Low Plasma Adiponectin in Risk of Type 2 Diabetes: Observational Analysis and One- and Two-Sample Mendelian Randomization Analyses in 756,219 Individuals

Adiponectin is a protein-hormone produced and secreted predominately by adipocytes (1–3). Since its discovery in the 1990s, the protein-hormone has attracted attention in the field of type 2 diabetes due to potential insulin-sensitizing, anti-inflammatory, and atherogenic properties suggested in experimental animal models (4–7). In humans, observational studies have unambiguously shown an inverse association between plasma adiponectin and risk of type 2 diabetes (8–12), but investigations regarding a causal relationship have been unclear (13–15). This is an important question to address, as adiponectin could be a potential new drug target for treatment of type 2 diabetes.

A Mendelian randomization (MR) approach takes advantage of natural randomization and uses genetic variants as proxies of modifiable exposures to investigate causal relationships (16). Since alleles are randomly distributed at conception, genetic variants are less likely to be susceptible to reverse causation and confounding (17,18). Genetic variants that specifically associate with low plasma adiponectin have been identified (13,19,20), providing an ideal framework to assess low plasma adiponectin as a causal risk factor for type 2 diabetes in an MR approach (21).

We tested the hypothesis that low plasma adiponectin is observationally and causally associated with increased risk of type 2 diabetes by using an MR approach (Supplementary Fig. 1). For this purpose, we used information on 30,045 individuals in observational analyses and on 96,903 individuals in individual-level one-sample MR from the Copenhagen General Population Study (Supplementary Fig. 2). Also, we used information on 659,316 individuals of European ancestry in two-sample MR from ADIPOGen, Genetic Epidemiology Research on Adult Health and Aging (GERA) Cohort, DIAbetes Genetics Replication And Meta-analysis (DIAGRAM), and UK Biobank (22) (Supplementary Fig. 2).

The Copenhagen General Population Study is a population-based cohort initiated in 2003 with ongoing enrollment (23–25). Individuals aged 20–100 years were randomly selected from the national Danish Civil Registration System to reflect the adult White population of Danish descent. All participants completed a comprehensive questionnaire, underwent a physical examination, and gave blood for biochemical and genetic analyses. Questionnaires were reviewed at the day of attendance by a health care professional together with the participant. The study was approved by a Danish ethics committee (approval no. H-KF-01-144/01; Copenhagen, Denmark) and conducted according to the Declaration of Helsinki. All participants provided written informed consent.

Measurements of total plasma adiponectin with a latex-enhanced turbidimetric immunoassay were conducted with the investigator blind to information on genotypes and type 2 diabetes (measurement range 0.5–40 µg/mL) on a cobas autoanalyzer (Roche). Plasma samples were collected in 2003–2015 and stored at −80°C until 2017–2018 before measurement. The interassay coefficient of variation on a monthly basis was 3.6–5.5% based on daily testing during 12 months of measuring. Information on plasma adiponectin was available for 30,045 individuals.

Genotyping for genetic variants specifically associated with plasma adiponectin concentrations in the ADIPOQ (KNG1 rs2062632, LINC02043 rs266717, ADIPOQ rs68100 75, ADIPOQ rs17366568) and CDH13 (CMIP rs2925979) loci was conducted with blinding to information on plasma adiponectin and type 2 diabetes. We used a targeted selection of genetic variants in the ADIPOQ and CDH13 loci that had the lowest P value and largest effect size in the association with plasma adiponectin according to genome-wide association studies (13,19,20) (Supplementary Table 1). A TaqMan-based method by LCG Genomics (Teddington, U.K.) was used to genotype for rs2062632, rs266717, and rs6810075, and the ABI PRISM 7900HT Sequence Detection System (Applied Biosystems) was used with TaqMan assays to genotype for rs17366568 and rs2925979. R² values for linkage disequilibrium between pairwise combinations of the genetic variants in and around the ADIPOQ locus were all <0.24 (25,26) (Supplementary Table 2). An aggregate genetic score for plasma adiponectin was generated by counting the number of plasma adiponectin decreasing alleles and divided into quartiles. The genetic adiponectin score was internally weighted. Information on all five genetic variants was available for 96,903 individuals.

