Association of cardiac and vascular changes with ambient PM_2.5 in diabetic individuals

Study population

Twenty-two volunteers, aged 48-78, with type 2 diabetes were identified through the University of North Carolina (UNC) Diabetes Clinic and through newspaper advertisements. Details on the inclusion and exclusion criteria have been previously published [17]. Potential participants had to have a diagnosis of type 2 diabetes, but without taking insulin; and be on a stable medication regimen throughout their participation. Volunteers were excluded if they had a recent vascular event or intervention such as coronary artery graft bypass surgery or percutaneous coronary intervention. Participants were asked to refrain from vigorous exercise on study-mornings and to refrain from medications like anti-oxidants (vitamins C and E), fish oil, niacin, arginine, over-the-counter vasoactive agents such as decongestants, anti-inflammatory agents such as ibuprofen, naproxen or aspirin unless it was prescribed as a daily medication (in which case it was continued) for the week prior to the study as well as the week of the study. Participants were also asked to refrain from use of phosphodiesterase enzyme inhibitors during the week of the study.

Each participant visited the U.S. Environmental Protection Agency's National Health Effects Environmental Laboratory, Environmental Public Health Division (EPHD) in Chapel Hill, NC, on five consecutive weekdays between November 2004 and December 2005. Data on health status, pulmonary and cardiac symptoms, medication, and smoking history were obtained at baseline and during four follow-up visits. Altogether a maximum of 88 observations were available for analysis, depending on the examined health outcome. All volunteers signed a written consent form and the study protocol was approved by the UNC Human Studies Biomedical Institutional Review Board as well as by the EPA.

Clinical measurements

Air pollution monitoring

Concentrations of fine ambient PM (average from 9 AM - 9 AM) were measured on each clinic day and for the four prior days with a 3000K Versatile Air Pollution Sampler (URG Corp., Chapel Hill, NC) [21] located on the EPHD rooftop approximately 30 meters above ground level. In addition, daily 24-hour concentrations (midnight to midnight) of fine ambient PM_2.5 mass were obtained from a monitoring station located approximately 44 kilometers (27 miles) east of the EPHD and operated by the State of North Carolina. The Spearman correlation between both PM_2.5 measurements was 0.85 across the 14 month study period and the network data was used to impute missing rooftop data based on a linear regression model. Of thirteen daily averages relevant for the health outcome analysis, eleven could be imputed and only two remained missing.

Continuous 2-minute measurements of air temperature, barometric pressure and relative humidity were obtained from the EPHD rooftop and 24-hour averages were calculated. Rooftop data was complete for the health outcome analysis, so no imputation was necessary.

Statistical analyses

The study was conducted as a panel study with four (three for 24-hour ECGs) repeated measurements per participant. Thus every person acted as his/her own control which limited the need for an adjustment for patient characteristics in the analysis.

Data were analyzed using the statistical package SAS Version 9.1 (SAS Institute, Inc., Cary, NC). For the analysis of the PM effect, additive mixed models with a random participant effect and "compound symmetry" covariance structure were used [22, 23]. Model building to identify meteorological and temporal determinants, as potential confounders, was done for each outcome variable separately. Model fit was assessed by the Akaike Information Criterion (AIC). Details on the confounder selection process can be found in Schneider et al. (2008) [17].

The exposure, in this case rooftop PM_2.5 mass, was considered as an immediate (lag 0) or a delayed (lag 1 to lag 4) linear effect. Effect estimates are presented as percentage changes of the mean outcome variable together with 95%-confidence intervals (CI) for a 10 μg/m³ increment in ambient PM_2.5 mass.

Sensitivity analyses were done by changing the covariance structure to first order autocorrelation as well as by repeating parts of the analysis with PM_2.5 values that were not imputed. In addition, analyses were repeated using network PM_2.5 data imputed with rooftop PM_2.5 data as exposure. Individuals that had experienced PM_2.5 values exceeding 35 μg/m³ during their relevant days of exposure were excluded to examine if health effects occurred also below current 24-hour U.S. National Ambient Air Quality Standard (NAAQS). Moreover, to assess the heterogeneity across study participants in this repeated measures study, we examined patient-specific, random slope-models for selected PM_2.5 results.

Participant characteristics

Air pollutant and meteorology measurements

The 24-hour PM_2.5 values over the study period were generally below the U.S. NAAQS of 35 μg/m³ and the mean over the study period was below annual average NAAQS of 15 μg/m³ (see Additional File 1: Table S1). The median of the within-individual PM_2.5 ranges was 7.9 μg/m³. Correlations between PM_2.5 and meteorology parameters as well as the time-course of PM_2.5 and air temperature during the study period can be found in Schneider et al. (2008) [17].

