UC Irvine UC Irvine Previously Published Works Title Air Pollution Exposures and Circulating Biomarkers of Effect in a Susceptible Population : Clues to Potential Causal Component mixtures and mechanisms

Background Mechanisms involving oxidative stress and inflammation have been proposed to explain associations of ambient air pollution with cardiovascular morbidity and mortality. Experimental evidence suggests that organic components and ultrafine particles (UFP) are important. Methods We conducted a panel study of 60 elderly subjects with coronary artery disease living in retirement communities within the Los Angeles, California, air basin. Weekly biomarkers of inflammation included plasma interleukin-6, tumor necrosis factor-α soluble receptor II (sTNF-RII), soluble platelet selectin (sP-selectin), and C-reactive protein (CRP). Biomarkers of erythrocyte antioxidant activity included glutathione peroxidase-1 and superoxide dismutase. Exposures included outdoor home daily particle mass [particulate matter < 0.25, 0.25–2.5, and 2.5–10 μm in aerodynamic diameter (PM0.25, PM0.25–2.5, PM2.5–10)], and hourly elemental and black carbon (EC–BC), estimated primary and secondary organic carbon (OCpri, SOC), particle number (PN), carbon monoxide (CO), and nitrogen oxides–nitrogen dioxide (NOx–NO2). We analyzed the relation of biomarkers to exposures with mixed effects models adjusted for potential confounders. Results Primary combustion markers (EC–BC, OCpri, CO, NOx–NO2), but not SOC, were positively associated with inflammatory biomarkers and inversely associated with erythrocyte anti-oxidant enzymes (n = 578). PN and PM0.25 were more strongly associated with biomarkers than PM0.25–2.5. Associations for all exposures were stronger during cooler periods when only OCpri, PN, and NOx were higher. We found weaker associations with statin (sTNF-RII, CRP) and clopidogrel use (sP-selectin). Conclusions Traffic-related air pollutants are associated with increased systemic inflammation, increased platelet activation, and decreased erythrocyte antioxidant enzyme activity, which may be partly behind air pollutant–related increases in systemic inflammation. Differences in association by particle size, OC fraction, and seasonal period suggest components carried by UFP are important.

Research Ambient mass concentrations of particulate matter (PM) air pollution < 2.5 µm (PM 2.5 ) and < 10 µm (PM 10 ) in aerodynamic diameter have been associated with hospital admissions and mortality due to cardiovascular causes in time series studies (Pope and Dockery 2006). Mechanisms involving oxidative stress and inflammation have been proposed to explain these associations (Mills et al. 2007) (Figure 1). In addition, a growing toxicol ogy literature suggests that ultrafine particles (UFP), < 0.1 µm in diameter, may have greater potential to induce oxidative stress and inflammation than larger particles that domi nate PM 2.5 and PM 10 mass (Ntziachristos et al. 2007). This is likely because compared with larger particles, UFP have a higher air way deposition efficiency, magnitudes higher particle number (PN) concentration and sur face area, and higher concentrations of organic components shown to induce oxidative stress responses (Li et al. 2003). The ability of UFP to translocate systemically from pulmonary sites makes them particularly relevant to the cardiovascular effects of inhaled PM (Elder and Oberdörster 2006).
We aimed to improve the characterization of PM exposure in order to yield clues to poten tially important pollutant sources and causal component mixtures not otherwise evident with ambient PM 2.5 and PM 10 mass, which are regulated by the U.S. Environmental Protection Agency   (Figure 1). For example, traffic (a common exposure source of redox active PM) increases spatial variability of UFP . Regional ambient data are thus likely to misrepresent personal exposure. In addition, ambient PM is made up of primary combustion aerosols, photochemi cally produced secondary organic aerosols, and mechanically generated crustal material. These particle types have different spatial and tempo ral variability. The organic component mix and size distribution differs as well between these two classes of particulate organic matter, with primary aerosols being more common in UFP and secondary aerosols more common in the accumulation mode (∼ 0.1-2.5 µm).
