Trace element concentrations in ambient air as a function of distance from road

ABSTRACT

Traffic-related air pollution is associated with various adverse health effects. In the absence of more complicated exposure assessment techniques, many environmental health studies have used the natural logarithm of distance to road as a proxy for traffic-related exposures. However, research validating this proxy and further explaining the spatial patterns and elemental composition of traffic-related particulate matter air pollution remains limited. In this study, we collected air samples using a mobile particle concentrator that allowed for high sample loading from major roadways in the Greater Boston Area. We found that concentrations of Cl, Ti, V, Cr, Mn, Fe, Co, Cu, Zn, Sr, Zr, Sn, Ba, and Pb were significantly associated with the natural logarithm of distance to road in coarse particulate matter, and total fine particulate mass concentrations of Al, Ca, Ti, Cr, Mn, Fe, Cu, and Zn were significantly associated with natural logarithm of distance to road in fine particulate matter. Road type (A1 or A2 [primary roads or highways] versus A3 [secondary and connecting roads]) was not a significant predictor of any traffic-related elements in particulate matter air pollution. Our results help identify traffic-related elements in particulate matter air pollution and support the use of logarithm of distance to road as a proxy for traffic-related particulate matter air pollution exposure assessment in epidemiological studies.

Introduction

Particulate matter (PM) air pollution is the leading cause of death in environmental health and has been linked to numerous health outcomes including respiratory disease, cardiovascular disease, stroke, and cancer (Chow et al. 2006; Dantzer and Keet 2015; Hamra et al. 2014; Pope, Schwartz, and Ransom 1992; Shah et al. 2015). Common exposure assessment techniques for air pollution include use of personal air monitors, mobile monitoring, indoor-outdoor sampling, and air pollution modeling (for example, air dispersion or land use regression models that may incorporate various traffic parameters and meteorological variables) (Han, Hu, and Bai 2017). Other studies employ biological sampling of air pollution components, for example blood sampling to detect heavy metals known to be present in ambient air (Zeng et al. 2016). In epidemiological research, individually quantifying exposure measurements for large cohorts can be difficult and costly. One of the simplest types of exposure assessment models for air pollution is a proximity model that assesses exposure based on proximity to an emissions source. The natural logarithm of road proximity is frequently used as a surrogate for traffic-related air pollution in epidemiological studies (Henry, Anthopolos, and Maxson 2013; Lue et al. 2013; Wilker et al. 2016). A previous study from our group showed indoor concentrations of various traffic-related elements were related to log distance to road (Huang et al. 2018). However, a detailed analysis on the spatial and temporal patterns of traffic-related air pollution has been limited by high costs and time requirements for accurate measurements.

Several studies have employed mobile platforms equipped with monitors to measure PM and criteria gases in order to examine the spatial patterns of traffic-related pollutants (Solomon and Sioutas 2008; Westerdahl et al. 2005). The conventional sampling methods used are costly and require hours of sampling due to low flow rates, so they cannot be scaled up for large sample sizes (Kam et al. 2012). The primary sources of non-tailpipe PM air pollution are mechanical wear of brakes, tires, and other car parts, and suspension of road surface particles (Ketzel et al. 2007; Kupiainen et al. 2005). Some effort has been made at studying resuspended road dust and developing prediction models. Several recent studies have collected re-suspended road dust samples or ambient PM samples and developed prediction models to attempt to capture PM exposure levels (Kauhaniemi et al. 2011, 2014; Lozhkina et al. 2016). Although progress has been made in measuring and modeling patterns of on-road traffic-related pollutants, less emphasis has been put on how these pollutants decay with increasing distance from road, especially for non-tailpipe pollutants. There is still significant work to be done in order to accurately capture both the spatial and temporal patterns of traffic-related PM and to understand how these patterns vary by chemical element.