As a sensitivity analysis, we also used a less restricted approach with eight genetic variants found in the ADIPOGen Consortium to be associated with plasma adiponectin and reaching genome-wide significance level (P value <5 × 10⁻⁸) and not in linkage disequilibrium (R² < 0.0001) (13,27) in a subpopulation from the Copenhagen General Population Study with available Illumina chip data (rs2062 632, rs1108842, rs17366568, rs731839, rs6810075, rs7955 516, rs7615090, and rs12051272). Information on all eight genetic variants was available for 11,449 individuals.

Information on a diagnosis of type 2 diabetes (ICD-8 250 and ICD-10 E11, E13, and E14) was collected from 1977 through December 2018 through review of all inpatient and outpatient hospital contacts identified through the national Danish Patient Registry. All diagnoses in the national Danish Patient Registry are recorded by physicians according to national Danish laws. A diagnosis of type 2 diabetes required a plasma glucose according to changing criteria over time (28). Measurement of nonfasting plasma glucose was also performed at baseline examination using standard hospital assays and was included as a secondary outcome.

BMI was calculated as measured weight in kilograms divided by the square of measured height in meters. Smoking status was categorized as never, former, or current smoker. Cumulative tobacco consumption was calculated in pack-years based on information on duration of tobacco smoking and daily amount of consumed tobacco. Alcohol consumption was reported in units per week (1 unit = 12 g). Socioeconomic status was based on years attending school and annual household income. Degree of leisure-time physical activity was self-reported.

We used genetic information from the ADIPOGen Consortium (13), GERA Cohort, DIAGRAM consortium, and UK Biobank (22) in a two-sample MR analysis. The ADIPOGen Consortium is an international collaboration that seeks to identify genetic variation associated with plasma adiponectin concentration and includes information on 29,347 Europeans (13). Information on type 2 diabetes was assessed from a large genome-wide association study (GWAS) using the GERA Cohort, DIAGRAM consortium, and UK Biobank (22). All genetic variants associated with plasma adiponectin that reached genome-wide significance threshold of P value <5 × 10⁻⁸ were included, and variants in linkage disequilibrium defined as R² > 0.001 were removed using the MR-Base software and MR_Practicals R package using information on linkage structure in 3,775 genomes from the 1000 Genomes Project, as done previously (27). One palindromic genetic variant (rs2980879) was removed due to difficulty harmonizing the effect allele in the exposure sample with the allele in the outcome sample. A total of 10 genetic variants specifically associated with life-long low plasma adiponectin were included from the ADIPOGen Consortium (13) (Supplementary Table 3). From GERA, DIAGRAM, and UK Biobank (22) we harmonized and tested these 10 genetic variants against type 2 diabetes risk for 659,316 Europeans, with 62,892 type 2 diabetes cases (27).

We also investigated whether plasma adiponectin could be a potential mediator between waist-to-hip ratio and risk of type 2 diabetes using multivariable MR (Supplementary Fig. 3) and information on waist-to-hip ratio (Genetic Investigation of ANthropometric Traits [GIANT] consortium) (30), plasma adiponectin (ADIPOGen) (13), and type 2 diabetes (GERA, DIAGRAM, and UK Biobank) (22). Multivariable MR (MVMR) is a variant of instrumental variable analysis which estimates the direct effect of multiple exposures on an outcome using genetic variants as instruments (31).

We used STATA/SE 15.1 and R 3.6.1 for Windows. In one-sample MR, deviation from Hardy-Weinberg equilibrium was investigated using Pearson’s χ² test. Associations with potential confounders were investigated using linear and logistic regressions.

