Physical Activity, Sedentary Behavior, and Type 2 Diabetes: Mendelian Randomization Analysis Shuai Yuan,1 Xue Li,2 Qianwen Liu,3 Zhe Wang,4 Xia Jiang,3 Stephen Burgess,5,6 and Susanna C. Larsson1,7 1Unit of Cardiovascular and Nutritional Epidemiology, Institute of Environmental Medicine, Karolinska Institutet, Stockholm, 17165, Sweden 2School of Public Health and the Second Affiliated Hospital, Zhejiang University School of Medicine, Hangzhou 310000, China 3Department of Clinical Neuroscience, Karolinska Institutet, Stockholm, 17165, Sweden 4Icahn School of Medicine at Mount Sinai, The Charles Bronfman Institute for Personalized Medicine, New York, NY 10029, USA 5MRC Biostatistics Unit, University of Cambridge, Cambridge CB2 1TN, UK 6Department of Public Health and Primary Care, University of Cambridge, Cambridge CB2 1TN, UK 7Unit of Medical Epidemiology, Department of Surgical Sciences, Uppsala University, Uppsala, 75185, Sweden Correspondence: Shuai Yuan, MD, MMedSc, Institute of Environmental Medicine, Karolinska Institutet, Nobelsväg 13, Stockholm, 17165, Sweden. Email: shuai. yuan@ki.se. Abstract Context: The causality and pathways of the associations between physical activity and inactivity and the risk of type 2 diabetes remain inconclusive. Objective: We conducted an updated mendelian randomization (MR) study to explore the associations of moderate-to-vigorous physical activity (MVPA) and leisure screen time (LST) with type 2 diabetes mellitus (T2DM). Methods: Genetic variants strongly associated with MVPA or LST with low linkage disequilibrium were selected as instrumental variables from a genome-wide meta-analysis including more than 600 000 individuals. Summary-level data on T2DM were obtained from the DIAbetes Genetics Replication And Meta-analysis consortium including 898 130 individuals. Data on possible intermediates (adiposity indicators, lean mass, glycemic traits, and inflammatory biomarkers) were extracted from large-scale genome-wide association studies (n = 21 758-681 275). Univariable and multivariable MR analyses were performed to estimate the total and direct effects of MVPA and LST on T2DM. Methylation MR analysis was performed for MVPA in relation to diabetes. Results: The odds ratio of T2DM was 0.70 (95% CI, 0.55-0.88; P = .002) per unit increase in the log-odds ratio of having MVPA and 1.45 (95% CI, 1.30-1.62; P = 7.62 × 10−11) per SD increase in genetically predicted LST. These associations attenuated in multivariable MR analyses adjusted for genetically predicted waist-to-hip ratio, body mass index, lean mass, and circulating C-reactive protein. The association between genetically predicted MVPA and T2DM attenuated after adjusting for genetically predicted fasting insulin levels. Two physical activity-related methylation biomarkers (cg17332422 in ADAMTS2 and cg09531019) were associated with the risk of T2DM (P < .05). Conclusion: The study suggests causal associations of MVPA and LST with T2DM that appear to be mediated by obesity, lean mass, and chronic low-grade inflammation. Key Words: inflammation, mendelian randomization, obesity, physical activity, sedentary behavior, type 2 diabetes Abbreviations: BMI, body mass index; CRP, C-reactive protein; ELPD, expected log pointwise posterior density; FG, fasting glucose; FI, fasting insulin; GWAS, genome-wide association study; HbA1c, glycated hemoglobin A1c; IL-1RA, interleukin-1 receptor antagonist; IL-6, interleukin-6; LD, linkage disequilibrium; LST, leisure screen time; mQTL, methylation quantitative trait loci; MR, mendelian randomization; MVPA, moderate-to-vigorous physical activity; OR, odds ratio; RCT, randomized controlled trial; SNV, single-nucleotide variation; T2DM, type 2 diabetes mellitus; WHR, waist-to-hip ratio. Received: 13 May 2023. Editorial Decision: 21 June 2023. Corrected and Typeset: 5 July 2023 © The Author(s) 2023. Published by Oxford University Press on behalf of the Endocrine Society. This is an Open Access article distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs licence (https://creativecommons. org/licenses/by-nc-nd/4.0/), which permits non-commercial reproduction and distribution of the work, in any medium, provided the original work is not altered or transformed in any way, and that the work is properly cited. For commercial re-use, please contact journals.permissions@oup.com Type 2 diabetes mellitus (T2DM) is a global pandemic that causes a heavy burden on the health system as well as the whole of society [1]. Population-based epidemiological stud- ies have revealed several modifiable lifestyle factors, such as cigarette smoking, unhealthy diets, and physical inactivity, for T2DM [2], which provide evidence bases for disease pre- vention [3]. Regarding physical activity and sedentary behav- iors, meta-analyses of observational studies found that higher levels of physical activity and lower levels of sedentary behav- iors were associated with a reduced risk of T2DM [4-6]. However, whether these associations are causal remains un- determined because of potential biases, such as reverse causal- ity and confounding, in observational studies. For the association between physical activity and diabetes risk, randomized controlled trials (RCTs) have generated conflict- ing evidence. A systematic review of RCTs found no firm evi- dence that physical activity alone compared to standard treatment influenced the risk of T2DM [7] even though some trials revealed significant associations of higher levels of physical activity with a lower risk incidence of diabetes Journal of the Endocrine Society, 2023, 7, 1–8 https://doi.org/10.1210/jendso/bvad090 Advance access publication 26 June 2023 Clinical Research Article D ow nloaded