Insulin Response to Oral Glucose and Cardiometabolic Disease: A Mendelian Randomization Study to Assess Potential Causality Free

Obesity is a major public health concern. Cardiovascular disease is the leading cause for mortality in people with obesity and is likely modulated by atherogenic cardiometabolic risk factors including insulin resistance (IR), type 2 diabetes (T2D), dyslipidemia (increased atherogenic triglyceride [TG]-rich lipoproteins and reduced HDL), nonalcoholic fatty liver disease (NAFLD), and hypertension (HTN) (1). Postprandial insulin hypersecretion is postulated to contribute to obesity by increasing adipogenesis and via indirect effects on satiety (2–6). Two-sample Mendelian randomization (MR) can be used to infer causality between an exposure and an outcome using genetic variants associated with the exposure as an instrument (7).

Findings from a prior MR study supported a causal association between insulin levels at 30-min post–glucose challenge and BMI (4). In this analysis insulin concentration was not adjusted for plasma glucose, which is a major stimulator of insulin secretion (4,8,9). Corrected insulin response (CIR) 30 min after glucose ingestion is a measure of insulin secretion independent of glucose at baseline, and that time point and is likely a better measure of insulin hypersecretion (9). Whether CIR increases BMI is not established, and its impact on cardiometabolic disease is also not established. Impaired adipogenesis and reduced adipose storage capacity underpin cardiometabolic disease, predisposing to ectopic lipid deposition in liver, skeletal muscle, and pancreas and consequently IR, T2D, dyslipidemia, and HTN, which are likely causal risk factors for coronary artery disease (CAD) (10–12). Reduced adipose storage and impaired adipogenesis phenotypically manifest with centripetal adiposity and/or reduced femoro-gluteal adiposity, i.e., increased waist-to-hip ratio (WHR) (10–12). As insulin stimulates adipogenesis (13), postprandial hyperinsulinemia may plausibly protect against cardiometabolic disease despite weight gain.

In a previous meta–genome-wide association study (GWAS), investigators identified genome-wide significant single nucleotide polymorphisms (SNPs) associated with CIR (14). Here, for our primary analyses, we have undertaken bidirectional MR analysis on summary-level data from participants of European descent to assess potential causal relationships between increased CIR, BMI, and cardiometabolic phenotypes. We also undertook secondary analyses to assess whether insulin level 30 min after glucose challenge (unadjusted for plasma glucose) and CIR adjusted for insulin sensitivity influence cardiometabolic phenotypes.

Details of the cohorts can be found in Table 1. Informed consent and institutional approval were previously obtained by the individual cohort investigators. Our main exposure (CIR) was assessed in participants of European descent. We therefore selected the largest GWAS undertaken exclusively in participants of European descent for outcomes.

The exposure included 13 discovery GWAS, which included Diabetes Genetics Initiative, Helsinki Birth Cohort Study, and Uppsala Longitudinal Study of Adult Men (ULSAM) (n = 6,382 participants of a total of 26,037 participants [24.5% of exposure cohort]), which were also part of the Genetic Investigation of ANthropometric Traits (GIANT) consortium (2.8% of GIANT consortium). Some U.K. participants from the Ely and Relationship between Insulin and Cardiovascular Study (RISC) (∼2,000 participants [∼0.4%]) may potentially have also participated in the UK Biobank.

We used bidirectional MR to assess causal associations between CIR (CIR = 100 × insulin at 30 min) / (glucose at 30 min × glucose at 30 min − 3.89) (14), and cardiometabolic phenotypes from summary statistics from published GWAS. We incorporated recommendations from recently published Strengthening the Reporting of Observational Studies in Epidemiology using Mendelian Randomization (STROBE-MR) reporting guidelines (15) (checklist in Supplementary Material). The instrument for CIR was based on eight genome-wide significant SNPs associated with CIR in the Meta-Analysis of Glucose- and Insulin-related traits Consortium (MAGIC) (14), in up to 26,037 participants without T2D from nine cohorts. The strands of the palindromic rs10830963 did not align, and therefore it was excluded, leaving seven SNPs in the instrument (Table 2). The instruments for the cardiometabolic phenotypes were derived from the cohorts listed in Table 1. We also assessed whether CIR affects WHR (adjusted and unadjusted for BMI), leg fat (assessed by impedance), liver fat (assessed by MR), T2D, plasma TG and HDL, HTN, and CAD (Table 1). A P value of <0.05 was considered significant for each analysis.