Observational association of plasma adiponectin with risk of type 2 diabetes was investigated using Cox proportional hazards regression with age as timescale (= age adjusted) and left truncation (delayed entry) at study entry for observational analyses, and at birth, or 1977, whichever came last, for genetic analyses. Restricted cubic spline was used, and number of knots for best fit was based on Akaike information criterion (32). Risk of type 2 diabetes was also investigated jointly with plasma adiponectin and other relevant risk factors, like sex, BMI, and waist circumference, where a likelihood ratio test was used to assess for potential effect modification (= interaction) (33). Observational association of plasma adiponectin with plasma glucose was investigated using multiple linear regression and graphically displayed using geometric means. Cumulative incidence of type 2 diabetes according to plasma adiponectin was calculated using competing risk analysis according to Fine & Gray with death and emigration as competing events (34). Observational analyses were adjusted for potential confounders, that is, age, sex, BMI, smoking status, cumulative tobacco consumption, alcohol consumption, education, income, and leisure-time physical activity. Measurement of plasma glucose was nonfasting and analyses with plasma glucose were therefore additionally adjusted for time since last meal. Since some of the participants lacked information on some potential confounders (missing covariates were 1.7%), we performed multivariate imputation using chained equations to fill in missing values; however, results were similar without imputation.

In genetic, causal analyses, associations of genetic variants with plasma adiponectin was investigated using linear regression and graphically displayed using geometric means; we evaluated the strength of the genetic variants as instruments by examining the F-statistic (F >10 indicates sufficient statistical strength) (35). Furthermore, the variance in plasma adiponectin explained by the genetic variants (R²) was investigated using the first stage regression. The association of genetic variants with risk of type 2 diabetes was investigated using logistic regression. In individual-level one-sample MR, instrumental variable analysis with an internally weighted genetic adiponectin score and two-stage least-squares regression was used to investigate the association of plasma adiponectin with plasma glucose using multiple linear regression and risk of type 2 diabetes using multiple logistic regression (36,37); internally weighted genetic adiponectin score means that for each of the five genetic variants, number of adiponectin decreasing alleles (0, 1, or 2) was multiplied by the β-coefficient for the association of each genetic variant on plasma adiponectin, with all five numbers each consisting of number of adiponectin decreasing alleles times β-coefficient then being added for each individual. Genetic analyses were adjusted for age and sex in the second stage regression only, as genes are largely unconfounded. To investigate potential pleiotropy in individual-level one-sample MR and to compare with results from two-sample MR summary data, different methods of instrumental variable analysis, including inverse-variance weighted (IVW), MR-Egger, weighted median estimates, and weighted mode regressions, were used (37,38). To have a comparable unit of plasma adiponectin from one- and two-sample MR analyses, we naturally log-transformed plasma adiponectin in the Copenhagen General Population Study from µg/mL to log(µg/mL).

For two-sample MR we used the MR_Practicals R package (including MRInstruments and TwoSampleMR) (27) for performing corresponding analyses.

Power calculations for individual-level and two-sample MR analyses were performed using an online power calculator to determine the causal effect we can detect with 80% power, as done previously (39).

We also investigated if plasma adiponectin could be a potential mediator between waist-hip-ratio and risk of type 2 diabetes using MVMR (Supplementary Fig. 3). First, we used the MR-Base platform and MR_Practicals R package (including MRInstruments and TwoSampleMR) (27) to retrieve information on waist-to-hip ratio (GIANT) (30), plasma adiponectin (ADIPOGen) (13), and type 2 diabetes (GERA, DIAGRAM, and UK Biobank) (22). Second, we used the full set of genetic variants associated with both waist-to-hip ratio and plasma adiponectin with information on both exposures and the outcome in an MVMR using the MVMR R package (31) to estimate the direct effect of waist-to-hip ratio and plasma adiponectin on type 2 diabetes. Third, we used univariate MR with a subset of genetic variants only associated with waist-to-hip ratio to estimate the total effect of waist-to-hip ratio on type 2 diabetes. Fourth, the indirect effect was found with subtraction of the direct effect from the total effect. Fifth, the proportion mediated through plasma adiponectin was found in dividing the indirect effect by the total effect.