from https://academ ic.oup.com /jes/article/7/8/bvad090/7207971 by U niversity of C am bridge user on 08 D ecem ber 2023 https://orcid.org/0000-0001-5055-5627 https://orcid.org/0000-0001-5878-8986 https://orcid.org/0000-0001-5365-8760 mailto:shuai.yuan@ki.se mailto:shuai.yuan@ki.se https://creativecommons.org/licenses/by-nc-nd/4.0/ https://creativecommons.org/licenses/by-nc-nd/4.0/ and insulin resistance [8, 9]. Other trials found a consistent protective effect of lifestyle intervention on diabetic incidence; however, the intervention usually simultaneously acted on multiple lifestyle factors, like diet, physical activity, and treat- ments [7, 10, 11], which hinders the assessment of the inde- pendent role of physical activity in diabetes. In addition, despite RCT being regarded as the gold-standard design for causal inference, it may be confined by a short intervention pe- riod and low adherence. Physical activity and sedentary lifestyle have been associ- ated with obesity [12], lean mass [13], glycemic status [14], and inflammation [15], which are vital factors for the develop- ment of T2DM. Nevertheless, whether and how these factors mediate the associations of physical activity and sedentary be- haviors with diabetes risk are unestablished. A clear appraisal of these pathways may benefit the development of more pre- cise preventive and therapeutic strategies. Mendelian randomization (MR) analysis is an epidemiologic- al approach that can reinforce causal inference using observa- tional genetic data [16]. Using genetic variants as an instrumental variable for an exposure (eg, moderate-to-vigorous physical activity [MVPA]), the MR approach can diminish con- founding and reverse causality since germline genotypes are usu- ally not associated with environmental or self-adopted factors and cannot be modified by the onset of disease [16]. A previous MR study examined the associations of accelerometer-based physical activity average accelerations, vigorous physical activ- ity, and sedentary behavior with T2DM and 4 glycemic traits and found none of the associations [17]. However, findings were conflicting in another MR study, in which the associations of moderate and vigorous physical activity and television watch- ing with T2DM were identified [18]. Epigenetics, including DNA methylation, plays an important role in human health and has been revealed to mediate the association between exer- cise and diabetes [19]. Therefore, we conducted this MR study based on larger data sets to 1) determine the associations of MVPA and leisure screen time (a long time presenting sedentary behavior) with the risk of T2DM; 2) explore potential pathways of these associations; and 3) explore how physical activity affects diabetes from the epigenetics perspective. Materials and Methods Study Design The present study was based on publicly available summary- level genome-wide association study (GWAS) data and thus required no ethical permit. Individual studies included in the GWASs have been approved by an ethical review board. We first estimated the genetic correlations between MVPA, LST, and T2DM. We then used MR-CAUSE analysis to compare the fit of genetic sharing and causal pathways [20]. To strengthen causal inference, we conducted MR analyses to ex- plore the associations of MVPA and LST with T2DM. To re- veal potential pathways, we examined the MR associations of MVPA and LST with possible intermediates (ie, obesity, lean mass, glycemic traits, and chronic inflammation biomarkers) and further performed multivariable MR analyses adjusting for associated traits to estimate the mediation effects. To re- veal possible methylation mechanisms, we also performed a DNA methylation MR analysis. T2DM was the primary out- come, and all intermediates were secondary outcomes. The MR design and its assumptions are displayed in Fig. 1. Genetic Instrumental Variable Selection Summary-level data for MVPA and LST were obtained from a genome-wide meta-analysis of up to 606 820 and 526 725 in- dividuals of European ancestry, respectively [21]. MVPA and LST were measured by domain- and intensity-specific ques- tions. MVPA was defined as a binary variable with being ac- tive (∼48% of studied participants) and inactive status after harmonizing the phenotype across more than 40 included studies that used different questionnaires. Detailed definitions of MVPA in each study can be found in the original GWAS [21]. LST included time spent watching television, playing videogames, sitting at the computers, etc and was treated as a continuous variable with a mean (SD) of 3.53 (1.87) hours/day. GWAS analysis was performed in each study by sex and with adjustment for age, age [2], genetic principal components, and study-specific covariates [21]. Single-nucleotide variations (SNVs; formerly single- nucleotide polymorphisms [SNPs]) associated with MVPA or LST at the genome-wide significance threshold (P < 5 × 10−8) were extracted from the corresponding GWAS data set. Linkage disequilibrium (LD) of selected SNVs for each trait was calculated based on the 1000 Genomes European reference panel [22]. To increase the power of the analysis by including more genetic variants, SNVs without or in low LD (defined as R2 < 0.1) were used as instrumental variables, leading to 22 SNVs for MVPA and 152 SNVs for LST. We also selected SNVs in lower LD (R2 < 0.01) as supplementary instruments. SNVs for MVPA and LST were combined and pruned using the same selection criteria to select instruments for multivariable MR analysis with mutual adjustment. Selected SNVs are shown in Supplementary Table S1 [23]. We also used rs1625595 as an instrumental variable for MVPA in the supplementary analysis to minimize