To address these questions, we conducted a panel study with repeated measurements of biomarkers and exposures in 60 elderly indi viduals with a history of coronary artery disease (CAD), a population potentially susceptible to adverse effects of air pollution (von Klot et al. 2005). We investigated the relationship of intensive measurements of outdoor home air pollutants to changes in circulating bio markers of inflammation, platelet activation, and antioxidant capacity (Figure 1). These bio markers may be risk factors for cardiovascular diseases (EspinolaKlein et al. 2007;Pai et al. 2004). We investigated the relative strength of biomarker associations between various PM size fractions. We also investigated differences in associations between exposure markers of trafficrelated primary PM compared with sec ondary PM. Results presented here add a sec ond year of data from 31 subjects to data used in a previous analysis of 29 subjects ).

materials and methods
Population. We recruited subjects from four large retirement communities in the Los Angeles (LA), California, air basin. Three were in the San Gabriel Valley, closer to downtown LA and thus closer to traffic sources, and one was further inland in Riverside, California. Eligibility criteria included a confirmed CAD history, ≥ 65 years of age, nonsmoker, and unexposed to environmental tobacco smoke. We clinically evaluated 105 potentially eligible subjects on site. Twentyone subjects were not eligible, and 20 dropped out or had too few blood draws (< 5 of 12 weeks), leaving 64 sub jects. Four subjects had insufficient biomarker data, mostly because of exclusions for frequent infections, leaving 60 subjects ≥ 71 years of age with 5-12 weekly blood draws (n = 578) ( Table 1). The study protocol was approved by the Institutional Review Board of the University of California, Irvine, and we obtained informed written consent from subjects.
Two communities were studied in 2005-2006 (29 subjects) and two communi ties were studied in 2006-2007 (31 subjects). We studied subjects in two periods to enhance known contrasts across the LA basin in particle composition and size distribution by season . In each community, we collected 6 weeks of data during a period of higher temperature (July−midOctober), and thus higher photochemical activity and mixing depths, and 6 weeks of data during a cooler period (midOctober−February), with more frequent periods of air stagnation and lower mixing heights (when trafficrelated primary air pollutants increase at ground level). Over a 7month period, each subject was followed weekly in these two 6week blocks with blood draws for circulating biomarkers of inflamma tion and antioxidant activity. Subjects com pleted daily diaries reporting medication use.
Biomarkers. Venous peripheral blood sam ples were drawn at the same time of day and day of week (Friday afternoons) and rapidly separated within 30 min into erythrocytes and plasma and frozen at our onsite mobile field laboratory. For the current analysis, we focused on biomarkers that were most informative in the previous analysis of firstyear data . Plasma samples stored at -80°C were thawed and assayed using 96well immuno assay kits for the proinflammatory cytokines inter leukin6 (IL6) and tumor necrosis factorα (TNFα) (Quantikine HS, R&D Systems, Minneapolis, MN), soluble TNFα receptor II (sTNFRII) (Quantikine, R&D Systems), the acutephase protein Creactive protein (CRP) (Zymutest, Hyphen BioMed, Neuvillesur Oise, France), and a marker of platelet activa tion, soluble platelet selectin (sPselectin) (Jurk and Kehrel 2005). Frozenthawed erythrocyte lysates were assayed spectrophotometrically for activities of two antioxidant enzymes, gluta thione peroxidase1 (GPx1) and copper-zinc superoxide dismutase (Cu,ZnSOD) (Cayman Chemical, Ann Arbor, MI), normalized to units per gram of hemoglobin (U/g Hb).
