Long-term exposure to iron and copper in fine particulate air pollution and their combined impact on reactive oxygen species concentration in lung fluid: a population-based cohort study of cardiovascular disease incidence and mortality in Toronto, Canada

Abstract

Background

Exposure to fine particulate (PM_2.5) air pollution is associated with increased cardiovascular disease (CVD), but less is known about its specific components, such as metals originating from non-tailpipe emissions. We investigated the associations of long-term exposure to metal components [iron (Fe) and copper (Cu)] in PM_2.5 with CVD incidence.

Methods

We conducted a population-based cohort study in Toronto, Canada. Exposures to Fe and Cu in PM_2.5 and their combined impact on the concentration of reactive oxygen species (ROS) in lung fluid were estimated using land use regression models. Incidence of acute myocardial infarction (AMI), congestive heart failure (CHF) and CVD death was ascertained using health administrative datasets. We used mixed-effects Cox regression models to examine the associations between the exposures and health outcomes. A series of sensitivity analyses were conducted, including indirect adjustment for individual-level cardiovascular risk factors (e.g. smoking), and adjustment for PM_2.5 and nitrogen dioxide (NO₂).

Results

In single-pollutant models, we found positive associations between the three exposures and all three outcomes, with the strongest associations detected for the estimated ROS. The associations of AMI and CHF were sensitive to indirect adjustment, but remained robust for CVD death in all sensitivity analyses. In multi-pollutant models, the associations of the three exposures generally remained unaltered. Interestingly, adjustment for ROS did not substantially change the associations between PM_2.5 and CVD, but attenuated the associations of NO₂.

Conclusions

Long-term exposure to Fe and Cu in PM_2.5 and their combined impact on ROS were consistently associated with increased CVD death.

Fine particulate, metals, non-tailpipe emission, cardiovascular disease

Key Messages

• There is limited evidence of the cardiovascular effects of specific components of fine particulate (PM_2.5), such as metal components originating from non-tailpipe emissions.

• In a large population-based cohort, we observed positive associations between long-term exposure to iron and copper in PM_2.5 and their combined impact on the concentration of reactive oxygen species (ROS) in lung fluid and the incidence of acute myocardial infarction, congestive heart failure and cardiovascular mortality.

• The observed associations were stronger for the estimated ROS than for iron and copper individually.

• Our findings suggest that PM_2.5 air pollution originating from non-tailpipe sources may have adverse cardiovascular effects.

Introduction

Particulate matter (PM) air pollution is a global public health concern responsible for approximately 4.6 million global deaths in 2017, predominantly due to cardiovascular and respiratory causes.¹ Although studies have consistently linked PM air pollution, especially fine particulate (PM with aerodynamic diameter < 2.5 µm; PM_2.5), to increased cardiovascular mortality, ^2–4 less is known about the specific PM components or sources that are more relevant to the adverse health effects of PM. Metal contents of PM are thought to play an important role with respect to adverse health effects because of their potential in producing reactive oxygen species (ROS),⁵ which has been hypothesized as a key mechanism linking PM and cardiovascular disease (CVD).⁶ However, current epidemiological evidence related the long-term health effects of exposure to PM components remains scarce as most previous studies only measured PM mass concentrations, assuming equal toxicity for all components. In addition, although a growing number of studies in the US and Europe have investigated the health effects of long-term exposure to PM components including metals on CVD mortality^7–13 and coronary heart events , ¹⁴ the results are inconsistent and there is limited agreement on which particular group(s) of metal components are more essential.

Traffic is a major source of PM air pollution in urban areas. In addition to tailpipe emissions, non-exhaust or non-tailpipe sources (e.g. brake wear, road wear, tire wear and road dust re-suspension) also contribute to a nontrivial portion of PM.¹⁵ Animal and In vitro experiments have shown the adverse health effects of non-tailpipe particles,¹⁶ but epidemiological evidence from human populations is still limited. Such evidence may help facilitate the development of targeted policies to reduce non-tailpipe emissions and thus to further reduce the population health impacts of air pollution. In the present study, we characterized the associations between long-term exposure to iron (Fe) and copper (Cu) in PM_2.5 and the incidence of acute myocardial infarction (AMI), congestive heart failure (CHF) and CVD death. Mainly originating from non-tailpipe emissions (e.g. brake/rail wear and engine abrasion), Fe and Cu in PM_2.5 have been widely considered as a measure of non-tailpipe traffic air pollution in several previous studies.^10–14 To better capture the combined exposure arising from ambient levels of Fe and Cu, we also considered a third exposure metric that estimated the total impact of Fe and Cu on ROS concentration in lung epithelial lining fluid and we examined its associations with CVD outcomes of interest. Furthermore, we sought to isolate the effects of Fe and Cu from that of PM_2.5 and nitrogen dioxide [NO₂, a valid marker for traffic related air pollution] and to each other.

