants could be used as instruments in the RG and 781 T2D MR analyses, respectively. For FEV1, 157 and 140 variants served as instruments in the 782 RG and T2D MR analyses, respectively. 783 Causal effects were estimated using the inverse-variance weighted method, which combines 784 the causal estimates of individual instrumental variants (Wald ratios) in a random -effects 785 meta-analysis95. As a sensitivity analysis, we employed MR-Egger regression to obtain causal 786 estimates that are more robust to the inclusion of invalid instruments96. 787 788 PRS analysis 789 We tested the ability of the RG genet ic effects to predict RG, T2D and HbA1c. We compared 790 that to the predictive power of T2D and FG genetic instruments by computing PRS for RG, T2D 791 and FG and assessing their performance in predicting RG, T2D and HbA1c. PRS analyses 792 require base - and target data from independent populations. The bas e datasets in our 793 analyses were UKBB-only estimates from the present RG GWAS, meta -analysis estimates of 794 32 studies f or T2D 97 and meta -analysis estimates from the MAGIC for FG 54. We u sed the 795 second largest cohort, the Vanderbilt Un iversity Medical Centre ( VUMC), as our target 796 dataset. PRS const ruction and model evaluation were done using the software PRSice 797 (v2.2.3)98. The PRS for an individual is the summation of the effect (trait -increasing) alleles 798 weighted by the effect size of the SNP taken from the base data. The SN Ps in the base data 799 are clumped so that they are largely independent of each other and thus their effects can be 800 summed. To asses s predictive power, PRS for RG, T2D and FG were regressed on to the 801 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 35 phenotypes of interest (i.e. RG, T2D and HbA1c ) providing the coefficient of determination 802 (R2) as an estimate for the correlation between the phenotype and the PRS in the VUMC 803 cohort. All models were adjusted for age, four principal components, sex and the cohort -804 specific batch effect. Since the optimal P-value th reshold for including SNPs in the PRS is 805 unknown a priori, PRS are calculated over a range of thresholds and regressed onto the 806 phenotype of interest, optimising prediction accordingly. The R2 estimates for each trait were 807 derived by subtracting the R 2 from the null model ( Phenotype ~ sex + age + 4 principal 808 components + batch) from the R 2 from the full model ( Phenotype ~ PRS + sex + age + 4 809 principal components + batch ) which contains the PRS at the best predicting P -value 810 threshold. 811 812 Clustering of the RG signals with results for 45 other phenotypes 813 We looked up the Z-scores (regression coefficient beta divided by the standard error) of the 814 distinct 143 RG signals in publicly available summary statistics of 45 relevant phenotypes. All 815 variant effects we re aligned to the RG risk allele. HapMap2 based summary statistics were 816 imputed using SS -Imp v0.5.5 65 to minimise missingness. Missing summary statistics values 817 were imputed via mean imputation. The resulting variant-trait association matrix was scaled 818 by the square root of the study’s mean sample size. We used agglomerative hierarchical 819 clustering with Ward’s method to partition the variants into groups by their effects on the 820 considered outcomes. The clustering analysis was performed in R using function hclust() from 821 in-built stats package. 822 823 Data availability 824 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 36 GWAS summary statistics for RG analyses presented in this manuscript will be deposited on 825 https://www.magicinvestigators.org/downloads/ and will be also be available through the 826 NHGRI-EBI GWAS Catalog https://www.ebi.ac.uk/gwas/downloads/summary-statistics. 827 828

829 1. Santos, R.L. et al. Heritability of fasting glucose levels in a young genetically isolated 830 population. Diabetologia 49, 667-72 (2006). 831 2. Almgren, P. et al. Heritability and familiality of type 2 diabetes and related 832 quantitative traits in the Botnia Study. Diabetologia 54, 2811-9 (2011). 833 3. Scott, R.A. et al. An Expanded Genome-Wide Association Study of Type 2 Diabetes in 834 Europeans. Diabetes 66, 2888-2902 (2017). 835 4. Vujkovic, M. et al. Discovery of 318 new risk loci for type 2 diabetes and related 836 vascular outcomes among 1.4 million participants in a multi-ancestry meta-analysis. 837 Nat Genet 52, 680-691 (2020). 838 5. Chen, J. et al. The Trans-Ancestral Genomic Architecture of Glycaemic Traits. bioRxiv, 839 2020.07.23.217646 (2020). 840 6. Dimas, A.S. et al. Impact of type 2 diabetes susceptibility variants on quantitative 841 glycemic traits reveals mechanistic heterogeneity. Diabetes 63, 2158-71 (2014). 842 7. Ingelsson, E. et al. Detailed physiologic characterization reveals diverse mechanisms 843 for novel genetic Loci regulating glucose and insulin metabolism in humans. Diabetes 844 59, 1266-75 (2010). 845 8. Scott, R.A. et al. Large-scale association analyses identify new loci influencing 846 glycemic traits and provide insight into the underlying biological pathways. Nat 847 Genet 44, 991-1005 (2012). 848 9. Deng, Y.N., Xia, Z., Zhang, P., Ejaz, S. & Liang, S. Transcription Factor RREB1: from 849 Target Genes towards Biological Functions. Int J Biol Sci 16, 1463-1473 (2020). 850 10. Piccand, J. et al. Rfx6 maintains the functional identity of adult pancreatic beta cells. 851 Cell Rep 9, 2219-32 (2014). 852 11. Tsuboi, T. et al. Glucagon-like peptide-1 mobilizes intracellular Ca2+ and stimulates 853 mitochondrial ATP synthesis in pancreatic MIN6 beta-cells. Biochem J 369, 287-99 854 (2003). 855 12. Bosma, K.J. et al. Pancreatic islet beta cell-specific deletion of G6pc2 reduces fasting 856 blood glucose. J Mol Endocrinol 64, 235-248 (2020). 857 13. Rutter, G.A., Georgiadou, E., Martinez-Sanchez, A. & Pullen, T.J. Metabolic and 858 functional specialisations of the pancreatic beta cell: gene disallowance, 859 mitochondrial metabolism and intercellular connectivity. Diabetologia 63, 1990-1998 860 (2020). 861 14. Hara, K. et al. Genome-wide association study identifies three novel loci for type 2 862 diabetes. Hum Mol Genet 23, 239-46 (2014). 863 15. Morris, A.P. et al. Large-scale association analysis provides insights into the genetic 864 architecture and pathophysiology of type 2 diabetes. Nat Genet 44, 981-90 (2012). 