The Impacts of Atmospheric PM_2.5 Components on Depression in Middle-Aged and Elderly People

1. Introduction

Population ageing has become a major global challenge. According to WHO, the proportion of individuals aged 65 and over will rise from 12% in 2015 to 22% in 2050, reaching nearly 2 billion [1]. It is estimated that more than 20% of the population aged 60 and over suffered from mental or neurological disorders, which accounted for 6.6% of the total Disability-Adjusted Life Years (DALYs) of this population. Depression was the most prevalent of these disorders in older population, accounting for approximately 7%.

Air pollution is indeed a significant global public health issue. Epidemiological studies about the association of fine particulate matter with depression were mainly focused on the ambient PM_2.5 mass. However, PM_2.5 is complex heterogeneous mixture of miscellaneous chemical composition, whose major components include organic matter (OM), black carbon (BC), nitrate (NO₃^-), sulfate (SO₄^2-), ammonium (NH₄⁺), soil particles (SOIL) and sea salt (SS). Previous studies were based on the assumption that the toxicity of PM_2.5 components was consistent, which overlooked the difference in potential toxicity of chemical components of PM_2.5, especially those derived from combustion processes [2,3]. Considering the facts of China's substantial elderly population and the continuous high levels of particulate matter, it is essential to assess the effect of exposure to various components of PM_2.5 on depression in middle-aged and elderly populations.

To fill these knowledge gaps, a national study was carried out, and the impact of long-term exposure to PM_2.5 and its components (including BC, OM, NO₃^-, SO₄^2-, NH₄⁺, SOIL, and SS) on depression among middle-aged and elderly people in China was investigated. And then, the modifying effects of social activity and green space on the association were evaluated.

2. Materials and Methods

2.1. Study Population

All of participants were drawn from the China Health and Retirement Longitudinal Study (CHARLS) launched by National School for Development (China Center for Economic Research). The CHARLS constructed a nationally representative subsets of Chinese older adults (participants aged 45 years or above) based on a four-stage and well-established sample design [4], in which it covered 125 cities from 28 provinces across China’s mainland. In each wave of survey, participants’ information including socioeconomic and demographic characteristics, lifestyle, behavior patterns, and health status was comprehensively collected through standard questionnaires by well-trained investigators, in the form of face-to-face interviews. More details are available elsewhere (http://charls.pku.edu.cn/en/data/User2018.pdf).To avoid the influence of COVID-19 on depression [5,6], this analysis was conducted using the CHARLS data of four waves (2011-2018). In addition, CHARLS only provides the city in which each participant resides, and detailed home addresses are not disclosed to the public.

In this study, 17,596 participants were involved at baseline. Then, adults under the age of 45 at baseline were excluded. Additionally, considering the focus on mental and neurological diseases, participants diagnosed with memory-related diseases (such as Alzheimer's Disease, Brain Atrophy, and Parkinson's Disease), with emotional or psychiatric issues, with cancer or malignant tumors (excluding minor skin cancers) or who had previously taken medication for mental illness were excluded. Individuals with missing outcome information and key personal information were excluded. Ultimately, a total of 9,725 participants were included in the analysis. Detailed inclusion and exclusion criteria were shown in Supplementary Figure S1.

2.2. Exposure Assessment

Monthly spatially-resolved estimates of surface-level total PM2.5 and its seven specific components, including organic matter (OM), black carbon (BC), sulfate (SO₄^2-), nitrate (NO₃^-), ammonium (NH₄⁺), soil particles (SOIL), and sea salt (SS) for the period 2000-2017, were obtained from the V4.CH.02 product, a dataset formulated by the Atmospheric Composition Analysis Group at Washington University [7,8]. Datasets for China was developed using the methodology described previously [9,10]. By well incorporating satellite-based aerosol optical depth observations, gridded estimates at 0.01° × 0.01° spatial resolution were predicted by utilizing the nested GEOS-Chem chemical transport model (http://geos-chem.org, accessed on November 15th, 2023) and geographically weighted regression. The predicted PM_2.5 estimates demonstrated substantial alignment with measurements from globally dispersed ground sites, evidenced by a high coefficient of determination (R² = 0.81) and a slope of 0.90. While further calibration for the compositional values was not undertaken due to the absence of available ground-based observations across mainland China, this methodology proved effective without calibration, as evidenced by its application in North America [8] and global estimates [10]. Currently, the V4.CH.02 product from China is extensively utilized for exposure assessment in the field of environmental epidemiology [3,11,12].

