Particulate matter components, sources, and health: Systematic approaches to testing effects

Abstract

Exposure to particulate matter (PM) is associated with adverse health outcomes. There has long been a question as to whether some components of the PM mixture are of greater public health concern than others so that the sources that emit the more toxic components could be controlled. In this paper, we describe the National Particle Component Toxicity (NPACT) initiative, a comprehensive research program that combined epidemiologic and toxicologic approaches to evaluate this critical question, partly relying on information from a national network of air quality monitors that provided data on speciated PM_2.5 (PM with an aerodynamic diameter <2.5 μm) starting in 2000. We also consider the results of the NPACT program in the context of selected research on PM components and health in order to assess the current state of the field. Overall, the ambitious NPACT research program found associations of secondary sulfate and, to a somewhat lesser extent, traffic sources with health effects. Although this and other research has linked a variety of health effects to multiple groups of PM components and sources of PM, the collective evidence has not yet isolated factors or sources that would be closely and unequivocally more strongly related to specific health outcomes. If greater success is to be achieved in isolating the effects of pollutants from mobile and other major sources, either as individual components or as a mixture, more advanced approaches and additional measurements will be needed so that exposure at the individual or population level can be assessed more accurately. Enhanced understanding of exposure and health effects is needed before it can be concluded that regulations targeting specific sources or components of PM_2.5 will protect public health more effectively than continuing to follow the current practices of targeting PM_2.5 mass as a whole.

Implications: This paper describes a comprehensive epidemiologic and toxicologic research program to evaluate whether some components and sources of PM may be more toxic than others. This question is important for regulatory agencies in setting air quality standards to protect people’s health. The results show that PM from coal and oil combustion and from traffic sources was associated with adverse health outcomes, but other components and sources could not definitively be ruled out. Thus, given current knowledge, the current practice of setting air quality standards for PM mass as a whole likely remains an effective approach to protecting public health.

Introduction

Ambient particulate matter (PM) is a complex mixture of solid and liquid particles suspended in air. The size, chemical composition, and other physical and biological properties of particles vary with location and time. This variability in pollutant levels derives from differences in pollutant sources. The sources may be natural, such as forest fires, or the result of human activities, such as driving vehicles and operating manufacturing or power plants. In addition, reactive species in the atmosphere combine to generate secondary particles, such as sulfates, that may constitute a significant fraction of total PM. Ambient PM levels in any particular location are also affected by local ambient mixtures of gaseous pollutants, meteorology, geography, and seasonal patterns.

Although the characteristics of PM differ from place to place, epidemiologic studies in diverse locations in the United States generally have reported small but consistent, statistically significant associations between increases in daily mortality and certain morbidity endpoints with increases in the concentrations of particles less than 10 µm in diameter (PM₁₀) and particles less than 2.5 µm in diameter (PM_2.5) (Health Effects Institute, 2001, 2002). Some scientists have hypothesized that even smaller, ultrafine particles (less than 0.1 µm in diameter), which dominate in terms of numbers of particles in ambient air, may be particularly toxic (reviewed in Utell and Frampton, 2000; Oberdörster et al., 2000; Health Effects Institute, 2013).

Findings from epidemiologic and controlled-exposure studies about the health effects of PM have led the U.S. Environmental Protection Agency (EPA) and other regulatory agencies to establish mass-based ambient air quality standards for PM within a specific size range. PM_2.5 is considered to be particularly important because of the consistently reported associations between PM_2.5 concentrations and cardiovascular mortality in large cohort studies and other studies cited in the updated American Heart Association (AHA) scientific statement on PM air pollution and cardiovascular disease (Lepeule et al., 2012; reviewed more broadly in Brook et al., 2010).

Because the composition of PM is complex, there has long been a question as to whether some components of the PM mixture are of greater public health concern than others. Obtaining this information would help focus efforts to reduce human exposure by enabling the control of those sources that contribute most of the toxic components in the PM mixture.

The major components of PM include metals, organic compounds (measured as organic carbon [OC]) including materials of biological origin, inorganic carbonaceous material (including black carbon [BC] and elemental carbon [EC]), and sulfate, nitrate, ammonium, and other ions. The sources and composition of larger particles generally differ from those of smaller particles: coarse PM (particles 2.5–10 µm in diameter) consists in large part of insoluble crust- derived minerals, biological material (such as pollen, endotoxins, fungi, and bacteria), and sea salts. By contrast, PM_2.5—which includes the ultrafine fraction—is derived mainly from combustion-related sources. PM_2.5 includes particles with a carbon core (with attached hydrocarbons and metals), hydrocarbons, and secondary particles formed from oxides of sulfur and nitrogen. In addition to PM, ambient air also contains a range of gaseous pollutants (e.g., ozone, carbon monoxide, sulfur dioxide, and nitrogen oxides) that are derived from some of the same sources as PM. Gaseous pollutants have health effects of their own and may act in concert with PM to cause health effects. Any consideration of the health effects of different components and sources of PM must consider how gaseous pollutants may affect the toxicity of PM constituents.

In the past decade, a number of studies have evaluated the toxicity of PM constituents, using both epidemiologic and toxicologic approaches. Controlled-exposure studies, particularly in nonhuman species, have provided information about the different sizes and physicochemical compositions and solubility of particles that induce adverse effects (reviewed in Health Effects Institute, 2002; EPA, 2004). However, such studies have used relatively high concentrations, and often used model particles (such as black carbon, titanium dioxide, or vanadium pentoxide) that do not represent the complexity of ambient PM. Studies that use concentrated ambient particles (CAPs) avoid some of those issues but are technically complex, involve exposure to PM/gas mixtures that are significantly changed from the ambient mixture, and have to account for daily variation in the composition of PM (Godleski et al., 2000).

Epidemiologic studies have also evaluated the effects of particles of different sizes and composition on mortality and morbidity (Pekkanen et al., 1997; Peters et al., 1997; Lippmann et al., 2000; Wichmann et al., 2000); however, the number of such studies have been limited due to the lack—until recently—of air quality monitors that measure ultrafine particles or PM composition. Although these efforts have contributed valuable insights, the ability of some studies to detect statistically significant pollution effects is sometimes limited, as relatively short study periods and modest populations may limit the available data, and relatively short study periods, modest populations, and high correlations among pollutants in any one city may preclude clear associations between health effects and any single component. Other studies have attempted to associate health effects directly with source exposures (Clarke et al., 2000; Laden et al., 2000; Riediker et al., 2004; reviewed more broadly in Stanek, 2011). The statistical approaches in these studies, which include factor analysis and principal component analysis, are based on assumptions about the groups of elements and compounds that characterize an emission source. Major questions remain about the specificity of the markers used to define particular pollutant sources. Recent reviews of the evidence on the relative toxicity of different PM components and sources by the EPA (Stanek, 2011) and the World Health Organization (WHO) European Office (WHO, 2013) have concluded that although there are associations found for a number of health outcomes and different PM components and sources, there is not a consistent set of findings pointing either to one or more components/sources that are more toxic, nor toward components/sources that are less toxic.