Methods

Study population

Study participants were from the Ontario Population Health and Environment Cohort (ONPHEC), a population-based cohort established to investigate the health effects of environmental stressors on chronic diseases in Ontario, Canada.¹⁷ In brief, ONPHEC comprised all Canadian-born adult residents of Ontario, who were registered with provincial health insurance on 1 April, 1996 onwards. The cohort was established through linking the records of population-based health administrative databases at ICES (formerly Institute for Clinical Evaluative Sciences) in Toronto. Those who were not born in Canada were excluded using the Immigration, Refugee and Citizenship Canada (IRCC) Permanent Resident Database. Use of the data in the present study was authorized under section 45 of the Personal Health Information Protection Act of Ontario, which does not require review by a Research Ethics Board.

Toronto is the provincial capital of Ontario, and the most populous city in Canada with a recorded population of 2.73 million in 2016.¹⁸ In the present study, we restricted our study population to the residents of the City of Toronto, who were aged 40 to 85 years on 1 January 2001, and had been living in Toronto for ≥ 3 years prior to cohort enrolment. We made the restriction on age because incidence of AMI and CHF is very low in younger adults.^19,20 Meanwhile, excluding immigrants and subjects living in Toronto for a short period could reduce the potential of exposure misclassification. A total of 834 445 participants were included in the analysis of CVD death. For AMI and CHF, participants with a history of the conditions under investigation at baseline were excluded, leaving 819 552 and 803 894 participants in the AMI cohort and CHF cohort, respectively.

Air pollution exposure assessment

We used land use regression models to estimate the mass concentrations of Fe and Cu in PM_2.5 and their combined impact on ROS concentration in lung epithelial lining fluid.²¹ These models were developed based on large-scale air monitoring campaigns conducted during the summer of 2016 and the winter of 2017 in Toronto, Canada. The cross-validation R² values for predicting the annual concentrations of Fe (ng/m³), Cu (ng/m³) and ROS (nM) were 0.51, 0.75, and 0.71, respectively. We assigned the 2016/2017 exposure surface of Fe, Cu and ROS to earlier years to calculate the historical annual average concentrations over our study period. We also estimated ambient PM_2.5 and NO₂ concentrations. Briefly, ground-level PM_2.5 was estimated using a combination of satellite and chemical transport model data to produce a high resolution surface at approximately 1 km × 1 km over 2000 to 2016. The models were validated against ground PM_2.5 concentrations measured at fixed monitoring stations across North America (R² = 0.80).²² NO₂ was estimated using a land use regression model which was developed based on the measurement of ground-level NO₂ concentrations at 95 locations across the City of Toronto in 2006 (R² = 0.69).²³ Yearly average concentrations of the air pollutants were estimated at the centroid of each participant’s annual six-character residential postal code over the study period, accounting for residential mobility. We calculated 3-year moving average concentrations of Fe and Cu and related ROS as indicators of long-term exposure. (See Supplementary Material, available as Supplementary data at IJE online for more details about exposure assessment.)

Outcome ascertainment

We ascertained incident cases of AMI, CHF and CVD death during the study period (2001 to 2016) using the Ontario Myocardial Infarction Database, the Ontario Congestive Heart Failure Database and the Ontario Registrar General Death Database (ORGD), respectively. These datasets were linked using unique encoded identifiers and analyzed at ICES. (See Supplementary Material, available as Supplementary data at IJE online for more details about these datasets.) All participants were followed until death, moving out of Toronto, first occurrence of the investigated conditions (for AMI and CHF incidence), or the end of the follow-up (31 December, 2016), whichever came first.

Covariates

We obtained demographic information (including age and sex) at baseline. Information on individual-level socio-economic status (SES) and lifestyle factors (e.g. physical activity and smoking) was not available in our dataset. Instead, we derived four SES indicators at the level of Canadian Census dissemination area, including income quintile, percentage of recent immigrants, percentage of population aged ≥ 15 years who had an educational level lower than high school, and percentage of population aged ≥15 years who were unemployed from the 2001, 2006 and 2011 Canadian Census. Previous studies including several Canadian studies^24–26 have suggested that the socioeconomic gradients associated with individual- and area-level SES measures have been generally consistent, with comparable trends in risk factors and health outcomes such as smoking and obesity. Therefore, adjustment for area-level SES may help to control the influence of individual-level SES and lifestyle risk factors.