pleiotropy. Rs1625595 is strongly associated with MVPA (P = 1.9 × 10−11) and located approximately 300 kb upstream of the ACTN3 gene that is highly expressed in skeletal muscle (GTEx; P = 6.6 × 10−5) [21]. A missense variant of this gene makes the α-actinin-3 filaments more flexible, resulting in low susceptibility to exercise-induced muscle damage and thereby facilitating a more active lifestyle [21]. Data Sources for Type 2 Diabetes Summary-level data on the DNA-diabetes associations were obtained from the DIAbetes Genetics Replication And Meta-analysis (DIAGRAM) consortium including 898 130 in- dividuals (74 124 patient cases and 824 006 controls) of European descent with a mean age of 55 years, 51.8% of whom are men [24]. The genetic associations were adjusted for genetic principal components, relatedness, and study- specific covariates [24]. Patient cases with T2DM were de- fined by previous clinical diagnosis (self-reported and medical records), antidiabetic medication, and cutoff of certain gly- cemic biomarkers in specific studies [24]. Data Sources for Adiposity, Lean Mass, Glycemic Traits, and Inflammatory Biomarkers Two adiposity indicators (waist-to-hip ratio [WHR] and body mass index [BMI]), lean mass, 3 glycemic traits (fasting insulin [FI], fasting glucose [FG], and glycated hemoglobin A1c [HbA1c]), and 3 inflammatory biomarkers (C-reactive protein [CRP], interleukin-1 receptor antagonist [IL-1RA], and interleukin-6 [IL-6]) were treated as possible intermediates 2 Journal of the Endocrine Society, 2023, Vol. 7, No. 8 D ow nloaded from https://academ ic.oup.com /jes/article/7/8/bvad090/7207971 by U niversity of C am bridge user on 08 D ecem ber 2023 [25]. Summary-level data on these phenotypes were obtained from the Genetic Investigation of ANthropometric Traits (GIANT) Consortium for WHR (n = 224 459) [26] and BMI (n = 681 275) [27], a genome-wide association analysis in the UK Biobank study (n = 450 243) for lean mass measured by the bioelectrical impedance analysis approach [28], the Meta-Analysis of Glucose and Insulin-related Traits Consortium (MAGIC, n = up to 46 613) for glycemic traits [29], the Neale Lab GWAS in UK Biobank for CRP (n = 343 524), and the Systematic and Combined AnaLysis of Olink Proteins (SCALLOP) consortium (n = up to 21 758 par- ticipants) for IL-1RA and IL-6 [30]. Blood DNA Methylation Marker Identification Blood DNA methylation markers (CpG sites) of MVPA were obtained from an epigenome-wide association study with 5513 individuals [31]. DNA methylation levels were meas- ured by Illumina Infinium HumanMethylation450 (HM450) BeadChip array, and the MVPA-methylation associations were estimated after adjustment for age, sex, birthplace, socio- economic status, smoking status, alcohol consumption, sam- ple type, and estimated white blood cell composition [31]. Seven CpG sites were found to be associated with MVPA at P < 1 × 10−5 [31]. We further identified methylation quantita- tive trait loci (mQTL) that regulate the methylation levels of identified MVPA-related CpG sites based on the Accessible Resource for Integrative Epigenomic Studies Genetics of DNA Methylation Consortium (GoDMC) [32]. The Matrix eQTL software was used to test the preliminary association of SNVs with CpG sites [32]. SNVs identified at the epigenome-wide association significance level were further an- alyzed using exact linear regression adjusted for covariates in PLINK 1.07, and genome-wide complex trait analysis was performed to obtain the most representative independent lo- cus in relation to each CpG site [33]. Methylation levels at MVPA-associated CpG sites were treated as the exposure, which was instrumented by their mQTLs. Given no epigenome-wide association study on LST, corresponding methylation MR analysis could not be performed. Data Analysis Genetic correlations were calculated using LDSC software [34]. MR-CAUSE [20] based on summary-level data was used to compare the fit of genetic sharing and causal pathways of the association between MVPA, LST, and T2DM. This method holds the assumption that the genetic relationship be- tween 2 phenotypes is a mixture both of causal and shared correlated pleiotropy. The analysis estimates posterior distri- butions of a null effect, a shared effect, and a causal effect, and compares the model fit by estimating Δ expected log pointwise posterior density (ELPD) [20]. SNVs were harmonized by effect alleles and corresponding allele frequency of the exposure genetic associations. SNVs with ambiguous matches (ie, allele mismatch or palindromic pair with minor allele frequency of 0.45-0.55) between the ex- posure and outcome data were removed from the analysis. We used the inverse variance weighted method with multiplicative random effects as the primary method. To detect the minimize horizontal pleiotropy, 3 sensitivity analyses, including the weighted median [35], MR-Egger [35], and MR-PRESSO [36], were used. The weighted median can provide accurate estimates when more than half of the SNVs are valid [35]. MR-Egger can generate estimates after correction for possible horizontal pleiotropy detected by its embedded intercept test even though the estimates are usually underpowered [35]. MR-PRESSO can detect SNV outliers that introduce plei- otropy and generate estimates after the removal of these out- lying SNVs [36]. Cochran’s Q test was used to assess the heterogeneity of SNV estimates. Multivariable MR analysis with mutual adjustment was performed to minimize the mu- tual influence of MVPA and LST on the outcomes. Further, we performed multivariable MR adjusting for related inter- mediates to estimate the direct effect of MVPA and LST on T2DM, where the same sets of SNVs as