Exposure assessment. Measurement methods are more thoroughly described in the Supplemental Material (doi:10.1289/ ehp.0800194.S1). Hourly outdoor home air pollutants were measured over 9 days before each blood draw as described elsewhere (Arhami et al. 2006;Polidori et al. 2007). These meas urements included pollutant gases [carbon monoxide, nitrogen oxides-nitrogen dioxide (NO x -NO 2 ), ozone], total PN (condensa tion particle counter model 3785; TSI Inc, Shoreview, MN), PM 2.5 organic carbon (OC) and elemental carbon (EC) (OC_EC analyzer model 3F; Sunset Laboratory Inc., Tigard, OR), and black carbon (BC) (aethalometer; Magee Scientific, Berkeley, CA). We also measured sizefractionated PM mass with the Sioutas per sonal cascade impactor sampler (SKC, Inc., Eighty Four, PA) over 24hr periods from midafternoon to midafternoon for 5 days before each blood draw (unlike hourly pollut ants meas ured 9 days before). This included particles 0-0.25 µm in diameter (PM 0.25 ), accumulationmode particles 0.25-2.5 µm in diameter (PM 0.25-2.5 ), and coarse mode par ticles 2.5-10 µm in diameter (PM 2.5-10 ). We refer to PM 0.25 as "quasiultrafine" because the upper size cutpoint for the ultrafine mode has varied from 0.1 to 0.2 µm, depending on loca tions and seasons. UFP are traditionally defined as those originating mostly from fresh emis sion sources and accounting for > 90% of the numberbased particle concentrations . A major fraction of accumulation mode PM originates from the ultrafine mode. Figure 1. Pathways from PM exposure to measured changes in systemic biomarkers and to hypothesized adverse cardiovascular health effects. The different source characteristics and particle-size fractions we measured represent different types and concentrations of redox active components, including organic chemicals (we measured OC), that may lead to the production of reactive oxygen species in various cells of the lungs, blood, and vascular tissues. This will induce oxidative stress and then antioxidant responses from enzymes involved in oxidant defense, either by interacting with a relatively fixed pool of erythrocyte enzymes (we measured GPx-1 and Cu,Zn-SOD) or through induction of genes coding these enzymes in nucleated cells. Proinflammatory effects (we measured IL-6, TNF-α, sTNF-RII, and CRP) then occur at higher levels of exposure-induced oxidative stress when antioxidant defenses are overwhelmed (Li et al. 2003). This is expected to lead to upregulation of adhesion molecules on vascular endothelium, circulating leukocytes, and platelets (we measured sP-selectin). Induction of endothelial dysfunction, inflammation, and thrombosis by these mechanisms would increase the risk of acute adverse cardiovascular outcomes. Repeated acute insults such as these likely contribute to atherogenesis.  This is unlike coarse particles and fine particles (PM 2.5 , or accumulation plus ultrafine), which are naturally divided by a cutpoint of 2.5 µm and have clearly different origins. We estimated outdoor secondary OC (SOC) and primary OC (OC pri ) from total OC as detailed in our recent publica tion (Polidori et al. 2007) and summarized in the Supplemental Material (doi:10.1289/ ehp.0800194.S1). OC pri is representative of particles emitted directly from combustion sources (mostly fossil fuels in the LA basin), whereas SOC represents semivolatile and low volatile products of photochemical reactions involving reactive organic gases from anthro pogenic and biogenic sources. The study aver age outdoor SOC accounted for 34% and 44% of total OC in the cooler and warmer phases, respectively.
Analysis. We used linear mixedeffects mod els to analyze relationships of biomarkers to air pollutant exposures (Verbeke and Molenberghs 2001). Because withinindividual repeated measures of outcomes are correlated, random effects were estimated at the subject level, nested within phase and community. The covariance structure observed from empiric variograms was representative of an autoregressive1 correlation, and models were fit as such.
Using mean centered exposures, we adjusted for betweensubject group and betweenphase exposure effects. Thus, the interpretation of reported estimates is at the subject level [see Supplemental Material (doi:10.1289/ehp.0800194.S1)].