Statistical analysis

We used mixed-effects Cox proportional hazard regression models with neighborhood as the random intercept to investigate the associations of Fe and Cu mass concentrations, and the measure of their combined impact on ROS generation with incident AMI, CHF and CVD death. There are 140 neighborhoods in the City of Toronto, defined based on the Statistics Canada census tract boundaries.²⁷ The mixed-effects model enabled us to account for the potential influence of unmeasured factors which are more similar among participants living in the same neighborhood but may vary across different neighborhoods, and it has been used in our previous studies.^28–30 First, we developed crude models with stratification by baseline age (in 1-year intervals) and sex, allowing for variations in baseline risks by age and sex. Then we developed adjusted models (main models) with adjustment for neighborhood-level SES factors. The exposures and neighborhood-level variables were included in the regression models as time-varying variables. Time-in-study (i.e. time of follow-up) was used as the timescale. Proportional hazard assumption was examined using plots of Schoenfeld residuals and no notable violation was found. We performed a complete case analysis, and participants with complete information on exposure and covariates at baseline were included in Cox models. Because Fe and Cu in PM_2.5 and related ROS were moderately to highly correlated (Supplementary Table S1, available as Supplementary data at IJE online), we used single-pollutant models to examine their associations with the study outcomes separately. Hazard ratios (HRs) were calculated for an interquartile range (IQR) increment in 3-year moving average concentrations of Fe (ng/m³) and Cu (ng/m³) in PM_2.5 and related ROS (nM). In addition to Fe, Cu and ROS, we also examined the associations between PM_2.5 mass and NO₂ and the incidence of the cardiovascular outcomes.

We performed a series of sensitivity analyses, including: (1) introducing Fe and Cu into regression models together for mutual adjustment; (2) using two-pollutant models which included Fe, Cu or ROS together with PM_2.5 or NO₂; (3) exploring alternative time windows of exposure (1-year and 5-year moving averages); (4) using alternative standard fixed-effects Cox regression models with adjustment for neighborhood (as a categorical variable); (5) using a recently developed method³¹ to indirectly adjust for three individual-level covariates including smoking, physical activity and body mass index; (See Supplementary Material, available as Supplementary data at IJE online for more details about this approach.) (6) introducing year as an additional covariate to account for potential time trend in disease diagnosis and incidence.

We also performed stratified analyses to examine the potential effect modification by age (< 65 years vs ≥ 65 years), sex (male vs female) and neighborhood-level SES levels (income quintile). Each modifying factor was examined by adding a multiplicative interaction term with exposure variable in separate models.

To characterize the shape of the concentration-response relationship between Fe and Cu in PM_2.5 and related ROS, and the three investigated health outcomes, we used a recently developed integrated modeling approach, the Shape Constrained Health Impact Function (SCHIF).³² Briefly, a class of functions were constructed by defining transformations of concentration as the product of either a linear or log-linear function of concentration multiplied by a logistic weighting function. Uncertainty in the risk functions is quantified using simulation methods. An ensemble model is then derived by a weighted average of all the shapes of associations examined, with weights defined by the likelihood function values. The model functions are flexible but monotonically non-decreasing and therefore are suitable for health effects assessment. (See Supplementary Material, available as Supplementary data at IJE online for more details about this approach.)

All statistical analyses were performed using SAS 9.4 (SAS Institute, Cary, NC, USA) and R 3.1.2 (R Core Team, Vienna, Austria). A two-tailed P value of < 0.05 defined statistical significance for both main effects and interactions.

Results

The general characteristics of the study population are summarized in Table 1. The average age was 57.8 [standard deviation (SD): 12.3], 57.5 (SD: 12.1) and 58.0 (SD: 12.3) years for participants in the AMI, CHF and CVD cohorts, respectively. There were fewer males than females in all three cohorts. Overall, the distribution of the general characteristics was similar across the three cohorts. During the study period of 2001 to 2016, we observed 25 657, 69 801 and 58 159 incident cases of AMI, CHF and CVD death, respectively.

Table 1

General characteristics of the study participants at baseline

Characteristics	AMI cohort	CHF cohort	CVD cohort
Number of participants	819 522	803 894	834 445
Individual-level variables
Age (year)	57.8 ± 12.3	57.5 ± 12.1	58.0 ± 12.3
Sex (male)	378 206 (46.2%)	372 132 (46.3%)	388 293 (46.5%)
Area-level variables
Income quintile
1-lowest	191 155 (23.3%)	186 627 (23.2%)	195 339 (23.4%)
2	192 647 (23.5%)	188 441 (23.4%)	196 305 (23.5%)
3	155 279 (19.0%)	152 477 (19.0%)	157 952 (18.9%)
4	113 057 (13.8%)	111 231 (13.8%)	115 111 (13.8%)
5-highest	167 384 (20.4%)	165 118 (20.5%)	169 738 (20.3%)
Percentage of recent immigrants (%)	8.5 ± 9.0	8.5 ± 9.0	8.5 ± 9.0
Percentage of population aged ≥15 years without employment (%)	6.5 ± 4.8	6.5 ± 4.8	6.5 ± 4.8
Percentage of population aged ≥15 years with education level lower than high school (%)	24.9 ± 14.1	24.9 ± 14.1	25.0 ± 14.1

Data are presented as mean ± SD for continuous variables and number (percentage) for categorical variables.