Data from the Copenhagen General Population Study can be made locally accessible under controlled conditions from the corresponding author upon reasonable request.

For two-sample MR analysis, data are available through the MR-Base platform for two-sample MR using summary data from the GWAS Catalog (27,40). In the current study we used data from ADIPOGen (identifier ieu-a-1) (13); from GERA, DIAGRAM, and UK Biobank (identifier ebi-a-GCST006867) (22); and from GIANT (identifier ieu-a-73) (30).

Plasma adiponectin was associated with potential confounders (Table 1, upper panel). In contrast, the genetic adiponectin score was not associated with any potential confounders (Table 1, lower panel). There was no evidence of deviation from Hardy-Weinberg equilibrium (all P values ≥0.05).

In observational analyses, 1,751 individuals were diagnosed with type 2 diabetes during a median follow-up time of 9.2 years (from 2003 until 2018) among 29,038 Copenhagen study individuals. In genetic analyses, 5,012 diagnosed with type 2 diabetes from 1977 to 2018 among 96,903 Copenhagen individuals.

Low plasma adiponectin was associated with elevated plasma glucose and increased risk of incident type 2 diabetes after multivariable adjustment for potential confounders (Figs. 1–3).

In comparisons with individuals with a median plasma adiponectin of 28.9 µg/mL (4th quartile), multivariable-adjusted hazard ratios (HRs) for type 2 diabetes were 1.42 (95% CI 1.18–1.72) for 19.2 µg/mL (3rd quartile), 2.21 (1.84–2.66) for 13.9 µg/mL (2nd quartile), and 4.05 (3.38–4.86) for individuals with a median plasma adiponectin of 9.2 µg/mL (1st quartile) (Fig. 2, right panel). Corresponding cumulative incidences for type 2 diabetes at age 70 years were 3%, 7%, 11%, and 20%, respectively (Fig. 3).

The genetic variants in the ADIPOQ and CDH13 loci were as expected associated with stepwise lower plasma adiponectin (Fig. 4); the genetic adiponectin score explained 2.0% of the variation in plasma adiponectin, with an F value of 577. Geometric mean plasma adiponectin was 19.9 µg/mL (95% CI 19.7–20.1) for individuals in the 1st quartile of the genetic adiponectin score, 18.9 µg/mL (18.7–19.1) for individuals in the 2nd quartile, 18.0 µg/mL (17.8–18.2) for individuals in the 3rd quartile, and 16.4 µg/mL (16.2–16.6) for individuals in the 4th quartile (Fig. 4). There was no association between individual genetic variants for plasma adiponectin and risk of type 2 diabetes except for rs2925979, which was associated with higher risk of type 2 diabetes (Fig. 4). In comparisons with individuals in the 1st quartile of the genetic adiponectin score, multivariable-adjusted HRs were 1.00 (95% CI 0.93–1.09) for individuals in the 2nd quartile, 1.03 (0.95–1.12) for individuals in the 3rd quartile, and 1.04 (0.96–1.12) for individuals in the 4th quartile (Fig. 4).

In instrumental variable analysis, a genetically determined 1 µg/mL lower plasma adiponectin was associated with a 0.004 mmol/L (95% CI −0.002 to 0.009) higher plasma glucose and a causal risk ratio of 1.01 (95% CI 0.99–1.03) for type 2 diabetes (Fig. 5 and Supplementary Fig. 4). Correspondingly, in observational analyses, a 1 µg/mL lower plasma adiponectin was associated with 0.011 mmol/L (95% CI 0.014–0.009) higher plasma glucose and a HR of 1.07 (95% CI 1.06–1.09) for type 2 diabetes (Fig. 5 and Supplementary Fig. 4). Genetic and observational results regarding risk were similar for women and men separately, genetic for those with BMI < and ≥25 kg/m², and genetic for those with waist circumference < and ≥88/102 cm (all P values for interaction ≥0.05); however, risk of type 2 diabetes by observational results was most pronounced in those with ≥25 kg/m² and waist circumference ≥88/102 cm (P values of 2 × 10⁻⁷ and 0.001, respectively) (Fig. 5). Albeit with lower statistical power, results were similar with use of five or eight genetic variants for plasma adiponectin (Supplementary Fig. 4).