In this study, we extracted data on the monthly total mass of PM_2.5 and the concentrations of its seven chemical components across mainland China for the period from 2000 to 2017. Consistent with prior studies [13,14], we employed a 1-year average of PM_2.5 mass and its seven components at the city level prior to the survey year as a measure of PM_2.5 influence in our primary analyses. For instance, for participants enrolled in 2018, we aligned the annual average concentrations for 2017 as their final exposure in the wave 4 survey.

2.3. Outcome Assessment

The depression status of the participants was measured using the 10-item Center for Epidemiologic Studies Depression scale (CES-D-10). The validity of CES-D-10 has been examined among Chinese adults [15]. Each question measured the frequency of a specific type of negative mood using a score of 0 (rarely or none), 1 (some days), 2 (occasionally), 3 (most of the time). The sum of score for each question was calculated as the depressive score to indicate the general depressive status for each participant. The depressive score ranged from 0 to 30, with a higher score indicating a higher severity of depressive symptoms.

2.4. Statistical Analysis

Baseline characteristics were summarized as proportions (%) for categorical variables and mean (SD) for continuous variables. The Spearman rank correlation analysis was applied to assess the relationship among PM_2.5 mass and each component in 125 cities in 2011.

Linear mixed effect models (LME) with time-varying exposures were established to investigate associations of PM_2.5 or its components with depression score. Our analysis was first performed without adjustments (M1) and then expanded by introducing additional covariates. M2 was adjusted for age and sex. M3 (main model) was further adjusted for sociodemographic background (education attainment, marital status, building type, cooking fossil type, location of residence), health status (cardiovascular related diseases, disabled), and behavior characteristics (smoking, alcohol drinking, social activity).

$$Y_{i,j}={\beta }_0+{\beta }_{1}P{M}_{i,j}+\mathit{\gamma }{\mathit{z}}_{\mathit{i},\mathit{j}}+cit{y}_i+\lambda \left(i\right)$$

Where i denoted the subject index; j denoted the visit index; ${\beta }_0$ denoted the intercept; ${\beta }_1$ denoted the regression coefficients for PM_2.5 and its components, respectively. ${\mathit{z}}_{\mathit{i},\mathit{j}}$ denoted a set of adjusted covariates and $\gamma$ denoted the corresponding coefficients; city_i denoted a fixed effect to control the unmeasured city-specific risk 7 factors, such as traditional culture [14,16]; $\lambda$ denoted random intercept to model the correlation between records from the same subject, respectively.

To check the exposure-response (ER) curve for the associations of exposure to PM_2.5 and its 7 components with depression score, we separately included a cubic spline with 4 knots for each exposure of interest based on generalized additive mixed models (GAMM). The significance of linear and nonlinear terms was obtained using the Wald test, while the likelihood ratio test was used to compare the merits of linear and nonlinear models. Subgroup analyses were performed stratified by whether participated in social activity (yes / no) and level of green space coverage (high / low), which based on the median Normalized Difference Vegetation Index (NDVI) of study cities. Furthermore, we merged these two factors, to compare the effect in population participated in social activity and in high greenness (Yes + High) with no social activity and in low greenness(No + Low). In line with prior studies, the 2-sample z-test were used to examine potential effect differences between subgroups [17]. The subgroup analyses were conducted only for PM_2.5 and components which were significantly associated with outcomes. To determine the effect of mixed exposure of PM_2.5, and identify the relative contribution of each component, weighted quantile sum (WQS) regression was conducted adjusting confounders which were included in the main model [18].