Abbreviations: AMI, acute myocardial infarction; CHF, congestive heart failure; COPD, chronic obstructive pulmonary disease; CVD, cardiovascular disease.

Table 2 shows the distribution of long-term concentrations (3-year moving averages) of Fe and Cu in PM_2.5 and their estimated combined impact on ROS generation as well as PM_2.5 mass and NO₂ at baseline. The distribution of the pollutants was broadly similar across the three cohorts. The mean concentrations of Fe, Cu, ROS, PM_2.5 and NO₂ in the CVD cohort were 84.2 (SD: 26.1) ng/m³, 0.53 (SD: 0.17) ng/m³, 59.3 (SD: 13.2) nM, 9.5 (SD: 0.5) µg/m³ and 21.4 (SD: 3.7) ppb, respectively. Fe, Cu and ROS had relatively greater spatial variations compared with PM_2.5 and NO₂ with greater coefficient of variations (CV). Fe and Cu in PM_2.5 were highly correlated with ROS (Spearman correlation coefficient > 0.75). Relatively weak correlations were observed between these three measures and PM_2.5 mass or NO₂, especially for ROS and PM_2.5 mass (Spearman correlation coefficient = 0.09). (Table S1, available as Supplementary data at IJE online)

Table 2

Distribution of long-term (3-year moving average) concentrations of air pollutants for study participants at baseline

Statistics	Fe (ng/m3)	Cu (ng/m3)	ROS (nM)	PM2.5 (µg/m3)	NO2 (ppb)
AMI cohort
Mean	84.1	0.53	59.3	9.5	21.4
Standard deviation	26.1	0.17	13.4	0.5	3.7
5th percentile	61.0	0.38	44.7	8.7	16.2
25th percentile	70.1	0.45	51.4	9.0	19.1
Median	77.8	0.50	56.5	9.5	21.0
75th percentile	89.3	0.56	63.6	9.9	23.2
95th percentile	126.8	0.78	84.4	10.3	28.3
Interquartile range	19.2	0.11	12.2	0.9	4.1
Coefficient of variation (%)	31.0	33.0	22.6	5.3	17.4
CHF cohort
Mean	84.1	0.53	59.3	9.5	21.4
Standard deviation	26.1	0.17	13.1	0.5	3.7
5th percentile	61.0	0.38	44.7	8.7	16.1
25th percentile	70.1	0.45	51.4	9.0	19.0
Median	77.7	0.50	56.5	9.5	21.0
75th percentile	89.2	0.56	63.6	9.9	23.2
95th percentile	126.8	0.78	84.4	10.3	28.3
Interquartile range	19.2	0.11	12.2	0.9	4.1
Coefficient of variation (%)	31.0	33.0	22.2	5.3	17.4
CVD cohort
Mean	84.2	0.53	59.3	9.5	21.4
Standard deviation	26.1	0.17	13.2	0.5	3.7
5th percentile	61.0	0.38	44.7	8.7	16.1
25th percentile	70.1	0.45	51.4	9.0	19.0
Median	77.8	0.50	56.5	9.5	21.0
75th percentile	89.3	0.56	63.6	9.9	23.2
95th percentile	126.8	0.78	84.4	10.3	28.3
Interquartile range	19.2	0.11	12.2	0.9	4.1
Coefficient of variation (%)	31.0	32.1	22.3	5.3	17.3

Abbreviations: AMI, acute myocardial infarction; CHF, congestive heart failure; CVD, cardiovascular disease; NO₂, nitrogen dioxide; PM_2.5, particulate matter with aerodynamic diameter < 2.5 µm; ROS, reactive oxygen species.

In the crude models controlling for age and sex only, Fe and Cu in PM_2.5 and their combined impact on ROS were positively associated with all three health outcomes. (Figure 1 and Supplementary Table S2, available as Supplementary data at IJE online) Further adjustment for neighborhood-level SES factors (i.e. main model in Figure 1 and Supplementary Table S2, available as Supplementary data at IJE online) attenuated the effect estimates slightly, but the associations remained statistically significant. Similarly, positive associations were also observed for incident CHF with slightly lower HRs. In contrast, we observed stronger associations for all three exposures with CVD death.