865 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 37 16. Mahajan, A. et al. Identification and functional characterization of G6PC2 coding 866 variants influencing glycemic traits define an effector transcript at the G6PC2-867 ABCB11 locus. PLoS Genet 11, e1004876 (2015). 868 17. Pullen, T.J. & Rutter, G.A. Roles of lncRNAs in pancreatic beta cell identity and 869 diabetes susceptibility. Front Genet 5, 193 (2014). 870 18. Benonisdottir, S. et al. Sequence variants associating with urinary biomarkers. Hum 871 Mol Genet 28, 1199-1211 (2019). 872 19. Teumer, A. et al. Genome-wide association meta-analyses and fine-mapping 873 elucidate pathways influencing albuminuria. Nat Commun 10, 4130 (2019). 874 20. Wuttke, M. et al. A catalog of genetic loci associated with kidney function from 875 analyses of a million individuals. Nat Genet 51, 957-972 (2019). 876 21. Spracklen, C.N. et al. Identification of type 2 diabetes loci in 433,540 East Asian 877 individuals. Nature 582, 240-245 (2020). 878 22. Wessel, J. et al. Low-frequency and rare exome chip variants associate with fasting 879 glucose and type 2 diabetes susceptibility. Nat Commun 6, 5897 (2015). 880 23. Tomkin, G.H. Treatment of type 2 diabetes, lifestyle, GLP1 agonists and DPP4 881 inhibitors. World J Diabetes 5, 636-50 (2014). 882 24. Koole, C. et al. Polymorphism and ligand dependent changes in human glucagon-like 883 peptide-1 receptor (GLP-1R) function: allosteric rescue of loss of function mutation. 884 Mol Pharmacol 80, 486-97 (2011). 885 25. Wan, Q. et al. Mini G protein probes for active G protein-coupled receptors (GPCRs) 886 in live cells. J Biol Chem 293, 7466-7473 (2018). 887 26. Jones, B. et al. Targeting GLP-1 receptor trafficking to improve agonist efficacy. Nat 888 Commun 9, 1602 (2018). 889 27. Deganutti, G. et al. Dynamics of GLP-1R peptide agonist engagement are correlated 890 with kinetics of G protein activation. bioRxiv, 2021.03.10.434902 (2021). 891 28. Venkatakrishnan, A.J. et al. Diverse GPCRs exhibit conserved water networks for 892 stabilization and activation. Proc Natl Acad Sci U S A 116, 3288-3293 (2019). 893 29. Yuan, S., Filipek, S., Palczewski, K. & Vogel, H. Activation of G-protein-coupled 894 receptors correlates with the formation of a continuous internal water pathway. Nat 895 Commun 5, 4733 (2014). 896 30. Zhao, P. et al. Activation of the GLP-1 receptor by a non-peptidic agonist. Nature 897 577, 432-436 (2020). 898 31. Karczewski, K.J. et al. The mutational constraint spectrum quantified from variation 899 in 141,456 humans. Nature 581, 434-443 (2020). 900 32. Hauser, A.S. et al. Pharmacogenomics of GPCR Drug Targets. Cell 172, 41-54 e19 901 (2018). 902 33. Sorli, C. et al. Efficacy and safety of once-weekly semaglutide monotherapy versus 903 placebo in patients with type 2 diabetes (SUSTAIN 1): a double-blind, randomised, 904 placebo-controlled, parallel-group, multinational, multicentre phase 3a trial. Lancet 905 Diabetes Endocrinol 5, 251-260 (2017). 906 34. Pers, T.H. et al. Biological interpretation of genome-wide association studies using 907 predicted gene functions. Nat Commun 6, 5890 (2015). 908 35. Timshel, P.N., Thompson, J.J. & Pers, T.H. Genetic mapping of etiologic brain cell 909 types for obesity. Elife 9(2020). 910 36. Ding, Q. et al. Genome-wide meta-analysis associates GPSM1 with type 2 diabetes, a 911 plausible gene involved in skeletal muscle function. J Hum Genet 65, 411-420 (2020). 912 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 38 37. Kurilshikov, A. et al. Genetics of human gut microbiome composition. bioRxiv, 913 2020.06.26.173724 (2020). 914 38. Lopera-Maya, E.A. et al. Effect of host genetics on the gut microbiome in 7,738 915 participants of the Dutch Microbiome Project. bioRxiv, 2020.12.09.417642 (2020). 916 39. Võsa, U. et al. Unraveling the polygenic architecture of complex traits using blood 917 eQTL metaanalysis. bioRxiv, 447367 (2018). 918 40. Sharapov, S.Z. et al. Defining the genetic control of human blood plasma N-glycome 919 using genome-wide association study. Hum Mol Genet 28, 2062-2077 (2019). 920 41. Clerc, F. et al. Human plasma protein N-glycosylation. Glycoconj J 33, 309-43 (2016). 921 42. Novokmet, M. et al. Changes in IgG and total plasma protein glycomes in acute 922 systemic inflammation. Sci Rep 4, 4347 (2014). 923 43. Schmidt, M.I. et al. Markers of inflammation and prediction of diabetes mellitus in 924 adults (Atherosclerosis Risk in Communities study): a cohort study. Lancet 353, 1649-925 52 (1999). 926 44. Dotz, V. et al. Plasma protein N-glycan signatures of type 2 diabetes. Biochim Biophys 927 Acta Gen Subj 1862, 2613-2622 (2018). 928 45. Keser, T. et al. Increased plasma N-glycome complexity is associated with higher risk 929 of type 2 diabetes. Diabetologia 60, 2352-2360 (2017). 930 46. Wittenbecher, C. et al. Plasma N-Glycans as Emerging Biomarkers of Cardiometabolic 931 Risk: A Prospective Investigation in the EPIC-Potsdam Cohort Study. Diabetes Care 932 43, 661-668 (2020). 933 47. Johswich, A. et al. N-glycan remodeling on glucagon receptor is an effector of 934 nutrient sensing by the hexosamine biosynthesis pathway. J Biol Chem 289, 15927-935 41 (2014). 936 48. Lemmers, R.F.H. et al. IgG glycan patterns are associated with type 2 diabetes in 937 independent European populations. Biochim Biophys Acta Gen Subj 1861, 2240-2249 938 (2017). 939 49. Liu, D. et al. Ischemic stroke is associated with the pro-inflammatory potential of N-940 glycosylated immunoglobulin G. J Neuroinflammation 15, 123 (2018). 941 50. Kopf, S. et al. Breathlessness and Restrictive Lung Disease: An Important Diabetes-942 Related Feature in Patients with Type 2 Diabetes. Respiration 96, 29-40 (2018). 943 51. Sonoda, N. et al. A prospective study of the impact of diabetes mellitus on restrictive 944 and obstructive lung function impairment: The Saku study. Metabolism 82, 58-64 945 (2018). 946 52. Abdi, A., Jalilian, M., Sarbarzeh, P.A. & Vlaisavljevic, Z. Diabetes and COVID-19: A 947 systematic review on the current evidences. Diabetes Res Clin Pract 166, 108347 948 (2020). 949 53. Zhu, L. et al. Association of Blood Glucose Control and Outcomes in Patients with 950 COVID-19 and Pre-existing Type 2 Diabetes. Cell Metab 31, 1068-1077 e3 (2020). 951 54. Lagou, V. et al. Sex-dimorphic genetic effects and novel loci for fasting glucose and 952 insulin variability. Nat Commun 12, 24 (2021). 953 55. Marullo, L., El-Sayed Moustafa, J.S. & Prokopenko, I. Insights into the genetic 954 susceptibility to type 2 diabetes from genome-wide association studies of glycaemic 955 traits. Curr Diab Rep 14, 551 (2014). 956 56. Mingrone, G. et al. Metabolic surgery versus conventional medical therapy in 957 patients with type 2 diabetes: 10-year follow-up of an open-label, single-centre, 958 randomised controlled trial. Lancet 397, 293-304 (2021). 959 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 39 57. Whang, A., Nagpal, R. & Yadav, H. Bi-directional drug-microbiome interactions of 960 anti-diabetics. EBioMedicine 39, 591-602 (2019). 961 58. Voight, B.F. et al. The metabochip, a custom genotyping array for genetic studies of 962 metabolic, cardiovascular, and anthropometric traits. PLoS Genet 8, e1002793 963 (2012). 964 59. Li, Y., Willer, C., Sanna, S. & Abecasis, G. Genotype imputation. Annu Rev Genomics 965 Hum Genet 10, 387-406 (2009). 966 60. Marchini, J., Howie, B., Myers, S., McVean, G. & Donnelly, P. A new multipoint 967