To assess more acute versus cumulative exposure-response relationships, we evaluated last 24hr averages of air pollutants (1 day) as well as cumulative exposures up to 9 days (or 5 days for particle mass) before the blood draw. We chose a set of averaging times that skipped over averages by 1 day to simplify the presenta tion while still presenting a view of associa tions across the span of averaging times (1day, 3day, 5day, 7day, and 9day averages).
We decided a priori to exclude person weeks with acute infectious illnesses, given their known impact on measured biomarkers. We controlled for temperature at the same averaging time as the air pollutant.
We hypothesized a priori that medication variables known to influence inflammation and oxidative stress would act as effect modifi ers. This included 3hydroxy3methylglutaryl coenzyme A (HMG CoA) reductase inhibi tors (statins), and angiotensinconverting enzyme (ACE) inhibitors or angiotensin II receptor blockers (ARB). We also tested effect modification of sPselectin associations by clopidogrel, a platelet aggregation inhibitor.
We tested differences in association by seasonal phase of study to gain clues regard ing underlying differences in potentially important air pollutant components. In addition, given the known differences in air pollution in the San Gabriel Valley (three communities) compared with Riverside (one community), we tested differences in associa tion between these regions. We planned these analyses in advance by designing the study to follow subjects during two seasonal phases and in different regions, known factors lead ing to differences in pollutant components and particle size distribution. All interactions (medications, phase, and group) were tested in product term models and all stratified results come from these models including all data.
Associations were more strongly posi tive in the San Gabriel Valley (44 subjects) than in Riverside (16 subjects We examined residual diagnostics to investigate deviations from standard linear mixedmodel assumptions (functional form of independent variables and covariance assumptions) and the presence of influential observations. Four influential high outliers for IL6 > 10 pg/mL were reset to 10 pg/mL, and one extreme influential outlier for sP selectin (221 ng/mL) was removed to obtain more representative estimates of association. Residuals for both CRP and TNFα exhibited a highly skewed distribution, primarily due to a cluster of subjects in the upper quartile of biomarker concentrations, and 2-3 high out liers > 3 SD above the mean. Outliers were reset to the next highest values, and secondary subgroup analyses were conducted among subjects in the upper quartile of mean CRP versus the lower three quartiles. Although this analysis was clearly data driven, similar sub group analyses have been previously reported (Dubowsky et al. 2006;Rückerl et al. 2006).  Mixedmodels analyses for both CRP and TNFα were stratified as such to show dif ferential risk by chronic inflammation and to express results for both variables in their measured units.
To identify influential subject clusters, we tested random slopes models as well as individual autoregressive models. Through this exploratory data analysis, we found that five subjects in erythrocyte Cu,ZnSOD models and three subjects in erythrocyte GPx1 models formed highly influential clusters with positive associations between air pollutants and biomarkers. One subject was a positive responder for both biomarkers [see Supplemental Material (doi:10.1289/ ehp.0800194.S1)]. Below, we present results for secondary analyses excluding these influ ential subjects. Because these results stem from a sensitivity analysis, the reported results should be interpreted conservatively.

Results
Descriptive statistics. Descriptive data for bio marker measurements are shown in Table 2. Table 3 gives descriptive statistics for exposures by phase of study. Exposures were generally similar across the two phases, except for nota bly higher concentrations of OC pri , PN, and NO x in phase 2 (colder phase), and higher concentrations of SOC and O 3 in phase 1 (warmer phase). High outdoor PM 0.25 relative to PM 2.5 are likely attributable to large impacts of local traffic in the LA basin compared with the eastern half of the nation with much larger contributions to PM 2.5 from accumulation mode sulfate aerosols. Table 4 shows expo sure correlations for combined phases. EC, BC, OC pri , NO x , and CO were strongly cor related with each other, likely because they are products of fossil fuel combustion. These correlations were stronger in phase 2 than in phase 1 (data not shown). PN and PM 0.25 concentrations were moderately correlated with these combustionrelated pollutants, and these correlations were stronger in the three San Gabriel Valley communities closer to traf fic sources than in Riverside (data not shown). There is a stronger correlation between PM 0.25 and PM 2.5-10 (coarse particles) than between PM 0.25 and PM 0.25-2.5 because PM 0.25 and coarse particles come from primary traffic sources in our study region. Whereas PM 0.25 is primarily a product of fresh emissions, PM 0.25-2.5 is a product of aging and photo chemical reactions.