those used in Figure 1. Multivariable mendelian randomization (MR) design overview. SNP, single-nucleotide polymorphism (now known as single-nucleotide variation [SNV]). Assumption 1 indicates that the used genetic variants as instrumental variables should be strongly associated with the exposure; assumption 2 indicates that the used genetic variants should not associated with confounders; and assumption 3 indicates the used genetic variants should affect the outcome merely via the exposure, not partly or entirely via other alternative pathways. Journal of the Endocrine Society, 2023, Vol. 7, No. 8 3 D ow nloaded from https://academ ic.oup.com /jes/article/7/8/bvad090/7207971 by U niversity of C am bridge user on 08 D ecem ber 2023 univariable MR were employed due to no interest in the effects of intermediates on T2DM [16]. Multivariable MR was per- formed for each intermediator in turn. Mediation was calcu- lated as the direct effect divided by the total effect and subtracted from 1. The Wald ratio method was used to esti- mate the methylation MR association where the odds ratio (OR) of T2DM was scaled per SD increase in MVPA-reduced methylation levels. Associations with P less than .05 were considered statistically significant. All statistical tests were 2-sided, and analyses were performed using CAUSE, TwoSampleMR, MendelianRandomization, and MR-PRESSO packages in R software (4.4.1). Results MVPA was inversely (rg = −0.24; P = 5.54 × 10−20) and LST positively (rg = 0.34; P = 4.41 × 10−55) correlated with T2DM. MR-CAUSE analysis revealed both genetic sharing and causal pathways of the associations of MVPA and LST with T2DM (P < .05) and supported that the causal pathways had a better fit (P = 1.70 × 10−6 for MVPA and 8.70 × 10−13 for LST; Table 1). Associations of Moderate-to-Vigorous Physical Activity and Leisure Screen Time With Diabetes Genetically predicted MVPA and LST were associated with the risk of T2DM (Fig. 2). The OR of T2DM was 0.70 (95% CI, 0.55-0.88; P = .002) per unit increase in the log-OR of having MVPA and 1.45 (95% CI, 1.30-1.62; P = 7.62 × 10−11) per SD increase in genetically predicted LST. The associations were stable in sensitivity analyses (see Fig. 2). High heterogeneity (Cochran’s Q = 83 for MVPA and 409 for LST) was observed but there was no indication of horizontal pleiotropy detected by MR-Egger intercept tests (P > .201). Five and 6 SNV outliers were identified by MR-PRESSO analyses; however, the association persisted after removal of these outliers. The association for MVPA became stronger in the supplementary analysis using rs1625595 with few pleotropic effects as the instrumental variable (OR 0.27; 95% CI, 0.15-0.48; P = 1.21 × 10−5). The associations were stable in the analysis based on SNVs in LD less than 0.01 (Supplementary Table S2) [23]. The asso- ciations attenuated slight but remained statistically significant in multivariable MR analysis with mutual adjustment (see Fig. 2). Associations of Moderate-to-Vigorous Physical Activity and Leisure Screen Time With Intermediators Genetically predicted MVPA and LST were associated with some of the possible intermediators (Fig. 3). Genetically pre- dicted MVPA was associated with lower levels of WHR (β = −0.14; 95% CI, −0.23 to −0.05; P = .004), FI (β = −0.12 pmol/L; 95% CI, −0.17 to −0.07; P = 7.21 × 10−6), and CRP (β = −0.52 mg/L; 95% CI, −0.76 to −0.28; P = 2.54 × 10−5), and higher levels of lean mass (β = 0.15 kg; 95% CI, 0.05-0.25; P = .004). One SD increase in genetically predicted LST was associated with higher levels of WHR (β = 0.21; 95% CI, 0.17-0.26; P = 1.78 × 10−18), BMI (β = 0.14; 95% CI, 0.11-0.18; P = 4.19 × 10−5), CRP (β = 0.75 mg/L; 95% CI, 0.42-1.07; P = 7.29 × 10−6), IL-1RA (β = 0.17 SD; 95% CI, 0.07-0.25; P = .001), and IL-6 (β = 0.13 SD; 95% CI, 0.03-0.23; P = .008), and less lean mass (β = −0.09 kg; 95% CI, −0.14-−0.05; P = 1.48 × 10−4). Moderate-to-high heterogeneity was observed in the associations of genetically predicted MVPA and LST with BMI, lean mass, glycemic traits, and CRP; however, there was no indication of horizontal pleiotropy (MR-Egger inter- cept P > .05) (Supplementary Table S3) [23]. These associa- tions remained stable in sensitivity analyses (Supplementary Table S3) [23] and attenuated but were overall consistent when MVPA and LST were mutually adjusted (Fig. 3). Direct Effects of Moderate-to-Vigorous Physical Activity and Leisure Screen Time on Diabetes and Mediation Effects The associations of genetically predicted MVPA and LST with T2DM attenuated and became nonsignificant in multivariable MR analysis adjusted for related intermediates (Table 2). Most of the association between MVPA and T2DM was medi- ated by WHR (66%), followed by CRP (63%), FI (43%), BMI (22%), and lean mass (9%) (see Table 2). BMI (53%) medi- ated the most effect of LST on T2DM, followed by WHR (43%), CRP (27%), IL-1RA (12%), and lean mass (10%) (see Table 2). Of note, because some included mediators are genetically correlated, the total mediation percentage ex- ceeded 100%. Associations of Moderate-to-Vigorous Physical Activity–related DNA Methylation With Diabetes Seven mQTLs were identified to be associated with MVPA-associated CpG sites and included in the methylation MR analysis (Table 3). Two CpG sites (cg17332422 in ADAMTS2 and cg09531019 in an unknown gene) were asso- ciated with the risk of T2DM (P < .05). Per SD increase in methylation levels associated with having no MVPA, the OR of diabetes was 2.78 (95% CI, 1.02-7.69) for cg17332422 and 2.86 (95% CI, 1.05-7.69) for cg09531019. Table 1. MR-CAUSE analysis on the associations of moderate-to-vigorous physical activity and leisure screen time with type 2 diabetes Model 1a Model 