for genome-wide association studies by imputation of genotypes. Nat Genet 968 39, 906-13 (2007). 969 61. Fuchsberger, C., Abecasis, G.R. & Hinds, D.A. minimac2: faster genotype imputation. 970 Bioinformatics 31, 782-4 (2015). 971 62. Loh, P.R., Kichaev, G., Gazal, S., Schoech, A.P. & Price, A.L. Mixed-model association 972 for biobank-scale datasets. Nat Genet 50, 906-908 (2018). 973 63. Loh, P.R. et al. Efficient Bayesian mixed-model analysis increases association power 974 in large cohorts. Nat Genet 47, 284-90 (2015). 975 64. Genomes Project, C. et al. A global reference for human genetic variation. Nature 976 526, 68-74 (2015). 977 65. Rueger, S., McDaid, A. & Kutalik, Z. Evaluation and application of summary statistic 978 imputation to discover new height-associated loci. PLoS Genet 14, e1007371 (2018). 979 66. Willer, C.J., Li, Y. & Abecasis, G.R. METAL: fast and efficient meta-analysis of 980 genomewide association scans. Bioinformatics 26, 2190-1 (2010). 981 67. Magi, R. & Morris, A.P. GWAMA: software for genome-wide association meta-982 analysis. BMC Bioinformatics 11, 288 (2010). 983 68. Chang, C.C. et al. Second-generation PLINK: rising to the challenge of larger and 984 richer datasets. Gigascience 4, 7 (2015). 985 69. Yang, J. et al. Conditional and joint multiple-SNP analysis of GWAS summary statistics 986 identifies additional variants influencing complex traits. Nat Genet 44, 369-75, S1-3 987 (2012). 988 70. Fang, Z. et al. The Influence of Peptide Context on Signaling and Trafficking of 989 Glucagon-like Peptide-1 Receptor Biased Agonists. ACS Pharmacol Transl Sci 3, 345-990 360 (2020). 991 71. Fang, Z. et al. Ligand-Specific Factors Influencing GLP-1 Receptor Post-Endocytic 992 Trafficking and Degradation in Pancreatic Beta Cells. Int J Mol Sci 21(2020). 993 72. Ward, R.J., Alvarez-Curto, E. & Milligan, G. Using the Flp-In T-Rex system to regulate 994 GPCR expression. Methods Mol Biol 746, 21-37 (2011). 995 73. Peng, T. et al. A BaSiC tool for background and shading correction of optical 996 microscopy images. Nat Commun 8, 14836 (2017). 997 74. Jaccard, N. et al. Automated method for the rapid and precise estimation of 998 adherent cell culture characteristics from phase contrast microscopy images. 999 Biotechnol Bioeng 111, 504-17 (2014). 1000 75. Kenakin, T. A Scale of Agonism and Allosteric Modulation for Assessment of 1001 Selectivity, Bias, and Receptor Mutation. Mol Pharmacol 92, 414-424 (2017). 1002 76. Harvey, M.J., Giupponi, G. & Fabritiis, G.D. ACEMD: Accelerating Biomolecular 1003 Dynamics in the Microsecond Time Scale. J Chem Theory Comput 5, 1632-9 (2009). 1004 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 40 77. Cuzzolin, A., Deganutti, G., Salmaso, V., Sturlese, M. & Moro, S. AquaMMapS: An 1005 Alternative Tool to Monitor the Role of Water Molecules During Protein-Ligand 1006 Association. ChemMedChem 13, 522-531 (2018). 1007 78. Wakefield, J. A Bayesian measure of the probability of false discovery in genetic 1008 epidemiology studies. Am J Hum Genet 81, 208-27 (2007). 1009 79. Pers, T.H., Timshel, P. & Hirschhorn, J.N. SNPsnap: a Web-based tool for 1010 identification and annotation of matched SNPs. Bioinformatics 31, 418-20 (2015). 1011 80. Genomes Project, C. et al. A map of human genome variation from population-scale 1012 sequencing. Nature 467, 1061-73 (2010). 1013 81. Tabula Muris, C. et al. Single-cell transcriptomics of 20 mouse organs creates a 1014 Tabula Muris. Nature 562, 367-372 (2018). 1015 82. Timshel, P.N., Thompson, J.J. & Pers, T.H. Mapping heritability of obesity by brain cell 1016 types. bioRxiv, 2020.01.27.920033 (2020). 1017 83. Barbeira, A.N. et al. Exploring the phenotypic consequences of tissue specific gene 1018 expression variation inferred from GWAS summary statistics. Nat Commun 9, 1825 1019 (2018). 1020 84. Gamazon, E.R. et al. A gene-based association method for mapping traits using 1021

transcriptome data. Nat Genet 47, 1091-8 (2015). 1022 85. Delaneau, O., Marchini, J., Genomes Project, C. & Genomes Project, C. Integrating 1023 sequence and array data to create an improved 1000 Genomes Project haplotype 1024

panel. Nat Commun 5, 3934 (2014). 1025 86. Iotchkova, V. et al. Discovery and refinement of genetic loci associated with 1026 cardiometabolic risk using dense imputation maps. Nat Genet 48, 1303-1312 (2016). 1027 87. Almgren, P. et al. Genetic determinants of circulating GIP and GLP-1 concentrations. 1028 JCI Insight 2(2017). 1029 88. Dobbyn, A. et al. Landscape of Conditional eQTL in Dorsolateral Prefrontal Cortex 1030 and Co-localization with Schizophrenia GWAS. Am J Hum Genet 102, 1169-1184 1031 (2018). 1032 89. Sharapov, S. et al. Genome-wide association summary statistics for human blood 1033 plasma glycome. (Zenodo, 2018). 1034 90. Finucane, H.K. et al. Partitioning heritability by functional annotation using genome-1035 wide association summary statistics. Nat Genet 47, 1228-35 (2015). 1036 91. Zheng, J. et al. LD Hub: a centralized database and web interface to perform LD score 1037 regression that maximizes the potential of summary level GWAS data for SNP 1038 heritability and genetic correlation analysis. Bioinformatics 33, 272-279 (2017). 1039 92. Fedko, I.O. et al. Genetics of fasting indices of glucose homeostasis using GWIS 1040 unravels tight relationships with inflammatory markers. bioRxiv, 496802 (2018). 1041 93. Shrine, N. et al. New genetic signals for lung function highlight pathways and chronic 1042 obstructive pulmonary disease associations across multiple ancestries. Nat Genet 51, 1043 481-493 (2019). 1044 94. Hemani, G. et al. The MR-Base platform supports systematic causal inference across 1045 the human phenome. Elife 7(2018). 1046 95. Burgess, S. et al. Guidelines for performing Mendelian randomization investigations. 1047 Wellcome Open Res 4, 186 (2019). 1048 96. Bowden, J., Davey Smith, G. & Burgess, S. Mendelian randomization with invalid 1049 instruments: effect estimation and bias detection through Egger regression. Int J 1050 Epidemiol 44, 512-25 (2015). 1051 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 41 97. Mahajan, A. et al. Fine-mapping type 2 diabetes loci to single-variant resolution 1052 using high-density imputation and islet-specific epigenome maps. Nat Genet 50, 1053 1505-1513 (2018). 1054 98. Choi, S.W. & O'Reilly, P.F. PRSice-2: Polygenic Risk Score software for biobank-scale 1055 data. Gigascience 8(2019). 1056 1057 1058