The first assumption is that the instrument is associated with the exposure; therefore, we used SNPs that were associated with the exposure at genome-wide significance. The second assumption is that the instrument does not influence the outcome via another pathway other than the outcome (horizontal pleiotropy). Third, there are no confounders associated with the instrument (7).

Univariable MR using an inverse variance–weighted (IVW) approach was conducted with the TwoSampleMR package in R (RStudio, version 1.3.1073, and R, version 4.0.3) (7) to assess potential causality between traits. If the SNP was not matched directly, a proxy in linkage disequilibrium (LD) (r² > 0.8) was selected. Plots were generated with the ggplot2 and metaphor packages. MR analyses were primarily undertaken with IVW and MR-Egger, with additional sensitivity analyses including weighted median and weighted mode. Cochran Q test was used to assess heterogeneity (7). Leave-one-out analyses were undertaken to assess if any MR estimate was biased by a single SNP potentially with horizontal pleiotropic effect (7).

IVW MR was used to assess causal associations through meta-analysis of the individual Wald ratio for each SNP. MR-Egger relaxes the assumption of no horizontal pleiotropy and adapts the IVW to permit a nonzero intercept. It returns an unbiased causal estimate, in the presence of horizontal pleiotropy, providing that the horizontal pleiotropic effects are not correlated with the SNP-exposure effects (InSIDE assumption) (7,16). For the weighted median MR, the median effect of all SNPs in the instrument was used for analysis permitting SNPs with greater effect to larger contribution by weighting the contribution of each SNP by the inverse variance of its association with the outcome (17). This is a robust method even if only 50% of the SNPs satisfy all three MR assumptions (17). In weighted-mode MR, SNPs are clustered into groups based on similarity of causal effects and the causal effect estimate is returned based on the cluster with the largest number of SNPs (18).

All data have been included in the manuscript file and Supplementary Material.

We did not find evidence that BMI, WHR, liver fat, TG, HDL, HTN, and CAD causally impact CIR. MR indicates that T2D reduces CIR (b = −0.22 ± 0.04, P = 1.3 × 10⁻⁷) (Supplementary File 1 and Supplementary Table 1).

Increased CIR may increase BMI (IVW b = 0.05 ± 0.02, P = 0.03, and MR-Egger b = 0.17 ± 0.05, P = 0.02), with evidence of heterogeneity (I² = 90.6%) (Table 3 and Fig. 1). Leave-one-out analyses indicate that exclusion of either rs11671664 or rs7756992 leads to a nonsignificant association. We conducted additional phenome-wide association studies analyses of these SNPs. rs11671664 is an expression quantitative trait locus (eQTL) for GIPR (encoding the receptor for glucose-dependent insulinotropic polypeptide) in Epstein-Barr virus–transformed lymphocytes as well as SYMPK (basal ganglia, thyroid, esophagus and tibial artery, testis), DMPK and DMWD (cerebellum), and EML2-AS1 (fibroblasts) (https://gtexportal.org/home/snp/rs11671664, accessed 12 December 2021) (19). The insulin-increasing allele is associated with increased BMI and reduced T2D and is in LD with rs10423928 (D′: 1, R² = 0.39) (https://ldlink.nci.nih.gov/?var1=rs10423928&var2=rs11671664&pop=GBR&genome_build=grch37&tab=ldpair, accessed 21 January 2022) (20), which is associated with lower 2-h glucose after oral glucose tolerance test (OGTT) (21) (Supplementary Table 2). The insulin-increasing allele of rs7756992 is associated with higher birth weight, reduced glycated hemoglobin (HbA_1c), and T2D at genome-wide significance (Supplementary Table 3). No significant eQTLs have been reported for rs7756992 (https://gtexportal.org/home/snp/rs7756992, date accessed December 12, 2021) (19).