2a Δ ELPDb SE (Δ ELPD) Z score P MVPA (2349 SNVs) Null Sharing −27.69 5.27 −5.26 7.30E-08 Null Causal −34.81 6.73 −5.17 1.10E-07 Sharing Causal −7.11 1.53 −4.65 1.70E-06 P value testing indicates that causal model is a better fit: 1.7E-06 LST (3297 SNVs) Null Sharing −73.49 9.45 −7.77 3.80E-15 Null Causal −81.44 10.51 −7.75 4.50E-15 Sharing Causal −7.95 1.13 −7.05 8.70E-13 P value testing indicates that causal model is a better fit: 8.7E-13 Abbreviations: ELPD, expected log pointwise posterior density; LST, leisure screen time; MVPA, moderate-to-vigorous physical activity; SNV, single-nucleotide variation. aModel 1 and model 2 refer to the models being compared (null, sharing, or causal). bModel fit is measured by Δ ELPD and negative values indicate that model 2 is a better fit. 4 Journal of the Endocrine Society, 2023, Vol. 7, No. 8 D ow nloaded from https://academ ic.oup.com /jes/article/7/8/bvad090/7207971 by U niversity of C am bridge user on 08 D ecem ber 2023 Discussion Using data from large-scale genetic studies, this updated MR study identified potential causal associations of MVPA and LST with T2DM. Our multivariable MR analysis revealed that adiposity, lean mass, and chronic inflammation mediated the associations. Insulin levels also appeared to partly mediate the association between MVPA and T2DM. Increased levels of DNA methylation at cg17332422 in the ADAMTS2 gene and cg09531019 in relation to low physical activity levels were associated with a higher risk of T2DM. Our findings are consistent with many observational associ- ations of high levels of physical activity and low levels of sed- entary behavior with reduced risk of T2DM [4-6, 37] and evidence from some RCTs [8, 9] even though a systematic re- view of RCTs summarized that physical activity plus diet in- stead of physical activity alone reduced or delayed the incidence of T2DM [7]. Our MR results strengthened the po- tential causality of the associations between physical activity and the risk of diabetes. Given that diabetes may be more effectively modified by multiple factors, a comprehensive rec- ommendation including physical activity, diet, and treatments should be made in the clinical setting to the population with a high risk of diabetes [38]. Additionally, our multivariable MR analysis identified obesity as an important mediator in these associations even though some clinical trials failed to observe significant body mass change after exercise training or a combined lifestyle modification including promoting physical activity in patients with T2DM [39, 40]. The possible explanation for this discrepancy may be the different obesity measurements used between studies. Body weight employed as an indicator of obesity in the aforementioned trials might not capture the change in body fat disposition caused by in- creased levels of physical activity [41]. Other factors, such as insufficient levels of physical activity as well as low adher- ence to the intervention, may also contribute to the observed disagreement. Except for adiposity, the present MR study also identified a small mediation by lean mass. Exercise and moderate-to-high levels of training maintained and improved fat-free body mass [42], and high levels of lean mass were found to be associated with a low risk of T2DM [43]. Chronic low-grade inflammation, featured by high CRP lev- els, is another mediator of the associations of MVPA and LST with T2DM in this study. This is in line with previous data [44]. For example, a meta-analysis of 83 controlled trials found that exercise training is associated with a decrease in CRP levels [45]. Likewise, sedentary behavior, like long-time television viewing, was associated with an increase in CRP lev- els [46]. Except for CRP, this MR analysis additionally found associations of genetically proxied longer LST with increased levels of 2 other inflammatory biomarkers, IL-1RA and IL-6, that have previously been shown to be associated with the risk of T2DM [25]. However, our mediation analysis indicated that only IL-1RA appeared to partly mediate the association between LST and T2DM. All these findings support the high potential of promoting physical activity and reducing seden- tary behavior for diabetes prevention and on the other side im- ply that dampening systemic inflammation in immobile individuals with low physical activity levels may be a strategy for lowering the risk of T2DM. Different types of physical activity, such as free-living walk- ing and high-intensity interval training, were associated with an improved glycemic profile in patients with T2DM [14, 47, 48]. However, the present MR study did not observe clear associations of genetically predicted MVPA and LST with any of the studied glycemic traits, except for a positive association between genetically predicted MVPA and FI levels. The reason for the discrepancy may be that GWASs on glycemic traits were conducted in individuals without T2DM. Our methylation analysis revealed 2 methylation signals (cg17332422 and cg09531019) in support of low levels of physical activity and high risk of diabetes. The cg17332422 site is in the ADAMTS2 gene. In line with our findings, an epigenome-wide association study found that another methy- lation site in ADAMTS2 was associated with T2DM in HIV-infected individuals [49]. CpG cg09531019 was not Figure 2. Associations of moderate-to-vigorous physical activity (MVPA) and leisure screen time (LST) with type 2 diabetes. OR, odds ratio. Journal of the Endocrine Society, 2023, Vol. 7, No. 8 5 D ow nloaded from https://academ ic.oup.com /jes/article/7/8/bvad090/7207971 by U niversity of C am bridge user on 08 D ecem ber 2023 annotated to a gene. Thus, more study is needed