1059 Airwave 1060 The Airwave Health Monitoring Study was funded by the UK Home Office (780- TETRA, 2003-1061 2018) and is currently funded by the MRC and ESRC (MR/R023484/1) with additional funding 1062 from the NIHR Imperial College Biomedical Research Centre (BRC) in collaboration with 1063 Imperial College NHS Healthcare Trust. We thank all Airwave participants for their 1064 contribution to the study. 1065 Personal support: Paul Elliott acknowledges support from the MRC and PHE (MR/L01341X/1, 1066 812 2014-2019) and currently from the MRC for the MRC Centre for Environment and Health 1067 813 (MR/S019669/1). Paul Elliott and Ioanna Tzoulaki are supported by the UK Dementia 1068 Research Institute which 814 receives funding from UK DRI Ltd funded by the UK Medical 1069 Research Council, Alzheimer’s 815 Society and A lzheimer’s Research UK. Paul Elliott is 1070 associate director of the Health Data 816 Research UK London funded by a consortium led by 1071 the UK Medical Research Council. 1072 1073 BRIGHT 1074 This work was funded by the Medical Research Council of Great Britain (grant number : 1075 G9521010D). The BRIGHT study is extremely grateful to all the patients who participated in 1076 the study and the BRIGHT nursing team. This work formed part of the research themes 1077 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 42 contributing to the translational research portfolio for the NIHR Barts Cardiov ascular 1078 Biomedical Research Centre. The funders had no role in study design, data collection and 1079 analysis. 1080 1081 deCODE 1082 We thank participants in deCODE genetic studies whose contribution made this work 1083 possible. 1084 1085 EMIL 1086 EMIL-Cohort, a population-based cohort; the study was approved by the ethical committee 1087 of the Chamber of Physicians Baden -Württemberg in the year 2002 (Registration Number 1088 133-02; dates 05.09.2002 and 24.09.2002). We thank Silke Rosinger, Simone Claudi -Boehm, 1089 Rosina Sing, Sabine Schilling and Angelika Kurkhaus for technical support. 1090 1091 EPIC-Norfolk 1092 The EPIC-Norfolk study (https://doi.org/10.22025/2019.10.105.00004) has received funding 1093 from the Medical Research Council (MR/N003284/1 and MC -UU_12015/1) and Ca ncer 1094 Research UK (C864/A14136). The genetics work in the EPIC-Norfolk study was funded by the 1095 Medical Research Council (MC_PC_13048). We are grateful to all the participants who have 1096 been part of the project and to the many members of the study teams at t he University of 1097 Cambridge who have enabled this research. 1098 1099 FINRISK87 1100 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 43 Support for FUSION was provided by NIH grants R01 -DK062370 (to M.B.), R01-DK072193 (to 1101 K.L.M.), and intramural project number 1Z01-HG000024 (to F.S.C.). Genome-wide genotyping 1102 was conducted by the Johns Hopkins University Genetic Resources Core Facility SNP Center 1103 at the Center for Inherited Disease Research (CIDR), with support from CIDR NIH contract no. 1104 N01-HG-65403. 1105 1106 Framingham Heart Study 1107 Also supported by National Institute for Diabetes and Digestive and Kidney Diseases (NIDDK) 1108 U01/UM1 DK078616 to Dr. Meigs 1109 1110 HUNT2/Tromson 1111 The Nord -Trøndelag Health Study (the HUNT study) is a collaboration between HUNT 1112 Research Centre (Faculty of Medicine, Norwegian University of Science and Technology 1113 NTNU), Nord-Trøndelag County Council, Central Norway Health Authority, and the Norwegian 1114 Institute of Public Health. University of Tromsø, Norwegian Research Council (project number 1115 185764). 1116 1117 InterAct 1118 We thank all EPIC participants and staff and the InterAct Consortium members for their 1119 contributions to the study. The InterAct project received funding from the European Union 1120 (Integrated Project LSHM-CT-2006-037197 in the Framework Programme 6 of the European 1121 Community). We thank staff from the technic al, field epidemiology and data teams of the 1122 Medical Research Council Epidemiology Unit in Cambridge, UK, for carrying out sample 1123 preparation, DNA provision and quality control, genotyping and data handling work. 1124 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 44 1125 KORA F3 1126 The KORA research platform (KORA, Cooperative Research in the Region of Augsburg) was 1127 initiated and financed by the Helmholtz Zentrum München –German Research Center for 1128 Environment and Health, which is funded by the German Federal Ministry of Education and 1129 Research and by the state of Bavaria. Furthermore, part of this work was supported by the 1130 German National Genome Research Network (NGFN) and the Munich Center of Health 1131 Sciences (MC Health) as part of LMUinnovativ. 1132 1133 Brisbane Adolescent Twin Study / SSAGA-NAG adult cohort 1134 Genotyping and phenotyping were supported by the Australian National Health and Medical 1135 Research Council (389891, 389892, 496739), the EU 5th Framework Programme 1136 GenomEUtwin Project (QLG2 -CT-2002-01254) and the U.S. National Institutes of Health 1137 (AA07535, AA13320, AA13321, AA13326, AA14041, DA12854). B.B. and G.W.M. are 1138 supported by National Health and Medical Research Council (NHMRC) Fellowship Schemes. 1139 Participants gave informed consent and the studies were approved by appropriate 1140 institutional review boards. 1141 1142 PPP-Botnia/Malmö Diet Cancer 1143 Personal support: Emma Ahlqvist was funded by grants from the Swedish Research Council 1144 (2017-02688, 2020-02191). 1145 1146 PROCARDIS 1147 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 45 PROCARDIS was supported by the European Community Sixth Framework Program (LSHM -1148 CT- 2007-037273), AstraZeneca, the British Heart Foundation, the Wellcome Trust (Contract 1149 No. 075491/Z/04), the Swedish Research Council, the Knut and Alice Wallenberg Foundation, 1150 the Swedish Heart -Lung Foundation, the Torsten and Ragnar Söderberg Foundati on, the 1151 Strategic Cardiovascular and Diabetes Programs of Karolinska Institutet and Stockholm 1152 County Council, the Foundation for Strategic Research and the Stockholm County Council. 