National-scale studies of the health effects of exposure to PM components were difficult to conduct until more population-focused nationwide air quality monitoring of PM components was in place. Although the EPA Interagency Monitoring of Protected Visual Environments (IMPROVE) network had been collecting some speciation information for some time, detailed and geographically comprehensive information on urban PM_2.5 composition began to be collected systematically in the year 2000 as part of the EPA’s Speciation Trends Network, which is currently called the Chemical Speciation Network (CSN). Concentrations of PM_2.5 components and gaseous pollutants at sites in the CSN and IMPROVE networks, as well as state, local, and tribal air monitoring stations in the United States have been made available for the years 2000 through 2013 via the Health Effects Institute (HEI) Air Quality Database (https://hei.aer.com). The data were subsequently used by several teams of researchers for harmonized epidemiologic and toxicologic studies of the effects of air pollutant components on health, including those funded under HEI’s National Particle Component Toxicity (NPACT) initiative described in this paper.

In this paper, we provide an overview and synthesis of the results of the NPACT initiative, bringing in evidence from other key studies and literature reviews that shed light on the question of whether certain components of PM may be more toxic than others. We discuss what evidence is consistent across the NPACT studies and other studies, and what research gaps remain to be addressed in future research. The evaluation of the two NPACT studies included here is based on an independent review by the HEI NPACT Review Panel.

The National Particle Composition Toxicity Initiative

HEI launched the NPACT initiative in view of emerging evidence that there are geographic differences in both the composition and toxicity of PM across the country. Under this research program, two major studies were funded (Lippmann et al., 2013; Vedal et al., 2013a) that combined coordinated efforts in (1) exposure assessment using advanced techniques, (2) epidemiology focusing on PM components and long-term health effects, and (3) toxicology focusing on endpoints that are relevant to the cardiovascular and other health effects observed in epidemiologic studies. Each of the two main studies comprised several studies, led by co-investigators, to evaluate different aspects of the questions regarding the cardiovascular and other health effects of short- and long-term exposure to PM components, using approaches that would complement each other.

While addressing the PM components question, a secondary goal of the research program was to advance the state of the science, in particular to advance exposure assessment and statistical approaches to evaluate the effects of multiple pollutants simultaneously. The program also tried to address knowledge gaps related to long-term effects of PM exposure and effects of air pollutants on clinical and subclinical cardiovascular outcomes. The following sections provide a brief summary of study designs (shown in Table 1), and key results, followed by a discussion of insights learned from the studies. Our comments reflect the view of the HEI NPACT Review Panel that conducted a thorough evaluation of both studies. For a full discussion of the results, we refer to the individual reports (Lippmann et al., 2013; Vedal et al., 2013a) and a recent review of the program (Lippmann, 2014).

Coordinated epidemiologic and toxicologic studies: Study design

Lippmann et al. (2013) conducted four parallel toxicologic and epidemiologic studies to determine short- and long-term health effects associated with PM and its components. The two epidemiologic studies were a cohort study (Thurston et al., 2013) to evaluate associations between long-term exposure to PM components and mortality from cardiovascular disease (CVD), respiratory disease, and lung cancer for participants in the Cancer Prevention Study II maintained by the American Cancer Society and a time-series analysis (Ito et al., 2013) of all-cause mortality and hospital admissions associated with specific source categories of PM_2.5 in 150 U.S. cities. One toxicologic study (Chen et al., 2013) analyzed heart rate variability (HRV) and atherosclerosis in mice exposed for 6 months by inhalation to fine concentrated ambient particles (CAPs) in five geographic regions of the United States. Another toxicologic study (Gordon et al., 2013) measured acute changes in markers of inflammation and oxidative stress in mice and human cell lines exposed to a large number of PM samples collected at the same five locations as in the other toxicologic study, focusing on metal composition and PM size classes (coarse, fine, and ultrafine). The Lippmann team used data from the EPA’s Chemical Speciation Network (CSN), both as elemental concentrations and using source apportionment techniques to evaluate which specific components and source categories might be contributing most to the health effects associated with exposure to PM.

Vedal et al. (2013a) hypothesized that the cardiovascular health effects associated with long-term exposure to PM_2.5 are driven in large part by traffic-related sources. They conducted parallel epidemiologic and toxicologic studies. The epidemiologic study (Vedal et al., 2013b) used data from the Multi-Ethnic Study of Atherosclerosis (MESA) and the Women’s Health Initiative—Observational Study (WHI-OS) cohorts to evaluate two subclinical markers of atherosclerosis, carotid intima-media thickness (CIMT) and coronary artery calcium (CAC), as well as adverse events and mortality from total CVD and from atherosclerotic and cerebrovascular disease (including stroke). The investigators used spatial models based on concentrations for PM_2.5, sulfur, OC, EC, and silicon (which they considered to be markers for specific source categories) from the CSN database to estimate long-term pollutant concentrations to which participants in both cohorts had been exposed (referred to as the national spatial model). They also used data from additional measuring campaigns in the MESA cities to estimate spatially and temporally resolved concentrations at the participants’ residences in the MESA cohort (referred to as the spatiotemporal model). The parallel toxicologic study (Campen et al., 2013) evaluated the role of mixed vehicular engine emissions (MVE), its gaseous components, and other PM components (e.g., sulfates, nitrates, and road dust) in contributing to adverse health effects of PM. Campen and colleagues generated a mixture of diesel and gasoline emissions and exposed mice that are prone to developing atherosclerotic plaques, assessing biomarkers of oxidative stress and vascular inflammation.

Table 1. Broad overview of NPACT study designs

Study Approach	Lippmann et al.	Vedal et al.
Exposure timescales	Short- and long-term	Long-term only
Health endpoints	Respiratory and cardiovascular	Cardiovascular only
Epidemiologic Studies
Study design	Multicity time-series analysis and one cohort	Two cohorts
Health endpoints	Acute: respiratory and cardiovascular mortality and hospitalizationsChronic: mortality	Chronic: Subclinical markers of atherosclerosis; cardiovascular disease events (including mortality)
PM components and exposure assessment	EPA CSN monitors; MSA averages; sources	Cohort-specific and EPA CSN and IMPROVE monitors; individual-level exposure predictions; two exposure models; focus on OC, EC, silicon, and sulfur; included some evaluation of other pollutants and PM components
Source apportionment goal	Assessing exposure	Interpretation of exposure health effect estimates
Toxicologic Studies
Study design	ApoE knockout mouse model (normal diet); 6-month exposures; FVB/N mice; 12-day and 100-day exposures	ApoE knockout mouse model (high-fat/ high-cholesterol diet); 50-day exposure
Biologic endpoints	Cardiovascular effects and markers of oxidative stress and inflammation	Vascular effects and markers of oxidative stress and inflammation
Animal and cell culture exposures	Concentrated ambient particles (in vivo) and ambient particles collected on filters (in vitro and in vivo); five air sheds	Laboratory-generated complex mixtures: combinations of mixed vehicle emissions and non-vehicular primary particles (in vivo)

Notes: ApoE = apolipoprotein E; CSN = Chemical Speciation Network; EC = elemental carbon; EPA = U.S. Environmental Protection Agency; IMPROVE = Interagency Monitoring of Protected Visual Environments; MSA = metropolitan statistical area; OC = organic carbon; PM = particulate matter.

