Using non-linear and stratified Mendelian randomisation to explore relationships between body mass index and the plasma proteome

Elevated body mass index (BMI) raises the risk of diseases such as Type 2 Diabetes (T2D). The circulating proteome has been explored as an intermediate between adiposity and disease, where alterations in growth hormone receptor (GHR) and insulin-like growth factor binding proteins (IGFBP1/2) likely mediate some risk. These associations are often modelled using linear methods but recent studies have indicated that such relationships may be non-linear. This is important when measuring the effect of BMI on protein levels as it may have implications for understanding how weight loss interventions influence disease risk reduction. ¬We estimated linear and non-linear associations between BMI and protein (SomaLogic, N>3600) levels using observational and Mendelian randomisation (MR) frameworks in INTERVAL, a UK blood donor cohort (N=2737). To estimate non-linear associations, an R package was developed (“glsmr” - generalized additive model (GAM) and linear stratified MR). Non-linear effects were identified by comparing the observational linear model and non-linear GAM models using an F-test. In addition, BMI stratified (BMI=18.5-25kg/m2, 25-30kg/m2 and 30-40kg/m2) linear observational and MR analyses were performed to estimate BMI-protein associations in each stratum. We observed attenuated estimates for BMI on levels of proteins including leptin, GHR and IGFBPs when comparing the obesity stratum with normal-weight and overweight strata. These analyses suggest weight loss interventions may be less effective at higher BMIs where plateauing associations between BMI and protein level is observed. This could have implications for weight loss interventions and disease risk reduction.