Several sensitivity analyses were performed to evaluate the robustness of our effect estimates. First, we further adjusted PM_2.5 or residuals of PM_2.5 to control for confounding among the components of PM_2.5 due to high correlation. Second, we further adjusted the random effect of community using a random intercept. Third, we switched the model from an individual random effects model to an individual fixed effects model (less efficient in estimation, but with smaller bias) [19], controlling for the same covariates as in the main model, although some covariates that do not vary across individuals were absorbed by the individual fixed effects. Fourthly, to explore how the measurement error influenced the estimated association between PM_2.5 and its constituents with depression, we used a bootstrap method to incorporate the data-generation procedure during model inference [20]. The detailed information for bootstrap method could be found in Supplementary Text S1.

Statistical analyses were conducted using R (version 4.2.1). This included the application of the "lme4", "lmerTest", and "gamm4" packages for the execution of linear mixed effect model and generalized additive mixed model analyses. Additionally, the "plm" package was employed for the analysis of the fixed effects model. The "gWQS" package was employed for identifying contributions of all components. Two-sided p-value< 0.05 was considered statistically significant.

3. Results

3.1. Descriptive Results

The distribution of annual average of PM_2.5 and its components at baseline were shown in Table S1 in the supplement. The concentration of PM_2.5 ranged from 4.32 to 82.41μg/m³, with an average [SD] concentration of 44.11[18.63]μg/m³. The average [SD] concentrations of BC, OM, NO₃^-, SO₄^2-, NH₄⁺, SOIL, and SS were 2.76 [1.59]μg/m³, 8.62 [4.05]μg/m³, 10.17 [5.45]μg/m³, 10.15 [3.53]μg/m³, 6.71 [2.78]μg/m³, 4.52 [5.42]μg/m³, and 0.15 [0.15]μg/m³, respectively. Notably, NH₄⁺, NO₃^-, BC, OM, and SO₄^2- demonstrated higher correlations with PM_2.5, with correlation coefficients of 0.944, 0.932, 0.917, 0.886, and 0.868, respectively. SOIL exhibited a moderately correlation with PM_2.5, with a coefficient of 0.428. SS presented negative correlation with PM_2.5 and most of other components, although these correlations were relatively weaker. The correlation coefficients among BC, OM, NO₃^-, SO₄^2-, and NH₄⁺ were varied among 0.735-0.967 (Table S2 in the supplement).

Table 1 showed the characteristics for population in this analysis (N=36,847). The age (mean [SD]) of participants at baseline was 57.63 [8.34] years, and 4,736 (48.7%) were male. At baseline (2011), a higher proportion of participants completed primary education (41.6%), got married (90.5%), never drank (65.6%), lived in single story (62.6%) and used fossil fuels (55.0%). A lower proportion of participants were smokers (39.9%), lived in urban area (38.1%), and had cardiovascular disease (35.2%) or diabetes (13.9%). By 2018, the proportion married, living in a single story and using fossil fuels had declined to 84.8%, 49.3% and 33.4%, respectively. The proportion smoking had increased to 43.1%, and the proportion with CVD and diabetes had increased to 59.2% and 38.4%, respectively.

3.2. Association of PM_2.5 and Its Components with Depression Score

The results from the primary linear mixed effect model analysis (M3) revealed significant associations between exposure to PM_2.5, BC, OM, SOIL, and SS and increased depression scores in the population. For each interquartile range (IQR) increase in exposure to these pollutants, depression score significantly rose by 0.90 (95% CI: 0.59, 1.21), 0.71 (95% CI: 0.34, 1.09), 0.94 (95% CI: 0.61, 1.26), 0.51 (95% CI: 0.38, 0.63) and 0.53 (95% CI: 0.33, 0.73), respectively. For each IQR increase in SO₄^2-, depression score rose by 0.18 (95% CI: -0.03, 0.38, p=0.088), although not significantly. No significant associations were observed of exposure to NO₃^- and NH₄⁺ with depression scores (Table 2).