Regression analysis. Many positive asso ciations were found for IL6, sPselectin, sTNFRII, TNFα, and CRP with markers of trafficrelated air pollution (EC, OC pri , BC, NO x , and CO). We also found inverse asso ciations of Cu,ZnSOD and GPx1 with the same pollutants. However, this was found only in the restricted subset of 55 (Cu,ZnSOD) and 57 subjects (GPx1) as presented below, whereas models including all 60 subjects were mostly nonsignificant [see Supplemental Material (doi:10.1289/ehp.0800194.S1)]. To simplify the presentation, we focus here on two biomarkers of inflammation (IL6 and sTNFRII) and present results for TNFα and CRP online [see Supplemental Material (doi:10.1289/ehp.0800194.S1)]. We also present results for two key markers of primary combustion (EC and OC pri ) that were strongly correlated with the other markers not shown (BC, NO x , and CO). We present pollutant averaging times here that best represent asso ciations across the span of time rather than the full set of selected averaging times. In many but not all cases, associations were strongest for longerterm averages out to the last 5 days and, in some cases, 9 days. Regression results for all pollutants and all selected lag averages are shown online [see Supplemental Material (doi:10.1289/ehp.0800194.S1)].
Biomarkers of systemic inflammation (IL6 and sTNFRII), but not sPselectin, were more strongly and significantly associated with quasiultrafine PM 0.25 than largersize fractions (Figure 2). We also found inverse associations of Cu,ZnSOD and GPx1 with PM 0.25 that were somewhat stronger than largersize fractions.
Across all biomarkers, we also found con sistently stronger associations for OC pri than for SOC, and this was generally reflected by

Figure 2.
Associations of biomarkers with outdoor size-fractionated particle mass on days 1, 3, and 5. Expected change in the biomarker (adjusted coefficient and 95% confidence interval) corresponds to an interquartile range change in the air pollutant (Table 3). Models for sTNF-RII and sP-selectin are restricted to 44 subjects living in the San Gabriel Valley.  intervals (CIs) for total OC that usually crossed zero (Figure 3).
Associations were generally stron ger in phase 2 than in phase 1 for IL6, Cu,ZnSOD, and sPselectin ( Figure 4). The panels were marginally different across phases, because subjects were followed in two phases with few exceptions (5 of 60 subjects). In models restricted to 55 subjects with data in both phases, phase differences were nearly unchanged (data not shown).
Furthermore, associations were stron ger among subjects not taking statins for sTNFRII and stronger among subjects not taking clopidogrel for sPselectin ( Figure 5). There were few consistent pollutant inter actions with ACE/ARB.
As previously reported and discussed for year 1 data  but not presented here, regression coefficients for ozone had opposite signs compared with other pollutants [see Supplemental Material (doi:10.1289/ehp.0800194.S1)] and were completely confounded by markers of primary combustion (EC, BC, OC pri , CO, NO x ) with which O 3 was inversely correlated (Table 4).

Discussion
Our results are largely consistent with recent repeatedmeasures studies showing associa tions between ambient air pollution and bio markers of systemic inflammation in healthy young adults (Chuang et al. 2007) and sus ceptible subjects with CAD (Dubowsky et al. 2006;Rückerl et al. 2006Rückerl et al. , 2007a2007b;Yue et al. 2007). We extended these previous find ings with data from intensive home exposure assessments and modeling that provided clues to potentially causal pollutant components. This is also the first study to show adverse effects of air pollutants on erythrocyte anti oxidant enzymes.