For the current study, we developed and used a Mobile Particle Concentrator Platform (MPCP), which allows one to collect sufficient masses of sample in one hour for trace element analysis using X-Ray Fluorescence (XRF). Here we characterize the size and composition of ambient particles at various distances from 21 roads across the Greater Boston area and examine the relationship between PM species and distance to road. The mobile platform used for our study collected PM at a very high flow rate compared to existing speciation samplers, reducing the time and cost of sample collection and allowing for speciation of non-tailpipe traffic pollutants. To our knowledge, this is the first time that particle concentrator techniques have been used to collect short term integrated samples that make possible the analysis of trace metals related to both exhaust and non-exhaust emissions.

Materials and methods

Samples were collected during the period of June 2018 through January 2019, at 21 different roads in the Greater Boston area. A mobile laboratory was assembled using a 16-foot box truck equipped with coarse and fine particle concentrators for ambient PM collection. At each road site, three hourly samples of fine PM (PM_0.2–2.5) and coarse PM (PM_2.5–10) were collected at three distance ranges from the road (0–50 m, 51–150 m, and 500–750 m) (Martins et al. 2020). All three locations relative to the same road were sampled on the same day and in a random order to minimize the effect of time of day and varying traffic density.

Road sampling and proximity

Three different types of roads were sampled, defined in the U.S. Census as A1, primary highways with limited access, A2, primary roads without limited access, and A3, secondary and connecting roads. Limited access generally refers to roads for high-speed traffic that are less accessible from nearby land and roads. Annual average daily vehicle count from the Massachusetts Department of Transportation Road Inventory data for each sampled road is shown in Table S1. Five roads were sampled two to three times on different dates and in different seasons for a total of 28 sets of samples. Of these sets, 10, 7, and 11 were collected at A1, A2, and A3 roads, respectively. Samples were collected on days without precipitation. Sample sites were carefully selected in order to avoid areas with non-traffic-related emissions sources nearby. In addition, sites were selected such that the closest A1, A2, or A3 road to intermediate and background sites was the road that was studied. Perpendicular distance from road, d, was calculated using the Street Map^TM North America ArcGIS 10 Data and Maps.

Measurement of trace element concentrations in ambient coarse and fine PM

This study used a modified Harvard Ambient Fine Particle Concentrator (HAFPC), originally a 5,500 LPM two-stage fine particle concentrator that has been described in detail in previous studies (Lawrence et al. 2004; Sioutas et al. 1997). The system was modified to require less power in order to accommodate the mobile platform. The modified system used two parallel two-stage concentrators whose outputs were combined and then separated to collect two separate samples on Teflon and Quartz fiber filters. Each sample was collected at an output flow of 45 LPM. For coarse particles, the Harvard Ambient Coarse Particle Concentrator (HACPC) was used. The HACPC, a 5,000 LPM two-stage coarse particle concentrator, is similar to the HAFPC, but with acceleration and collection slits designed for coarse particles. The output flow is approximately 50 LPM and was divided to collect two samples, one on Teflon and one on Quartz fiber filters, each at 25 LPM.

For each location, coarse and fine PM samples were collected using particle concentrators for 1 hour as described above. Total mass concentrations were determined using the weight difference of Teflon filters before and after ambient air collection. Elemental analysis for samples collected on Teflon filters was conducted for 28 elements using XRF at HSPH with a PANalytical Epsilon Model 5 XRF analyzer. Detailed procedures are explained in detail in a methods paper (Martins et al. 2020).

Other covariates

The effects of season were examined in our primary model, as there is more dust on the ground and more resuspension of dust in the summer due to the dryness of the ground surface. June through September was considered the warm season, and October through January was considered the cold season. Wind speed and wind direction during sample collection were considered in additional models. For each sample, wind speed and direction were measured using a Davis Vantage Vue Weather Station.