Figure 1

Associations of long-term exposure to iron and copper in PM_2.5 and related reactive oxygen species with incident acute myocardial infarction, congestive heart failure and cardiovascular disease death

Hazard ratio was calculated for an interquartile range increment (Fe: 19.2 ng/m³; Cu: 0.11 ng/m³; ROS: 12.2 nM) in 3-year moving average concentration of exposure (results from main models are marked in red). ‘Crude’: crude model—mixed-effects Cox regression model stratified by age and sex; ‘Main’: main model—crude model with further adjustment for area-level covariates, including income quintile, immigrants, unemployment rate and education; ‘Two-pollutant’: two-pollutant model with Fe and Cu introduced into Cox models simultaneously; ‘+PM_2.5’: main model with additional adjustment for PM_2.5 mass; ‘+NO₂’: main model with additional adjustment for NO₂; ‘1-yr Mean’ and ‘5-yr Mean’: instead of the 3-year moving average concentrations of exposure used in main analysis, 1-year or 5-year moving averages were used; ‘Fixed-effects’: fixed-effects Cox model with adjustment for neighborhood (as categorical variable); ‘Indirect’: indirectly adjusted for smoking, physical activity and body mass index which were not available in the dataset; ‘+Year’: main model with additional adjustment for year to account for potential time trend.

Abbreviations: AMI, acute myocardial infarction; CI, confidence interval; CHF, congestive heart failure; CVD, cardiovascular disease; Cu, copper; Fe, iron; HR, hazard ratio; NO2, nitrogen dioxide; PM_2.5, particulate matter with aerodynamic diameter < 2.5 µm; ROS, reactive oxygen species.

In sensitivity analyses, when Fe and Cu were included in the models simultaneously, Fe remained a significant predictor for the incidence of all three outcomes, whereas Cu was no longer associated with AMI and was negatively associated with CHF and CVD death. The associations observed in the main analyses were robust to adjustment for PM_2.5 mass, but weakened after adjustment for NO₂, especially for Cu. Also, varying exposure time windows did not change the results materially. Furthermore, standard fixed-effects Cox regression models generated similar results. For AMI and CHF, the significant associations with all three exposure measures diminished after the indirect adjustment for individual-level lifestyle factors, whereas the positive associations with CVD death remained. Finally, adjustment for year had little influence on the results. (Figure 1 and Supplementary Table S2, available as Supplementary data at IJE online)

In stratified analyses, we found stronger associations between the exposure and AMI and CHF among younger individuals compared with the elderly (P for interaction <0.05). For CVD death, this pattern was observed for Fe, but not Cu or ROS. (Supplementary Tables S3–S5, available as Supplementary data at IJE online) We observed no consistent pattern supporting a clear effect modification by sex or neighborhood-level SES.

The shape of the estimated concentration-response relationship between Fe and Cu in PM_2.5 and related ROS and the three health outcomes is demonstrated in Figures 2–4. We found some evidence of a sub-linear concentration-response relationship between AMI and Fe and ROS as the curves were steeper at higher concentrations. For Cu, the curve was nearly linear. A similar pattern was observed for CHF. For CVD death, we observed a supra-linear concentration-response shape for all three exposure measures with a relatively flatter curve at higher concentrations.

Figure 2

Concentration-response relationships between iron (a) and copper (b) in PM_2.5 and responding reactive oxygen species concentration in lung fluid (c) and incident acute myocardial infarction

The concentration-response curves are presented as solid lines with 95% uncertainty bounds (shaded area) based on ensemble models, estimated using mixed-effects Cox regression models controlling for age, sex and neighborhood-level income quintile, percentage of recent immigrants, unemployment rate and education

Abbreviations: PM_2.5, particulate matter with aerodynamic diameter < 2.5 µm; ROS, reactive oxygen species.

Figure 3

Concentration-response relationships between iron (a) and copper (b) in PM_2.5 and responding reactive oxygen species concentration in lung fluid (c) and incident congestive heart failure

The concentration-response curves are presented as solid lines with 95% uncertainty bounds (shaded area) based on ensemble models, estimated using mixed-effects Cox regression models controlling for age, sex and neighborhood-level income quintile, percentage of recent immigrants, unemployment rate and education.

Abbreviations: PM_2.5, particulate matter with aerodynamic diameter < 2.5 µm; ROS, reactive oxygen species.

Figure 4

Concentration-response relationships between iron (a) and copper (b) in PM_2.5 and responding reactive oxygen species concentration in lung fluid (c) and incident cardiovascular disease death

The concentration-response curves are presented as solid lines with 95% uncertainty bounds (shaded area) based on ensemble models, estimated using mixed-effects Cox regression models controlling for age, sex and neighborhood-level income quintile, percentage of recent immigrants, unemployment rate and education.