Description of health outcomes

A description of the patient means for the analyzed blood and ECG markers can be found in the Additional File 1: Tables S2-4. The majority of the blood parameters exhibited only low to moderate correlations. The medians of the patient-specific Spearman correlation coefficients within repolarization as well as HRV parameters of the 5-minute ECG were also on average low to moderate. HRV parameters of the 24-hour ECG correlated moderately to highly. As expected, BPsys and BPdia showed a high correlation. Correlation tables for all health outcomes can be found in the additional file (see Additional File 1: Tables S5-9).

Association of PM2.5 with health outcomes

Blood parameters

The pro-inflammatory cytokines IL-6 and TNFα showed a strong increment in association with an increment in PM_2.5 with a lag of two days, which persisted for one more day and then fell back to null (Figure 1). The acute-phase reactants CRP and fibrinogen did not exhibit as strong a response, but did increase slightly, however non-significantly, with a lag of two days. Albumin, a negative acute-phase protein, showed a decrease with a lag of one and two days. No associations with PM_2.5 were found with blood parameters representing endothelial function and platelet activation (vWf, sICAM-1, sVCAM-1 and E-selectin) (data not shown). Markers of the coagulation pathway also showed slight, non-significant changes in association with PM_2.5. In addition to the increase of fibrinogen with a lag of two days, protein C also increased slightly with the same time lag whereas factor IX remained unchanged. In contrast, factor VII showed a slight decrease after one day and also fibrinogen showed a decreasing tendency with a lag of 3 days (Figure 1). The fibrinolysis pathway marker PAI-1 increased with a lag of one day (Figure 1). Plasminogen, D-dimer (a fibrin degeneration product) and t-PA were not associated with PM_2.5(data not shown).

Decrements in total cholesterol (TC), high density lipoprotein (HDL) and low-density lipoprotein (LDL) were associated with PM exposure with a lag of three to four days (Figure 2). Red blood cells (RBC) showed a decrement with a lag of one to three days. Consistent with the RBC associations, mean corpuscular volume (MCV) and total hemoglobin also had decrements with a similar lag (Figure 2). Circulating eosinophils (absolute number and % of WBC) were positively associated with PM_2.5 with a lag of two days, followed by a decrement at lag 3 (Figure 2). None of the other blood panel markers showed meaningful associations with PM_2.5 (data not shown).

ECG parameters and blood pressure

Analysis of ECG markers included an interaction term for the intake of beta-blockers, since it is known that they can alter ECG response to air pollutants. In the 5-minute resting ECGs, we did not find any association between heart rate or frequency domain parameters of HRV PM_2.5. The 24-hour ECG time domain parameters showed an association with PM_2.5; pNN50 increased with a lag of 2 and 3 days, and RMSSD increased with a lag of 3 days. SDNN showed a slight immediate increase for the non-beta-blocker subgroup (Figure 3).

QTc, the QT interval corrected for heart rate, decreased in association with air pollution at lag 0; and T wave complexity, a still relatively new marker to study ventricular repolarization, increased with a lag of 1 and 2 days for participants without beta-blocker intake (Figure 3). T wave amplitude, a marker for the morphology of the T wave, showed a strong association with PM_2.5 in individuals not on beta-blockers: a decrease starting at lag 0 and persisting for lag 1 and 2 and ending with an increase at lag 4 (data not shown). Exclusion of one patient (female, Caucasian, 63 years, BMI = 20 kg/m², taking diuretics and estrogens) who showed a very strong association with PM_2.5 at lag 0 in the random slopes analysis, resulted in a loss in the lag 0-association, but led to much stronger effects for lag 1 and lag 2 and a weaker increase for lag 4 (Figure 3). The other repolarization markers (QT norm, QTVI, variability of T wave complexity) did not show any association with air pollution.

BPsys showed an immediate decrease at lag 0 (-2.7% [-5.5%; 0.1%]) and together with BPdia an increase with a delay of four days (BPsys: 2.6% [0.0%; 5.3%]; BPdia: 2.6% [0.2%; 5.0%]) in association with ambient PM_2.5.