Coordinated epidemiologic studies: Key results and interpretation

Key results from the NPACT epidemiolgy studies are presented in Table 2. In the American Cancer Society (ACS) cohort, Thurston et al. (2013) found the strongest associations for all-cause mortality with the Coal Combustion and Traffic source categories and with sulfur—which is linked to both source categories—and EC, which is linked to Traffic. They also found strong associations between ischemic heart disease mortality and Coal Combustion and sulfur. The associations of Traffic and EC with mortality were, however, highly sensitive to the inclusion of ecologic covariates in the analyses and to the use of a random-effects Cox model instead of a standard Cox proportional hazards model. The investigators concluded that long-term exposure to PM_2.5 and the Coal Combustion source category explained most of the associations of exposure to PM_2.5 with all-cause, ischemic heart disease, and lung cancer mortality (but not respiratory mortality). However, the Traffic source category and EC, with their large spatial variation within cities, are not well captured by the limited number of CSN monitors in a city. It is also unclear how a decreasing trend in coal combustion emissions over the past decades may have influenced the results. Thus, the relative contribution of Coal versus Traffic to the observed long-term effects on all-cause mortality remains unresolved.

Table 2. Approaches and key findings of the epidemiologic studies^a,b

Study Approach	New York University/Ito		New York University/Thurston	University of Washington/Vedal^c
Study design	Time series (short-term)		Cohort (long-term)	Cohort (long-term)		Cohort (long-term)
Population	U.S. MSAs		ACS CPS-II cohort	MESA		WHI-OS
Cities	150 cities		100 cities	6 cities		45 cities
Participants	Population >100 million		~450,000 people	~6,800 people		~90,000 people
Health endpoints	Hospitalization	Mortality	Mortality	CIMT	CAC	Time to first event (MI, stroke, cardiac procedures, and CVD deaths)	Mortality
PM components and source categories associated with health outcomes	Cold season: PM2.5, NO2, CO, EC, OC, and Cu	Warm season: PM2.5, NO3, and sulfate	Components: PM2.5, As, Se, sulfur, Cl, Fe, Pb, and EC	Best evidence for OC and sulfur; little evidence for EC or silicon; some evidence for Cu	Some evidence for OC only	Best evidence for sulfur and less for OC; little evidence for EC or silicon	Best evidence for OC and less for sulfur; little evidence for EC or silicon
	Modified^d by PM2.5: Cu, Ni, and V	Modified^d by PM2.5: sulfate	Not associated: OC, silicon, and K
	Sources: vehicle exhaust	Sources: secondary aerosols	Source categories: coal combustion and possibly traffic; little evidence for soil or biomass combustion

Notes: ^aACS CPS-II = American Cancer Society Cancer Prevention Study II; As = arsenic; CAC = coronary artery calcification; Cl = chlorine; CIMT = carotid intima-media thickness; CO = carbon monoxide; Cu = copper; CVD = cardiovascular disease; EC = elemental carbon; Fe = iron; K = potassium; MESA = Multi-Ethnic Study of Atherosclerosis; MI = myocardial infarction; MSA = metropolitan statistical area; Ni = nickel; NO₂ = nitrogen dioxide; NO₃ = nitrate; OC = organic carbon; Pb = lead; PM = particulate matter; PM_2.5 = particulate matter <2.5 µm in aerodynamic diameter; Se = selenium; V = vanadium; WHI-OS = Women’s Health Initiative—Observational Study. ^bEpidemiologic studies at New York University were headed by George Thurston and Kazuhiko Ito and at the University of Washington by Sverre Vedal. ^cSource apportionment results supported the focus on four main components for evaluation in the Vedal study: EC and OC as markers of vehicle exhaust and other combustion emissions; OC also as a marker of secondary organic aerosol; silicon as a marker of crustal PM; and sulfur as a marker of secondary PM. ^dRisk estimates for these components were substantially modified when PM_2.5 was included in a two-pollutant model.

In their analyses of the WHI-OS cohort, Vedal et al. (2013b) reported that total deaths from CVD and from atherosclerotic disease showed the strongest associations with OC; associations with PM_2.5 and EC were marginal. Associations between deaths from cerebrovascular disease and exposure to OC were significant but less strong; they were not significant for PM_2.5 or any of the other components. Associations of total CVD events with PM_2.5 and sulfur were statistically significant, although small; negative and marginal associations were found for silicon. The only significant association for coronary heart disease events was with sulfur. Cerebrovascular disease events were significantly associated with OC and with PM_2.5; significant negative associations were observed with silicon. Additional analyses to compare the relative contributions of within- and between-city variances found mixed results.

Vedal et al. (2013b), in their analyses within the MESA cohort, reported that CIMT was significantly associated with exposure to PM_2.5, OC, and sulfur in both the spatiotemporal and national spatial models, although the risk estimates were generally small. The relative risks for OC and sulfur were higher than for PM_2.5 for the spatiotemporal model, but in the national spatial model this was true only for the city-adjusted model for sulfur. The investigators reported few significant associations of CAC with PM_2.5 or the four components in either model, with the exception of OC in the national spatial model. The analysis of subclinical cardiovascular effects is a promising direction for air pollution epidemiology. However, the longitudinal analyses of CAC and CIMT (i.e., over several follow-up visits) were hampered by the short period of time between evaluations, leaving only the cross-sectional evaluation (i.e., at one time point across cities) with interpretable results. Further follow-up of the MESA cohort would be useful, including, if feasible, analyses of subclinical endpoints not included in the current study (e.g., markers of inflammation and coagulation and other biomarkers).

In city-specific time-series analyses, Ito et al. (2013) reported many associations for daily all-cause mortality and hospital admissions for respiratory and cardiovascular causes across a variety of statistical models, although associations with individual PM_2.5 components were not particularly consistent. The most consistent associations were with total PM_2.5 mass itself and with the Traffic source category. However, this could be in part because PM_2.5 was measured more frequently than its components were, and Traffic was identified as a source category in more cities than were other categories. The results support associations of daily all-cause mortality and hospital admissions for respiratory and cardiovascular causes with both traffic-related pollutants and secondary aerosols. However, it should be noted that a high proportion of the data for important PM components (e.g., nickel, arsenic, copper, and vanadium) was below the limit of detection or had low monitor-to-monitor correlations within cities.

Coordinated toxicologic studies: Key results and interpretation

Key results from the NPACT toxicology studies are presented in Table 3. Chen et al. (2013) observed greater plaque development in the arteries of mice exposed to CAPs for 6 months at Manhattan (New York), Tuxedo (New York), and East Lansing (Michigan), compared with mice exposed to filtered air. Mice exposed at Manhattan and Tuxedo also showed acute increases in heart rate and decreases in heart rate variability. In contrast, no such differences were seen at Seattle (Washington) and Irvine (California). Chen and colleagues concluded that the effects on plaque progression were most likely attributable to a Coal Combustion source category, and that the Residual Oil Combustion, Coal Combustion, and Traffic source categories contributed most to the observed acute cardiac effects. Presumably, the effects observed at Tuxedo resulted from long-range transport of pollutants from other areas. Surprisingly, few changes were observed at Seattle and Irvine, two major urban areas dominated by traffic-related pollution. However, it remains unclear to what extent the larger responses observed in some locations might have reflected higher CAP exposures, rather than differences in PM composition.

