Particulate Air Pollution, Oxidative Stress Genes, and Heart Rate Variability in an Elderly Cohort

1Particulate air pollution (PM) is associated with increased risk of hospitalization and death from cardiovascular disease, but the mechanisms underlying such effects are not fully understood. Reductions in heart rate variability (HRV), a noninvasive measure that independently predicts cardiovascular mortality, have been related to PM exposure, particularly to fine-particulate air pollution of < 2.5 μm in aerodynamic diameter (PM2.5).

2Animal experiments indicate that reactive oxygen species (ROS), which have established relevance in the pathogenesis of cardiovascular disease, are potential mediators for particle effects on HRV and other cardiovascular end points. While animal models can identify potential mechanisms of particle effects, the relative importance of these pathways in humans at lower doses is not clear and may be determined by examining subjects with genetically determined differences in oxidative-stress defenses. In elderly subjects living in the Boston, Massachusetts, metropolitan area, we recently showed that PM2.5 levels during the 48 hr before the study were associated with decreased HRV in individuals with the glutathione S-transferase M1 (GSTM1) deletion, but had no effect in subjects with GSTM1 present.

3Studies showing the effect of just one polymorphism are unlikely to correctly represent the complex etiology of common diseases, and failure to account for gene–gene interactions in the search for susceptibility genes has been widely suggested to explain the persisting difficulties in replicating significant findings. Particle exposure induces both heme oxygenase-1 (HMOX-1) and GSTM1 expression through activation of the genetic antioxidant response element (ARE). A high number of microsatellite (GT)n dinucleotide repeats in 5′-flanking region may reduce HMOX-1 inducibility by ROS and has been associated with increased risk of coronary artery disease in high-risk groups with hyperlipidemia, diabetes, or current smoking. Consequently, individuals with a high number of (GT)n repeats may be more susceptible to the effects of airborne particles.

4We hypothesized that the gene encoding HMOX-1, which is involved in various aspects of responses against oxidative stress, may a) directly modify the effect of ambient PM on HRV and b) interact with GSTM1 to determine which subjects are susceptible to airborne particle effects. To establish the role of the antioxidant response pathway in determining the cardiovascular effects of airborne particles, we examined in the present study the association of PM2.5 with HRV in a repeated measure study of elderly subjects from the Boston metropolitan area, and evaluated how that association was affected by genetic variation in the HMOX-1 and GSTM1 loci.

5Our study population consisted of 539 white males from the Normative Aging Study (NAS), a longitudinal study of aging established in 1963 by the U.S. Veterans Administration. Between January 2000 and June 2005, all participants still presenting for examination (n = 676) were evaluated for HRV. Of these, 137 subjects were excluded because of heart arrhythmias, measurement time < 3.5 min, or missing potential confounding variables or HMOX-1 data. Among the remaining 539 subjects, GSTM1 data were available for 476 subjects, who had one (n = 314) or two (n = 162) HRV measurements. In subjects with multiple HRV measurements, the time interval between measurements was approximately 3 years. This study was conducted in compliance with all applicable requirements of the U.S. and international regulations (including institutional review board approval). All subjects gave written informed consent prior to the study.

6HRV was measured at rest during normal breathing for 7 min using a two-channel (five-lead) ECG monitor (Trillium 3000; Forest Medical, East Syracuse, NY) while the subject was seated. Standard deviation of normal-to-normal intervals (SDNN), high frequency (HF) (0.15–0.4 Hz), and low frequency (LF) (0.04 –0.15 Hz) were computed with a fast Fourier transform using software (Trillium 3000 PC Companion Software; Forest Medical) complying with established guidelines. In the analysis, we used the 4 consecutive minutes of ECG reading that included the lowest number of artifacts.

7Continuous PM2.5 was measured at a stationary monitoring site on the roof of Countway Library of Harvard University in downtown Boston using a Tapered Element Oscillating Microbalance (TEOM; Model 1400A, Rupprecht & Patashnick Co., East Greenbush, NY). Meteorologic data was obtained from the Boston airport weather station. The 48-hr moving average of PM2.5 before each HRV measurement was used as the exposure index, as this exposure period has shown the strongest association in previous studies.