1153 Ethical permission was granted by the local ethical board for each centre. 1154 Personal support: Anuj Goel has received support from the BHF, European Commission 1155 [LSHM-CT- 2007-037273, HEALTH -F2-2013-601456] and TriPartite Immunometabolism 1156 Consortium [TrIC]- NovoNordisk Foundation [NNF15CC0018486]. Hugh Watkins has received 1157 support from Wellcome Trust core awards (090532/Z/09/Z, 203141/Z/16/Z) and is member 1158 of the Oxford BHF Centre for Research Excellence (RE/13/1/30181) . Rona J Strawbridge is 1159 supported by a UKRI Innovation-HDR-UK Fellowship (MR/S003061/1). 1160 1161 Rotterdam Study 1162 The gener ation and management of GWAS genotype data for the Rotterdam Study is 1163 supported by the Netherlands Organisation of Scientific Research NWO Investments (nr. 1164 175.010.2005.011, 911-03-012). This study is funded by the Research Institute for Diseases in 1165 the El derly (014 -93-015; RIDE2), the Netherlands Genomics Initiative (NGI)/Netherlands 1166 Organisation for Scientific Research (NWO) project nr. 050 -060-810. We thank Pascal Arp, 1167 Mila Jhamai, Marijn Verkerk, Lizbeth Herrera and Marjolein Peters for their help in cr eating 1168 the GWAS database, and Karol Estrada and Maksim V. Struchalin for their support in creation 1169 and analysis of imputed data. The Rotterdam Study is funded by Erasmus Medical Center and 1170 Erasmus University, Rotterdam, Netherlands Organization for the Hea lth Research and 1171 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 46 Development (ZonMw), the Research Institute for Diseases in the Elderly (RIDE), the Ministry 1172 of Education, Culture and Science, the Ministry for Health, Welfare and Sports, the European 1173 Commission (DG XII), and the Municipality of Rotterdam. The authors are grateful to the study 1174 participants, the staff from the Rotterdam Study and the participating general practitioners 1175 and pharmacists. 1176 1177 The Section of Endocrinology and Investigative Medicine is funded by grants from the MRC, 1178 BBSRC, NIHR, and is supported by the NIHR Biomedical Research Centre Funding Scheme. The 1179 views expressed are those of the author(s) and not necessarily those of the any of the funders, 1180 the NHS, the NIHR or the Department of Health. 1181 1182 UK Biobank 1183 We thank UK Biobank for data availability, project number 37685. 1184 Personal support: Marika A Kaakinen, Anna Ulrich, Zhanna Balkhiyarova and Inga Prokopenko 1185 are in part funded by the European Union’s Horizon 2020 research and innovation 1186 programme LONGITOOLS, H2020-SC1-2019-874739. Marika A Kaakinen is also funded by the 1187 European Foundation for the Study of Diabetes (EFSD) Albert Renold Travel Fellowship. 1188 1189 Vanderbilt 1190 The samples and/or dataset(s) used for the analyses described were obtained from Vanderbilt 1191 University Medical Center’s BioVU which is supported by numerous sources: institutional 1192 funding, private agencies, and federal grants. These include the NIH funded Shared 1193 Instrumentation Grant S10OD017985 and S10RR025141; and CTSA grants UL1TR002243, 1194 UL1TR000445, and UL1RR02 4975. Genomic data are also supported by investigator -led 1195 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 47 projects that include U01HG004798, R01NS032830, RC2GM092618, P50GM115305, 1196 U01HG006378, U19HL065962, R01HD074711; and additional funding sources listed at 1197 https://victr.vumc.org/biovu-funding/”. 1198 Personal support: NIH Grants R01 HL146588 and R01 HL146588-01S1. 1199 1200 Individual acknowledgements 1201 Alejandra Tomas was supported by MRC project grant MR/R010676/1 and grants from 1202 Diabetes UK and the European Federation for the Study of Diabetes. 1203 Alessia David is funded by Wellcome Trust grant 104955/Z/14/Z. 1204 Ben Jones is supported by an Imperial Post-CCT Post-Doctoral Fellowship, MRC project grant 1205 MR/R010676/1, and grants from the European Federation for the Study of Diabetes, Academy 1206 of Medical Sciences, Imperial NIHR Biomedical Research Centre, Engineering and Physical 1207 Sciences Research Council, Society for Endocrino logy and British Society for 1208 Neuroendocrinology. 1209 Christopher Reynolds is a Royal Society Industry Fellow. 1210 Inês Barroso was funded by an “Expanding excellence in England” award from Research 1211 England. 1212 Inga Prokopenko is funded by the World Cancer Research Fund (WCRF UK) and World Cancer 1213 Research Fund International (2017/1641), the Wellcome Trust (WT205915), the European 1214 Union’s Horizon 2020 research and innovation programme (DYNAhealth, project number 1215 633595) the LONGITOOLS, H2020-SC1-2019-874739 and the Royal Society (IEC\R2\181075). 1216 Mark McCarthy was a Wellcome Investigator and an NIHR Senior Investigator (Niddk. U01-1217 DK105535, Wellcome: 090532, 098381, 106130, 203141, 212259). 1218 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 48 Patrick M Sexton and Denise Wootten: The work was supported by NHMRC Ideas #1184726 1219 and NHMRC program grant #1150083. Patrick M Sexton is a NHMRC Senior Principal Research 1220 Fellow (#1154434) and Denise Wootten is an NHMRC Senior Research Fellow (#1155302). 1221 Vasiliki Lagou was supported by a fellowship from the Research Foundation-Flanders (FWO). 1222 Victoria Salem is the recipient of a Diabetes UK Harry Keen Clinician Scientist Fellowship. 1223 Sharapov Sodbo was supported by a grant from the Russian Science Foundation 1224 (RSF) No. 19-15-00115. 1225 Tricia M Tan acknowledges support from the Nationa l Institute for Health Research (NIHR) 1226 Imperial Biomedical Research Centre (BRC) Funding Scheme. 1227 1228 Author contributions 1229 First author: V.L., L.J., A.U., Central analysis and writing group : V.L., L.J., A.U., L.Z., K.G., Z.B., A.F., 1230 L.M., Additional analyses junior lead : S.C., P.T., S.S., A.David, R.M., R.R., E.Ahlqvist, T.M.T., A.T., 1231 V.Salem, GWAS cohort analyst: G.T., Η.G., Ε.