We focused on elucidating the role of pol lutants closely associated with traffic, including EC and PM 0.25 . This was accomplished with extensive measurements of exposures in the immediate outdoor community microenviron ments of subjects, including sizefractionated PM, OC fractions, and measurements across seasons that helped us characterize differences in response potentially due to particle size distribution and chemical composition. The approach likely enabled us to detect stronger associations with PM 0.25 than PM 0.25-2.5 (Figure 2). Such differences in these two parti clesize fractions of regulated fine PM (PM 2.5 ) have not been as clearly demonstrated in pre vious panel studies that have relied on central site data. This finding may be attributable to the higher deposition fraction of the unmeas ured UFP fraction (PM 0.1 ) of PM 0.25 than accumulationmode particles and the ability of UFP to translocate systemically to poten tially induce oxidative stress and inflammation (Elder and Oberdörster 2006;Li et al. 2003).
The approach also enabled us to demon strate for the first time that associations of IL6, sPselectin, and SOD with PM mark ers of primary combustion (EC, OC pri ), PN, and PM 0.25 were stronger in a cooler 6week period (phase 2) than a warmer 6week period (phase 1) (Figure 4). We did not find any positive associations with SOC (Figure 3), with most regression coefficients being nega tive and nonsignificant at p < 0.05. OC pri and PN concentrations were higher in cooler months (typically characterized by air stag nation and lower secondary particle forma tion). Interestingly, concentrations of other Figure 3. Associations of biomarkers with indoor and outdoor OC on days 1, 3, and 5: differences by OC pri and SOC fractions. Expected change in the biomarker (adjusted coefficient and 95% CI) corresponds to an interquartile range change in the air pollutant (Table 3). Models for sTNF-RII and sP-selectin are restricted to 44 subjects living in the San Gabriel Valley. Total OC OC pri SOC Total OC OC pri SOC Total OC OC pri SOC Total OC OC pri SOC Total OC OC pri SOC 3 5 1 3 5 1 3 5 1 3 5 1 3 5 1 3 5 1 3 5 1 3 5 1 3 5 1 3 5 1 3 5 1 3 5 1 3 5 Moving averages of air pollutants over the previous 5 days Figure 4. Associations of biomarkers with outdoor air pollutants on days 1, 5, and 9: differences by phase of study. (A) IL-6 (pg/mL). (B) sP-selectin (ng/mL). (C) Cu,Zn-SOD (U/g Hb). Expected change in the biomarker (adjusted coefficient and 95% CI) corresponds to an interquartile range change in the air pollutant (Table 3). Phase 1 is a warmer period of greater photochemical activity, and phase 2 is a cooler period of greater air stagnation. Results for sP-selectin are for 44 subjects living San Gabriel Valley. All data Phase 1 Phase 2

Moving averages of air pollutants over the previous 9 days
Moving averages of air pollutants over the previous 9 days Moving averages of air pollutants over the previous 9 days pollutants also more strongly associated with biomarkers in the cooler phase (EC, BC, and PM 0.25 ) were not higher, suggesting that dif ferences in particle composition or size dis tribution were important, perhaps as better reflected by OC pri and PN, respectively. This is an important finding because particle mass alone does not provide sufficient information about composition or sources. It is conceiv able, for example, that our findings for PM 0.25 and PN are attributable partly to nanoparti cles. It has been shown that particles 6-12 nm were much higher in the winter than in the summer near a Los Angeles freeway, but larger particles 50-100 nm showed the opposite trend (Zhu et al. 2004). The potential impor tance of trafficrelated particles is supported by stronger positive associations for sTNF RII and sPselectin in the San Gabriel Valley communities closer to traffic sources than in the Riverside community. Associations were also generally stronger for OC pri (Figure 3) than for PM 0.25 (Figure 2). OC pri and related EC are mostly associated with UFP (Li et al. 2003). Because PM 0.25 includes some accu mulationmode particles, it likely represents both fresh and aged trafficrelated particles. Based on these results, both primary organic components and quasiultrafine or smaller size fractions appear to be important. Semivolatile organic components associated with particles may also have been important, given that biomarker associations were similarly robust for the correlated gases CO and NO x [see Supplemental Material (doi:10.1289/ ehp.0800194.S1)]. This included generally stronger associations with gases in phase 2 than in phase 1 when NO x concentrations were lower. These gases were unlikely causal at the observed low concentrations (Devlin et al. 1999;Thom et al. 2006) but instead served as markers for other traffic emission components.