Effect modification

Interaction analysis for the inflammatory cytokine IL-6 revealed that stronger associations were observed in the subgroup with higher BMI, higher HbA1c value, and lower adiponectin levels, all markers associated with insulin resistance. Additionally, stronger associations were observed in those individuals with higher blood ferritin levels as well those carrying the GSTM1 deletion on both alleles (Figure 4). However, associations between TNFα and PM_2.5 were strongest in the subgroup with lower BMI, lower HbA1c value, higher adiponectin level, and with at least one functional GSTM1 allele. Both cytokines showed a similar effect modification by ferritin (see Additional File 1: Figure S1). The decrease in factor VII was found mainly in the subgroup of participants with high BMI and high HbA1c (data not shown). Decreases in albumin were mainly found in individuals with higher BMI, higher HbA1c and lower adiponectin level (see Additional File 1: Figure S2). The interaction analysis with aspirin revealed only for protein C that the lag 2 response was much more pronounced for individuals without aspirin intake. Associations between PM_2.5 and RBC or total hemoglobin were strongest in the subgroup with high BMI, high HBA1c, low adiponectin and the null GSTM1 genotype, as was seen for IL-6 (Figure 5 and Additional File 1: Figure S3). Eosinophils (absolute number count as well as % of WBC) increased slightly more at lag 2 in the subgroup of elevated BMI and HbA1c (data not shown). In contrast to our previously published work [17], interaction analysis with MPO levels did not show any notable effect modifications. No appreciable effect modification by age or by statins intake was found.

Since the models for the ECG marker analysis included an interaction term for the intake of beta-blockers, further effect modification analysis was not possible due to small sample size.

Sensitivity analyses

Repeating the analysis without imputed PM values showed that the effect estimates didn't change appreciably. Health effects were observed below the current U.S. NAAQS 24-hour standard for PM_2.5 as reducing the dataset to individuals that never experienced a day above 35 μg/m³ during our relevant exposure period did not change the estimated effects. Changing the analyzed exposure from PM_2.5 values measured on the roof of the EPHD in Chapel Hill, NC, to values obtained from fixed site community monitors resulted in similar effect estimates for the different health outcomes. The estimated effects were not sensitive to changing the covariance matrix from compound symmetry to first-order autocorrelation. Random slopes showed consistent PM_2.5 associations across all study participants for IL-6 and TNFα (see Additional File 1: Figure S4), QTc, T wave amplitude, and T wave complexity (data not shown).

Inflammation and coagulation blood parameters

One of the more robust findings of this study is the association between PM and increases in TNFα and IL-6 levels in the blood. IL-6 acts as both a pro- and anti-inflammatory cytokine and plays a role in the acute phase response. TNFα plays a key role in the initiation of systemic inflammation and regulation of immune cells. The fact that both of these markers of inflammation are elevated in association with PM concentrations, as well as blood eosinophils, CRP, and fibrinogen (the latter both not significantly) indicates that ambient levels of PM are capable of inducing an inflammatory response in the blood.

Dubowsky et al.(2006) [11] also reported PM-associated increases in IL-6 and CRP in individuals with diabetes, with significance being achieved with a 3-4 day moving average of PM. Interestingly, in our study inflammatory markers were observed with a lag of two days. The effect size of the slight, but non-significant increases in CRP and fibrinogen with lag 2 in our study is in agreement with other air pollution epidemiological studies [14, 15]. Our albumin results should be interpreted cautiously as albumin is often used as a measure of blood density and is an abundant blood protein that normally does not vary much. However, it can also be interpreted as an anti-acute phase protein and therefore the decreases at lag 1 and lag 2 fit well within the blood marker interplay of an inflammatory reaction. The strong delayed increase in eosinophils, an inflammatory leukocyte cell type, adds to this picture. However, the small increase in eosinophils in our non-asthmatic individuals is of uncertain significance regarding the inflammatory status and may also be related to an allergic or Th2 immune response governed by interleukin-5.

Erythrocytes

We observed a decrease in RBC with a lag of 1-3 days, as well as decreased levels of hemoglobin and MCV. These findings are consistent with those of Seaton et al. (1999) [24] and Rückerl et al. (2007) [25], who saw significant decreases in both parameters in a similar time-frame. They and others have speculated that PM may cause constriction of small blood vessels or alter the adhesive properties of RBC, thus increasing the time these cells spend in systemic capillaries and not in larger vessels. Since increased oxidative stress is known to decrease the life span of RBC, it is also possible that PM-associated increases in oxidative stress which have been reported by many, could also result in fewer RBCs or could, perhaps, injure these sensitive cells causing leakage of hemoglobin and decreased volume.

