Identifying effects of genetic obesity exposure on leukocyte telomere length using Mendelian randomization

Study design

The summary-level data of obesity-related traits, including body mass index (BMI), whole body fat mass, leg fat mass (right), leg fat mass (left), arm fat mass (right), arm fat mass (left), and trunk fat mass were extracted from the IEU Open GWAS database (https://gwas.mrcieu.ac.uk/). The summary statistical data of GWAS associated with LTL were also obtained from the IEU Open GWAS database. The data were from large sample studies and were used in conducting the two-sample MR investigation and exploring the causal relationship between obesity-related traits and LTL.

Assumptions of MR study

The fundamental assumptions and design of the MR investigation are shown in Fig. 1. (1) Relevance assumption: The genetic variants (instrumental variables) must be strongly correlated with obesity-related traits (exposure). (2) Independence assumption: no unpredictable confounders of the correlations between genetic variants and LTL (outcome) are present. (3) Exclusion restriction assumption: the genetic variants influence LTL only via obesity-related traits.

Data sources

The GWAS summary statistical data of obesity-related traits were obtained from European populations. The information included whole body fat mass with 330,762 participants, leg fat mass (right) with 331,293 populations, leg fat mass (left) with 331,275 populations, arm fat mass (right) with 331,226 individuals, arm fat mass (left) with 331,164 people, and trunk fat mass with 331,093 volunteers; all above obesity-related traits were obtained from UK Biobank cohort of the Neale lab. Single-nucleotide polymorphisms (SNPs) associated with obesity-related traits were identified as instrumental variables based on these parameters: p < 5 × 10⁻⁸ as a genome-wide statistical significance, independence among SNPs in linkage disequilibrium (r² < 0.001; clump window, 10,000 kb). The F statistic was used in assessing the instrumental variables’ power in the MR and the computational method described by a previous study (Pierce, Ahsan & Vanderweele, 2011). F statistic > 10 was defined as the minimum required threshold. The GWAS summary data associated with LTL from European ancestry (472,174 individuals) were downloaded from the IEU Open GWAS database. All participants in the obesity-related trait research projects were not screened for the LTL cohort.

Statistical analysis

The IVW was used as a primary analysis algorithm in estimating the causal effect size of obesity-related traits on LTL (Burgess, Butterworth & Thompson, 2013). The MR-Egger (Bowden, Davey Smith & Burgess, 2015), maximum likelihood (Xue, Shen & Pan, 2021), MR-pleiotropy residual sum outlier (MR-PRESSO) (Verbanck et al., 2018), and robust adjusted profile score (MR-RAPS) (Zhao et al., 2020) algorithms were utilized in validating the reliability and robustness for the causal relationship between obesity traits and LTL. An available online tool (https://shiny.cnsgenomics.com/mRnd/) (Brion, Shakhbazov & Visscher, 2013) was used to calculate the statistical power of MR analysis, and a power greater than 80% was deemed as a good value. The directionality that obesity-related traits cause LTL was verified via the MR Steiger test (Hemani, Tilling & Davey Smith, 2017). p < 0.05 indicated statistical significance.

Sensitivity analysis

The heterogeneity of SNPs was inspected via the IVW method and MR-Egger regression. MR-PRESSO, MR-Egger, and IVW approaches were used in identifying and removing outliers. The MR-Egger algorithms were utilized to detect potential pleiotropy. A single SNP’s influence on the total effect of IVW was evaluated through leave-one-out permutation analysis.

All MR analyses were conducted using TwoSampleMR (version 0.5.6) and MRPRESSO packages in R software (version 4.1.2; R Core Team, 2021).

Causality between BMI and LTL

To understand the causal relationship between obesity and LTL preliminarily, we analyzed the effect of BMI on LTL by using the IVW method in the two-sample MR study. After outliers were removed, all 268 independent SNPs associated with BMI were included in the MR for the computation of the pool effect size of BMI on LTL (File S1). The result of the IVW approach showed that a genetically determined 1-SD increase in BMI was correlated with decreased LTL, and the odds ratio (OR) was 0.957 (95% confidence interval [CI] = 0.942–0.973, p = 1.17E−07; Table 1 and Fig. 2A). In the analysis, the F statistic of the SNPs was 61.0, and the statistical power was 100%, indicating the absence of weak-instrument bias and high credibility, respectively (Table 1).