Ε., B.B., R.S., A.I., J.Z., S.M.W., T.J., C.G., H.G., C.M., M.M., 1232 R.J.S., A.G., D.R., J.D., Y.S.A., M.A.K., Metabochip cohort analyst : E.Albrecht, A.U.J., H.M.S., Cohort 1233 sample collection, genotyping, phenotyping or additional analysis : I.R.C., F.E., V.Steinthorsdottir, 1234 A.G.U., P.B.M., M.J.B., S.J., O.H., B.T., K.H., T.W., K.L.M., Metabochip cohort PI : W.Kratzer, H.M., 1235 W.Koenig, B.O.B., GWAS cohort PI: J.T., M.B., J.C.F., A.Hamsten, H.Watkins, I.N., H.Wichmann, M.J.C., 1236 K.K., C.v.D., A.Hofman, N.J.W., C.L., J.B.W., N.G.M., G.M., I.T., P.E., U.T., K.S., E.L.B., J.B.M., Additional 1237 analyses senior: P.M.S., D.W., L.G., G.D., A.Demirkan, T.H.P., C.A.R., Other senior author (analysis and 1238 writing group): Y.S.A., Other senior author (writing): I.B., C.S., M.I.M., P.F., J.D., J.B.M., Senior author: 1239 M.A.K., B.J., I.P. 1240 1241 Competing interests 1242 Alejandra Tomas has received grant funding from Sun Pharmaceuticals. 1243 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 49 Ivan R Corrêa, Jr is an employee of New England Biolabs, Inc., a manufacturer and vendor of 1244 reagents for life science research. 1245 Mark J Caulfield is Chief Scientist for Genomics England, a UK Government company. 1246 The views expressed in this article are those of the author(s) and not necessarily those of the 1247 NHS, the NIHR, or the Department of Health. Mark McCarthy has served on advisory panels 1248 for Pfizer, NovoNordisk and Zoe Global, has received honoraria from Merck, Pfizer, Novo 1249 Nordisk and Eli Lilly, and research funding from Abbvie, Astra Zeneca, Boehringer Ingelheim, 1250 Eli Lilly, Janssen, Merck, NovoNordisk, Pfizer, Roche, Sanofi Aventis, Servier, and Takeda. As 1251 of June 2019, MMcC is an employee of Genentech, and a holder of Roche stock. 1252 Patrick M Sexton receives grant funding from Laboratoires Servier. 1253 Toby Johnson is now a GSK employee. 1254 Yurii S Aulchenko is owner of Maatschap PolyOmica and PolyKnomics BV, private 1255 organizations providing services, research and development in the field of computational and 1256 statistical, quantitative and computational (gen)omics. 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 50 Tables 1268 Table 1. Novel loci for glycaemic traits discovered through i) a GWAS meta -analysis of RG 1269 levels in up to 479,482 Europeans without diabetes, and ii) a trans-ethnic meta-analysis of 1270 up to 496,036 Europeans and individuals of other ancestries (Black, Indian, Pakistani, 1271 Chinese) in UKBB. Loci showing sex -dimorphic effects on glycaemic trait l evels for the first 1272 time are also shown. 1273 1274 Signal Nearest gene(s) Lead variant Chr Position Type Alleles (effect/ other) EAF Effect (SE) P-value P het N European KDM4A rs3791033 1 44,134,077 primary T/C 0.68 -0.0018 (0.00030) 4.5x10-9 0.59 476,655 European FAM46C rs1966228 1 118,144,332 primary A/G 0.74 0.0032 (0.00030) 5.8x10-22 0.13 472,798 European, nonsyn EDEM3 rs78444298 1 184,672,098 primary A/G 0.019 0.0073 (0.0011) 1.5x10-11 0.61 439,856 European ACVR1C rs146418816 2 158,432,811 primary A/G 0.057 -0.0034 (0.00060) 5.6x10-8 0.055 475,174 European ACVR1C rs2509903 2 158,514,510 secondary T/C 0.14 0.0022 (0.00040) 6.7x10-8 0.14 478,582 European RBMS1 rs12692596 2 161,265,910 primary T/C 0.37 0.0018 (0.00030) 1.3x10-9 0.84 478,570 European G6PC2 rs143869345 2 169,708,322 secondary A/G 0.98 -0.0152 (0.0012) 1.7x10-37 1.00 401,810 European, nonsyn NEUROD1 rs8192556 2 182,542,998 primary T/G 0.024 0.0053 (0.00090) 2.8x10-8 0.50 439,856 European CACNA2D3 rs34222465 3 55,123,055 primary A/G 0.56 -0.0019 (0.00030) 5.5x10-10 0.052 439,856 European MBNL1 rs4679997 3 152,396,466 secondary C/G 0.33 0.0017 (0.00030) 9.3x10-8 0.28 473,926 European MBNL1 rs78482374 3 152,492,522 secondary A/T 0.037 -0.0042 (0.00080) 6.7x10-8 0.22 455,510 European TRIM59,KPNA4 rs56394279 3 160,171,092 primary T/C 0.52 -0.0018 (0.00030) 1.2x10-9 0.068 474,089 European MECOM rs73174306 3 169,194,244 primary A/T 0.96 -0.0057 (0.00070) 1.4x10-14 0.095 432,212 European LCORL rs75631642 4 18,049,216 secondary T/C 0.78 -0.0017 (0.00040) 2.1x10-6 0.13 466,061 European LCORL rs6840504 4 18,205,102 primary T/C 0.45 0.0018 (0.00030) 1.2x10-9 0.15 475,423 European ADRB2 rs71584073 5 148,149,418 primary T/C 0.93 0.0035 (0.00060) 3.3x10-10 0.44 439,856 European SYNGAP1 rs9461856 6 33,395,199 primary A/G 0.48 -0.00030 (0.00030) 0.33 0.091 457,070 European ITPR3 rs1830873 6 33,620,397 primary C/G 0.57 0.00070 (0.00030) 0.021 0.87 452,301 European ARMC2,SESN1 rs118126621 6 109,304,170 primary A/G 0.025 0.0039 (0.0010) 5.0x10-5 0.049 432,212 European POP7,EPO rs534043 7 100,312,724 primary A/G 0.11 -0.0031 (0.00050) 1.7x10-11 0.37 475,631 European PRKAR2B rs3801969 7 106,711,492 primary T/G 0.43 0.0016 (0.00030) 2.2x10-8 0.22 478,580 European A1CF rs61856594 10 52,637,925 primary A/G 0.71 0.0022 (0.00030) 1.6x10-11 0.58 473,354 European PRKG1 rs4415704 10 53,561,613 primary T/C 0.42 -0.0016 (0.00030) 5.6x10-8 0.82 474,069 European LMO1 rs9667977 11 8,541,291 secondary T/C 0.46 -0.0013 (0.00030) 4.5x10-6 0.71 457,903 European USP47 rs34718245 11 11,863,080 primary A/G 0.15 -0.0023 (0.00040) 4.3x10-8 0.63 470,144 European PDE3B rs141521721 11 14,763,828 primary A/C 0.023 0.0050 (0.0010) 1.8x10-7 0.0059 439,856 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 51 European PDHX rs75479466 11 34,961,066 primary A/G 0.083 0.0029 (0.00050) 2.1x10-8 0.29 472,109 European OR4A5 rs72913090 11 50,653,357 primary A/C 0.92 0.0031 (0.00050) 1.0x10-8 0.13 418,793 European TRIM48 rs150587121 11 55,036,391 primary T/C 0.91 0.0029 (0.00050) 7.7x10-8 0.14 435,903 European OR8K3,OR8K1 rs2170441 11 56,095,739 primary A/G 0.076 -0.0031 (0.00060) 1.7x10-8 0.28 422,873 European SOX5 rs12581677 12 24,060,732 primary A/G 0.91 0.0031 (0.00050) 1.2x10-9 0.036 477,019 European MANSC4,KLHL42 rs11049144 12 27,931,511 primary A/C 0.22 -0.0021 (0.00040) 1.5x10-9 0.010 455,032 European MANSC4,KLHL42 rs10492373 12 27,959,998 primary A/G 0.19 -0.0022 (0.00040) 2.3x10-9 0.0095 479,267 European RNF6 rs12874929 13 26,781,607 primary A/G 0.77 -0.0027 (0.00030) 1.1x10-14 0.97 476,730 European KL rs488166 13 33,554,352 primary C/G 0.18 0.0042 (0.00040) 3.0x10-27 0.064 478,334 European ZC3H13 rs12429980 13 46,550,138 primary A/C 0.30 -0.0018 (0.00030) 7.3x10-9 0.40 474,764 European SPRY2 rs1359790 13 80,717,156 primary A/G 0.28 -0.0019 (0.00030) 4.2x10-9 0.0010 477,640 European HECTD1,HEATR5A rs727675 14 31,733,642 primary A/G 0.57 0.0017 (0.00030) 7.5x10-9 0.86 477,060 European WARS rs45617834 14 101,295,801 secondary C/G 0.97 0.0043 (0.00090) 1.2x10-6 0.79 432,212 European HERC1 rs67507374 15 64,038,340 primary A/T 0.30 -0.0023 (0.00030) 4.3x10-13 0.20 475,691 European ITFG3,RAB11FIP3 rs111811257 16 541,818 secondary T/C 0.040 -0.0046 (0.00070) 5.5x10-10 0.76 432,212 