to verify this association. The merits of this study include data sources from larger-scale studies, the combination of traditional and methy- lation MR analyses, as well as the utilization of multivariable MR analysis to explore the mediation pathways. Limitations need to be considered when interpreting our findings. First, the present analysis was based on European populations, which limits the generalizability of our results to other popu- lations. Second, glycemic trait data were obtained from indi- viduals without T2DM, which confines the generalizability of these findings to diabetic populations. Third, we used a comparative relaxed LD threshold to select genetic instru- ments to increase the power, which might inflate the rate of type 1 error. However, the associations were consistent in the sensitivity analysis using stringent instrument selection cri- teria. Fourth, we might have overlooked other important methylation sites linking physical activity and T2DM given the small sample size of the epigenome-wide association study on physical activity [31]. Fifth, the study could not differenti- ate the comparative effects of different types of physical activ- ity on the risk of T2DM based on the summary-level data. Sixth, we could not examine the sex- or ethnicity-specific asso- ciations of physical activity or sedentary behaviors with the risk of T2DM because of data unavailability. Seventh, differ- ent genotyping arrays/methods were used in the included GWASs, which might influence our MR results, even though this concern is limited since different genotyping arrays have been found not to be related to the usability of GWAS [50]. In summary, this updated MR study found associations of genetically predicted MVPA and LST with the risk of Figure 3. Associations of moderate-to-vigorous physical activity (MVPA) and leisure screen time (LST) with obesity, glycemic traits, and inflammatory biomarkers. BMI, body mass index; CRP, C-reactive protein; FG, fasting glucose; FI, fasting insulin; HbA1c, glycated hemoglobin A1c; IL-1RA, interleukin-1 receptor antagonist; IL-6, interleukin-6; OR, odds ratio; WHR, waist-to-hip ratio. Table 2. Associations of moderate-to-vigorous physical activity and leisure screen time with type 2 diabetes after adjusting for intermediates and mediation effect Adjustment Association after adjusting for intermediate Mediation OR 95% CI P MVPA None 0.70 0.55-0.88 .002 - WHR 0.89 0.60-1.31 .540 66 BMI 0.75 0.54-1.05 .098 22 Lean mass 0.72 0.50-1.03 .071 9 FI 0.81 0.61-1.09 .166 43 CRP 0.87 0.64-1.19 .396 63 LST None 1.45 1.30-1.62 7.62E-11 - WHR 1.23 1.06-1.43 .006 43 BMI 1.19 1.01-1.41 .037 53 Lean mass 1.40 1.25-1.56 2.45E-09 10 CRP 1.31 1.17-1.47 3.55E-06 27 IL-1RA 1.39 1.24-1.55 1.11E-08 12 IL-6 1.45 1.29-1.63 2.59E-10 0 Abbreviations: BMI, body mass index; CRP, C-reactive protein; FI, fasting insulin; IL-1RA, interleukin-1 receptor antagonist; IL-6, interleukin-6; LST, leisure screen time; MVPA, moderate-to-vigorous physical activity; OR, odds ratio; SNV, single-nucleotide variation; WHR, waist-to-hip ratio. Table 3. Associations of moderate-to-vigorous physical activity– related CpG sites and risk of type 2 diabetes CpG site SNV β SE OR 95% CI P cg17332422 1 1.03 0.52 2.78 1.02-7.69 .046 cg18954163 1 −0.04 0.26 0.96 0.58-1.59 .866 cg00174845 1 0.32 0.55 1.37 0.47-4.00 .559 cg09531019 1 1.06 0.51 2.86 1.05-7.69 .039 cg10868567 1 −0.50 0.77 0.61 0.13-2.78 .516 cg17757602 1 −0.57 0.41 0.56 0.25-1.25 .160 Abbreviations: OR, odds ratio; SNV, single-nucleotide variation. 6 Journal of the Endocrine Society, 2023, Vol. 7, No. 8 D ow nloaded from https://academ ic.oup.com /jes/article/7/8/bvad090/7207971 by U niversity of C am bridge user on 08 D ecem ber 2023 T2DM, and these associations were mediated by adiposity, chronic inflammation, and lean mass. Our findings suggest promoting physical activity and lowering sedentary behavior may be a strategy for diabetes prevention. Acknowledgments The authors thank the DIAbetes Genetics Replication and Meta-analysis (DIAGRAM) consortium, the Neale lab, and all cited GWAS for sharing data for the analysis. Funding S.C.L. is supported by research grants from the Swedish Cancer Society (Cancerfonden), the Swedish Research Council for Health, Working Life and Welfare (Forte; grant No. 2018-00123), and the Swedish Research Council (Vetenskapsrådet; grant No. 2019-00977). X.L. is supported by research grants for Distinguished Young Scholars of Zhejiang Province (No. LR22H260001). The funders had no role in the design and conduct of the study; collection, man- agement, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; or the decision to sub- mit the manuscript for publication. Author Contributions All authors read and approved the final manuscript and au- thor contributions statement using CRediT with degree of contribution: S.Y. (conceptualization: lead; methodology: lead; formal analysis: lead; data curation: equal; and writing– review and editing: lead); X.L. (conceptualization: equal; meth- odology: equal; formal analysis: equal; data curation: equal; and writing—original draft: equal); Q.L. (conceptualization: equal; methodology: equal; formal analysis: equal; and writ- ing—original draft: equal); Z.W. (conceptualization: equal; data curation: equal; and writing—review and editing: equal); X.J. (conceptualization: equal; methodology: equal; formal analysis: equal; and writing—original draft: equal); S.B. (con- ceptualization: equal; methodology: equal; data curation: equal; and writing—review and editing: leading); and S.C.L. (conceptualization: equal; methodology: equal; data curation: equal; and writing—review and editing: lead) Disclosures The authors have nothing to disclose. Data Availability All data analyzed in this study can be obtained by reasonable request to corresponding authors. References 1. Chatterjee S, Khunti K, Davies MJ. Type 2 diabetes. Lancet. 