Statistical analysis

Primary model

A mixed-effects regression model with repeated measurements to account for between-road and within-road variation was used to analyze the relationship between particle mass or element concentration and distance to road. Because of best fit with data and because previous epidemiological studies examining the health effects of traffic have used natural logarithm of distance to road, we chose to use the same metric. Analyses were performed in Stata using the following model:

$ln\left(C_{ij}\right)={\beta }_{0j}+y_{ij}+{\beta }_{1j}\times ln\left(d_j\right)+{\beta }_{2j}\times season+e_{ij},$

where $C_{ij}$ is the concentration of element j (or total particle mass) at road i ($\mu$g/m³); ${\beta }_{0j}$ is the fixed intercept (which should represent the regional background contribution); $y_{ij}$ is the random intercept for road date (several roads were sampled on more than one day)i; $ln\left(d_j\right)$ is the natural log of perpendicular distance from sample location to road (m); ${\beta }_{1j}$ is the slope for the distance to road parameter; $season$ is a dichotomous variable for the season when the sample was collected; ${\beta }_{2j}$ is the slope for the season parameter; and $e_{ij}$ represents random error. We chose to transform the outcome variable to log of species concentration because residuals are approximately normally distributed, and it provided better fit than the untransformed concentration. The model was fit separately for coarse and fine particles for each element and for total particle mass. In this analysis, we included only elements whose sample concentrations were greater than sample uncertainties for at least 60% of the samples.

Effect of road type

A dichotomous variable for road type, $r$, was used to examine the effect of road type. In order to increase power, A1 and A2 roads were grouped together as A12 and were compared to A3 roads. Interaction between road type and log distance to road was also considered. The model including both road type main effect and interaction terms is as follows:

$ln\left(C_{ij}\right)={\beta }_{0j}+y_{ij}+{\beta }_{1j}\times ln\left(d_j\right)+{\beta }_{2j}\times season+{\beta }_{3j}\times r+{\beta }_{4j}\times r\times ln\left(d_j\right)+e_{ij},$

where $r$ is a dichotomous variable for road type; ${\beta }_{3j}$ represents the slope of the road type parameter; and ${\beta }_{4j}$ represents the slope for the interaction between road type and ln(d) to road.

Effect of wind variables

Wind speed and direction were analyzed as contributing parameters. In order to keep models parsimonious, wind speed and direction were each added to the model with ln(d) to road and season, individually. Wind speed was included as a continuous variable. Wind direction was controlled for as a categorical variable, assigned as upwind, downwind, or perpendicular, based on the wind position of the sample collected compared to the road. For example, if wind was blowing from the road toward the sample location, the sample was collected downwind of the road. The indicator variables were assigned such that when wind ran downwind of the road $w{d}_d$ was set to one and $w{d}_p$ to zero, when wind ran parallel to the road $w{d}_p$was set to one and $w{d}_d$ to zero, and when wind ran upwind of the road both $w{d}_d$ and $w{d}_p$ were set to zero. The models including wind speed and wind direction, respectively, are as follows:

$ln\left(C_{ij}\right)={\beta }_{0j}+y_{ij}+{\beta }_{1j}\times ln\left(d_j\right)+{\beta }_{2j}\times season+{\beta }_{3j}\times ws+e_{ij},$

where $ws$ is a continuous variable for wind speed and ${\beta }_{3j}$ represents the slope of the wind speed parameter;

$ln\left(C_{ij}\right)={\beta }_{0j}+y_{ij}+{\beta }_{1j}\times ln\left(d_j\right)+{\beta }_{2j}\times season+{\beta }_{3j}\times w{d}_d+{\beta }_{4j}\times w{d}_p+e_{ij},$

where $w{d}_d$ is an indicator variable that equals one when the wind ran downwind perpendicular to the road (from the road to the sample location); ${\beta }_{3j}$ represents the slope comparing samples collected downwind of the road to samples collected upwind of the road; $w{d}_p$ is an indicator variable that equals one when the wind ran parallel to the road; ${\beta }_{4j}$ represents the slope comparing samples collected while wind ran parallel to the road to samples collected upwind of the road.

All analyses were performed using Stata software. Results were considered significant when p < 0.05.

Results

Twenty-two elements in coarse PM and 19 elements in fine PM had sample concentrations greater than sample uncertainties for at least 60% of samples and were included in the analysis. A summary of sample characteristics is shown in Table 1. Values for mean concentrations and standard deviations are shown for each element in Table 2.