We did not find evidence that CIR causally impacts WHR (IVW b = 0.01 ± 0.02, P = 0.69) (Supplementary File 2). Similar results were found for BMI-adjusted WHR (Supplementary File 2).

Increased CIR may increase left leg fat percentage (IVW b = 0.03 ± 0.01, P = 0.02, and MR-Egger b 0.09 ± 0.003, P = 0.03), with evidence of heterogeneity (I² = 80.4%). Leave-one-out analyses indicate that exclusion of either rs11671664 or rs7756992 leads to a nonsignificant association. Similar results were found for right leg fat percentage (Supplementary File 2, Table 3, and Fig. 2).

We did not find evidence that CIR causally impacts liver fat (IVW b = −0.08 ± 0.04, P = 0.06) (Supplementary File 2).

Increased CIR may reduce T2D (IVW b = −0.73 ± 0.15, P = 6 × 10⁻⁷, and odds ratio [OR] 0.52 [95% CI 0.42–0.66]). MR-Egger analyses was not significant: b = −1.3 ± 0.55, P = 0.08, and OR 0.27 (0.09–0.8). There was evidence of heterogeneity (I² = 94.5%) (Table 4 and Fig. 3).

Increased CIR may reduce TG (IVW b = −0.07 ± 0.02, P = 0.003), but the association was not significant with MR-Egger analyses (b = −0.17 ± 0.07, P = 0.05). Increased CIR may increase HDL (IVW b = 0.04 ± 0.017, P = 0.006), but MR-Egger analyses were nonsignificant (b = 0.12 ± 0.05, P = 0.05). There was evidence of heterogeneity for both TG and HDL (I² = 83.0% and 83.4%, respectively) (Table 3 and Fig. 4).

We did not find evidence that CIR causally impacts HTN (IVW b = 0.001 ± 0.004, P = 0.7, and OR 1.00 [95% CI 0.99–1.01]) (Supplementary File 2). Similar nonsignificant results were found for systolic and diastolic blood pressure (Supplementary File 2).

We did not find evidence that CIR causally impacts CAD (IVW b = −0.002 ± 0.002, P = 0.48, and OR 0.99 [95% CI 0.81–1.21]) (Supplementary File 2). A similar nonsignificant association was found for major coronary heart disease event (Supplementary File 2).

The CIR-reducing allele at rs933360, near GRB10, has been noted to have stronger effects in women for CIR (women b = −0.11 ± 0.02, P = 1.52 × 10⁻⁸, men b = −0.04 ± 0.01, P = 0.001; sex heterogeneity P = 0.002) with greater effects seen if inherited from the father (14). The outcomes were not significantly different with leave-one-out analyses for BMI, leg fat, T2D, lipids, and liver fat.

We also undertook additional bidirectional MR analyses of CIR adjusted for insulin sensitivity and cardiometabolic phenotypes. Although CIR-associated SNPs are nominally associated with CIR adjusted for insulin sensitivity (14), with a P value threshold of <5 × 10⁻⁸ only three SNPs were associated with this phenotype after LD pruning and harmonization. Of these, two were also associated with CIR, with similar effect size (rs4502156, rs7756992). An additional SNP (rs1111875) is in LD with rs7923866 (D′: 0.97, R² = 0.79) in the U.K. population (https://ldlink.nci.nih.gov/?tab=ldpair, accessed 12 April 2022) (Supplementary Table 4), which is associated with CIR at genome-wide significance. IVW analyses were concordant with the analyses with CIR, with potential causal association between increased CIR and increased leg fat and reduced T2D and TG, while T2D likely reduces CIR adjusted for insulin sensitivity (Supplementary Files 3 and 4, Supplementary Tables 5–7). We did not find evidence that increased CIR adjusted for insulin sensitivity increased BMI consistent with the leave-one-out MR analysis assessing the effect of CIR on BMI: rs11671664 was excluded from the instrument for CIR adjusted for insulin sensitivity as it was not associated with this parameter at genome-wide significance.