Blood lipids

In this study we observed an early decrease in HDL followed by decreases in HDL, LDL, and total cholesterol with a lag of 3-4 days. This is in contrast to an earlier study in which we observed increased levels of triglycerides and LDL associated with exposure of asthmatics to coarse PM [26]. However, it is in partial agreement with a study in which we reported decreased HDL/LDL ratio, VLDL, and triglycerides in healthy volunteers exposed to ultrafine particles [19]. One explanation for a PM-associated decrease in blood lipids is that PM-induced oxidative stress might modify blood lipids, making them more prone to adhere to vascular walls and removing them from the circulation. Kunitomo et al. (2009) [27] reported that exposure of mice to tobacco smoke resulted in oxidatively modified LDL. Araujo et al. [28] reported that exposure of mice to ultrafine and fine particles resulted in oxidized blood lipids as well as an inhibition of the anti-inflammatory capacity of HDL. Soares et al. (2009) [29] also reported that in mice exposed to urban air pollution LDL was oxidized and the lipid content of arterial plaques was increased. Another possible explanation for the observed lipid decreases is that they are related to changes known to occur during acute inflammation and the acute phase response. According to a review by Khovidhunkit et al. (2004) [30] infection and inflammation lead either to no change or to a decrease in serum cholesterol, LDL as well as HDL levels during the acute-phase response in humans.

Other studies have reported PM-associated changes in blood lipids. Suwa et al. (2002) [31] demonstrated in Watanabe heritable hyperlipidemic rabbits that exposure to PM₁₀ increased the total amount of lipids in aortic lesions. Sun et al. (2005) [32] showed in a murine model that the lipid content in the aortic arch increased 1.5-fold in mice fed a high-fat chow diet and exposed to PM versus filtered air. In an epidemiologic case-control study of traffic-exposed police officers, the average values of HDL cholesterol and triglycerides were elevated in the exposed group versus an unexposed control group [33].

ECG parameters and blood pressure

In our study, we found a slight immediate increase in SDNN and delayed positive associations between RMSSD, pNN50 and air pollution exposure, with the latter suggesting increased efferent vagal activity. Although many studies report a decrease in HRV associated with increasing PM concentration [7, 8, 34–38] and an exacerbation of these adverse PM effects in people with insulin resistance or diabetes [39, 40], positive associations between PM and HRV have also been reported. An occupational study of NC State Highway Patrol troopers reported positive associations between PM and heart beat cycle length, HRV and ectopic beats on the next morning after their shift [41]. Wheeler et al. (2006) [42] reported a positive relationship between SDNN and PM_2.5 in patients with pulmonary disease, whereas an inverse association was observed in patients with cardiovascular disease. Moreover, healthy [19] and asthmatic [43] volunteers exposed to concentrated ambient particles (CAPs) showed an increase in HF and LF approximately 18 hours after exposure.

In contrast to HRV, repolarization parameters have only recently been investigated in air pollution studies. In our study we observed an immediate negative association between PM_2.5 exposure and QTc, indicative of vagal stimulation [44]. In addition, we observed changes in T wave amplitude. Decrements in T wave amplitude corresponded with increments in T wave complexity and vice versa. These findings are in agreement with other studies. In a study of male coronary artery disease patients, Henneberger et al. (2005) [45] reported an increase of QTc associated with organic carbon as well as a decrease of T wave amplitude and an increase of T wave complexity associated with ultrafine particles (UFP) and PM_2.5. Frampton et al. (2004) [46] and Samet et al. (2009) [19] reported reduced QT intervals in healthy individuals breathing air and UFP with intermittent exercise. Zareba et al. (2009) [47] demonstrated small changes in QT interval, T wave amplitude and variability of T wave complexity in subjects exposed to UFP. We observed both early and late changes in T wave amplitude. It is possible that changes at earlier lags may be induced directly by changes via the autonomic nervous system or via changes in endothelial function (which was observed in our previous work [17] in the same study population), and changes at later lags may be triggered by inflammatory reactions.

Recent epidemiological and controlled exposure studies have reported small increases in BPsys and BPdia in association with PM at the same day as well as during a 5-day-average up to a 30-day-average [48, 49]. The diabetic patients in our study showed an immediate decrease in BPsys and then a much delayed increase of BPsys and BPdia at a lag of 4 days. However, Ibald-Mulli et al. (2004) [50] also found small significant decreases in systolic as well as diastolic blood pressure in association with PM. We have not observed changes in blood pressure in studies in which volunteers were exposed to CAPs [51] and Brook et al. (2002) [52] did not observe changes in blood pressure in healthy subjects in a controlled exposure study with inhalation of fine CAPs and ozone. However, Urch et al. [53] reported an early increase in BPdia after controlled exposure to CAPs and ozone with the magnitude of the BPdia change being associated with the PM_2.5 carbon content.