European TAOK1,ABHD15 rs9894551 17 27,880,124 primary A/T 0.17 -0.0027 (0.00040) 5.2x10-11 0.71 415,229 European HNF1B rs10908278 17 36,099,952 primary A/T 0.52 -0.0017 (0.00030) 5.2x10-9 0.0041 439,856 European, syn NMT1 rs2239923 17 43,176,804 primary T/C 0.29 0.0019 (0.00030) 4.1x10-9 0.62 478,582 European, nonsyn WIPI1 rs883541 17 66,449,122 primary A/G 0.77 -0.0024 (0.00030) 4.4x10-12 0.24 477,006 European SKA1,MAPK4 rs2957989 18 48,075,733 primary A/G 0.82 0.0021 (0.00040) 2.2x10-8 0.72 458,445 European RALY rs6059497 20 32,446,960 primary C/G 0.54 -0.0017 (0.00030) 9.2x10-9 0.85 464,409 European HNF4A rs2267850 20 43,524,963 primary T/C 0.27 -0.0019 (0.0003) 3.8x10-9 0.92 458,445 European TSHZ2 rs2255805 20 51,627,634 primary T/C 0.57 -0.0018 (0.00030) 5.5x10-10 0.99 457,514 European STX16-NPEPL1 rs2296529 20 57,282,381 primary T/C 0.77 0.0020 (0.0003) 5.5x10-9 0.12 455,859 European STX16-NPEPL1 rs73129529 20 57,404,701 secondary C/G 0.11 0.0025 (0.00050) 1.0x10-7 0.47 439,856 European EEF1A2,PPDPF rs6122466 20 62,139,177 primary A/G 0.85 -0.0027 (0.00040) 1.6x10-10 0.75 443,482 European MTMR3,HORMAD2 rs5763882 22 30,597,426 primary A/G 0.092 -0.0028 (0.00050) 2.9x10-8 0.39 451,947 European MTMR3,HORMAD2 rs6006399 22 30,598,516 primary T/G 0.88 0.0025 (0.00040) 2.4x10-8 0.79 478,119 European, UKBB only PEX7 rs7756291 6 137235325 primary T/C 0.55 -0.00080 (0.00030) 0.0084 0.64 456,157 European, UKBB only SLC38A4 rs74832478 12 47193148 primary T/G 0.07 0.0031 (0.00060) 7.0x10-8 0.03 476,132 European, UKBB only INAFM2,C15orf52 rs4143838 15 40622374 primary T/C 0.95 -0.0036 (0.00070) 2.3x10-7 0.12 418,793 European, UKBB only ADCY9,SRL rs2018506 16 4227922 primary C/G 0.85 -0.0021 (0.00040) 1.5x10-7 0.45 461,733 European, UKBB only ERN1 rs57676627 17 62203128 primary T/C 0.15 0.0022 (0.00040) 4.0x10-7 0.03 432,212 European, UKBB only CELF5,NFIC rs55740449 19 3334232 primary T/C 0.17 0.0020 (0.00040) 5.1x10-7 0.79 439,856 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 52 European, UKBB only RFX1 rs2305780 19 14083761 primary T/C 0.54 0.0015 (0.00030) 2.5x10-7 0.24 439,856 Trans- ethnic M-A GATAD2B rs10908526 1 153,883,169 primary C/T 0.51 0.0017 (0.00030) 1.2x10-8 0.35 479,064 Trans- ethnic M-A RRNAD1 rs3806415 1 156,698,265 primary C/T 0.68 -0.0017 (0.00030) 4.8x10-8 0.30 480,890 Trans- ethnic M-A PPP1CB,SPDYA rs111502507 2 29,009,180 primary A/G 0.99 0.0064 (0.0011) 2.6x10-8 0.97 439,427 Trans- ethnic M-A MINPP1,PAPSS2 rs11202473 10 89,378,838 primary G/A 0.63 -0.0017 (0.00030) 1.2x10-8 0.53 490,575 Trans- ethnic M-A EPS8 rs6488794 12 15,816,675 primary A/G 0.030 0.0048 (0.00080) 1.4x10-8 0.54 482,276 Trans- ethnic M-A SLC38A4 rs74832478 12 47,193,148 primary G/T 0.93 0.0033 (0.00060) 7.7x10-9 0.026 492,686 Trans- ethnic M-A FOXN3 rs12892260 14 89,580,986 secondary T/C 0.94 -0.0033 (0.00060) 1.5x10-8 0.23 448,766 Sex-dim: men PRDM16 rs60330317 1 3,107,547 primary G/A 0.82 0.0011 (0.00050) 0.021 0.0026 233,066 0.82 0.0035 (0.00060) 6.1x10-9 194,008 Sex-dim: women SGIP1 rs7532598 1 66,998,624 primary C/A 0.84 -0.0029 (0.00051) 1.2x10-8 0.030 233,066 0.84 -0.0012 (0.00062) 0.053 194,008 Sex-dim: men THADA rs149290349 2 43,451,957 in LD with primary G/A 0.92 0.0035 (0.00074) 2.2x10-6 3.6x10-4 233,066 0.92 0.0076 (0.00089) 1.0x10-17 194,008 Sex-dim: men G6PC2 rs13431652 2 169,753,415 in LD with primary T/C 0.7 0.016 (0.00042) 1.0x10-1374 5.6x10-4 233,066 0.7 0.018 (0.00050) 3.4E-286 194,008 Sex-dim: men TRIM59,KPNA4 rs56394279 3 160,171,092 primary C/T 0.49 0.0012 (0.00038) 0.0015 0.0075 233,066 0.49 0.0028 (0.00046) 8.7x10-10 194,008 Sex-dim: men SOGA3,RSPO3 rs2800734 6 127,417,035 primary G/A 0.71 0.0011 (0.00042) 0.0077 0.0032 233,066 0.71 0.0031 (0.00051) 1.5x10-9 194,008 Sex-dim: men DGKB,AGMO rs1974619 7 15,065,300 primary C/T 0.45 -0.0037 (0.00038) 8.2x10-22 1.2x10-5 233,066 0.45 -0.0063 (0.00046) 1.0x10-42 194,008 Sex-dim: women SRRM3 rs11773850 7 75,824,961 in LD with primary G/A 0.98 -0.0087 (0.0013) 4.2x10-11 0.0014 233,066 0.98 -0.0021 (0.0016) 0.18 194,008 Sex-dim: men POP7,EPO rs534043 7 100,312,724 primary A/G 0.11 -0.0019 (0.00060) 0.0014 0.0016 233,066 0.11 -0.0048 (0.00072) 1.6x10-11 194,008 Sex-dim: women R3HDM2 rs7484541 12 57,714,803 in LD with primary A/T 0.78 0.0032 (0.00046) 5.3x10-12 0.030 233,066 0.78 0.0016 (0.00056) 0.0037 194,008 Sex-dim: men RMST rs6538804 12 97,848,910 primary C/G 0.60 0.0016 (0.00040) 7.3x10-5 6.0x10-4 233,066 0.60 0.0037 (0.00048) 7.8x10-15 194,008 Sex-dim: men FBRSL1 rs11146926 12 133,125,450 primary G/A 0.78 -0.0014 (0.00046) 0.0035 0.016 233,066 0.78 -0.0031 (0.00056) 2.3x10-8 194,008 Sex-dim: women FAM234A rs9929922 16 294,749 in LD with primary A/G 0.82 0.0042 (0.00049) 4.8x10-18 0.033 233,066 0.82 0.0026 (0.00059) 9.6x10-6 194,008 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 53 Sex-dim: women SLC43A2 rs56405641 17 1,528,464 primary C/T 0.91 -0.004 (0.00067) 2.2x10-9 2.7x10-4 233,066 0.91 -0.00020 (0.00080) 0.81 194,008 1275 Chr: chromosome; Pos: Position GRCh37; nonsyn: non-synonymoys; sex-dim: sex-dimorphic; 1276 EAF: allele frequ ency of the random glucose (RG) raising allele . A signal was annotated as 1277 “European” if it had reached genome -wide significance ( P<5x10-8) in the meta -analysis of 1278 European cohorts in either of our two models of interest with adjustment for age, sex with or 1279 without time sin ce last meal (where available) along with exclusion of extreme 1280 hyperglycaemia (RG>20 mmol/L) or in their combination . A signal was annotated as 1281 “European, UKBB only” if it had reached genome-wide significance (P<5x10-8) in UKBB in any 1282 of the six RG models (Methods). The EAF and P-values reported here are from the combined 1283 RG model. Heterogeneity among studies was assessed using the I 2 index. The Cochran's Q -1284 test (for sex heterogeneity representing the differences in allelic effects between sexes ) P-1285 value is also shown. Sex-dimorphic effects and P-values are presented first for women. 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 54 Figures 1297 Figure 1. Summary of all RG loci identified in this study. 