Abbreviations: PM_2.5, particulate matter with aerodynamic diameter < 2.5 µm; ROS, reactive

Both PM_2.5 mass and NO₂ were positively associated with incident AMI, CHF and CVD death, although the PM_2.5-AMI association was not statistically significant. (Table 3) Additional adjustment for ROS had little influence on the associations of PM_2.5. However, associations of NO₂ were attenuated after adjustment for ROS. Consistently, the concentration-response curves of PM_2.5 did not significantly change after adjustment for ROS. The curves of NO₂ were more sensitive to adjustment for ROS as the effect estimates decreased, but the supra-linear shape remained (Supplementary Figures S1–S6, available as Supplementary data at IJE online).

Table 3

Associations of long-term exposure to PM_2.5 and NO₂ with incident acute myocardial infarction, congestive heart failure and cardiovascular disease death

Models	AMI	CHF	CVD Mortality
	HR (95% CI) a	HR (95% CI) a	HR (95% CI) a
PM2.5
Main modelb	1.042 (0.961,1.130)	1.211 (1.148,1.278)	1.145 (1.070,1.226)
Main + ROS	1.042 (0.961,1.131)	1.211 (1.147,1.277)	1.126 (1.050,1.206)
NO2
Main modelb	1.027 (1.007,1.048)	1.041 (1.025,1.054)	1.060 (1.044,1.077)
Main + ROS	1.005 (0.983,1.028)	1.030 (1.016,1.045)	1.019 (1.001,1.038)

a

Hazard ratio was calculated for an interquartile range increment (PM_2.5: 1.5 µg/m³; NO₂: 4.1 ppb) in 3-year moving average concentration of exposure.

b

Mixed-effects Cox regression model stratified by age and sex, and adjusted for area-level covariates, including income quintile, percentage of recent immigrants, percentage of population aged ≥15 years without employment and percentage of population aged ≥15 years with education level lower than high school.

Abbreviations: AMI, acute myocardial infarction; CI, confidence interval; CHF, congestive heart failure; CVD, cardiovascular disease; HR, hazard ratio; NO₂, nitrogen dioxide; PM_2.5, particulate matter with aerodynamic diameter < 2.5 µm; ROS, reactive oxygen species.

Discussion

In this large population-based cohort in Toronto, we found that long-term exposure to Fe and Cu in PM_2.5 and their combined impact on ROS were associated with increased incidence of AMI, CHF and CVD death. Stronger associations were observed for ROS than for Fe and Cu individually. The observed associations were robust for CVD death in all sensitivity analyses, whereas the associations for AMI and CHF were more sensitive to the indirect adjustment for smoking, physical activity and body mass index.

Current epidemiological evidence on the long-term health effects of PM components, including metals, is still scarce. In a study of 70 000 veterans in the US, all-cause mortality was found to be associated with nickel (Ni), vanadium (V) and elemental carbon (EC) in PM_2.5, but not Fe or Cu.³³ Similar findings were reported in an extended follow-up study as Ni and EC remained the most robust predictors of all-cause mortality.³⁴ However, this study only assessed all-cause mortality while mortality due to CVD was not investigated. In the California Teachers Study, ischemic heart disease (IHD) mortality was associated with several PM_2.5 components studied, including Fe.^7,8 The recent analysis of this cohort reported positive associations between IHD mortality and certain components of PM_2.5 and ultrafine particles, including Fe and Cu.¹¹ In the American Cancer Society Study, Fe and a number of other non-metal components in PM_2.5 [e.g. Silicon (Se), Arsenic (As) and sulfur (S)] were associated with increased IHD death.^9,12 In the ESCAPE study in Europe, long-term concentrations of eight elements [Fe, Cu, potassium (K), Ni, S, Se, V and zinc (Zn)] in both PM_2.5 and PM with an aerodynamic diameter < 10 µm (PM₁₀) were measured.¹⁴ Fe, K and Si were associated with an increased risk of incident coronary heart disease.¹⁴ In other analyses using the ESCAPE cohorts, none of the eight elements were associated with CVD mortality,¹⁰ and only S was associated with natural-cause mortality.³⁵ More recently, in an Italian study which used the ESCAPE models for exposure assessment, all eight estimated PM elements were associated with mortality (non-accidental, CVD and IHD mortality) with higher effects being observed for tracers of non-tailpipe emissions, including Fe, Cu and Zn.¹³ For subclinical outcomes, in the Multi-Ethnic Study of Atherosclerosis in the US, Cu in PM_2.5 was reportedly associated with increased carotid intima-media thickness, a measure of atherosclerosis.³⁶ Although most previous studies suggest a tendency for increased CVD morbidity and mortality in association with exposure to metal components of PM, there is limited agreement on which specific metals are most relevant. The inconsistency of current findings could be attributable to various factors, such as heterogeneity of study populations, differences in PM concentration and composition as well as methodological differences related to exposure assessment (i.e. different amounts of measurement error for different components) and statistical analysis (i.e. different techniques to handle the correlations among components). For example, the estimated concentrations of the PM metal components differed in different regions. The concentrations of Fe and Cu in our study were generally lower than those reported in the US ^7–9,11,12 and several cities in southern Europe (e.g. Rome, Turin and Athens).^10,13,14,35 Comparatively, cities in north Europe such as Stockholm and Helsinki generally reported lower concentrations , ^10,14,35 which are more comparable to our study. As for exposure assessment methods, some studies only used exposure derived from central monitoring stations,^7,8,12,33 which have a limited capacity to capture spatial variation in exposure. Moreover, because of the correlations among PM components, most studies including the present study largely relied on single-pollutant or single-component analysis approaches. Separating the individual effects of each of the components in PM still remains challenging. Further research with advanced exposure assessment and analytical methods are warranted.