Figure 1 showed the nonlinear associations of PM_2.5 and its components with depression scores. The exposure-response relationship for depression scores indicated that the nonlinear models for PM_2.5, BC, OM, and SS did not significantly improve over the linear models (pcompare > 0.05). The associations of NO₃^-, SO₄^2-, NH₄⁺ were not significant in either linear or nonlinear models (plinear > 0.05, pnonlinear >0.05). The nonlinear model for SOIL showed a significant improvement over the linear model, but the association between SOIL and depression scores was almost linear within 95% of the concentration distribution.

3.3. Contribution of PM_2.5 Components to Depression

The combined effects and contribution of the seven components were shown in Figure 2. For each IQR increase in the mixture of all components, depression score increased by 1.104 (95%CI: 0.901, 1.307), and BC held the largest index weight (33.6%), followed by SOIL (28.59%) and SS (25.05%). Compared with single exposure model, the contribution of OM was weaker in the mixture, with index weight of 7.38%.

3.4. Subgroup Analyses

For this analysis, stratified analyses were conducted for PM_2.5, BC, OM, SOIL, SS, which were significantly associated with depression, as well as for SO₄^2-, which was marginally significant. Social activity may be a potential modifier of PM_2.5, OM, and SO₄^2-. In those did not participated in social activities, depression scores increased by 0.99 (95% CI: 0.68, 1.31), 1.04 (95% CI: 0.71, 1.37), and 0.26 (95% CI: 0.04, 0.48) for each IQR increase of these three pollutants, respectively. The harmful effects of these three pollutants were lower among those who participated in social activities, with depression scores increasing by 0.81 (95% CI: 0.50, 1.13), 0.85 (95% CI: 0.51, 1.18), and 0.11 (95% CI: -0.11, 0.32), respectively. Green space may be a potential modifier of OM, SO₄^2-, SOIL and SS. In those exposed to low level green space, depression scores increased by1.18 (95% CI: 0.83, 1.53) and 0.22 (95% CI: 0.00, 0.44) for each IQR increase of OM and SO₄^2-, respectively. The harmful effects of these pollutants were even lower among those exposed to higher level green space, with depression scores changing by 0.82 (95% CI: 0.49, 1.16) and -0.01 (95% CI:-0.27, 0.25), respectively. Furthermore, compared with population with “social activity + high greenness”, those with “no social activity + low greenness” showed higher depression score cause by these pollutants. We also found that individuals living in high levels green space were more susceptible to SOIL and SS. For each IQR increase in SOIL or SS, the increase in depression scores was 0.81 (95% CI: 0.61, 1.01) and 0.65 (95% CI: 0.44, 0.86) in high level green space, higher than 0.45 (95% CI: 0.31, 0.58) and 0.48 (95% CI: 0.28, 0.68) for low level green space (Figure 3 and Table S3 in the supplement).

3.5. Sensitivity Analyses

The sensitivity analysis results were shown in Table 2. After adjustment for within-community correlations, the results for each pollutant remained robust. After further adjustment for residual of PM_2.5, or replacing individual random intercepts with individual fixed effects, the point estimate for the association between NO₃^- and depression scores changed from negative to positive, although not significantly. The results after bootstrap error correction were all comparable to the estimates obtained directly from regression in the two analysis.

4. Discussion

In this research, we found that exposure to PM_2.5 mass and its components (BC, OM, SOIL, SS) were associated with higher depression levels in middle aged and older population in China. We further identified the relative contribution of each components, and found the protective effect of social activity and green space. The results would potentially contribute to a more nuanced understanding of the health impact of harmful components in ambient particulate pollution.