Table 1. Characteristics of ambient air samples

	n	%
Total elements detected	28
Elements analyzed
Fine	19	67.9
Coarse	22	78.6
Roads Sampled	28
A1	10	35.7
A2	7	25.0
A3	11	39.3
Seasons
Warm	16	57.1
Cold	12	42.9

Table 2. Mean particle mass or concentration and standard deviation values for each element above analyzed in coarse and fine particulate matter (PM). *Element not included in the data analysis because less than 60% of samples had sample concentration greater than sample uncertainty

	Coarse			Fine
Element	Mean ($n$g/m^3)		SD	Mean ($n$g/m^3)	SD
PM	6,000		4,000	9,000	5,000
Na	86.2		179	91.4	153
Mg	30.0		29.2	33.6	33.2
Al	156		149	186	169
Si	368		360	170	208
S	32.8		25.9	373	365
Cl	154		415	82.6	244
K	64.8		56.2	55.9	44.1
Ca	150		201	75.6	15.6
Ti	18.0		22.0	12.4	12.5
V	0.528		0.878	na
Cr	0.551		0.716	10.9	19.6
Mn	3.58		3.80	4.39	4.01
Fe	219		263	205	193
Co	0.429		0.623	na
Ni	na			3.97	7.61
Cu	5.99		9.82	9.25	7.92
Zn	5.54		5.59	27.8	24.9
Br	0.500		0.866	2.72	2.13
Sr	1.26		1.34	2.67	2.03
Zr	2.78		4.92	*	-
Sn	1.74		2.21	6.84	7.30
Ba	11.6		24.3	*	-
Pb	1.14		0.937	5.23	4.59

Concentrations of trace elements

Concentrations were higher during the warm season than the cold season for all elements in coarse PM, although differences were not all statistically significant. For elements in fine PM, concentrations were higher in the warm season for all elements except Ca, Cr, Mn, Fe, Ni, and Zn.

Associations between concentrations of particle mass and trace elements and road proximity

Of the elements analyzed, 14 in coarse PM and 9 in fine PM were found to be significantly related to ln(d) to nearest road. Results from the mixed-effects analysis of the association between ln(d) to road and log of element concentration, controlling for season, are shown in Table 3. In coarse PM, elements significantly associated with ln(d) to road included Cl, Ti, V, Cr, Mn, Fe, Co, Cu, Zn, Sr, Zr, Sn, Ba, and Pb. For fine PM, total fine mass concentration, Al, Ca, Ti, Cr, Mn, Fe, Cu, and Zn were significantly associated with ln(d) to road. Several elements, Ti, Cr, Mn, Fe, Cu, and Zn, were significantly related to distance to road in both coarse and fine PM. For all of these elements, decay slopes with ln(d) to road were steeper in coarse PM than fine PM, as expected from the relationship between particle size and deposition velocity, although these differences did not reach statistical significance.

Table 3. Log distance coefficients for traffic-related pollutants and estimated standard errors (SE), adjusted for season. *Element not included in the data analysis because less than 60% of samples had sample concentration greater than sample uncertainty. Slopes for log distance to road significant at $\alpha =0.05$ shown in bold

