Heterogeneous associations of multiplexed environmental factors and multidimensional aging metrics

Abstract

Complicated associations between multiplexed environmental factors and aging are poorly understood. We manipulated aging using multidimensional metrics such as phenotypic age, brain age, and brain volumes in the UK Biobank. Weighted quantile sum regression was used to examine the relative individual contributions of multiplexed environmental factors to aging, and self-organizing maps (SOMs) were used to examine joint effects. Air pollution presented a relatively large contribution in most cases. We also found fair heterogeneities in which the same environmental factor contributed inconsistently to different aging metrics. Particulate matter contributed the most to variance in aging, while noise and green space showed considerable contribution to brain volumes. SOM identified five subpopulations with distinct environmental exposure patterns and the air pollution subpopulation had the worst aging status. This study reveals the heterogeneous associations of multiplexed environmental factors with multidimensional aging metrics and serves as a proof of concept when analyzing multifactors and multiple outcomes.

Fig. 1. Overview of the study design.

A The complex associations with multiplexed environmental factors and multidimensional aging metrics. B Weighted quantile sum regression (WQS) and self-organizing maps (SOM) were performed to deal with the high dimensionality and collinearity of multiplexed environmental factors and figured out the individual and joint effects of multiplexed environmental factors on aging, respectively. Five subpopulations with specific environmental exposure patterns were distinguished and were reflected in exact locations on the UK map. Cartoon figures can be freely downloaded at https://www.iconfont.cn/.

Fig. 2. Heatmaps of Spearman’s correlations among multiplexed environmental factors.

(A) and multidimensional aging metrics (B). PM₁₀, particulate matter with aerodynamic diameter ≤10 µm; PM_2.5, particulate matter with aerodynamic diameter ≤2.5 µm; PM_2.5–10, particulate matter with aerodynamic diameter between 2.5 and 10 µm; NO₂, nitrogen dioxide; NO_x, nitrogen oxides. We used Spearman’s correlations to assess the correlations among multiplexed environmental factors (A) and multidimensional aging metrics (B). Source data are provided as a Source Data file.

Fig. 3. Relative individual contributions of multiplexed environmental factors to aging metrics.

PM₁₀, particulate matter with aerodynamic diameter ≤10 µm; PM_2.5, particulate matter with aerodynamic diameter ≤2.5 µm; PM_2.5–10, particulate matter with aerodynamic diameter between 2.5 and 10 µm; NO₂, nitrogen dioxide; NO_x, nitrogen oxides. We used WQS to evaluate the relative individual contribution of multiplexed environmental factors to aging metrics. Source data are provided as a Source Data file.

Fig. 4. Associations of each environmental factor with volumes of aging-related regions and subcortical areas, and cognitive performances.

We used linear regression to evaluate the association of each environmental factor with aging-related regions, subcortical areas, and cognitive performances. Benjamini–Hochberg procedure was used to control the family-wise error rate in the main analyses (n = 285). Only associations with an FDR < 0.05 were displayed. The height, color, and size of each data point indicate the coefficient (β) between each environmental factor and one aging metric. The horizontal dashed line denotes the positive and negative correlation boundary. Source data are provided as a Source Data file.

Fig. 5. The characteristics (A) and distributions of identified subpopulations (B, C), and the associations of subpopulations with various aging metrics (D).

SOM, self-organizing map. We used SOM analyses to recognize group structure. Populations were differentiated into air subpopulation (red, specific name was used only to refer to the main environmental exposure feature, not all features), green space subpopulation (green), rural–urban fringe subpopulation (yellow), noise subpopulation (black), blue space subpopulation (blue), and others (gray). A Characteristics of subpopulations. A larger sector size represents the larger amount of a specific environmental factor. B Distributions of subpopulations and C Distributions separately. We used multiple linear regression models to evaluate the associations of subpopulations with various aging metrics (D). Linear regression models were performed to examine the associations of subpopulations with various aging metrics. All models were adjusted for age, sex, ethnicity, neighborhood socioeconomic status (nSES), smoking status, BMI (category variable), alcohol intake frequency, regular exercise, healthy diet, history of cardiovascular disease (CVD), and cancer at baseline. Benjamini–Hochberg procedure was used to control the family-wise error rate in the main analyses (n = 285). Two-sided P value of <0.05 was considered statistically significant (values are represented as a coefficient ± standard error of the mean. *P < 0.05, **P < 0.01, ***P < 0.005; different subpopulations vs. green space subpopulation). The blank map of the UK can be freely downloaded from GADM version 4.1 (https://www.gadm.org/). Source data are provided as a Source Data file.