Fig 2.

Panel A. Indirect mediation model shows BMI had an effect on AN-q through HOMA-IR. Panel B. Partial mediation model, the effect of HOMA-IR was completely attenuated to explain AN-q. This model fits the data well (The numbers correspond to environmental correlations and underlined numbers correspond to genetic correlations). Panel C and D are two models of partial mediation for BMI effects on metabolic and PFS traits. The thick arrows show standardized beta-coefficients greater than 0.3; thin lines show those beta coefficients between 0.3 and 0.1. Coefficients lower than 0.1 are not shown. The boxes indicate observed variables, circles indicate latent endogenous variables, those with a letter “e” are error terms. Panel C (genetic partial mediation) shows a web of genetic relationships. HOMA-IR and AN-q are mediators of BMI, perhaps activating other genes to have an effect on inflammation, lipids and physical fitness. Panel D (environmental partial mediation) supports the hypothesis that BMI is the main variable (with a small mediation effect of HOMA-IR on triglyceride concentrations) that influences the metabolic and PFS traits. This model had the best goodness-of-fit tests compared with other models of indirect effect of BMI by HOMA-IR or AN-q as mediators; * p-value < 0.01. BloodP: Blood pressure (latent factor). The dotted lines between AN-q and PFS show bidirectional (correlation) association between these two variables in an independent model.