A previous MR analysis had indicated that increased insulin at 30-min post–glucose challenge unadjusted for glucose increases BMI (4). The three genome-wide significant SNPs associated with 30-min insulin are in LD with three SNPs associated with CIR in Europeans: rs2287019 and rs11671664 (D′ = 0.7694; R² = 0.2906), rs742642 and rs7756992 (D′ = 0.9551, R² = 0.3277), and rs1111875 and rs7923866 (D′ = 0.9631, R² = 0.7539) (https://ldlink.nci.nih.gov/?tab=ldpair, accessed 21 December 2021) (20). We undertook additional MR analyses with an instrument comprising these three SNPs and did not find any evidence of adverse cardiometabolic traits associated with increased 30-min insulin post–glucose challenge (Supplementary File 5).

Impaired adipogenesis and adipose storage, in the setting of weight gain, underlies cardiometabolic disease (10,12,22). Although postprandial hyperinsulinemia has been implicated in weight gain (4), it can increase adipogenesis (13), which may protect against cardiometabolic disease. Consistent with that, our data suggest that increased CIR might modestly increase BMI with favorable fat distribution (increased leg fat mass with no change in WHR or liver fat), reduce T2D, and improve lipids (reduced TG and increased HDL) with neutral effects on HTN and CAD. The phenotype of increased BMI with improved metabolic parameters is also seen with PPARγ agonist treatment, which promotes adipogenesis/adipose storage capacity with favorable fat deposition (23). Further, genetic variants that reduce adipose storage capacity increase risk of metabolic disease at lower BMI (10,12,22), underscoring the role of adipose storage as a major determinant of cardiometabolic disease.

Accounting for multiple testing and potential pleiotropic/indirect effects of SNPs in the instrument, caution is warranted in interpreting these data. The cardiometabolic outcomes assessed were hypothesis driven and correlated. There are currently no consensus methods for assessment of significance thresholds in this setting. With use of a Bonferroni-corrected P value threshold of 0.0042 (after correction for 12 tests), which is likely overconservative, the results of the analyses support potential causal association between increased CIR and reduced T2D and TG. Leave-one-out analyses suggest that the strongest effects on BMI were mediated by rs11671664 and rs7756992, of which the latter may indirectly influence adult BMI via effects on birth weight (24,25). rs11671664 is an eQTL for GIPR, a target for weight loss pharmacotherapy (26). GIP can increase weight through insulin-stimulated adipogenesis (27). This loci is also associated with 2-h glucose post-OGTT (21), which if decreased might increase hunger and food intake and thus indirectly increase BMI (28). Notably, carriers of a protein-truncating variant in the skeletal muscle–specific isoform of TBC1D4 pArg684Ter (rs61736969, minor allele frequency 17% among Inuits in Greenland) (29) manifest marked postprandial hyperinsulinemia and hyperglycemia but not increased BMI, suggesting lower postprandial glucose might be an indirect contributor to hyperinsulinemia-mediated weight gain. Further highlighting the complexity and pleiotropic metabolic effects of the GIPR locus, the A allele of rs10423928, which is in LD with rs11671664, is associated with reduced insulin secretion in vitro and in vivo in response to glucose and GIP and reduced BMI. However, despite decreased insulin secretion, the risk of T2D is almost completely attenuated, likely due to improved insulin sensitivity, which may be due to reduced expression of the proinflammatory cytokine osteopontin in adipose tissue (30,31).

The reduction in T2D with increased CIR, as assessed with MR (with IVW, weighted median, weighted mode, and simple mode), is expected, as the insulin-reducing alleles in the MR instrument are associated with T2D and/or increased glycemia at genome-wide significance (14). Further, our analysis suggests that T2D reduces CIR. These analyses are in line with previous genetic data that indicate that increased insulin secretion protects against T2D (14,32). MR Egger assessment did not support a causal association between increased CIR and reduced T2D and improved lipids. This method can yield an unbiased causal estimate in the presence of horizontal pleiotropy, providing the horizontal pleiotropy is not related to the SNP-exposure effects (InSIDE assumption) (7,16): this assumption may not be valid as enhanced CIR may favorably influence T2D risk and lipids by increasing thigh fat. Some of the SNPs used in the CIR instrument may also have effects on T2D independent of CIR: rs933360 and rs4502156, for example, are associated with fasting glucose (33,34).