It is noteworthy that the associations between NO₂ and CVD outcomes observed in our study were attenuated significantly after adjustment for ROS. Ambient NO₂ primarily originates from the combustion of fuels, and road traffic, particularly engine emissions, is a major source of NO₂ in urban areas.³⁷ NO₂ is widely used as surrogate measure of traffic-related air pollution and it has been associated with various adverse health outcomes in ample epidemiological studies.³⁸ The observed NO₂-CVD associations indicate an association between CVD and traffic-related air pollution. We are not sure about the underlying mechanisms responsible for the attenuation of these associations after controlling for ROS. We hypothesized that the cardiovascular effects of air pollution are mainly through oxidative stress, for which individual metals or their impact on key mechanisms (ROS generation) are likely to be a better marker than NO₂. Nonetheless, our findings imply that the observed NO₂-CVD associations could be partially attributable to non-tailpipe emissions. In other words, in addition to emissions from fuel combustion (measured by NO₂), non-tailpipe emissions (measured by ROS) may also play a role in inducing adverse cardiovascular effects of air pollution. Our findings warrant replications. By contrast, the PM_2.5–CVD outcome associations remained nearly unchanged when ROS was included in the models, suggesting that the observed health effects of PM_2.5 may be less sensitive to metals from non-tailpipe emissions. This may be explained by the fact that Fe and Cu make up a small portion of overall PM_2.5 mass concentrations. Unlike NO₂ which mainly originates from local traffic emissions, a large fraction of ambient PM_2.5 in southern Ontario is from non-local non-traffic sources.³⁹ A previous study suggested that in Toronto, only 30%–45% of the PM_2.5 is from local sources, primarily vehicle emissions.⁴⁰ Components from both traffic emissions, regardless of tailpipe or non-tailpipe emissions, and other sources may contribute to the adverse health effects of PM_2.5. Therefore, adjustment for ROS may have less impact on the observed effects of PM_2.5.

We found some suggestive evidence that there were stronger associations of cardiovascular outcomes with Fe and Cu and related ROS generation in younger participants than the elderly, although it was not consistent across all exposures and outcomes. Some other studies also reported a similar pattern as greater effect estimates of air pollution were detected in younger individuals.^41,42 We hypothesized that this might be related to depletion of susceptible individuals in the older group.⁴³ As disturbance of autonomic nervous system has been cited as one of the mechanistic pathways underlying the cardiovascular effects of PM air pollution, the reduced responsiveness to autonomic nervous system stimuli along with aging may also partially explain these findings.⁴⁴ Our results on the modifying effects of neighborhood-level SES were also inconsistent. More studies are needed to better address this issue.

Our study has several important strengths. First, the large sample size and the long study period ensured sufficient power to characterize the associations between air pollution, which generally has relatively smaller effects compared with some traditional cardiovascular risk factors such as smoking and obesity, and CVD outcomes. Second, the detailed information on participants’ residential history allowed us to account for the influence of residential mobility on exposure. An additional major strength is that we used a newly developed marker, the production of ROS in human lung epithelial lining fluid, to estimate the combined health impact of Fe and Cu in PM_2.5. This marker may provide an alternative approach to deal with the correlations among PM_2.5 components when assessing the individual effects of these components. In our study, stronger associations with cardiovascular outcomes were detected for ROS than for Fe and Cu in PM_2.5 individually. More importantly, the associations of ROS were robust across different sensitivity analyses including those with additional adjustment for PM_2.5 mass and NO₂, implying the potential use of ROS concentration as an independent measure to assess the health effects of air pollution. The PM_2.5 levels in our study were relatively low (mean = 9.5 µg/m³) with limited spatial variation (IQR = 0.9 µg/m³, CV= 5.3% at baseline). In contrast, the spatial variations in ROS generation in response to Fe and Cu in PM_2.5 were much higher (IQR = 12.2 nM, CV = 22.3%). The findings based on ROS, which ultimately may be more biologically relevant, could help to shed light on the health effects observed at low PM_2.5 mass concentrations.