The variance in plasma adiponectin explained by the genetic variants was 8.0% in two-sample MR with use of ADIPOGen Consortium for the exposure cohort and GERA, DIAGRAM, and UK Biobank for the outcome cohort.

In two-sample MR, a 1-unit log-transformed lower plasma adiponectin was associated with a causal risk ratio of 1.26 (95% CI 1.01–1.57) for type 2 diabetes with inverse variance–weighted analysis, 1.00 (0.68–1.48) with MR-Egger, 1.13 (0.99–1.26) with weighted median, and 1.12 (0.99–1.27) with weighted mode (Fig. 6, upper panel). Corresponding causal risk ratios in individual-level one-sample MR were 1.13 (0.83–1.53), 0.72 (0.34–1.50), 1.07 (0.73–1.57), and 1.00 (0.66–1.51), respectively (Fig. 6, lower panel). The I²_GX, a measure of the variability in instrument strength across variants, was 97.2%, indicating that the MR-Egger estimate was less likely to be biased (41).

The MR-Egger intercept in two- and individual-level one-sample MR was, respectively, −0.02 (P value = 0.20) and −0.03 (P value = 0.19), thereby indicating that potential average pleiotropic effects of the genetic variants were balanced (Supplementary Fig. 5).

Power calculation indicated that we had 80% power to detect a causal risk ratio of 1.05 per SD change in plasma adiponectin in two-sample MR analyses and 1.35 per SD change in plasma adiponectin in individual-level MR analyses, respectively (Supplementary Table 4).

With data from GIANT, ADIPOGen, GERA, DIAGRAM, and UK Biobank, the proportion of genetically determined waist-to-hip ratio mediated through plasma adiponectin on risk of type 2 diabetes was 4.8% (95% CI 2.3%, 12%) (Supplementary Tables 5–7).

Plasma adiponectin was inversely associated with plasma glucose and risk of type 2 diabetes in observational analyses. At age 70 years, type 2 diabetes incidence was 20% vs. 3% in individuals with a median plasma adiponectin of 9.2 vs. 28.9 µg/mL. Although the evidence to support a genetic, causal relationship from plasma adiponectin to plasma glucose or risk of type 2 diabetes in individual-level one-sample MR analyses in the Copenhagen General Population Study was not convincing, using two-sample MR analysis with ADIPOGen, GERA, DIAGRAM, and UK Biobank we cannot exclude a causal association between low plasma adiponectin and increased risk of type 2 diabetes. However, this potential causal association (from univariable two-sample MR) attenuated/lost significance with inclusion of waist-to-hip ratio into the model using MVMR. This is a novel finding, as is subgroup analyses in our Copenhagen studies with use of individual participant data documenting the strongest association observationally in overweight and obese individuals.