We did not find evidence that increased CIR has a primary role in cardiometabolic disease. However, in the context of IR, we posit that hyperinsulinemia may have deleterious effects (22,35,36). IR is associated with increased lipid flux to liver, which likely contributes to NAFLD and increased TG/reduced HDL, a phenotype not seen in patients with severe IR due to loss-of-function mutations in the insulin receptor (36). This indicates that hyperinsulinemia in the presence of increased lipid flux is likely necessary for NAFLD and dyslipidemia in more common forms of IR (36). Prior work has also demonstrated that postglucose insulin hypersecretion (after adjustment for insulin sensitivity) in people without T2D was associated with increased BMI and increased risk of future T2D (37). However, these individuals had hepatic IR, higher steady-state free fatty acid, higher fasting TG, and reduced glucose tolerance at baseline (37). This hypothesis awaits confirmation with mechanistic studies. Given the heterogeneity of human metabolic disease, we cannot definitively exclude a primary role of postprandial insulin hypersecretion in metabolic disease in a subset of patients.

The neutral effects of increased CIR on CAD despite potential beneficial effects on T2D and lipids are also noteworthy. Some aspects of insulin signaling are increased and some reduced in the vasculature in animal models of IR, and it is postulated that some of the deleterious effects are mediated by hyperinsulinemia (38). Whether the effects of hyperinsulinemia attenuate the athero-protective effects of reduced T2D and dyslipidemia merits further investigation.

Strengths of the study include the detailed analysis of cardiometabolic phenotypes associated with CIR, in conjunction with the large sample sizes. There was no substantial overlap between participants in the exposure and outcome cohorts.

The study has a number of limitations. It included individuals of European descent, and therefore findings may not apply to other ethnic groups, especially as some develop cardiometabolic disease at lower BMI (39). We did not undertake analysis by sex. There is growing evidence of the modulation of genetic variants influencing body composition and metabolic parameters by sex (33). Prior data have shown that the effect of rs933360 on insulin secretion is greater in women and the effect on glycemic parameters is influenced by whether its inheritance is paternal or maternal (14). Leave-one-out analyses indicate that this variant did not significantly impact outcomes in this study. Unlike other complex traits, relatively few SNPs have been associated with post-OGTT insulin (14), which is likely reflective of smaller sample sizes due to a more in-depth and time-consuming procedure. Identification of more genome-wide significant variants associated with CIR through deployment of larger sample sizes with updated imputation (40) may permit more robust inferences, given the pleiotropic effects of some SNPs in the MR instrument. Further, we do not have longitudinal data on participants. Many people with increased BMI transition to more unhealthy metabolic phenotypes over time (41). We did not assess the interaction between dietary macronutrient intake and CIR. Further studies exploring this association will be informative.

In summary, we did not find evidence that increased CIR has deleterious cardiometabolic effects. Increased CIR might increase BMI with favorable fat distribution, reduce T2D, and improve lipids. It has neutral effects on liver fat, HTN, and CAD. Based on this and previous data, we suggest that postprandial hyperinsulinemia per se is likely not deleterious to cardiometabolic health in the absence of compromised adipose storage and IR.

Acknowledgments. The Genotype-Tissue Expression (GTEx) Project was supported by the Common Fund of the Office of the Director of the National Institutes of Health and by National Cancer Institute, National Human Genome Research Institute, National Heart, Lung, and Blood Institute, National Institute on Drug Abuse, National Institute of Mental Health, and National Institute of Neurological Disorders and Stroke: https://gtexportal.org/home/snp/rs11671664 (accessed 21 December 2021).

Funding. S.D. is funded by CIHR, Heart & Stroke Foundation of Canada, Diabetes Canada, and Banting & Best Diabetes Centre (DH Gales Family Charitable Foundation New Investigator Award and a Reuben & Helene Dennis Scholar in Diabetes Research).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. A.N., R.K., A.D.P., and S.D. designed the study. All authors analyzed data. R.K., A.M., and S.D. wrote the manuscript, and D.R. and A.D.P. read and edited the manuscript. S.D. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.