Table 3. Approaches and key findings of the toxicologic studies^a,b

Study Approach	New York University/Chen		New York University/Gordeon		University of Washington/Campen
Study design	Short-term (daily time-series)	Long-term (6 months)	Short-term (12 days)	Long-term (100 days)	Medium-term (50 days)
Exposures	Inhalation of CAPs at five locations (Manhattan, Tuxedo, East Lansing, Seattle, and Irvine)		Aspiration of PM collected on filters at five locations (Manhattan, Tuxedo, Ann Arbor, Seattle, and Los Angeles area); three size fractions		Inhalation of combinations of MVE or MVE gases with non-vehicular PM2.5 (sulfate, nitrate, or road dust)
PM concentrations	60–138 μg/m3		In vitro: 50–100 μg/mL		100 or 300 μg/m3
			In vivo: 50 μg
Model	ApoE knockout mice (normal diet)		FVB/N mice; epithelial cells, endothelial cells, cardiomyocytes	FVB/N mice	ApoE knockout mice (high-fat/highcholesterol diet)
	Heart rate and heart rate variability	Aorta plaque growth; serum biomarkers; heart rate and heart rate variability	Mice: lung inflammation	Lung inflammation	Aorta: lipid peroxidation, vascular function and remodeling, plaque growth and inflammation
			Cells: viability, ROS, inflammatory markers, beat frequency
PM components and sources associated with biologic findings	Effects after exposure to Ni (residual oil combustion) > to Al, EC, and P (traffic) > to sulfur (coal combustion) > to PM2.5	Plaque growth in Tuxedo, Manhattan, and East Lansing, but not in Irvine and Seattle; attributed to coal combustion	Complex interaction of particle size and composition (location and season); nothing ruled out	Source apportionment not reported	Effects after exposure to MVE > to MVE gases; fewer effects of nitrate and sulfate; no effects of road dust
					Effects after exposure to non-vehicular PM combined with MVE > non-vehicular PM without MVE

Notes: ^aAl = aluminum; CAPs = concentrated ambient particles; MVE = mixed vehicular engine emissions; Ni = nickel; P = phosphorus; PM = particulate matter; ROS = reactive oxygen species. ^bToxicologic studies at New York University were headed by Lung-Chi Chen and by Terry Gordon and for the University of Washington study by Matthew Campen at the University of New Mexico and Jacob McDonald at the Lovelace Respiratory Research Institute

Gordon et al. (2013) observed small differences in the production of reactive oxygen species (ROS) in human epithelial and endothelial cell lines according to location, season, and size fraction, with the highest ROS production for samples from Manhattan and Los Angeles (California). ROS responses to ultrafine PM samples from all sites were higher than responses to coarse and fine PM samples (on an equal-mass basis); responses were higher in summer for fine and ultrafine samples but higher in winter for coarse samples. Strong correlations were observed between ROS production and copper, antimony, vanadium, cobalt, beryllium, and nickel. The investigators observed an increase in neutrophils, a sign of inflammation, in the lungs of PM-exposed mice. They noted a larger neutrophil response to the coarse fraction of PM than to the fine and ultrafine fractions, but those changes did not correlate well with in vitro ROS production for the same PM sample. The investigators concluded that the composition of PM samples pointed to the Traffic and Residual Oil Combustion source categories as contributors to the observed effects. It should be noted that the differences were relatively small; therefore, the possible toxicity of any particular components or size classes could not be ruled out. A limitation of the study by Gordon and colleagues was that it did not evaluate any organic carbon (OC), elemental carbon (EC), or other organic components of PM.

Campen et al. (2013) reported that lipid peroxidation, a marker of oxidative stress, was increased in aortic tissue of mice exposed to various atmospheres, with the largest increase observed after exposure to motor vehicle emissions (MVE). Removing the particles from the atmosphere reduced these effects but did not fully eliminate them. In contrast, exposures to nonvehicular PM alone did not produce an effect. Infiltration of atherosclerotic plaques by macrophages increased after exposure to MVE and to MVE gases combined with either sulfate or nitrate. In contrast, plaque formation increased only after exposure to nitrate alone or nitrate combined with MVE gases, but not to the other atmospheres. The investigators reported less consistent changes in the other endpoints.

The results suggested that the PM in MVE played a significant role in the induction of aortic lipid peroxidation, more so than MVE gases. These findings differ from those of previous studies from this laboratory, which found that the gaseous components in diesel or in gasoline exhaust induced oxidative stress. However, in the absence of exposures using MVE particles alone (i.e., without the gases), the role of MVE particles by themselves remains unclear.

Several caveats suggest a cautious interpretation of these results, including possible variability in aortic tissues because of sample collection procedures, small group sizes for certain endpoints resulting in insufficient power to find an effect, and some subjectivity in the method for assessing plaque densities. Although the multiple additive regression trees (MART) analysis used by Campen and colleagues is an interesting approach, the interpretation remains limited because the number of independent atmospheres was small compared with the number of components measured, and because daily variability in composition was not assessed.

Common issues and considerations

PM composition data

The NPACT studies—and similar studies of PM composition in the United States—rely on the availability of the CSN data and highlight the importance of the CSN for research on the health effects of components of air pollution and for air quality management; they also highlight, however, some of the limitations of that network. First, the network is composed of about 200 locations nationally and does not capture finer-scale spatial gradients in chemical components within cities. Second, most CSN locations collect samples only once every 3 or 6 days, limiting researchers’ ability to evaluate associations of PM components with daily health outcomes in short-term study designs. Third, concentrations of many of the components measured in the CSN network, especially metals, are below their limits of detection on a large number of sampling days, limiting analyses to only those components that can be detected repeatedly and reliably. Fourth, because the accuracy of measured concentrations of EC and OC depends on the methods used to measure these components, there is considerable uncertainty associated with their measurement, making comparisons across studies difficult. These issues affect some of the chemical components most important to the NPACT studies.

Vedal et al. (2013b) addressed the sparseness of the monitoring network and noncontinuous sampling by adding extra monitors in additional locations to measure EC, OC, and the other PM components measured by the CSN, and by calculating average concentrations over longer (2-week) time periods. However, they did not use the same measurement approach in their additional monitoring as was employed by the CSN, and their results did not agree well with measurements from collocated CSN monitors. Thus, although the increased spatial information provided by the additional monitoring might have reduced exposure measurement error, the different approach and sampling time used by the additional monitoring campaign might have actually enhanced such error. However, supplemental exposure measurement and the use of spatial modeling are seeing increasing use. A recent review by Hoek and colleagues (Hoek et al., 2013: see tables 1, 3, and 5) revealed a variety of exposure assignments using nearest monitor, city-level, county-level, and land use regression (LUR)-based approaches.

The uncertainties in EC and OC measurements are important because these components are used to help identify traffic as a source of PM. The Vedal team focused on these components in accordance with their hypothesis that traffic-related air pollutants drive the effects of PM on health. Source apportionment analyses conducted by the Lippmann team were also sensitive to these two components, because they were used in the estimation of traffic-related source categories. In addition to EC and OC being operationally defined (EC and OC are complementary fractions of total carbon, and their respective concentrations depend on the methods used for sampling and measuring carbonaceous material; see above), they are known to be subject to strong spatial and temporal gradients, making it likely that the small number of observations made at central monitoring stations do not adequately represent the highly variable concentrations observed across an entire urban area. In their analysis of mortality and particle component exposure in the California Teachers Study cohort, Ostro and colleagues both mitigated and verified such spatial variability issues with these components by restricting their analyses to participants living within 8 and 30 km of a Speciation Trends Network (STN) monitor (Ostro et al., 2010). Nonetheless, EC and OC continue to be important components to characterize in studies that evaluate the health impacts of PM components, particularly when there is an interest in traffic-related effects.