4.1. Main Findings

Consistent with our findings, a meta-analysis encompassing 16 studies [21] identified significant positive correlations between PM_2.5 exposure and depression within the middle-aged and older populations in China. A quasi-experimental study utilizing the CHARLS database found that long-term exposure to PM_2.5 led to a linear increase in depression score [14]. A cohort study based on China Family Panel Studies (CFPS) found that long-term exposure to PM_2.5 components were associated with the existence of depressive symptoms in adults, and for each standard unit increase in BC, OM, NH₄⁺, NO₃^- and SO₄^2-, the odds ratio (OR) was 1.118 (95%CI: 1.020, 1.225), 1.134 (95%CI: 1.028, 1.252), 1.127 (95%CI: 1.011, 1.255), 1.107(95%CI: 9.981, 1.248) and 1.117 (95%CI: 1.020, 1.224), respectively [12]. Our research based on the CHARLS cohort offered results for older population and about additional components. We further found the potential harmful effect of natural sources of PM_2.5, such as SOIL and SS. Although we did not find significant association between water-soluble ions (NO₃^-, SO₄^2-, NH₄⁺) and depression, we found that after further adjusting for residual of PM_2.5, or applying fixed-effect model, the harmful effects tended to be significant, which suggested that the associations we did not initially find may be influenced by confounder. The evaluation of non-linear relationships indicated that linear models in our main analysis were able to simulate the actual associations of pollutants with outcomes, as the application of non-linear models did not significantly improve model performance in comparison to linear models. Despite of the high variability in SOIL, the exposure-response relationship within the 95% concentration range essentially maintained a linear association with the outcomes.

4.2. Contributions of PM_2.5 Components and Underlying Biological Mechanism

The toxicity of PM_2.5 mixtures may be depended on its components. However, there is currently insufficient evidence on the toxicity of the various components of PM_2.5 on depression. We observed that BC had the largest hazardous contribution to depression. BC is smaller and more likely to enter the circulation and brain parenchyma through olfactory or sensory nerve pathways, inducing neurotoxic effects. Animal experiments showed that BC may cause inflammation and oxidative stress and its molecular precursors, polycyclic aromatic hydrocarbons (PAHs), could also cause neurotoxicity [22,23]. We observed that the effect of OM on depression in the single exposure model was higher than other pollutants, while in the mixed exposure model the contribution of OM was relative weaker. It may be attribute to the similar sources of OM and BC, and the high correlation possibly affect the result of single exposure model for OM. BC and OM predominantly originates from the incomplete combustion of fossils, biofuels, and biomass, including motor vehicle emissions and coal combustion [24,25].

We found the relative high contribution of SS and SOIL to depression. SOIL particles often contain heavy metals and silica dioxide [26,27], substances that can pose toxicity risks to human health, and are mainly suspended in the air due to mechanical processes, including wind erosion and transportation, as well as industrial activities such as mining and cement production. A Boston study found the association of Nickel (Ni) in PM_2.5 with reduced verbal memory, recognition, and executive function [28]. A cross-sectional study found that potassium (K) and iron (Fe) were associated with poorer cognitive function [29]. SS is primarily composed of elements such as sodium (Na+), chloride (Cl-), and potassium (K+), along with other typical ions in seawater. The concentration of these naturally occurring particulate matters is influenced by wind speed, temperature, and humidity levels [30]. It was reported that SS was associated with hypertension [3] and dementia [31]. There were almost no epidemiological studies or mechanistic studies about the effects of sea salt on depression, but both our study and that of Li et al. have found that SS may have a detrimental effect on brain function, which deserved to be brought to the attention of future researchers.

NO₃^-, SO₄^2-, NH₄⁺ are secondary inorganic aerosols present in the atmosphere, predominantly generated from photochemical reactions involving gases released from the combustion of fuels like coal and petroleum. These reactions transpire under the influence of ultraviolet light, ozone, and hydroxyl radicals. Although we did not observe significant associations between these ions and depression, previous studies indicated that they could be neurotoxic agents [32]. Sulphate was able to create an acidic environment in the microcirculation and promotes the absorption of metallic components of particulate matter, allowing toxic compounds to settle deep in the lungs [33]. NH₄⁺ and NO₃^- could lead to inflammation, oxidative stress and mitochondrial impairment [34,35].