Other findings support the hypothesis that effects of air pollution on cardiovascular health are secondary to proinflammatory properties of redox active and other pollutant compo nents . Associations for sTNFRII were stronger in subjects not taking statins, which have antiinflammatory proper ties. These findings are consistent with reports of weaker associations between air pollutants and CRP among statin users in two other panel studies of susceptible elderly subjects (Dubowsky et al. 2006;Rückerl et al. 2006).
Our new finding for sPselectin is con sistent with a panel study showing an asso ciation of ambient UFP with another platelet activation marker (soluble CD40 ligand) in people with CAD (Rückerl et al. 2007b). Our study is the first to show a protective effect of clopidogrel. This finding supports the plausi bility of a pollutant effect on platelet activa tion, because this medication blocks platelet aggregation and is associated with decreased sPselectin (Xiao and Théroux 2004). Our findings are relevant to the potential for air pollution to affect CAD, because sPselectin activates both leukocytes and endothelial cells and induces adhesion of leukocytes to plate lets and to endothelial cells (Jurk and Kehrel 2005). Therefore, if air pollutants acutely activate platelets as suggested by our finding, this could increase the risk of a potentially fatal thrombotic event in the coronary arteries ( Figure 1). Platelet selectin is also critical to the development of neointimal formation after arterial injury (Wang et al. 2005). Potentially relevant findings by two epidemiologic studies are evidence of increased risks of athero sclerosis development with exposure to trafficrelated air pollution near the home (Hoffmann et al. 2007;Künzli et al. 2005).
Other novel findings are the inverse asso ciations between air pollutants and two anti oxidant enzymes (GPx1 and Cu,ZnSOD). Experimental results show that urban UFP can induce a positive antioxidant response represented by hemoxygenase1 in epithelial and macrophage cell cultures (Li et al. 2003). However, erythrocytes do not have nuclei and thus have a relatively fixed amount of antioxidant enzymes after maturation from reticulocytes. The findings for GPx1 and Cu,ZnSOD in most of the elderly subjects studied suggest enzyme inactivation within erythrocytes by pollutant components or PM 0.25 . There is experimental evidence to support this hypothesis (Hatzis et al. 2006;Pigeolet et al. 1990;Shinyashiki et al. 2008) as well as evidence showing that quasiUFP ≤ 0.2 µm in diameter and nanoparticles, but not larger particles, freely enter the erythrocyte (RothenRutishauser et al. 2006). This may be a clue to an important pathway involving primarily UFP (Elder and Oberdörster 2006) and related organic and inorganic compo nents that may enter the circulation to then target erythrocytes as well as other cells.
Erythrocytes are critical in protecting the body against oxidative stress (Tsantes et al. 2006). Therefore, it is conceivable that erythrocyte antioxidant enzyme inactivation is partly responsible for pollutantrelated increase in biomarkers of inflammation and thrombosis. This is supported by our finding of withinsubject inverse associations of IL6 with GPx1, and sPselectin with Cu,ZnSOD in mixed models. For an interquartile range decrease in GPx1 of 10.4 U/g hemoglobin, IL6 increased 0.25 pg/mL (95% CI, −0.03 to 0.53) or 10% of mean IL6. Similarly, for an interquartile range decrease in SOD of 2,026 U/g hemoglobin, sPselectin increased 5.8 ng/mL (95% CI, 3.3 to 8.3), or 13% of mean sPselectin. Biomarkers of inflamma tion were generally positively associated with each other, and GPx1 was positively asso ciated with Cu,ZnSOD (data not shown). Furthermore, erythrocyte antioxidant enzyme inactivation may modulate the putative effects of air pollutants on endothelial dysfunc tion. Erythrocytes have been shown to pro tect cultured endothelial cells against oxidant damage. Inhibitors of either the erythrocyte glutathione system or membrane transport of superoxide into erythrocytes significantly reduced this protection (Richards et al. 1998).