On the other hand, sulfate (measured as elemental sulfur) is well captured by the CSN. Sulfur concentrations are typically well above detection limits, are measured with relatively high certainty, and have relatively low spatial variability. Therefore, exposure measurement error associated with sulfate is expected to be low. Selenium, arsenic, vanadium, and nickel, which are key components for identifying coal burning and fuel-oil combustion, are often below the stated limit of detection for CSN measurements. The low concentrations of these pollutants, which have been decreasing over the past decades, hinder assessment of whether they might be linked to health impacts. However, as reported by the Lippmann team in the current and prior studies, in some locations (notably New York City) concentrations of vanadium and nickel were sufficiently high that it has been possible to identify associations of these elements with health outcomes.

Estimating exposure using air quality data

The Lippmann team’s approach assumed that the monitored concentrations (or source apportionment results estimated for each city based on a single monitor or a few central monitors) can be used directly. The Vedal team, on the other hand, developed a more elaborate spatiotemporal exposure model, which estimated exposures at the individual level (i.e., the outdoor concentrations at participants’ residences) for the MESA cohort. This approach was made possible by the intensive supplemental monitoring conducted by the team in the six cities of the MESA study. The Vedal team also constructed a national spatial exposure model, which estimated component concentrations at participants’ homes for their analyses of both the MESA cohort and the WHI-OS cohort.

There were challenges associated with estimating EC and OC concentrations at the individual level. For instance, there were only small differences between EC concentrations measured at roadside locations and those at urban background locations, raising questions about the ability of the spatiotemporal model to accurately assign exposure at participants’ residences. There are additional concerns, such as the varying R² values for the different components across the models (an indication of model accuracy in model validation) and the potential loss of volatile components over the longer sampling period of 2 weeks. At the same time, there is a more general challenge facing the primary alternative to such spatiotemporal modeling, which is the reliance on observations from just a few sites to characterize potential population wide intraurban exposures to pollutants such as EC, OC, and other primary pollutants (in much the same way the Lippmann team proceeded). Although using one or a few sites to characterize individual and population wide exposures to certain secondary PM components, such as sulfate, may be sufficiently accurate, using this approach to estimate exposures to primary pollutants—such as metals—introduces larger uncertainties, potentially biasing the results.

Linking PM components and sources to health outcomes

For their epidemiologic analyses, the two NPACT teams adopted somewhat different approaches on the use of source apportionment to link health outcomes to PM components. The Lippmann team conducted component-specific analyses but also applied a source apportionment approach that they had developed previously to link source categories directly to health outcomes in their epidemiologic analyses. In contrast, Vedal and colleagues used source apportionment to assist in the interpretation of their health effects estimates and to support their focus on OC, EC, silicon, and sulfur as markers of specific sources of health outcomes. An underlying question is which approach provides better information about which sources of PM components most affect health risks: Is it better to use source apportionment results, which may represent the combined effects of multipollutant atmospheres, but which require more effort and introduce additional uncertainties and assumptions, or is it better to use individual components that are typically linked to one or more specific sources?

All current source apportionment approaches introduce uncertainty (Balachandran et al., 2012). Although some approaches may decrease uncertainty by reducing temporal variability, other approaches that produce source categories may increase temporal variability as compared with approaches using concentrations of individual components. Using an approach based on factor analysis methods that they had developed previously, the Lippmann team found differences among locations in terms of which components contributed to similar source categories, providing indications that source emissions vary spatially and temporally and that the factor analytic approaches are sensitive to measurement uncertainties. The investigators also did not account for how uncertainties in the component measurements affect the certainty of the source categories and that many of the measured concentrations were below the limit of detection. Furthermore, it is not apparent which chemical components drive the associations between the source categories that they identified and health outcomes, although the Lippmann team came to consistent interpretations when they did include individual components in their analyses.

The Vedal team applied positive matrix factorization (PMF), a widely used source apportionment approach, to provide reassurance that the selected components (which were logical choices) generally covaried with the factors. The multiple exposure estimates used in the MESA study provided a good opportunity to gain new insights into how the choice of exposure model affected the results. However, the ability of the models to predict national-scale patterns does not necessarily translate into an ability to predict patterns within a city. It was not surprising that Vedal and colleagues found that the regionally varying pollutants sulfur and OC were more prominently associated with outcomes than more locally variable pollutants, such as EC. However, the lack of statistically significant results for such locally variable pollutants is not necessarily evidence of a lack of associations, given the study design and high correlations between components (particularly, EC and OC).

The question of how (or whether) to use source apportionment to identify which PM components have strong associations with adverse health outcomes is an important one. It is generally preferable to use both source categories and component concentrations directly in the health analyses, if the study design permits, with a focus on examining consistencies and differences between the two approaches. When source apportionment results are used for health analyses, researchers should recognize, discuss, and—if possible—address the uncertainties introduced by this method.

Single-pollutant and multipollutant models

When associations of PM_2.5 components and health outcomes are analyzed in single-pollutant models, potential interactions or high correlations between the components could affect the analysis and lead to misattribution of risk. Furthermore, other constituents of inhaled atmospheres—such as gaseous pollutants—might complicate the assessment of causality. The Lippmann team attempted to address these issues by employing source apportionment in all mass and a total-risk-impact approach in their cohort study. The Vedal team made simple comparisons between the results for individual components and those for PM_2.5 mass in their epidemiologic study and carried out sensitivity analyses involving two-pollutant models. They performed a more sophisticated analysis (viz., a MART analysis) in their toxicologic study (Campen et al., 2013), in which they related the hundreds of compounds measured in their complex exposure atmospheres to biological markers. Although both teams put considerable effort in their methods, any future research using PM component data needs to more directly address appropriate analyses for multipollutant atmospheres in the statistical design.

As governmental agencies are debating how to best approach the task of reducing exposure to a mixture of air pollutants, multipollutant statistical approaches are developing rapidly to fill a much needed research gap (Dominici et al., 2010; Greenbaum and Shaikh, 2010; Johns et al., 2012). In the near future, we expect that further advances in statistical approaches and multipollutant exposure metrics will provide additional insight into the question of how to separate health effects associated with one pollutant from effects of another pollutant that may be highly correlated (Austin et al., 2012; Barzyk et al., 2012; Levy et al., 2012; Pachon et al., 2012; Sun et al., 2013; Oakes et al., 2014; Park et al., 2014).

Studies using experimental inhalation

The two NPACT teams exposed apolipoprotein E (ApoE) knockout mice to exposure atmospheres with pollutant concentrations that were by design higher than typical North American ambient concentrations, although such concentrations can be found in developing countries or some occupational settings. The teams used different approaches to generate the pollutant mixtures, making it possible to compare responses to concentrated ambient PM and predetermined laboratory mixtures in a similar animal model. Chen et al. (2013) used fine PM concentrators (PM smaller than 2.5 μm), which did not exclude (or concentrate) gaseous pollutants or particles smaller than 0.1 µm (ultrafine PM). Thus, the resulting CAP exposure atmosphere is similar in pollutant composition to the ambient air, but the mixture can be altered substantially in terms of particle concentration and the relative ratios of gases to particles, and the potentially reduced ability of the reactive gaseous species to combine with the particles. Despite the potential shortcomings, this is an appropriate approach given the focus on PM components in the NPACT initiative and the fact that much of the mass of ambient PM is within the size range (PM_2.5) that is being concentrated and is of great interest regarding its health effects. Chen and colleagues measured about 30 components in the CAP atmospheres.