8The GSTM1 locus (UniGene Hs.301961; ) was amplified at exons 4 and 5 by polymerase chain reaction (PCR) as previously described to differentiate between the null polymorphism and the presence of one or more copies of the gene. The HMOX-1 (UniGene Hs.517581; ) microsatellite (GT)n length assay was designed per Yamada and coworkers. Briefly, the HMOX-1 locus was amplified by PCR at the 5′ promoter flanking region containing (GT)n repeats with primers as described by Yamada, and the sizes of the PCR products were analyzed with a laser-based automated DNA sequencer (AB 3100; Applied Biosystems, Foster City, CA). Although the exact cutoff for HMOX-1 modulation is still unknown, constructs with lengths of > 25 repeats showed reduced HMOX-1 basal promoter activity and decreased transcriptional upregulation in response to various stimuli like H2O2 compared with lengths < 25 repeats. In the analysis of the data, we used the 25-repeat cutoff to categorize the study subjects in two categories [< 25 (GT)n repeats in both alleles or ≥ 25 (GT)n repeats in at least one allele] based on the HMOX-1 microsatellite length.

9HRV measurements were log10-transformed to improve normality. The following potential confounders were chosen a priori and included in the analysis: age, body mass index (BMI), mean arterial pressure, fasting blood glucose, cigarette smoking (never/former/current), alcohol consumption (≥ 2 drinks a day, yes/no), use of beta-blockers, calcium channel blockers, acetylcholinesterase (ACE) inhibitors, room temperature, season, and 48-hr moving average of outdoor apparent temperature. Potential nonlinearity between apparent temperature and HRV was accounted for using a linear and quadratic term.

10Because our data included repeated measures of HRV for many participants, our data may lack independence. To deal with this, we fit a mixed effects model (PROC MIXED in SAS version 9.0; SAS Institute Inc., Cary, NC). We assumed:

11where Yit is the logarithm of HRV in subject i at time t, b0 is the overall intercept, and ui is the separate random intercept for subject i. In the above, X1—Xp are the covariates measured at each of the visits in which the HRV measurements were taken. This captures the correlation among measurements within the same subject.

12Table 1 shows the levels and distribution of the variables used in this study, overall, and for the different combinations of GSTM1 genotype and HMOX 1 microsatellite repeat length. The study participants were all male, with average age of 72.8 years (SD = 6.6 years) at the first HRV measurement. No differences among the subpopulations defined by the combinations of GSTM1 genotype [wild-type or null] and HMOX-1 microsatellite repeat length [< 25 (GT)n repeats in both alleles or ≥ 25 (GT)n repeats in at least one allele] were found in age, BMI, systolic blood pressure, diastolic blood pressure, mean arterial pressure, heart rate, fasting blood glucose, total cholesterol, high-density lipoproteins (HDL), triglyceride, smoking status, alcohol intake (≥ 2 drinks/day), history of coronary heart disease (CHD), diabetes, hypertension, or stroke, and use of beta-blockers, calcium channel blockers, or ACE inhibitors (Table 1). The two genes were not associated with each other (p > 0.67).

13Table 2 shows the results of the analyses for the association of PM2.5 with changes inHRV for the entire population (model 1), by GSTM1 genotype (model 2), and by HMOX-1 microsatellite (GT)n repeat length (model 3).

14For the entire population, we found that a 10 μg/m3 increase in ambient PM2.5 in the 48 hr before the HRV measurement was associated with a 6.8% decrease in SDNN [95% confidence interval (CI), −12.9 to −0.2; p = 0.043] and with a 17.3% decrease in HF (95% CI, −30.0 to −2.3; p = 0.026). Ambient PM2.5 concentrations were also negatively associated with LF (estimated change = −11.2%, 95% CI, −22.8 to 2.2), but the result was not statistically significant (p = 0.10).

15The PM2.5–HRV association was modified by GSTM1 genotype, with PM2.5 concentrations negatively associated with SDNN, HF, and LF in GSTM1-null subjects, whereas no association between PM2.5 and HRV was found in GSTM1–wild-type carriers. In subjects with the GSTM1-null deletion, a 10-μg/m3 increase in PM2.5 was associated with a 10.5% decrease in SDNN (95% CI, −18.2 to −2.2; p = 0.015), a 24.2% decrease in HF (95% CI, −39.2 to −5.5; p = 0.014), and a 17.0% decrease in LF (95% CI, −31.0 to −0.2; p = 0.048). In GSTM1–wild-type subjects, the estimated decreases in HRV for a 10-μg/m3 increase in PM2.5 were 2.0% (95% CI, −11.3 to 8.3; p = 0.69) for SDNN, 4.0% (95% CI, −24.8 to 22.6; p = 0.74) for HF, and 0.6% (95% CI, −19.0 to 22.0; p = 0.95) for LF. However, the p-values for statistical interactions between PM2.5 and GSTM1 genotype were not significant.