	Coarse			Fine
Element	Slope, ${\beta }_{1j}$	SE	p-value	Slope, ${\beta }_{1j}$	SE	p-value
PM	−0.295	0.181	0.104	−0.037	0.018	0.036
Na	−0.260	0.156	0.096	−0.003	0.043	0.951
Mg	−0.200	0.150	0.181	0.007	0.033	0.842
Al	−0.270	0.159	0.091	−0.194	0.046	< 0.001
Si	−0.273	0.165	0.098	−0.033	0.033	0.315
S	−0.275	0.150	0.066	−0.022	0.021	0.283
Cl	−0.307	0.152	0.044	−0.044	0.036	0.209
K	−0.244	0.153	0.111	−0.019	0.023	0.415
Ca	−0.269	0.158	0.090	−0.069	0.028	0.013
Ti	−0.340	0.150	0.023	−0.192	0.035	< 0.001
V	−0.366	0.141	0.010	*
Cr	−0.373	0.135	0.006	−0.222	0.050	< 0.001
Mn	−0.281	0.139	0.043	−0.141	0.052	0.007
Fe	−0.364	0.164	0.027	−0.218	0.032	< 0.001
Co	−0.562	0.165	0.001	*
Ni	*			−0.126	0.075	0.092
Cu	−0.420	0.143	0.003	−0.160	0.029	< 0.001
Zn	−0.319	0.140	0.023	−0.144	0.035	<0.001
Br	−0.199	0.153	0.194	−0.028	0.0549	0.609
Sr	−0.270	0.134	0.043	0.056	0.043	0.196
Zr	−0.560	0.139	<0.001	*
Sn	−0.308	0.147	0.036	0.028	0.061	0.651
Ba	−0.502	0.179	0.005	*
Pb	−0.267	0.138	0.053	−0.055	0.050	0.269

Effect of season

Total coarse mass concentration and 11 of the 22 elements analyzed in coarse PM, Mg, Al, K, Ti, V, Mn, Fe, Co, Cu, Zr, and Sn, were significantly associated with season at $\alpha =0.05$. In fine PM, total fine mass concentration and 5 out of the 19 elements analyzed, Al, K, Ti, Cr, and Ni, were associated with season. Although season was not significant for all elements, we believe it to be a relevant confounder due to differences in surface dryness and resuspension of dust and, therefore, included it in all of our models.

Effect of road type

Road type was not significant at $\alpha =0.05$ for any of the traffic-related elements when added to the model. Road type was significantly associated only with concentration of Sr in fine PM, although this element was not associated with distance to road. Interaction between road type and ln(d) was also not significant. Neither of these variables were included in the final models for analysis.

Effect of wind variables

Wind speed was significantly negatively associated with total coarse mass concentration and element concentration for all elements analyzed in coarse PM except Na, Cl, Cr, and Co. Wind direction was not significantly associated with coarse mass concentration or any elements in coarse PM. For fine PM, wind speed was significantly associated with total fine mass concentration, S, K, Ca, and Zn, and wind direction was significantly associated with Na, Cr, Ni, and Br concentration. For Na in fine PM, the coefficient comparing parallel wind to upwind and the coefficient comparing downwind to upwind were significant. For Cr, Ni, and Br, only the coefficient comparing parallel wind to upwind was significant. In models for element concentration in coarse PM, adjustment for wind variables changed estimates for the coefficient for ln(d) by less than 6%. In models for element concentration in fine PM, adjustment for wind variables changed estimates for the coefficient for ln(d) by 7% to 28%. Results for ln(d) slope estimates after adjustment for wind variables are shown in Table 4. Estimates for ln(d) after adjustment for wind variables are shown only for wind parameters that were significant predictors of element concentration.

Table 4. Log distance coefficient in the basic model adjusted for season compared to the coefficient in the model additionally adjusted for wind variables. Results are only shown for elements that were significantly associated with wind variables. Slopes for log distance to road significant at $\alpha =0.05$ shown in bold

	Coarse		Fine
Element	Slope, ${\beta }_{1j}$ Basic Model	+ Wind Speed	Slope, ${\beta }_{1j}$ Basic Model	+ Wind Speed	+ Wind Direction
PM	−0.295	−0.293	−0.037	−0.027
Na			−0.003		−0.041
Mg	−0.200	−0.197
Al	−0.270	−0.261
Si	−0.273	−0.265
S	−0.275	−0.278	−0.022	−0.008
K	−0.244	−0.241	−0.019	−0.007
Ca	−0.269	−0.267	−0.069	−0.050
Ti	−0.340	−0.343
V	−0.366	−0.387
Cr			−0.222		−0.246
Mn	−0.281	−0.277
Fe	−0.364	−0.366
Ni			−0.126		−0.159
Cu	−0.420	−0.431
Zn	−0.319	−0.321	−0.144	−0.134
Br	−0.199	−0.268	−0.028		−0.028
Sr	−0.270	−0.268
Zr	−0.560	−0.582
Sn	−0.308	−0.310
Ba	−0.502	−0.497
Pb	−0.267	−0.282