A small subset of subjects showed positive GPx1 and Cu,ZnSOD associations with air pollutants. Compared with the 53 nega tive responders, these seven subjects showed no notable differences in the distributions of characteristics listed in Table 1 except that none took clopidogrel and only one had a history of myocardial infarction. We spec ulate that these might be healthier subjects because of their ability to increase antioxi dant enzyme activity, perhaps by mounting a rapid bone marrow response to PM exposure, as suggested in several experimental stud ies (Goto et al. 2004;Mukae et al. 2001), including increased reticulocytes (Rivero et al. 2005). In mature erythrocytes, activities of Cu,ZnSOD and GPx1 are steadily eroded Figure 5. Associations of biomarkers with outdoor air pollutants on days 1, 3, 5, and 9: differences by medication use among 44 subjects living in the San Gabriel Valley. (A) sTNF-RII (pg/mL). (B) sP-selectin (ng/mL). Expected change in the biomarker (adjusted coefficient and 95% CI) corresponds to an interquartile range change in the air pollutant (Table 3). sTNF-RII (pg/mL) sP-selectin (ng/mL) 1 5 9 15 PN EC PM 0.25 9 15 9 1 5 3 1 5 9 15 9 1 5 9 1 5 3 1 5 3 1 5 9 15 PN EC PM 0.25 9 15 9 1 5 3 1 5 9 15 9 1 5 9 1 5 3 1 5 3 Consequently individuals with robust erythro poietic activity and greater proportion of newly released erythrocytes would be expected to have higher erythrocyte Cu,ZnSOD and GPx1 activities than sicker individuals. Study limitations include the following: Despite the biological plausibility, effect mod ification by medication use could have been secondary to other unmeasured characteristics of subjects. The home exposure data may be subject to exposure error because of differ ences with personal exposure. However, sub jects stayed at home 88% of the time (from diary data). Although we believe that the present exposure data represent key sources and components, we cannot link exposure to specific sources, nor can we identify specific component classes such as polycyclic aromatic hydrocarbons as being responsible for asso ciations. Nevertheless, the major source of fossil fuel emissions in the LA basin is motor vehicle exhaust, and because EC, BC, OC pri , CO, and NO x are linked to these emissions, our data suggest that vehicular pollutants are behind the reported associations.

Conclusion
Our results suggest that pollutant compo nents linked to emission sources of primary PM 2.5 OC, quasiUFP (PM 0.25 ), and PN con centrations are associated with increased sys temic inflammation, platelet activation, and decreased circulating erythrocyte antioxidant enzyme activity in elderly people with CAD. Inactivation of antioxidant enzymes may be one mechanism of air pollutant-related increases in systemic inflammation. These effects may be partly behind reported morbidity and mortality associations with ambient PM 2.5 mass concen trations (Pope and Dockery 2006). Stronger associations during the cooler phase of study, despite similar PM 0.25 mass concentrations in cooler and warmer phases, further support the view that the greatest impacts on systemic responses may be attributable to nanoparticles not adequately represented by the present par ticle mass measurements as well as to unmea sured toxic air pollutants that increase near ground level in the winter. Our related experi mental work using particles collected in the LA air basin at the Southern California Particle Center suggests that this might include redox active and electrophilic organic components of traffic exhaust particles in the ultrafine range (Araujo et al. 2008;Gong et al. 2007;Li et al. 2003;Ntziachristos et al. 2007;Shinyashiki et al. 2008).