Some limitations should also be noted. First, the land use regression models used for exposure assessment were developed based on short-term monitoring campaigns, and we assigned the exposure to a long study period (from 2001 to 2016). This may have resulted in exposure misclassification. However, it has previously been shown that spatial patterns of traffic-related air pollution are constant over a long period.^45–47 In our study, the concentrations of Fe and Cu in PM_2.5 as well as the estimated ROS levels were predominately explained by traffic-related factors.²¹ Therefore, the land use regression models could provide a reliable estimate of historical spatial differences in exposure. Second, we only estimated the ROS production in response to Fe and Cu in PM_2.5 but not the total ROS generated in response to the entire pollutant mixture. This may have led to an underestimation of the total oxidative burden caused by inhaled PM_2.5 mixture, consequently resulting in possibly underestimated health effects. Third, information on individual-level SES and lifestyle factors was unavailable in our study cohorts. We included a number of neighborhood-level SES indicators, which may have partially controlled the influence of individual SES and behaviors because the gradients in individual- and area-level SES measures are generally consistent and they are associated with comparable trends in health risk factors.^24–26 In addition, the observed effects of Fe, Cu and ROS on incident AMI and CHF were relatively small and thus more likely to be susceptible to residual confounding due to unmeasured factors. After indirect adjustment of smoking, physical activity and body mass index, the significant associations of AMI and CHF diminished. Comparatively, the associations of CVD death were stronger and robust to sensitivity analyses including the indirect adjustment. A possible reason for the significant associations with CVD death but not AMI or CHF could be that Fe and Cu in PM_2.5 might also contribute to the development of some other cardiovascular conditions such as stroke, which may in turn increase the risk of CVD death. Further investigation and biological evidence from toxicological studies are required to support this hypothesis

Despite the limitations, our findings suggest that long-term exposure to PM_2.5 metal components, particularly Fe and Cu, and the responding generation of ROS in lung fluid can lead to adverse cardiovascular effects. This supports the hypothesis that metal components, especially transition metals, are likely to contribute to the toxicity of PM because of their ability to induce oxidative stress.⁵ From a policy perspective, our findings also highlight the importance of considering traffic-related air pollution from non-tailpipe sources. More importantly, the associations between Fe and ROS remained pronounced after adjustment for NO₂, a measure of combustion-related air pollution from mobile sources, implying plausibility of the independent effects of non-tailpipe traffic air pollution on cardiovascular health.

Ethics approval

Use of the data in the present study was authorized under section 45 of the Personal Health Information Protection Act of Ontario, which does not require review by a Research Ethics Board. All procedures performed in the present study were in accordance with the ethical standards of the Declaration of Helsinki.

Supplementary data

Supplementary data are available at IJE online.

Funding

This study was funded by Health Canada (4500381545 and 4500370370). This study was supported by ICES, which is funded by an annual grant from the Ontario Ministry of Health and Long-Term Care (MOHLTC) This study was also supported by Public Health Ontario (PHO). Parts of this report are based on Ontario Registrar General information on deaths, the original source of which is Service Ontario. The views expressed therein are those of the authors and do not necessarily reflect those of ORG or Ministry of Government Services. Parts of the material are based on data and information complied by the Canadian Institute for Health Research (CIHI). Any interpretation or conclusion related to this manuscript does not necessarily represent the views of Health Canada, ICES, MOHLTC, PHO or CIHI. All copyrights and intellectual property conducted at Health Canada belong to the Crown, not to the authors. MS acknowledges funding from the Health Effects Institute (HEI) (Walter A. Rosenblith New Investigator Award, No. 4964-RFA17-3/18–6), an organization jointly funded by the United States Environmental Protection Agency (EPA) (Assistance Award No. CR-83590201) and certain motor vehicle and engine manufacturers.

Data availability

The dataset from this study is held securely in coded form at ICES. While data sharing agreements prohibit ICES from making the dataset publicly available, access may be granted to those who meet pre-specified criteria for confidential access, available at www.ices.on.ca/DAS. The full dataset creation plan and underlying analytic code are available from the authors upon request, understanding that the computer programs may rely upon coding templates or macros that are unique to ICES and are therefore either inaccessible or may require modification.

Author contributions

H.C. and S.W. initiated the study. R.T.B and H.C. developed the statistical methodology and software. Z.Z. and L.B. conducted data analysis. S.W., M.H., M.J., A.D. and R.M conducted exposure assessment. Z.Z., H.C. and S.W. interpreted the results. Z.Z drafted the manuscript. H.C. and J.C.K. supervised the study. All authors participated in critical revisions of the manuscript and approved the final version.

Conflict of interest

None declared.