Discussion

In this study, we analyzed the relationship between road proximity and concentrations of 19 elements in fine and 22 elements in coarse PM based on ambient air samples collected at 21 roads in the Greater Boston Area. We found that Cl, Ti, V, Cr, Mn, Fe, Co, Cu, Zn, Sr, Zr, Sn, Ba, and Pb in coarse PM, and total fine mass concentration, Al, Ca, Ti, Cr, Mn, Fe, Cu, and Zn in fine PM were significantly related to road proximity.

Most of the elements found to be related to distance from road can be traced to various traffic-related sources. Ti is most likely from re-suspended road dust (Amato et al. 2011). Cr, Fe, Zn, Zr, Sn, and Ba are related to brake and engine wear (Amato et al. 2011; Garg et al. 2000; Lough et al. 2005). Mn and Ca are associated with combustion of additives used in gasoline (Kaiser 2003; Lytle, Smith, and Mckinnon 1995). The elements that decay fastest with distance from road, such as Co, Cu, Zr, and Ba, are primarily from non-tailpipe emissions. These elements are found in the air due re-suspension of road dust, brake, engine, tire, and road wear, as well as other direct non-tailpipe sources. Co, Zr, and Ba were only related to distance to road in coarse particles, and we expect these larger particles to settle faster than fine particles released from combustion processes. As expected, we observed more negative slopes for elements in coarse PM. Mn and Cr, for example, were more strongly associated with road proximity in fine PM and decayed at a slower rate.

For elements associated with distance to road in both fine and coarse PM, decay with distance to road was steeper for coarse PM, although differences were not significant. This result is expected as coarse particles tend to settle faster while fine particles travel farther. Although wind speed was a significant predictor of most elements in coarse PM, control for wind variables had a stronger effect on ln(d) effect estimates for elements in fine PM. Total fine mass concentration had the greatest change in effect estimate after adjustment for wind parameters, while total coarse mass concentration was less affected. Wind speed and wind direction may change effective distance from road; however, control for wind variables did not significantly change the relationship between element concentration and distance to road for most traffic-related elements.

Elemental concentrations and their decay with distance did not depend on road type. This has important implications for health effects related to non-tailpipe emissions. People often worry about living close to busy roads with high traffic volume, but results from this study suggest that, with respect to non-tailpipe sources of PM air pollution, living near a secondary connecting road could result in trace element exposures comparable to those from living near a highway. This result could be related to our relatively small sample size, or to the fact that all study samples were collected in a relatively urbanized environment. Our analyses on road dust composition, presented in a different article, showed that several traffic-related elements were found at significantly higher mass fractions near A12 roads as compared to A3 roads, likely due to a difference in traffic density (Huang et al. 2020). Although we often worry about the size and concentration of PM mass in ambient air, the elemental composition may be important for biological consequences (Kelly and Fussell 2012; Stanek et al. 2011). For this reason, detailed particle speciation and analysis of particle spatial variability are of great importance to epidemiological studies.

Due to the difficulty in selectimg roads, obtaining permits to sample roads, and limited funding, we were only able to sample at 28 different sites. In addition, because another part of this project involved sampling road dust, we were unable to sample on days with significant precipitation. Since the sample size was relatively small, we were unable to study the effects of land use variables. However, we met our most important objective: to study the decay of concentrations of metals as a function of distance from road.

We have developed new methodologies to collect short-term PM samples near roads using concentrator technologies that allowed for elemental analysis using integrated samples of one hour. We were able to compare these integrated samples at different distances from the road. Even with a small dataset, we were able to observe significant results for the relationship between ln(d) to road and particle elemental concentration. This study contributes evidence to validate the use of ln(d) to road as a proxy for measuring exposure to fine and coarse PM for traffic-related elements in epidemiological studies.