Campen et al. (2013) generated controlled atmospheres by mixing diluted and cooled exhaust from a gasoline and a diesel engine to provide a base pollutant mixture (i.e., MVE) and then removing PM from the mixture or adding different types of PM. This approach was driven by their general focus on PM components derived from traffic (vehicular) sources for both the epidemiologic and toxicologic studies. Campen and colleagues measured close to 500 compounds (metals and many organic compounds in the particle and gas phases) in their complex exposure atmospheres.

The inhalation exposures used in the Campen study did not include secondary PM components that are formed by atmospheric processes (e.g., secondary organic aerosols). However, sulfate and nitrate ions, which are major PM components in ambient air, were added as primary particles, allowing the team to investigate their health effects. In a typical city, secondary sulfate particles would form by oxidation of gaseous sulfur dioxide emissions from coal or oil burning, whereas secondary nitrate particles would be formed by oxidation of nitrogen oxides emitted by vehicles and other combustion sources. A unique feature of the Campen study was the addition of road dust particles in the fine fraction. In contrast, the animal exposure atmospheres used in the Chen study included secondary aerosols by design, although the extent to which this occurred likely varied by location.

In addition to the animal inhalation exposures in the two studies, the Lippmann team also used intratracheal aspiration of particles collected on filters (in the Gordon study), which allowed them to investigate the differences in biological responses in mice exposed to different PM size ranges. This approach excluded gaseous components altogether. In addition, this approach delivers a “bolus” concentration of material to the cells, which is quite different from prolonged exposure at lower concentrations during inhalation exposures. The investigators analyzed endotoxin content of the filter samples and elemental composition but did not analyze OC, EC, or other organic compounds.

Because Chen et al. (2013) conducted inhalation studies in five locations with different ambient air pollution mixtures, they performed source apportionment to link their exposures back to source categories, such as emissions from mobile and stationary sources. Therefore, the animal exposure strategies of both teams had the potential to link biological endpoints to similar types of sources, such as traffic, power generation, and dust, as well as to secondary aerosols (sulfates and nitrates). Furthermore, the parallel epidemiologic studies used similar markers for mobile-source emissions (EC and OC), although their methods could not separate PM derived from gasoline engines from PM derived from diesel engines based on EC and OC concentrations.

Other investigators have also used CAP exposures to study mechanisms of toxicity. By locating the concentrator close to specific sources and using cut plates for different PM sizes, a variety of exposures can be created. For example, Kleinman et al. (2007) located a concentrator at varying distances from a major roadway in Los Angeles to capture the effects of traffic-related fine and ultrafine PM. They reported effects on allergic responses in mice exposed 50 m from the roadway as opposed to mice exposed at 150 m from the roadway, indicating a role for traffic-related air pollution in the observed effects. Wagner et al. (2012) located concentrators in neighborhoods with traffic sources (Grand Rapids [Michigan]) or traffic plus industrial sources (Detroit [Michigan]) and found stronger effects on airway symptoms in asthmatic rats exposed to fine CAPs in Detroit, pointing to adverse effects from industrial rather than traffic sources. In another study, Kamal et al. (2011) found that effects of fine CAPs on cardiovascular outcomes in hypertensive rats exposed in Steubenville (Ohio) depended on wind direction, pointing to incineration, lead, and iron/steel sources and, to a lesser extent, mobile sources. Concentrators have also been used to expose human volunteers to CAPs under controlled conditions. Recent studies showed that fine and coarse CAPs increased blood pressure in healthy adults (Bellavia et al., 2013) and that ultrafine CAPs can cause cardiovascular changes in people with metabolic syndrome (Devlin et al., 2014).

CAP exposures can also be mixed with other pollutants, allowing comparison of biological responses to the mixture. This approach was used by Xu et al. (2012) in which mice were exposed to nickel sulfate, CAPs, or a combination of nickel sulfate and CAPs; they showed that nickel and CAPs had synergistic effects on some of the metabolic and inflammatory endpoints measured; Huang et al. (2012) observed synergistic effects of NO₂ and fine CAPs on cardiovascular responses in healthy young volunteers.

In the approach used by Campen et al. (2013), MVE was a reasonable representation of mobile-source emissions for toxicologic studies that allowed a more direct comparison of the toxicologic results with epidemiologic results for non-source-specific estimates of traffic-related exposures. On the other hand, the sulfate added to the MVE exposures was a primary rather than a secondary particle and did not include other components (e.g., selenium, arsenic, vanadium, or nickel) that are often found in emissions from sources that emit sulfur dioxide and was thus less representative of real-world conditions.

The approach by Campen and colleagues was in line with a broader strategy employed by the National Environmental Respiratory Center (NERC) to evaluate specific source mixtures, as summarized in a series of recent reviews (Mauderly, 2014; Mauderly and Seilkop, 2014; Mauderly et al., 2014a, 2014b). The NERC program comprised a systematic set of inhalation exposures of rats and mice to diesel engine exhaust, gasoline engine exhaust, hardwood smoke, and coal emissions. The investigators evaluated a comprehensive set of biological endpoints (histopathology, respiratory effects and lung lavage, vascular effects, DNA damage and oxidative stress), after exposure for 7 days, or 6 months, with a few endpoints evaluated after 3 or 50 days (Mauderly, 2014). About 700 compounds were measured that were lumped into about 60 classes for data analysis that were further narrowed down to about 45 exposure or “predictor” variables in the MART analysis described above (Mauderly and Seilkop, 2014). Mauderly (2014) states that “All four combustion-related mixtures caused multiple statistically significant responses at the higher exposure levels, although none caused responses in all measured endpoints. No exposure caused overt illness, neutrophilic lung inflammation, increased circulating micronuclei or histopathology visible by light microscopy in major organs. The number and magnitude of significant responses varied among the mixtures, but the results did not support an overall ranking of relative toxicity.”

The TERESA (Toxicological Evaluation of Realistic Emission Source Aerosols) study (Godleski et al., 2011) was another comprehensive program that evaluated specific source mixtures using primary and secondary (aged) emissions from power plants. Their results showed relatively mild responses to the inhaled aerosols studied; they noted that “complex scenarios which included oxidized emissions and α-pinene to simulate biogenic secondary organic aerosol tended to induce more statistically significant responses than scenarios of oxidized and non-oxidized emissions alone. Relating adverse effects to specific components did not consistently identify a toxic constituent. These findings are consistent with most of the previously published studies using pure compounds to model secondary power plant emissions.” The results of the NERC and TERESA programs as well as the Campen NPACT study demonstrate that in addition to particles, gaseous components may have significant effects. The specific roles of particles versus gases and how they may interact is a complicated issue that warrants further research.

General Discussion

The NPACT studies, which are to date the most systematic effort to combine epidemiologic and toxicologic analyses of these questions, found associations of secondary sulfate and, to a somewhat lesser extent, traffic sources with health effects. The HEI NPACT Review Panel concluded that the NPACT data do not provide compelling evidence that any specific source, component, or size class of PM may be excluded as a possible contributor to PM toxicity.