16Similarly, we found that that PM2.5–HRV association was modified by HMOX-1 genotypes. Ambient PM2.5 concentrations were negatively associated with all three HRV outcomes in carriers of at least one allele with ≥ 25 microsatellite (GT)n repeats in the HMOX-1 promoter region, whereas no association between PM2.5 and HRV was present in carriers of < 25 repeats in both alleles. In subjects with at least one allele with ≥ 25 microsatellite (GT)n repeats, a 10-μg/m3 increase in PM2.5 was associated with a 8.5% decrease in SDNN (95% CI, −14.8 to −1.8; p = 0.014), a 20.1% decrease in HF (95% CI, −32.9 to −5.0; p = 0.012), and a 14.0% decrease in LF (95% CI, −25.7 to −0.5; p = 0.043). The p-value for statistical interactions between PM2.5 and GSTM1 genotype was marginally significant (p = 0.059) when the SDNN component of HRV was considered but was not significant for HF (p = 0.14) and LF (p = 0.11).

17We further evaluated the interrelationship between PM2.5, GSTM1, and HMOX-1 by estimating the effect of PM2.5 on HRV within each combination of the GSTM1 genotypes and HMOX-1 microsatellite repeat length categories (Table 3). These results indicate a clear trend of increasingly negative coefficients as we move across gene categories. In carriers of both the GSTM1-null deletion and at least one allele with ≥ 25 HMOX-1 microsatellite (GT)n repeats, PM2.5 was negatively associated with all three HRV outcomes, whereas no significant association was found in subjects with any other combinations. In subjects carrying the GSTM1-null deletion and at least one allele with ≥ 25 HMOX-1 microsatellite (GT)n repeats, a 10-μg/m3 increase in PM2.5 in the 48 hr before the HRV measurement was associated with a −12.7% decrease in SDNN (95% CI, −20.6 to −3.9; p = 0.0059), a 27.8% decrease in HF (95% CI, −43.0 to −8.5; p = 0.0073), and a 20.1% decrease in LF (95% CI, −34.5 to −2.7; p = 0.0261). GSTM1 genotypes and HMOX-1 microsatellite repeat lengths had a combined effect on the association between PM2.5 and HRV, as shown in Table 3 by the significant three-way interaction term between GSTM1, HMOX-1, and PM2.5 (coded as PM2.5 × a trend variable that is 1 for GSTM1 present and both short repeats, 2 for GSTM1 null and both short repeats, 2 for one long repeat and GSTM1 present, and 4 for GSTM1 null and at least one long repeat).

18To test whether the effect modification by genotype was driven by a few individuals or represented a more general shifting of the distribution of PM2.5–HRV slopes, we refit our mixed models, dropping the interaction terms with genotype but allowing for random subject-specific slopes. Figure 1 shows the distribution of subject-specific PM2.5 slopes for SDNN by three categories of genotype (HMOX-1 < 25 repeats and GSTM1 wt, either HMOX-1 ≥ 25 repeats or GSTM1 null, or both HMOX-1 ≥ 25 repeats and GSTM1 null). The shift to the left of the distributions does not appear to be driven by outliers. The same shift in the distributions was observed also for the HF and LF components (data not shown).

19Goodness of fit of the three-way interaction models was evaluated using the Akaike Information Criterion (AIC). AICs for models including the three-way interaction term were lower than those obtained from corresponding models including the main effects of GSTM1, HMOX-1, and PM2.5, as well as the two-way interactions between GSTM1 and PM2.5 and HMOX-1 and PM2.5, thus indicating better goodness of fit.

20Our study, based on an elderly population in Boston, showed that functional genetic variations in GSTM1 and HMOX-1, both of which are related to defenses against oxidative stress, modify the effects of PM2.5 on HRV. In the present work, we have extended our previous results examining the modification of the PM2.5–HRV association by GSTM1 to include other HRV outcomes and repeated measures on subjects, to show effect modification by HMOX-1, and to show a three-way interaction between the two genes and combustion particles.

21This work is part of a series of studies seeking to examine the potential pathways by which particles affect HRV. Specifically, we are looking at oxidative stress and endothelial function as potential pathways to this outcome. We hypothesize that if a pathway is important in the effect of PM2.5 on HRV, then factors that modify that pathway, either genes or drugs, may modify the PM2.5 response. We are also looking at metal-processing pathways as an indirect test of the hypothesis that metals on the PM2.5 particles play an important role in the HRV response.

22In our previous work on the same population we showed that ambient PM2.5 concentrations averaged over the 48 hr before the examination were associated with a reduction HF, with negative, albeit nonsignificant, associations were seen with SDNN and LF. In the present work, based on longer follow-up and additional HRV measurements, we were also able to show a significant effect on SDNN, as well as a more pronounced, although still nonsignificant, negative association with LF.