Our results showed the high contribution of BC, SOIL, and SS to depression in Chinese middle-aged and elder population, suggesting precise control of PM_2.5 emissions in the following ways. Firstly, we should continue to promote the use of clean energy, strictly control new production capacity in high-energy-consuming and high-polluting industries, and strengthen the control of industrial and traffic emissions. Secondly, the prevention and control of soil pollution and groundwater pollution should be promoted, including the monitoring of soil heavy metals and other harmful substances, pollution risk control and remediation, and the control of pollution sources in arable land and industrial and mining enterprises. Thirdly, establishing a cooperative management system for sea salt in coastal cities, improving the monitoring of the exposure level and composition of sea salt, and strengthening health education for local residents [35].

4.3. Modification Effect of Social Activity and Greenspace

Previous studies have shown that frequent social activity was associated with lower risk of depression in older adults [36,37,38]. In this study, we found the associations of PM_2.5 and some of its components with depression were lower in middle-aged and elderly people who regularly participated in social activities. We found that the associations of OM、SO₄^2- with depression were lower in those living at high green space levels. The findings also indicated that those who both engaged in social activities and resided in areas with high green space levels exhibited the least adverse effect of PM_2.5 components on depression. Previous research has indicated that green space was associated lower risk of depression and anxiety in the elderly population [39]. The potential mechanisms for the effect of green remains unclear, but may be explained by improving air quality, providing places for physical and social activity, and reducing stress. We also found that the harmful effect of SOIL on depression was higher in people in high level green space. It may be attributed to the different components of SOIL particles in different areas of China [40,41] and from different sources. Previous studies have found the effect of residential, transport and power generation on iron and lead concentrations in PM_2.5 [42]. A cross-sectional study found that the adverse effect of TC, NH₄⁺, K, and Fe was higher in people living in areas with lower greenspace, while the effects of Ca²⁺, Na⁺ and V were lower in lower greenspace [29]. More researches about the effect of different SOIL components in various area are needed to achieve precise control of air quality. In conclusion, this study calls for multi-pronged social support to reduce the risk of depression more effectively in the elderly population. On the one hand, for the individual, older people should be encouraged to go out of their homes and participate in social activities. On the other hand, at the environmental level, more outdoor activity areas and social facilities should be provided, such as pocket park and forest park.

4.4. Strengths and Limitations

This study has several major strengths. Primarily, it was the first nationwide epidemiological investigation exploring the relationship between PM_2.5 and its components with depression in China's middle-aged and older populations utilizing a longitudinal design. We identified several components that play a major contribution, which would provide theoretical support for precise control of PM_2.5 pollution, particularly in low- and middle-income regions experiencing high levels of air pollution. Secondly, our study employed a range of sensitivity analyses to verify the robustness of our findings. These methods encompassed adjustments for all available confounding factors, additional adjustments for PM_2.5 residual to assess the independent effects of each component, the use of fixed-effect models that offer reduced bias albeit with lower power, bootstrap simulations to model the impact of exposure errors, and accounting for community correlation.

However, our study also acknowledges some limitations. First, due to lack of detailed residential addresses of participants, we assigned exposure at the city level. Previous study suggested that bias might result in the investigated associations towards null effects and generally underestimated the risk [43]. However, the results kept robust after the simulation of exposure errors using the bootstrap methods. What’s more, because this was an observational study, it was not possible to determine causal associations. To more definitively ascertain the health impacts of PM_2.5 and its components on mental and neurodegenerative diseases, larger population samples, more accurate exposure measurements, and enhanced study methodologies are needed in future research.

5. Conclusions

In conclusion, exposure to PM_2.5 and its components were associated with depression, and BC, SOIL and SS may make major contributions. More social activity and higher greenspace may help to reduce the depression symptoms caused by PM_2.5 and its components. These findings hold substantial relevance for enhancing the understanding of the harmful effects of PM_2.5 components and offer epidemiological evidence particularly pertinent to low- and middle-income regions experiencing high levels of pollution.