How do the two NPACT studies compare with others in the published literature? Quite a few investigators have performed smaller-scale studies and analyses to identify which PM components and sources are associated with a variety of adverse health outcomes. Not surprisingly, the results of those studies have been mixed, if only because of the differences in the selection of PM components and health outcomes of interest, study time frames (short- versus long-term), and the imprecision of estimates because of the difficulties in obtaining truly large data sets on PM composition and sources.

Although the NPACT studies made a substantial contribution to the literature on chronic health effects and long-term exposure to PM_2.5 components, there are other large research studies and meta-analyses that have also made important contributions. Ostro and colleagues analyzed mortality data from the California Teachers Study (CTS) with PM_2.5 and component concentrations for EC, OC, sulfate, nitrate, Fe, K, Si, and Zn from monitors in the STN (Ostro et al., 2010). Although they noted statistically significant associations between nearly all types of mortality and components for CTS participants residing within 8 and 30 km of an STN monitor (tables 4 and 5 of their paper), they found particularly strong associations between all-cause, cardiopulmonary, and ischemic heart disease mortality and average (2002–2007) concentrations of PM_2.5, OC, sulfate, nitrate, K, and Si for both 8- and 30-km buffers. These findings are consistent with findings reported in NPACT for OC and S by Vedal and colleagues (2013b).

Recent papers on long-term exposure to PM and PM components analyzed data from two groups of European cohorts, Transport related Air Pollution and Health impacts – Integrated Methodologies for Assessing Particulate Matter (TRANSPHORM) and European Study of Cohorts for Air Pollution Effects (ESCAPE), and explored associations between cardiovascular mortality and a group of eight elements (Cu, Fe, K, Ni, S, Si, V, and Zn) said to represent major sources of pollution (Wang et al., 2014). Although they found no statistically significant associations between any of the element concentrations and cardiovascular mortality, they did note elevated hazard ratios for PM_2.5, Si, and S in this large data set. In a comprehensive survey and meta-analysis of long-term exposure and mortality studies of PM_2.5, coarse PM, NO₂, and EC/BC/black smoke, Hoek and colleagues provided both pooled estimates and summaries and evaluations of multiple studies and their results (Hoek et al., 2013). The authors pooled data from different groups of studies, depending on the particulate metrics and components under investigation, and provided tables exploring the differences between studies, such as the spatial scale used to assess exposure, years of follow-up, etc. For their pooled estimates, they reported statistically significant associations between all-cause and cardiovascular mortality and PM_2.5, and between all-cause mortality and EC. Although the EC exposure category incorporated studies that estimated EC from BC and black smoke levels, and was based on data from three studies, the findings are consistent with those reported in NPACT by Thurston et al. (2013).

Several other recent studies confirm the toxicity of particles of all size classes. For example, concentrated coarse particles in a rural area were shown to have cardiovascular effects in healthy adults (Brook et al., 2014). Further toxicologic work by the group at New York University confirmed the findings of NPACT regarding toxicity of fine, ultrafine, and coarse particles collected in rural and urban areas of New York and the role of chemical composition (Mirowsky et al., 2013). The recent reviews of the NERC program (see above) indicate the complexity of linking specific components of complex source mixtures to specific health outcomes. Overall, the results did not support a relative toxicity ranking of the four source mixtures evaluated, i.e., diesel exhaust, gasoline exhaust, hardwood smoke, and coal emissions (Mauderly, 2014). Also, recently there have been other reviews of the evidence (Lippmann, 2014; Grahame et al., 2014), although some of these (Grahame et al., 2014) are limited by their focus on certain components of the PM mixture without a systematic review of the full range of components and sources.

Short-term studies of PM_2.5 components and health outcomes are more commonly found in the literature. In one of the earliest studies to use speciated PM_2.5 data, Laden et al. (2000) used data from pollution monitors in the Harvard Six Cities Study to explore relationships between PM_2.5 components and daily mortality. When daily mortality was compared with both component concentrations and source factors derived from them, significant associations were noted between all-cause mortality and a “motor vehicle exhaust” factor, a “coal combustion” factor, nickel, lead, and sulfur. Bell (2012) used daily Medicare hospitalization data to evaluate the effects of short-term exposures to various components of the PM_2.5 mixture on daily morbidity. She focused on the average values of seven PM_2.5 components (those accounting for >1% of PM_2.5 mass in the CSN) in 187 U.S. counties, using national, regional, and seasonal models. For her all-year analysis of data from the entire United States, Bell reported strong and statistically significant increases in the association between cardiovascular hospitalizations and interquartile range increases in EC, nickel, and vanadium (Bell, 2012). Considering the many years between these studies and their different approaches (and quantities of data), they are fairly consistent with each other and with some of the short-term NPACT findings reported by Ito et al. (2013).

There have been several recent systematic reviews and meta-analyses of associations between short-term exposures to components and/or source factors and health effects worth noting here. A meta-analysis of time-series studies and reanalysis of data from the Speciation Trends Network and the Medicare system in the United States covering the years 2000–2008 provides some important insights (Levy et al., 2012). In a pooled analysis, the study found statistically significant associations between daily all-cause mortality and both PM_2.5 and sulfate. In addition, a sophisticated multipollutant analysis indicated that, when controlled for all other pollutants, EC was “more toxic” than the other components or PM_2.5 with respect to hospital admissions for cardiovascular disease, whereas OC was “more toxic” than the other components with respect to hospital admissions for respiratory disease (Levy et al., 2012). Although this study investigated a smaller range of PM components than Ito et al. (2013), these results are broadly consistent. A recent systematic review of the findings of animal toxicology, human chamber, and field epidemiology studies (Stanek et al., 2011) presents results from five epidemiologic studies on total mortality (see table 3 of that paper), which among them found that soil, sea salt, local sulfur dioxide, secondary sulfate, motor vehicle emissions, coal burning, wood smoke, biomass combustion, copper smelter emissions, residual oil combustion, and incinerator emissions were associated with health outcomes. The recent review by WHO Europe (WHO, 2013) arrived at similar conclusions.

Overall, the current body of evidence identifies some of the components found to have associations with multiple outcomes in the NPACT results, such as EC, OC, and sulfate. However, few studies were as comprehensive with regard to the number of PM_2.5 components and source factors studied as the NPACT studies were. The results of the research and reviews discussed here support the conclusion of Stanek et al. (2011) that “apportionment methods have linked a variety of health effects to multiple groups of PM components and sources of PM, but the collective evidence has not yet isolated factors or sources that would be closely and unequivocally related to specific health outcomes.” In other words, neither the NPACT studies nor other key studies in the literature have definitively identified specific components that are of greater concern for either long- or short-term health effects than PM_2.5 mass itself.

Remaining questions and recommendations for next steps in PM research

The NPACT studies and other recent research mark important progress in understanding the toxicity of PM and its components. We conclude that it is difficult to point to any PM component or PM size class that explains a majority of the observed adverse health effects. Although the current body of research has made many technical and statistical advances, it is clear that further advances are needed; the main recommendations for additional work are discussed below.