1298 1299 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 55 (a) Circular Manhattan plot summarising findings from the present study. Outermost layer: 1300 Gene names of the 162 distinct RG signals are labelled with different colours indicating three 1301 clusters defined in cluster analysis: 1a/b=metabolic syndrome, 2a/b=insulin release versus 1302 insulin action (with additional effects on inflammatory bowel disease for cluster 2a), 1303 3=defects of insulin secretion ( Methods). Asterisks annotate n ovel for glycaemic traits RG 1304 signals. Track 1: RG Manhattan plot reporting -log10(P-value) for RG-GWAS meta-analysis, 1305 signals reaching genome -wide significance ( P-value<510-8) are coloured in red. Crosses 1306 annotate genome-wide significant loci that show evidence of sex heterogeneity ( Methods): 1307 blue crosses indicate signals with larger effects in men, green crosses – signals with larger 1308 effects in women. Track 2: Effects of RG genome-wide significant on four GIP/GLP-1-related 1309 traits GWAS. The colours of the dotted lines indicate four GIP/GLP-1-related traits, grey dot - 1310 signals reaching P-value<0.01 for a GIP/GLP-1-related trait, red dot – lead SNP has significant 1311 effect on GIP/GLP-1-related trait (Bonferroni-corrected P-value<110-4). Track 3: Effects [-1312 log10(P-value)] of lead RG variants in 113 glycan PheWAS. Blue dots - RG lead SNPs, red dots 1313 - lead SNPs reaching P-value<10-4. Track 4: Effects [-log10(P-value)] of lead RG variants in 204 1314 gut-microbiome PheWAS. Light green dots - RG lead SNPs, red dots - variants with significant 1315 effects at P-value<10-4. Track 5: MetaXcan results for 10 selected tissues for RG GWAS meta-1316 analysis ( Methods), signals colocalising with genes ( P-value<510-6) are plotted for each 1317 tissue. (b) Credible set analysis of RG associations in the European meta -analysis. Variants 1318 from each of the RG signal credible sets are grouped based on their posterior probability (the 1319 percentiles labelled on the sides of the bar). SNP variants with posterior probability >80%, 1320 along with their locus names are provided. All variants from the credible set of the primary 1321 signals are highlighted in bold. 1322 1323 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 56 Figure 2. Functional and structural analysis of coding GLP1R variants. 1324 1325 (a) Weighted regression of AST20 β RG estimated in the UKBB study on GLP1R variant mini-Gs 1326 response to GLP-1 stimulation, with correction for variant surface expression, n=5-13. Size of 1327 dots is proportional to the weight (minor allele frequency) in the regression model (Methods). 1328 Error bars represent standard errors for β RG and mini -Gs coupling in response to GLP -1 1329 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 57 stimulation. The grey shaded area corresponds to the 95% confidence interval of the slope of 1330 the regression analysis (β=-0.027, 95%CI[-0.036 –{-0.016}], P-value=0.0001), which explained 1331 65% of the variance in these associations . Variants in red showed no detectable surface 1332 expression (NDE) and are not included in regression analysis. ( b) GLP1R variant mini -Gs 1333 coupling and receptor endocytosis, with surface expression correction, in response to GLP-1, 1334 oxyntomodulin (OXM), glucagon (GCG), exendin-4 (Ex4), semaglutide (Sema) and tirzepatide 1335 (TZP), n=6. Positive deviat ion indicates variant gain -of-function, with statistical significance 1336 inferred when the 95% confidence intervals shown do not cross zero. Responses are also 1337 compared between pathways by unpaired t -test, with * indicating statistically significant 1338 differences. (c) Architecture of the complex formed between the agonist -bound GLP-1R and 1339 Gs; the likely effect triggered by residues involved in GLP -1R isoforms A316T, G168S, and 1340 R421W (in magenta) are reported. ( d) Distributions of the distance between Y242 3.45 side 1341 chain and P3125.42 backbone computed during MD simulations of GLP-1R WT and A316T; the 1342 cut-off distance for hydrogen bond is shown. ( e) Difference in the hydrogen bond network 1343 between GLP1-R WT and A316T. ( f) Analysis of water molecules within the TM D of GLP1 -R 1344 WT and A316T suggests minor changes in the local hydration of position 5.46 (unperturbed 1345 structural water molecule). ( g) Distributions of the distance between position 168 1.63 and 1346 Y1782.48 during molecular dynamics simulations of GLP -1R WT and G168S. ( h) During MD 1347 simulations the GLP-1R isoform S168G showed increased flexibility of ICL1 and H8 compared 1348 to WT, suggesting a different influence on G protein intermediate states. (i) Contact 1349 differences between Gs and GLP -1R WT or W421R; the C termin al of W421R H8 made more 1350 interactions with N terminal segment of Gs  subunit. (j) Mini-Gs and GLP-1R endocytosis 1351 responses to 20 nM exendin-4, plotted against surface GLP-1R expression, from 196 missense 1352 GLP1R variants transiently transfected in HEK293T c ells (n=5 repeats per assay), with data 1353 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 58 represented as mean ± standard error after normalization to wild -type response and log 10-1354 transformation. Variants are categorised as “LoF1” when the response 95% confidence 1355 interval falls below zero or “LoF2” where expression-normalised 95% confidence interval falls 1356 below zero. (k) GLP-1R snake plot created using gpcr.com summarizing the functional impact 1357 of missense variants; for residues with >1 variant, classification is applied as 1358 LoF2>LoF1>tolerated. 1359 1360 Figure 3. Deterioration of glucose homeostasis progressing into type 2 diabetes (T2D) and 1361 leading to complications in multiple organs and tissues - established (left, in peach colour) 1362 and new (right, in green). 1363 1364 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 59 1365 (a) A human figure illustrating the main causes of hy perglycemia (a combination of lifestyle 1366 and genetic factors), and how hyperglycemia affects many organs and tissues. Complications 1367 on the left panel are well established for T2D. Those on the right panel are emerging ones and 1368 are supported by our current analyses. (b) Functional annotation of the RG GWAS results with 1369 DEPICT ( Methods). ( c) Functional annotation of the RG GWAS results with CELLECT 1370 (Methods). 1371 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint 60 Figure 4. Genome-wide genetic correlation between RG and a range of traits and diseases1372 1373 X axis provides the rg genetic correlation values for traits or diseases (Y axis) reaching at 1374 least nominal significance. Correlations reaching a P-value<0.01 are labelled with “ ‘ “, and 1375 those P-value<0.05/239 are labelled with “ * “. 1376 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted April 20, 2021. ; https://doi.org/10.1101/2021.04.17.21255471doi: medRxiv preprint