Relationship of diet and physical activity with genetic susceptibility to obesity: a longitudinal analysis in adults in South Africa

Abstract

Background Obesity-related disease conditions are a major public health concern in South Africa, exerting a healthcare cost of between ZAR 30 million and ZAR 36 million, the bulk of which is due to hypertension and type 2 diabetes. Moreover, evidence reveals that women in South Africa have higher BMI compared to men, yet men exhibit less insulin sensitivity and reduced beta cell function as well as stronger associations of adiposity with type 2 diabetes compared to women. The mechanisms underlying these sex differences are not known. BMI is highly polygenic in nature; however, genetic prediction of BMI has mostly been conducted using data from European ancestry populations that have poor predictive capacity in African ancestry populations. Moreover, the relationship of polygenic risks and proteomic profiles with regards to susceptibility to obesity and related cardiometabolic traits is yet to be explored in African populations. It has also been reported that using variants associated with the statistical variance of quantitative traits (vQTLs) like BMI aids in the depiction of components of BMI genetic susceptibility, which interacts with environmental factors such as diet and exercise. However, such studies are limited in continental Africans.

Aim This thesis, sought to determine the interplay of diet and physical activity with BMI genetic susceptibility. The specific objectives were: 1. To determine the association of physical activity with BMI in middle-aged black South African men and women. 2. To develop a highly predictive genetic risk score for BMI and test its longitudinal predictive ability in middle-aged black South African men and women. 3. To determine gene x lifestyle (GXE) interactions that influence BMI in Black South African adult men and women.

Methods Data from 11853 adult men and women in the African-Wits-INDEPTH partnership for Genomic studies (AWI-Gen) Cohort was used to fulfil objective 1. To fulfil objectives 2 and 3, data from 5921 AWI-Gen cohort participants in the three South African (SA) sites and a sub-study of AWI-Gen focusing on the factors influencing the risk of type 2 diabetes mellitus among middle-aged black South African men and women (GSK) was used. For objective 1, a sex-stratified meta-analysis of cross-sectional data from the study participants was used to assess the association of physical activity with BMI. The PRS-CSx method was used to develop a multi-ancestry PRS for BMI and evaluate its longitudinal prediction of severe obesity to meet objective 2. For objective 3, the Levene’s test, implemented in the OCSA Package, was used to determine candidate gene-interacting variants that exhibited trait variance heterogeneity in the study population. Detailed methods are in the relevant sections for each objective.

Results Meeting the recommended weekly physical activity levels of at least 150 minutes was associated with a BMI that was 0.80kg/m2 lower in men (95% CI = -1.14; -0.47) and 0.68kg/m2 lower in women (95% Ci = -1.03; -0.33). Sex and site-specific differences were also observed in domains of physical activity with an inverse relationship between transport-related physical activity and BMI being observed among men in Agincourt (beta = -1.15 kg/m2, 95% CI = -2.26; -0.04) and Nanoro (beta = -0.79 kg/m2, 95%CI = -1.25; -0.33). Work related physical activity was associated with lower BMI in Navrongo men (beta = -0.76 kg/m2, 95% CI=-1.25; -0.27) and Nanoro women (beta = -0.90 kg/m2, 95%CI = -1.44; -0.36).

The multi-ancestry PRS demonstrated superior predictive ability, explaining approximately 1.9% of variance in BMI compared to 0.7% and 1.2% explained by two scores developed using single ancestry methods. In addition, over a period of ten years, the multi-ancestry PRS was associated with repeated measures of BMI (β = 1.51 p = < 0.001) and there was significant longitudinal PRS * sex interaction (Pinteraction = 0.029), prompting subsequent sex-stratified analysis. In the combined analysis of men and women, being in the top 20% of the PRS distribution (top 20) was associated with three times greater hazard of severe obesity (hazard ratio = 2.98, 95% CI = 1.75 - 5.07, p = 5.33e-05) compared to being in the bottom 20% of the PRS distribution (bottom 20). This observation was shown to be driven by women, where being a woman in the top 20 was associated with 3.5 greater hazard of severe obesity (hazard ratio 3.48, 95% CI = 1.96 – 6.16, p = 1.94e-05) compared to being in the bottom 20 while the associations were not significant in men (hazard ratio = 1.13, 95% CI = 0.24 – 5.37, p = 0.878). Comparison of the associations of dysglycaemia with PRS, BMI and the proteomic score revealed no apparent sex differences in the association between BMI PRS and dysglycaemia for most of the glycaemic markers except for Matsuda Index though men exhibited lower insulin sensitivity compared to women. The proteomic score predicted higher insulin resistance in women than in men. Gene x lifestyle interaction analysis revealed novel interactions between three genetic variants with diet and lifestyle factors. The effect of the rs557505940 variant on BMI was accentuated by higher fruit intake (betainteraction = 0.03, Pinteraction = 0.04) in the combined analysis of men and women while higher SES, carbohydrate intake and self- reported physical activity attenuated the effect of rs527747185 (betainteraction = -0.349, Pinteraction = 0.037), rs3016751 (betainteraction = -0.056, Pinteraction = 0.035) and rs188275749 (betainteraction = -0.048, Pinteraction = 0.0001) respectively on BMI in men.

Conclusions Sex and geographical differences exist in associations between domains of physical activity and BMI. In addition, genetic risk better predicts incident severe obesity in women than in men while proteomic profiles have a weak correlation with PRS and show heterogenous associations with dysglycaemia, fat distribution, nutrient patterns and physical activity between men and women. Novel GXE interactions were also observed. These results underscore the need for further inquiry into the sex differences in genetic risk and environmental factors associated with BMI. Furthermore, a precision approach to obesity prevention and control, paying attention to the sex differences and contextual factors may be more efficient.