Associations of air pollution with obesity and body fat percentage, and modification by polygenic risk score for BMI in the UK Biobank

Abstract

Air pollution has consistently been associated with cardiometabolic outcomes, although associations with obesity have only been recently reported. Studies of air pollution and adiposity have mostly relied on body mass index (BMI) rather than body fat percentage (BF%), and most have not accounted for noise as a possible confounder. Additionally, it is unknown whether genetic predisposition for obesity increases susceptibility to the obesogenic effects of air pollution. To help fill these gaps, we used the UK Biobank, a large, prospective cohort study in the United Kingdom, to explore the relationship between air pollution and adiposity, and modification by a polygenic risk score for BMI. We used 2010 annual averages of air pollution estimates from land use regression (NO₂, NO_X, PM_2.5, PM_{2.5absorbance}, PM_2.5-10, PM₁₀), traffic intensity (TI), inverse distance to road (IDTR), along with examiner-measured BMI, waist-hip-ratio (WHR), and impedance measures of BF%, which were collected at enrollment (2006-2010, n = 473,026) and at follow-up (2012-2013, n = 19,518). We estimated associations of air pollution with BMI, WHR, and BF% at enrollment and follow-up, and with obesity, abdominal obesity, and BF%-obesity at enrollment and follow-up. We used linear and logistic regression and controlled for noise and other covariates. We also assessed interactions of air pollution with a polygenic risk score for BMI. On average, participants at enrollment were 56 years of age, 54% were female, and 32% had completed college or a higher degree. Almost all participants (~95%) were white. All air pollution measures except IDTR were positively associated with at least one continuous measure of adiposity at enrollment. However, NO₂ was negatively associated with BMI but positively associated with WHR at enrollment, and IDTR was also negatively associated with BMI. At follow-up (controlling for enrollment adiposity), we observed positive associations for PM_2.5-10 with BMI, PM₁₀ with BF%, and TI with BF% and BMI. Associations were similar for binary measures of adiposity, with minor differences for some pollutants. Associations of NO_X, NO₂, PM_{2.5absorbance}, PM_2.5 and PM₁₀, with BMI at enrollment, but not at follow-up, were stronger among individuals with higher BMI polygenic risk scores (interaction p <0.05). In this large, prospective cohort, air pollution was associated with several measures of adiposity at enrollment and follow-up, and associations with adiposity at enrollment were modified by a polygenic risk score for obesity.

Keywords: Air pollution; BMI; Epidemiology; Gene by environment interactions; Obesity.

Figure 1. Associations (ß, 95% CI) of Air Pollution With Continuous Measures of Adiposity at enrollment and at follow-up

N=473,026 at enrollment for NO and NOx models, N=440,193 for PM models. N=19,518 at follow-up. IDTR = log Inverse distance to road; TI = log Traffic Intensity. Air pollutants are scaled such that a one unit increase represents a one inter-quartile range increase, although inverse distance to road and traffic density are logged. All models controlled for race/ethnicity, age, assessment centre, Townsend Deprivation Index, education, genetic risk score for BMI, noise, and sex. BMI, WHR, and Body Fat percentage are modeled as inverse normalized variables. Associations with measures at follow-up additionally controlled for the relevant body composition measure at enrollment.

Figure 2. Associations (Odds Ratios & 95% CIs) of Air Pollution With Binary Measures of Adiposity at Enrollment and Follow-Up

At enrollment, N=473,026 for NOx and NO2 models, and N = 440,193 for PM models. At follow-up, N=19,518 and N= 19,514. BMI indicates BMI-based obesity; WHR indicates WHR-based obesity; and BodyFat% indicates body fat percentage-based obesity. IDTR = log Inverse distance to road; TI = log Traffic Intensity; these were modeled as log variables. Other air pollutants are scaled such that a one unit increase represents a one inter-quartile range increase. Models control for race/ethnicity, age, assessment centre, Townsend Deprivation Index, education, genetic risk score for BMI, noise, and sex. Associations with measures at follow-up are incident associations and exclude participants with the condition at enrollment. Note that the scales are different for enrollment and follow-up, for visualization purposes.

Figure 3. Associations of Air Pollution with BMI at Enrollment, by Genetic Risk Score for BMI

At enrollment, N = 445,257 for NOx and NO2 models, N = 440,193 for PM models. Models are restricted to those of British/Irish white race/ethnicity and control for sex, age, enrollment centre, education, noise, and the Townsend Deprivation Index. Interaction p values represent the likelihood ratio test p-value for models of interactions between the air pollutant and the continuous genetic risk score for BMI.