2017;389(10085):2239-2251. 2. Zheng Y, Ley SH, Hu FB. Global aetiology and epidemiology of type 2 diabetes mellitus and its complications. Nat Rev Endocrinol. 2018;14(2):88-98. 3. Yamaoka K, Tango T. Efficacy of lifestyle education to prevent type 2 diabetes: a meta-analysis of randomized controlled trials. Diabetes Care. 2005;28(11):2780-2786. 4. Aune D, Norat T, Leitzmann M, Tonstad S, Vatten LJ. Physical ac- tivity and the risk of type 2 diabetes: a systematic review and dose-response meta-analysis. Eur J Epidemiol. 2015;30(7): 529-542. 5. Smith AD, Crippa A, Woodcock J, Brage S. Physical activity and in- cident type 2 diabetes mellitus: a systematic review and dose-response meta-analysis of prospective cohort studies. Diabetologia. 2016;59(12):2527-2545. 6. Patterson R, McNamara E, Tainio M, et al. Sedentary behaviour and risk of all-cause, cardiovascular and cancer mortality, and inci- dent type 2 diabetes: a systematic review and dose response meta- analysis. Eur J Epidemiol. 2018;33(9):811-829. 7. Hemmingsen B, Gimenez-Perez G, Mauricio D, Roqué IFM, Metzendorf MI, Richter B. Diet, physical activity or both for pre- vention or delay of type 2 diabetes mellitus and its associated com- plications in people at increased risk of developing type 2 diabetes mellitus. Cochrane Database Syst Rev. 2017;12(12):CD003054. 8. Kriska AM, Rockette-Wagner B, Edelstein SL, et al. The impact of physical activity on the prevention of type 2 diabetes: evidence and lessons learned from the Diabetes Prevention Program, a long- standing clinical trial incorporating subjective and objective activity measures. Diabetes Care. 2021;44(1):43-49. 9. Davis CL, Pollock NK, Waller JL, et al. Exercise dose and diabetes risk in overweight and obese children: a randomized controlled tri- al. JAMA. 2012;308(11):1103-1112. 10. Knowler WC, Fowler SE, Hamman RF, et al. 10-year follow-up of diabetes incidence and weight loss in the Diabetes Prevention Program Outcomes Study. Lancet. 2009;374(9702):1677-1686. 11. Knowler WC, Barrett-Connor E, Fowler SE, et al. Reduction in the incidence of type 2 diabetes with lifestyle intervention or metfor- min. N Engl J Med. 2002;346(6):393-403. 12. Keating SE, Johnson NA, Mielke GI, Coombes JS. A systematic review and meta-analysis of interval training versus moderate- intensity continuous training on body adiposity. Obes Rev. 2017;18(8):943-964. 13. Ramsey KA, Rojer AGM, D’Andrea L, et al. The association of ob- jectively measured physical activity and sedentary behavior with skeletal muscle strength and muscle power in older adults: a system- atic review and meta-analysis. Ageing Res Rev. 2021;67:101266. 14. Karstoft K, Winding K, Knudsen SH, et al. The effects of free-living interval-walking training on glycemic control, body composition, and physical fitness in type 2 diabetic patients: a randomized, con- trolled trial. Diabetes Care. 2013;36(2):228-236. 15. Leiva-Valderrama JM, Montes-de-Oca-Garcia A, Opazo-Diaz E, et al. Effects of high-intensity interval training on inflammatory bi- omarkers in patients with type 2 diabetes. A systematic review. Int J Environ Res Public Health. 2021;18(23):12644. 16. Burgess S, Thompson SG. Mendelian Randomization: Methods for Using Genetic Variants in Causal Estimation. Interdisciplinary Statistics Series. Chapman and Hall/CRC; 2015. 17. Meisinger C, Linseisen J, Leitzmann M, Baurecht H, Baumeister SE. Association of physical activity and sedentary behavior with type 2 diabetes and glycemic traits: a two-sample Mendelian randomiza- tion study. BMJ Open Diabetes Res Care. 2020;8(2):e001896. 18. Deng MG, Cui HT, Lan YB, Nie JQ, Liang YH, Chai C. Physical activity, sedentary behavior, and the risk of type 2 diabetes: a two- sample Mendelian randomization analysis in the European popula- tion. Front Endocrinol (Lausanne). 2022;13:964132. 19. Ling C, Groop L. Epigenetics: a molecular link between environ- mental factors and type 2 diabetes. Diabetes. 2009;58(12): 2718-2725. 20. Morrison J, Knoblauch N, Marcus JH, Stephens M, He X. Mendelian randomization accounting for correlated and uncorre- lated pleiotropic effects using genome-wide summary statistics. Nat Genet. 2020;52(7):740-747. 21. Wang Z, Emmerich A, Pillon NJ, et al. Genome-wide association analyses of physical activity and sedentary behavior provide in- sights into underlying mechanisms and roles in disease prevention. Nat Genet. 2022;54(9):1332-1344. Journal of the Endocrine Society, 2023, Vol. 7, No. 8 7 D ow nloaded from https://academ ic.oup.com /jes/article/7/8/bvad090/7207971 by U niversity of C am bridge user on 08 D ecem ber 2023 22. Clarke L, Zheng-Bradley X, Smith R, et al. The 1000 Genomes Project: data management and community access. Nat Methods. 2012;9(5):459-462. 23. Yuan S. Supplementary material for “Physical activity, sedentary behavior, and type 2 diabetes: mendelian randomization analysis.” Deposited April 18, 2023. doi:10.17605/OSF.IO/4KERU. https:// osf.io/4keru/ 24. Mahajan A, Taliun D, Thurner M, et al. Fine-mapping type 2 dia- betes loci to single-variant resolution using high-density imputation and islet-specific epigenome maps. Nat Genet. 2018;50(11): 1505-1513. 25. Yuan S, Larsson SC. An atlas on risk factors for type 2 diabetes: a wide-angled Mendelian randomisation study. Diabetologia. 2020;63(11):2359-2371. 26. Shungin D, Winkler TW, Croteau-Chonka DC, et al. New genetic loci link adipose and insulin biology to body fat distribution. Nature. 2015;518(7538):187-196. 