Sponsored by:
Background: A higher body mass index (BMI) is associated with a shorter telomere length (TL), an indicator of cellular ageing. However, it remains unclear whether observational associations reflect causal relations, or instead, are due to confounding by unobserved factors or reverse causality. We assessed the robustness of observational associations to confounding.
Methods: We estimated BMI-TL associations for white adults aged ≥47 years in the (i) Swedish Twin Registry (STR, n=10,328), (ii) National Health and Nutrition Examination Surveys (NHANES, n=2,171) and (iii) Health and Retirement Study (HRS, n=3,960). TL was measured in DNA extracted from whole blood using qPCR in the STR and NHANES, and from saliva samples using qPCR in the HRS. Observational associations were estimated using Ordinary Least Squares. We estimated the association between a BMI polygenic risk score (PRS) and TL in the STR and HRS, which can provide evidence for the existence of a causal relationship. To assess the influence of confounding, we further estimated BMI-TL associations using (i) within-twin models, (ii) one sample Mendelian randomization (MR) in the STR and HRS, and (iii) two sample MR using summary statistics from genome wide association studies. We meta-analyzed observational, within-twins, and MR estimates to assess the overall association across datasets and methods.
Results: A higher BMI was associated with a shorter TL in the STR and NHANES, but with a longer TL in the HRS. A higher BMI PRS predicted a shorter TL in cross-sectional analyses in the STR and HRS, but not within-dizygotic twin pairs. Within-monozygotic twins estimates indicated negative associations between standardized BMI and standardized TL, but the 95% confidence interval was wide (β: -0.0239; 95% CI: -0.1258, 0.0779). One sample MR estimates showed negative associations between standardized BMI and standardized TL in the STR (β: -0.0763; 95% CI: -0.1443, -0.0082) and HRS (β: -0.0905; 95% CI: -0.1890, 0.0081), as did the two sample MR inverse variance weighted estimate (β: -0.05; 95% CI: -0.08, -0.02). The random-effects meta-analysis of within-twins and MR estimates showed that a one SD increase in BMI was associated with a 0.035 SD decrease in TL (95% CI: -0.0521, -0.0181).
Conclusion: The triangulation of evidence across different methods and datasets suggests that observation associations are robust to unobserved confounding.