Mechanistically, adiponectin may play a role in the pathogenesis of type 2 diabetes. The development of type 2 diabetes is characterized by a decline in both insulin production in the β-cells and insulin sensitivity in the peripheral tissues. Experimental animal models indicate that adiponectin may affect glucose and lipid metabolism, inflammation, and oxidative stress through activation of 5′ AMPK and peroxisome proliferator–activated receptor α (PPARα) in signal transduction by the adiponectin receptors AdipoR1 and AdipoR2 located on liver and skeletal muscle cells, thereby suppressing gluconeogenesis in liver cells, enhancing fatty acid oxidation in liver and muscle cells, and increasing glucose uptake in muscle cells (4–7). However, the evidence to support that plasma adiponectin plays a causal role in plasma glucose variation or risk of type 2 diabetes in humans according to MR analyses in the Copenhagen General Population Study is weak, which may be explained by a lower power to detect a causal association in the individual-level setting. In two-sample MR with ADIPOGen Consortium and GERA, DIAGRAM, and UK Biobank, the genetic evidence trended toward a causal relationship, i.e., the inverse variance–weighted analysis indicated a protective effect of plasma adiponectin, while other methods including MR-Egger, weighted median, and weighted mode did not support a causal association. The MR-Egger method takes potential pleiotropy into account by letting the estimated regression line be different from zero. However, the intercept was not significantly different from zero, indicating that potential average pleiotropic effects of the genetic variants were balanced.

The strong observational association may also be explained by residual confounding in the form of plasma adiponectin being a marker for another causal risk factor in type 2 diabetes pathogenesis such as adiposity or insulin resistance (15,42). Recently, we found no evidence that BMI, often used as a marker for general adiposity, is causally associated with plasma adiponectin in bidirectional one and two-sample MR analyses in 460,397 individuals (25). However, in another study that used bidirectional two-sample MR analyses including 210,088 individuals from GIANT and ADIPOGen, individuals genetically predisposed to abdominal fat had a lower plasma adiponectin concentration, whereas individuals genetically predisposed to gluteofemoral fat accumulation had a higher plasma adiponectin concentration (42). Higher waist circumference has previously been causally linked to an increased risk of type 2 diabetes (43). Two-sample MVMR to estimate the proportion of effect of waist-to-hip ratio on type 2 diabetes and mediated through plasma adiponectin did not indicate a mediating effect of plasma adiponectin. Thus, the strong inverse observational association between plasma adiponectin and risk of type 2 diabetes could be explained by plasma adiponectin being a biomarker of abdominal fat in type 2 diabetes (42). However, MR studies have also demonstrated an association between low plasma adiponectin and high insulin resistance in 1,157 individuals, measured by HOMA of insulin resistance, and between high plasma adiponectin and high insulin sensitivity in 942 individuals, measured by euglycemic insulin clamp (14,44). In contrast, in a larger MR study in 15,788 individuals, investigators found no evidence of a causal effect of low plasma adiponectin on insulin resistance. However, they were not able to exclude that high fasting insulin decreases plasma adiponectin, suggesting that the observed association between fasting insulin and plasma adiponectin may be due to a causal effect of high fasting insulin on plasma adiponectin and not vice versa (15).

Previous observational studies have unanimously shown an inverse association between plasma adiponectin and risk of type 2 diabetes (8–12,45–48). In a GWAS on plasma adiponectin in 45,891 individuals, a multi–single nucleotide polymorphism genotypic risk score based on plasma adiponectin decreasing alleles in 22,044 individuals was associated with increased risk of type 2 diabetes (13). However, the multi–single nucleotide polymorphism genotypic risk score suggested association through horizontal genetic pleiotropy and not through plasma adiponectin. Recently, in a two-sample MR analysis using summary data from five consortia including ADIPOGen, the Meta-Analyses of Glucose and Insulin-related traits Consortium (MAGIC), DIAGRAM, GERA, and the UK Biobank with a total of 659,316 individuals, investigators found no evidence of a causal relationship of plasma adiponectin with glucose homeostasis, as measured by fasting glucose, fasting insulin, HOMA of insulin resistance, and HOMA of β-cell function, or type 2 diabetes; however, more than one-half of the genetic variants were proxies, which may produce unreliable results (49). Taking these findings together and including the present findings from the Copenhagen General Population Study and two-sample MR using ADIPOGen, GERA, DIAGRAM, and UK Biobank, we cannot exclude a causal effect of low plasma adiponectin on increased risk of type 2 diabetes. Thus, even larger MR studies on this important field are warranted.