The elevated concentrations of non-tailpipe pollutants near roadways in the Boston area have important implications as engineers work to build cleaner cars. For health effects related to non-tailpipe pollutants from brake or engine wear or re-suspended road dust, reducing exhaust emissions will not solve all problems. Because of regulations on tailpipe emissions, it is estimated that non-tailpipe emissions will become the dominant source of urban PM by 2020 (Amato 2018). This study highlights the importance of furthering research on the sources and composition of non-tailpipe PM and calls for innovation in car manufacturing to reduce toxic emissions from brake, engine, and tire wear.

Conclusion

Even with a small data set, we observed significant relationships between ln(d) to road and coarse and fine particle elemental concentrations for several traffic-related elements from tailpipe and non-tailpipe sources. Increased distance from roadway was associated with lower ambient concentration of traffic-related elements in both coarse and fine particles, but decay was steeper for coarse particles. This study contributes to evidence validating the use of ln(d) to road as a proxy for measuring exposure to fine and coarse PM for traffic-related metals in epidemiological studies. In addition, our findings provide insight on other parameters, such as road type, which we found to be a less important predictor of concentration of these species near roads in the urban environment, and particle size, which can influence the rate of removal of specific trace elements. Speciation and spatial analysis of non-tailpipe emissions may help elucidate the ways in which people are affected by air pollution in urban and rural areas. The finding that road type is not a significant factor in this urban environment may be used in other areas to help better estimate human exposures to non-tailpipe emissions in communities.

Acknowledgment

Research described in this article was conducted under contract to the Health Effects Institute (HEI), an organization jointly funded by the United States Environmental Protection Agency (US EPA) (Assistance Award No. CR-83467701) and certain motor vehicle and engine manufacturers. The contents of this article do not necessarily reflect the views of HEI, or its sponsors, nor do they necessarily reflect the views and policies of the EPA or motor vehicle and engine manufacturers. In addition, this publication was made possible by US EPA grant RD-83587201. Its contents are solely the responsibility of the grantee and do not necessarily represent the official views of the US EPA. Further, US EPA does not endorse the purchase of any commercial products or services mentioned in the publication. The authors thank the China Section of the Air & Waste Management Association for the generous scholarship they received to cover the cost of page charges, and make the publication of this paper possible.

Disclosure statement

No potential conflict of interest was reported by the authors.

Supplementary material

Supplemental data for this article can be accessed on the publisher’s website.

Additional information

Funding

This work was supported by the Environmental Protection Agency [CR-83467701,RD-83587201].

Notes on contributors

Emily Silva

Emily Silva Department of Environmental Health, Harvard T.H. Chan School of Public Health, Boston, MA, USA.

Shaodan Huang

Shaodan Huang is Ph.D., Research fellow, Department of Environmental Health, Harvard T.H. Chan School of Public Health, Boston, MA, USA. Research area: Indoor Air Quality, Ambient Air Pollution, Health Effect of Air Pollution.

Joy Lawrence

Joy Lawrence is Research associate, Department of Environmental Health, Harvard T.H. Chan School of Public Health, Boston, MA, USA. Research area: Air Pollution, Ambient Environment.

Marco A.G. Martins

Marco A.G. Martins is Ph.D., Research associate, Department of Environmental Health, Harvard T.H. Chan School of Public Health, Boston, MA, USA. Research area: Air Pollution, Ambient Environment.

Jing Li

Jing Li is Ph.D., Research fellow, Department of Environmental Health,Harvard T.H. Chan School of PublicHealth, Boston, MA, USA. Research area: Ambient Air Pollution, Atmospheric Chemistry.

Petros Koutrakis

Petros Koutrakis is Ph.D., Professor, Department of Environmental Health, Harvard T.H. Chan School of Public Health, Boston, MA, USA. Research area: Air Pollution and its Health Effects.