A series of sensitivity analysis approaches was used in verifying the robustness and reliability of the above result. First, as shown in Fig. 2B, four methods (MR-Egger, maximum likelihood, MR-PRESSO, and MR-RAPS) consistently exhibited causal direction from BMI to LTL. The result suggested that the causality from BMI to LTL was stable. Second, the results from Cochran’s Q test in the IVW model (p_-het = 4.76E−07) and MR-Egger model (p_-het = 4.48E−07) indicated heterogeneity among the SNPs, which may have been caused by the random allocation of alleles. Third, the statistical result of the MR-Egger intercepts indicated no directional pleiotropy in the MR analysis (p_-intercept = 0.448). Fourth, leave-one-out analysis showed that no single SNP significantly influenced the causal association between BMI and LTL (p < 0.05; File S2). Finally, statistical evidence from the MR Steiger test suggested that BMI influencing LTL was a correct causal direction (p < 0.001; Table 1).

Causality between six body fat indexes and LTL

Previous studies reported some bias when only BMI was used as an indicator for measuring obesity and estimating the correlation between obesity and diseases (Nimptsch, Konigorski & Pischon, 2019; Piche, Tchernof & Despres, 2020; Vecchie et al., 2018) because BMI cannot reflect body fat distribution. Therefore, to avoid the bias of causality, six body fat indexes (whole body fat mass, right leg fat mass, left leg fat mass, right arm fat mass, left arm fat mass, and trunk fat mass) were used to confirm the casual assumption of obesity impact on LTL. After outliers were deleted, 247, 242, 245, 240, 236, and 247 independently available SNPs were associated with whole body fat mass, right leg fat mass, left leg fat mass, right arm fat mass, left arm fat mass, and trunk fat mass, respectively (File S1). The SNPs were then utilized in proving the genetically predicted causality between obesity and LTL. The result of the IVW algorithm indicated that enhanced body fat indexes are causally associated with a decrease in LTL. The effect size of six body fat indexes’ influences on LTL were as follows: whole body fat mass: OR = 0.949 (95% CI [0.932–0.967]; p = 3.69E−08), right leg fat mass: OR = 0.932 (95% CI [0.910–0.954]; p = 6.98E−09), left leg fat mass: OR = 0.931 (95% CI [0.910–0.952]; p = 3.79E−10), right arm fat mass: OR = 0.955 (95% CI [0.939–0.972]; p = 3.67E−07), left arm fat mass: OR = 0.949 (95% CI [0.932–0.965]; p = 4.63E−09), and trunk fat mass: OR = 0.964 (95% CI [0.947–0.981]; p = 5.53E−05; Table 1). In the analyses, the F statistics of the SNPs were larger than 10.0 (range: 58.8–61.1), indicating the absence of potential weak instrument bias, and all statistical power rates were approximately 100%, indicating high credibility (Table 1).

Likewise, sensitivity analysis methods were used in testing the robustness and reliability of the above result. First, MR-Egger, maximum likelihood, MR-PRESSO, and MR-RAPS were utilized in validating the stability of the causal hypothesis. The results uniformly showed that the causality between per body fat index and LTL was stable (Figs. 3A–3F). We analyzed the heterogeneity of per body fat index-related SNPs by using Cochran’s Q test in the IVW model and the MR-Egger model. The results suggested heterogeneity among SNPs (all p_-het < 0.05; Table 1). Furthermore, uncorrelated horizontal pleiotropy was detected via the MR-Egger method, and the result showed that the MR analyses had no uncorrelated horizontal pleiotropy (all p_-intercept < 0.01; Table 1). Moreover, we used the iterative leave-one-out analysis method to determine whether a single SNP significantly modifies the pool effect value of the IVW. The result indicated that no single SNP significantly disrupted the combined effect of IVW (all p < 0.05; File S2). Finally, the causal hypothesis of the six body fat indexes’ influences on LTL was examined using the MR Steiger algorithm. The results suggested that the six body fat indexes affecting LTL was the correct causal direction (all p < 0.001).