Potential limitations in MR analyses include population stratification bias, genetic pleiotropy, linkage disequilibrium, and weak instrument bias. However, in the two-sample MR both the exposure and outcome cohorts were of European ancestry with no sample overlap and in the Copenhagen General Population Study we had an ethnically homogenous population; hence, population stratification bias seems unlikely. Since the distribution of genetic variants did not appear to differ from Hardy-Weinberg equilibrium, genotyping and population sampling errors in the Copenhagen cohort also seem unlikely. Furthermore, there was no evidence of linkage disequilibrium between the chosen genetic variants. Nonetheless, weak instrument bias cannot be completely excluded, as our genetic variants explained ∼2% of the variation in plasma adiponectin concentration in the Copenhagen cohort, which could introduce a potential type 2 error. By virtue of using large consortia in two-sample MR with a higher number of individuals and type 2 diabetes cases, and 10 genetic variants explaining ∼8% of the variation in plasma adiponectin, a potential type 2 error was likely minimized. Further, it has been suggested that MR-Egger regression performs poorly in a one-sample setting with bias of the MR-Egger estimate toward that of the confounded observational association (41). However, this bias is attenuated when I²_GX is high (>97%), as in the case of the Copenhagen cohort. Also, in the Copenhagen cohort the direction of the MR-Egger estimate was opposite that of the observational association. A limitation in two-sample MR is the use of summary data without individual participant data, which can hinder subgroup analyses and further adjustments. However, individual participant data from the Copenhagen General Population Study were used in individual-level MR.

Strengths of the current study include use of both individual-level one-sample and two-sample MR analyses, thereby circumventing confounding and reverse causation, and the relatively large sample sizes, thereby reducing risk of spurious associations. It is also a strength that the observational and individual-level MR analyses were conducted in a single homogenous cohort where plasma adiponectin and type 2 diabetes diagnosis were ascertained with the same method for all individuals. Further, the advantage of having individual participant data in the Copenhagen cohort made it possible to investigate subgroup associations, while two-sample summary data provided greater power with a large sample size and number of type 2 diabetes cases, along with genetic variants explaining ∼8% of the variation in plasma adiponectin. Furthermore, two methods were used for selection of genetic variants: a targeted approach in selecting variants likely to mediate the effect via a functional effect on plasma adiponectin in one-sample MR versus a broad untargeted approach using genetic variants associated with plasma adiponectin from GWAS in two-sample MR and sensitivity analysis.

In conclusion, low plasma adiponectin is associated with increased risk of type 2 diabetes, an association that could represent a causal relationship.

Acknowledgments. The authors thank participants and staff of the Copenhagen General Population Study.

Funding. The submitted project and M.B.N. are supported by the Research Foundation for Health Research of the Capital Region of Denmark and Managing Director Kurt Bønnelycke and Mrs. Grethe Bønnelyckes Foundation.

Duality of Interest. Y.Ç. reports personal fees from Boehringer Ingelheim, AstraZeneca, and Sanofi Genzyme outside of the submitted work. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. M.B.N., Y.Ç., M.B., and B.G.N. contributed to the study concept and design. M.B.N., Y.Ç., M.B., and B.G.N. collected, analyzed, or interpreted the data. M.B.N. wrote the draft manuscript. M.B.N. and Y.Ç. performed the statistical analyses. M.B.N., Y.Ç., M.B., and B.G.N. revised the manuscript for important intellectual content. M.B.N. and B.G.N. obtained funding. B.G.N. provided administrative, technical, and material support. Y.Ç., M.B., and B.G.N. supervised the study. M.B.N., Y.Ç., and B.G.N. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in E-poster form with preliminary results at the European Atherosclerosis Society Congress 2021, 30 May–2 June 2021. Parts of this study were also presented in poster form at the Danish Diabetes Academy (DDA) Summer School on Diabetes & Metabolism 2021, 30 August to 2 September 2021.