27. Yengo L, Sidorenko J, Kemper KE, et al. Meta-analysis of genome- wide association studies for height and body mass index in ∼700000 individuals of European ancestry. Hum Mol Genet. 2018;27(20):3641-3649. 28. Pei YF, Liu YZ, Yang XL, et al. The genetic architecture of appen- dicular lean mass characterized by association analysis in the UK Biobank study. Commun Biol. 2020;3(1):608. 29. Chen J, Spracklen CN, Marenne G, et al. The trans-ancestral gen- omic architecture of glycemic traits. Nat Genet. 2021;53(6): 840-860. 30. Folkersen L, Gustafsson S, Wang Q, et al. Genomic and drug target evaluation of 90 cardiovascular proteins in 30 931 individuals. Nat Metab. 2020;2(10):1135-1148. 31. Van Roekel EH, Dugué PA, Jung CH, et al. Physical activity, televi- sion viewing time, and DNA methylation in peripheral blood. Med Sci Sports Exerc. 2019;51(3):490-498. 32. Min JL, Hemani G, Hannon E, et al. Genomic and phenotypic in- sights from an atlas of genetic effects on DNA methylation. Nat Genet. 2021;53(9):1311-1321. 33. Gaunt TR, Shihab HA, Hemani G, et al. Systematic identification of genetic influences on methylation across the human life course. Genome Biol. 2016;17(1):61. 34. Bulik-Sullivan B, Finucane HK, Anttila V, et al. An atlas of genetic correlations across human diseases and traits. Nat Genet. 2015;47(11):1236-1241. 35. Burgess S, Bowden J, Fall T, Ingelsson E, Thompson SG. Sensitivity analyses for robust causal inference from Mendelian randomization analyses with multiple genetic variants. Epidemiology. 2017;28(1): 30-42. 36. Verbanck M, Chen CY, Neale B, Do R. Detection of widespread horizontal pleiotropy in causal relationships inferred from Mendelian randomization between complex traits and diseases. Nat Genet. 2018;50(5):693-698. 37. Li X, Zhou T, Ma H, Liang Z, Fonseca VA, Qi L. Replacement of sedentary behavior by various daily-life physical activities and structured exercises: genetic risk and incident type 2 diabetes. Diabetes Care. 2021;44(10):2403-2410. 38. Magkos F, Hjorth MF, Astrup A. Diet and exercise in the preven- tion and treatment of type 2 diabetes mellitus. Nat Rev Endocrinol. 2020;16(10):545-555. 39. Boulé NG, Haddad E, Kenny GP, Wells GA, Sigal RJ. Effects of ex- ercise on glycemic control and body mass in type 2 diabetes melli- tus: a meta-analysis of controlled clinical trials. JAMA. 2001;286(10):1218-1227. 40. Jenum AK, Brekke I, Mdala I, et al. Effects of dietary and physical activity interventions on the risk of type 2 diabetes in South Asians: meta-analysis of individual participant data from randomised con- trolled trials. Diabetologia. 2019;62(8):1337-1348. 41. Wewege M, van den Berg R, Ward RE, Keech A. The effects of high- intensity interval training vs. moderate-intensity continuous training on body composition in overweight and obese adults: a systematic review and meta-analysis. Obes Rev. 2017;18(6):635-646. 42. Weinheimer EM, Sands LP, Campbell WW. A systematic review of the separate and combined effects of energy restriction and exercise on fat-free mass in middle-aged and older adults: implications for sarcopenic obesity. Nutr Rev. 2010;68(7):375-388. 43. Yeung CHC, Au Yeung SL, Fong SSM, Schooling CM. Lean mass, grip strength and risk of type 2 diabetes: a bi-directional Mendelian randomisation study. Diabetologia. 2019;62(5):789-799. 44. Koskeridis F, Evangelou E, Said S, et al. Pleiotropic genetic architec- ture and novel loci for C-reactive protein levels. Nat Commun. 2022;13(1):6939. 45. Fedewa MV, Hathaway ED, Ward-Ritacco CL. Effect of exercise training on C reactive protein: a systematic review and meta- analysis of randomised and non-randomised controlled trials. Br J Sports Med. 2017;51(8):670-676. 46. Hamer M, Poole L, Messerli-Bürgy N. Television viewing, C-reactive protein, and depressive symptoms in older adults. Brain Behav Immun. 2013;33:29-32. 47. Jelleyman C, Yates T, O’Donovan G, et al. The effects of high- intensity interval training on glucose regulation and insulin resist- ance: a meta-analysis. Obes Rev. 2015;16(11):942-961. 48. Shah SZA, Karam JA, Zeb A, et al. Movement is improvement: the therapeutic effects of exercise and general physical activity on gly- cemic control in patients with type 2 diabetes mellitus: a systematic review and meta-analysis of randomized controlled trials. Diabetes Ther. 2021;12(3):707-732. 49. Mathur R, Hui Q, Huang Y, et al. DNA methylation markers of type 2 diabetes mellitus among male veterans with or without human immunodeficiency virus infection. J Infect Dis. 2019; 219(12):1959-1962. 50. Verlouw JAM, Clemens E, de Vries JH, et al. A comparison of geno- typing arrays. Eur J Hum Genet. 2021;29(11):1611-1624. 8 Journal of the Endocrine Society, 2023, Vol. 7, No. 8 D ow nloaded from https://academ ic.oup.com /jes/article/7/8/bvad090/7207971 by U niversity of C am bridge user on 08 D ecem ber 2023 https://osf.io/4keru/ https://osf.io/4keru/ Physical Activity, Sedentary Behavior, and Type 2 Diabetes: Mendelian Randomization Analysis Materials and Methods Study Design Genetic Instrumental Variable Selection Data Sources for Type 2 Diabetes Data Sources for Adiposity, Lean Mass, Glycemic Traits, and Inflammatory Biomarkers Blood DNA Methylation Marker Identification Data Analysis Results Associations of Moderate-to-Vigorous Physical Activity and Leisure Screen Time With Diabetes Associations of Moderate-to-Vigorous Physical Activity and Leisure Screen Time With Intermediators Direct Effects of Moderate-to-Vigorous Physical Activity and Leisure Screen Time on Diabetes and Mediation Effects Associations of Moderate-to-Vigorous Physical Activity–related DNA Methylation With Diabetes Discussion Acknowledgments Funding Author Contributions Disclosures Data Availability References