Using Mendelian randomization to investigate a possible causal relationship between adiposity and increased bone mineral density at different skeletal sites in children

Abstract

Background: Lean mass is positively associated with bone mineral density (BMD). However, the relationship between adiposity and BMD is more controversial. In particular, it is unclear if the observational association between the two reflects a causal effect of fat mass on BMD. Previous Mendelian randomization (MR) studies using variants in the FTO and MC4R genes as genetic instruments for adiposity have suggested that fat mass does indeed causally influence BMD. However, it is possible that these genetic variants pleiotropically influence lean mass and affect BMD through pathways independent of adiposity, invalidating one of the core assumptions of MR and complicating interpretation of the analysis.

Methods: To investigate whether adiposity causally affects BMD, we investigated the relationship between fat mass and BMD at the skull (SK), upper limbs (UL) and lower limbs (LL), spine (SP) and pelvis (PE), using 32 body mass index (BMI)-associated SNPs, including a variant near ADCY3 that was strongly associated with fat but not lean mass in our sample. Dual-energy X-ray absorptiometry (DXA) scans and genetic data were available for 5221 subjects (mean age 9.9 years) from the Avon Longitudinal Study of Parents and Children. We performed a series of MR analyses involving single BMI-associated SNPs and allelic scores of these SNPs. We used new extensions of the MR method including MR Egger regression and multivariable MR, which are more robust to possible confounding effects due to horizontal pleiotropy and, in the case of multivariable MR, specifically account for the effect of lean mass in the analysis. Bidirectional Mendelian randomization analysis was also performed to examine whether BMD causally affected BMI and adiposity.

Results: Observationally, fat mass was strongly positively related to BMD at all sites, but more weakly at the skull. Instrumental variables (IV) analyses using an allelic score of BMI SNPs suggested that fat mass was causally related to LL-BMD, UL-BMD, SP-BMD and PE-BMD but not SK-BMD. Multivariable MR, Egger regression and IV analyses involving the ADCY3 variant suggested a positive causal effect of adiposity on all sites except the skull, and that an effect was present even after taking lean mass into account. Finally, IV analyses using BMD allelic scores showed no evidence of reverse causality between BMD and fat mass.

Conclusions: Our results suggest that adiposity is causally related to increased BMD at all sites except the skull, perhaps reflecting positive effects of loading on bone formation at weighted but not unweighted sites. In contrast, we found no evidence for BMD causally affecting BMI or measures of adiposity. Our results illustrate how MR can be used profitably to investigate clinical questions relevant to osteoporosis.

Keywords: ALSPAC; Bone; Mendelian randomization; body mass index; genetics.

Figure 1

Directed acyclic graph illustrating core instrumental variable assumptions of the Mendelian randomisation approach. The SNP/allelic score used as an instrumental variable (Z) is (1) associated with the exposure of interest (X), (2) independent of unmeasured confounders (U), and (3) independent of the outcome (Y) given the exposure and unmeasured confounding factors. Estimates of the causal effect of the exposure on the outcome can be obtained using a number of estimators including the ratio of the estimated instrumental variable and outcome association to the instrumental variable and exposure association.

Figure 2

Funnel plots displaying the strength of association between each of 32 SNPs ($\widehat{\mathbf{\gamma }}$) with BMI (Panel A) and Fat mass (Panel B) plotted against the causal estimate ($\widehat{\mathrm{\beta }}$_IV) of each SNP on BMD measured at the skull (SK) and lower-limbs (LL). The inverse-variance weighted and MR Egger causal effect estimates are represented by a red and blue line respectively.

Figure 3

Scatter plots displaying estimates of the association between each SNP and the relevant BMD outcome ($\widehat{\mathbf{\Gamma }}$) against effect estimates of each SNP with the relevant exposure [i.e BMI (panel A) and Fat mass (panel B)]. The slope of the blue line through the plot represents the MR Egger regression estimate ($\widehat{\mathrm{\beta }}$_IV) of the causal effect of the exposure on the outcome. The y-intercept of the blue regression line denotes the estimate of the degree of directional pleiotropy in the dataset ($\widehat{\mathrm{\beta }}$₀). The inverse-variance weighted causal effect estimate is represented by the slope of the red line.

Figure 4

Directed acyclic graphs illustrating three scenarios that potentially account for the causal relationship between adiposity and bone mineral density (BMD), in addition to the shared associations with lean mass. Panel A depicts a scenario in which the causal influence of adiposity on BMD is mediated by lean mass. Panel B depicts a scenario in which SNPs affecting adiposity also directly influence lean mass via horizontal pleiotropy, and both fat mass and lean mass have direct causal effects on BMD. Panel C depicts a combination of the scenarios illustrated by panel A and B.