Measuring PM components

Several measurement issues remain. First, concentrations of many components are often quite low, resulting in many nondetects. Second, certain measurements are collected only once every 3 or 6 days. Third, there are technical difficulties in reliably measuring certain compounds on filters, as is discussed above for EC and OC. Further efforts would be helpful to characterize EC, OC, and metals (i.e., combustion- and traffic-related components). OC, in particular, is composed of many different compounds from very different sources (e.g., traffic, biomass burning, and secondary aerosol) that likely have different impacts on health, and it is likely that the relationships between OC and health endpoints will vary in time and space. Until such issues are better resolved, it remains difficult to compare results across studies. Fourth, monitoring efforts are largely focused on the criteria pollutants, with less attention paid to air toxics; consequently, our knowledge about the potential contribution of air toxics is very limited.

Exposure measurement error

Central monitoring networks provide limited coverage nationwide and focus mostly on large urban areas, with rural areas being underrepresented. The limited number of central monitors in urban areas does not capture the spatial variation in many pollutant concentrations, which is an important issue for traffic-related air pollutants that have steep gradients with higher concentrations near busy roads. EPA’s recent launch of a near-road monitoring network is an important step forward (EPA, 2012) that will allow researchers to better capture roadside exposures and use those improved exposure estimates to better estimate the health effects associated with traffic-related air pollution. However, the loss of some CSN monitoring sites due to cost cutting measures is a worrisome trend that will negatively affect future research. In the meantime, a better understanding of the spatial gradients for components and their predictors will be required for improved spatiotemporal modeling of population exposures for epidemiologic studies.

Components versus mixtures

Several studies are now underway to improve statistical approaches to evaluating complex mixtures, as described earlier in the paper, but it is not yet clear how successful these new approaches are, and whether one approach may be distinctly better in tackling the difficult challenges of correlated and possibly interacting pollutants issue than other approaches. We propose that a next step would be to use a common data set (or data sets) to compare and contrast these new methods; one conclusion may be that further methods development is needed, or that existing methods capture some aspects of the problem sufficiently well to remain useful for certain applications.

Chronic exposure to air pollution and long-term health

The relatively short history of systematic, nationwide monitoring of PM component and sizes has hampered the study of long-term effects of exposure to air pollution. In particular, there have been few epidemiologic studies of the chronic effects of exposure to ultrafine particles. The chronic health effects of exposure to coarse particles are also understudied. Given the extensive efforts to reduce ambient concentrations of PM_2.5 and the potential for increasing numbers of particles in modern engine emissions, there is a need for long-term studies of different particle size classes, as well as differences in particulate composition.

Sulfate—often quantified as elemental sulfur—is much more consistently associated with health effects across studies. The high fraction of sulfate found in PM combined with its relatively large spatial scale of variation makes it difficult to disentangle its effects from that of PM_2.5 itself, and this has led some researchers to dismiss sulfate as a possibly causal component in the component mixture (Grahame et al., 2014). However, some researchers have postulated that sulfate in PM_2.5 leads to an acidic environment that activates other PM_2.5 components, such as metals, resulting in the observed adverse health effects (Ghio et al., 1999; Rubasinghege et al., 2010). In any case, the observed associations between sulfates in PM_2.5 and health effects should be evaluated further and either verified or explained through toxicologic studies of sulfates, sulfate mixtures, and sulfate interactions with gases.

There remains a disconnect between epidemiologic studies, which associate exposure to ambient mixtures that include both primary particles and secondary pollutants and gaseous components, and toxicologic studies, which most often use single pollutants or fresh source mixtures. Only fairly recently have researchers started to investigate the effects of aged mixtures (e.g., Sexton et al., 2004; Zielinska et al., 2010; Godleski et al., 2011; Papastopoulo et al., 2011). This type of research should be encouraged to provide a better understanding of, and link to, what humans are exposed to in the real word, and elucidate the mechanisms of effect that may drive observed epidemiologic associations. In particular, the relative role of PM versus gases and the extent to which their effects may be additive (or more than additive) needs further research. As the NERC studies have shown (Mauderly, 2014), the gaseous and oxidative components of the ambient pollutant mixture play an important role.

More comprehensive recommendations for additional research can be found in the NPACT Executive Summary and in the individual NPACT Review Panel commentaries for the Lippmann and Vedal reports (Lippmann et al., 2013; Vedal et al., 2013a, 2013b).

Conclusion

The NPACT studies, which are to date the most systematic effort to combine epidemiologic and toxicologic analyses of these questions, found associations of secondary sulfate and, to a somewhat lesser extent, traffic sources with health effects. But these studies, together with the wider body of evidence from other studies, do not provide compelling evidence that any specific source, component, or size class of PM may be excluded as a possible contributor to PM toxicity. If greater success is to be achieved in isolating the effects of pollutants from mobile and other major sources, either as individual components or as a mixture, more advanced approaches and additional measurements will be needed so that exposure at the individual or population level can be assessed more accurately. Such enhanced understanding of exposure and health will be needed before there is general agreement that regulations targeting specific sources or components of PM_2.5 will protect public health more effectively than continuing to follow the current practice of targeting PM_2.5 mass as a whole.

Acknowledgment

The authors would like to thank the members of the HEI NPACT Review Panel for their thorough evaluation of the NPACT reports. They would also like to thank the original investigators, Mort Lippmann and Sverre Vedal, and their teams for their efforts in designing and conducting these complex studies.

Funding

This document was produced with partial funding by the U.S. Environmental Protection Agency under Assistance Award CR-83234701 to the Health Effects Institute; however, it has not been subjected to the Agency’s peer and administrative review and therefore may not necessarily reflect the views of the Agency, and no official endorsement by it should be inferred. The contents of this document also have not been reviewed by private party institutions, including those that support the Health Effects Institute; therefore, it may not reflect the views or policies of these parties, and no endorsement by them should be inferred.

Additional information

Funding

This document was produced with partial funding by the U.S. Environmental Protection Agency under Assistance Award CR-83234701 to the Health Effects Institute; however, it has not been subjected to the Agency’s peer and administrative review and therefore may not necessarily reflect the views of the Agency, and no official endorsement by it should be inferred. The contents of this document also have not been reviewed by private party institutions, including those that support the Health Effects Institute; therefore, it may not reflect the views or policies of these parties, and no endorsement by them should be inferred.

Notes on contributors

Kate Adams

Kate Adams is a senior scientist, Daniel S. Greenbaum is President, Rashid Shaikh is Director of Science, and Annemoon M. van Erp is managing scientist at the Health Effects Institute.

Daniel S. Greenbaum

Kate Adams is a senior scientist, Daniel S. Greenbaum is President, Rashid Shaikh is Director of Science, and Annemoon M. van Erp is managing scientist at the Health Effects Institute.

Rashid Shaikh

Kate Adams is a senior scientist, Daniel S. Greenbaum is President, Rashid Shaikh is Director of Science, and Annemoon M. van Erp is managing scientist at the Health Effects Institute.

Annemoon M. van Erp

Kate Adams is a senior scientist, Daniel S. Greenbaum is President, Rashid Shaikh is Director of Science, and Annemoon M. van Erp is managing scientist at the Health Effects Institute.

Armistead G. Russell

Armistead (Ted) G. Russell is the Howard T. Tellepsen Chair of Civil and Environmental Engineering, School of Civil and Environmental Engineering, Georgia Institute of Technology. Dr. Russell served as co-chair of the HEI NPACT Review Panel.