Effect modification

Studies [11, 54, 55] have shown that patient characteristics such as high BMI or diseases like diabetes or hypertension increase the associations between PM and health outcomes. Obesity represents a pro-inflammatory state and as such may increase susceptibility to air pollution by increasing the response to inflammatory stimuli. We previously reported that markers of insulin resistance such as obesity and poor glycemic control as indicated by elevated HbA1c and low adiponectin were associated with a larger PM-associated decrease in endothelial dysfunction [17]. Here we show a similar finding for IL-6, RBC and total hemoglobin. Somewhat surprisingly, we observed that, while these patient characteristics were associated with increased PM-associated levels of IL-6, TNFα was increased more in those patients with lower BMI and better insulin control. Although both of these cytokines are associated with inflammation, they do not always travel together. They can be secreted by different cells under different control mechanisms and for different purposes. Blood levels of IL-6, but not TNFα have also been reported to be an independent predictor of type 2 diabetes after adjustment for BMI [56].

We observed that individuals with elevated levels of blood ferritin had increased levels of IL-6 and TNFα associated with PM exposure. Serum ferritin is an iron storage protein and considered a surrogate marker of total body iron levels. Higher levels of iron, often found in diabetic individuals, are hypothesized to make one more prone to inflammation and also to increase oxidative stress [57]. Excessive absorption and storage of dietary iron is also thought to contribute to the pathogenesis of insulin resistance, type 2 diabetes and its complications [58, 59]. Ghio et al. (2008) [60] reported that inhaled tobacco smoke particles were able to mobilize endogenous iron present in lung lining fluid, resulting in altered iron homeostasis, increased metal-dependent oxidative stress, and ultimately increased levels of pulmonary inflammation. It is possible that patients with higher ferritin-levels would have more iron available for complexing with any particles that enter the vascular system and would thus have more oxidative stress leading to higher levels of inflammatory markers. Consistent with this hypothesis, in our study individuals with ferritin levels above 60 ng/ml showed much stronger increases in IL-6 and TNFα than those with lower ferritin levels.

Glutathione-S-transferase is a phase II enzyme which can scavenge oxygen-free radicals, metabolize reactive oxygen species and detoxify xenobiotics present in PM. People with a GSTM1 null allele are thought to handle oxidative stress less effectively and may be more responsive to agents such as PM which increase oxidative stress. The potential of this genotype to make carriers more susceptible to PM exposure has already been shown repeatedly [55, 61–64]. We reported earlier that individuals with diabetes with the null GSTM1 polymorphism have the greatest PM-associated decreases in endothelial dysfunction. Here, we extend these findings by showing that individuals with diabetes carrying the null allele for GSTM1 also have the most pronounced changes in IL-6, RBC, total hemoglobin and MCV.

Strengths and limitations

Excellent adherence to the protocol by the volunteers who attended all clinical visits is one of the study's strengths. Yet, the design with four successive visits in one week might not be optimal to detect changes in biomarkers having longer half-lives, such as fibrinogen. However, using first order autocorrelation as a covariance matrix to account for that autocorrelation between the data from consecutive days did not show strong influence on the effect estimation. Daily variations could have an influence on the health outcomes but models were adjusted for day of the week if AIC proved it to be necessary. Circadian variation was controlled by design, as measurements on every individual were taken at the same time of day.

A limitation of the study is the small number of individuals studied. While the numbers are adequate to observe associations between PM and pro-inflammatory cytokines, there was not enough power to exclude significant changes in other biomarkers related to inflammation and coagulation where only trends appeared. A major strength of the study was the simultaneous measurement of interacting ECG and blood markers, thereby establishing the temporal relationship of changes in physiological and biochemical responses to PM_2.5 exposure, and providing insight into the potential mechanisms of air pollution effects. However, coagulation factors were measured as antigen levels, which don't necessarily reflect functional levels. As with any study that examines many different end points, multiple testing and as a consequence the potential for false-positive results can be a problem, especially with so many effect modifiers.

In this study we used daily PM_2.5 mass values from a monitoring station located on the rooftop of the EPHD. In addition, PM_2.5 values from a state monitoring station located 27 miles away from the EPHD were obtained. There was a strong correlation between these two sets of measurements, suggesting that we were accurately assessing ambient exposure of the study participants all living in a 30 mile radius of Chapel Hill. Thus, while ambient exposures are measured with some imprecision, these exposure measurement errors are non-differential with respect to either true ambient exposure or to the clinical measurements. As such, these Berkeson measurement errors will tend to bias our results towards the null. However, we have to acknowledge the possibility that the reported effect estimates might be modified by co-pollutants that were not measured. A further strength of this study is the investigation of air pollution effects in a particularly susceptible subgroup. On the other hand, the results cannot be generalized to the whole population.