23As part of our examination of oxidative stress we have previously shown that particles had no effect on HRV in subjects with the functional GSTM1 polymorphism (GSTM1–wild-type) but had a substantially increased effect in those with the deletion (GSTM1-null). Similarly, we showed that statin use and obesity, which both modify ROS production, altered PM2.5 effects on HRV, thus confirming the critical role of oxidative stress pathways. In this article, we extend those results by showing a three-way interaction with genetic modifiers of response to oxidative stress.

24Although particle exposure has also been linked with activation of inflammatory pathways, alterations in blood coagulation, endothelial injury and dysfunction, and alterations in the autonomic control of the heart, our findings suggest that genetic variations in oxidative stress pathways play a critical role in the cardiovascular effects of airborne particles.

25Rodents exposed to concentrated urban particles evinced increased reactive oxygen species in both the lung and the heart, an effect muted by preadministration of N-acetylcysteine, a glutathione precursor and potent antioxidant. Inhalation of particles produces oxidative stress directly or via acute pulmonary inflammation, thus causing a series of events, such as the production of proinflammatory mediators, an increase of extracellular calcium influx, and the disruption of nitric oxide regulation, that may impair autonomic function and hence HRV. Diesel particles have also been shown to increase oxidative stress in endothelial tissue, inducing the production of HMOX-1. The viability of cell cultures of microvascular endothelial cells was impaired by diesel particles with an accompanying large increase in induction of HMOX-1; this process was blunted by N-acetylcysteine. Woodsmoke particles have also been shown to deplete intracellular glutathione and upregulate HMOX-1 activity in endothelial cells. Our results showing interactions of particles with GSTM1 deletion and microsatellite (GT)n repeat length in the gene coding for HMOX-1 are consistent with these laboratory findings that suggest a prominent role of ROS in particle toxicity.

26HMOX-1, the inducible heme oxygenase isoform, is expressed in multiple tissues, including vascular smooth muscle and endothelial cells. HMOX-1 expression has been shown to be upregulated in rat heart microvessel endothelial cells exposed to organic extracts of diesel exhaust particles, an effect that is likely to represent a response directed against ROS production. Large individual differences in the ability to modulate the quantitative level of HMOX-1 activity in response to a given stimulus have been described, which correlate with differences in the length of a microsatellite (GT)n repeat in the 5′ flanking region of the HMOX-1 gene. The purine–pyrimidine alternating sequence in the (GT)n repeat has the potential to assume Z-DNA conformation, a left-handed double-helix structure that is thermodynamically unfavorable compared with B-DNA conformation and has been described as negatively affecting transcriptional activity. Yamada and co-workers demonstrated by transient-transfection assay in cultured cell lines that the larger the number of (GT)n repeats in the HMOX-1 gene promoter, the lower is the HMOX-1 inducibility by ROS.

27In our study, HMOX-1 microsatellite (GT)n repeat length appeared to modulate the effects of PM2.5 on autonomic function, as measured by HRV variability. PM2.5 exhibited a negative correlation with SDNN, HF, and LF in individuals with ≥ 25 repeats, whereas no effect was seen in subjects with < 25 repeats. Furthermore, the strongest effects of PM2.5 were found in subjects that were lacking efficiency in antioxidant responses due to the combination of the GSTM1 deletion and ≥ 25 microsatellite (GT)n repeats in the HMOX-1 promoter. While the p-values for statistical interaction between PM2.5 and GSTM1 and between PM2.5 and HMOX-1 were not significant, the tests for the three-way interaction among PM2.5, GSTM1, and HMOX-1 were highly significant, confirming that the stronger effect modification of PM2.5 effects on HRV is seen when both genes are considered. The statistical modeling we used to fit the three-way interaction tested whether the size of PM2.5 effects in carriers of GSTM1 null and at least one long HMOX-1 repeat was 2 times larger than that observed in subjects carrying either GSTM1 null or one long HMOX-1 repeat. Thus, our data strongly suggest that particle exposure interacts with individual variations in the antioxidant response pathway to determine its effects on HRV.

28A potential limitation of this study is that we used ambient PM2.5 concentrations from a single monitoring site as a surrogate for recent exposure to PM2.5. A recent study comparing ambient concentrations at this site with personal exposures in Boston has shown a high longitudinal correlation between the two measurements; the study also reported that PM2.5 concentrations were spatially homogeneous over the Boston area. This suggests that our use of ambient concentrations is reasonable and the resulting exposure error is likely to be nondifferential. In our analyses, we considered several potential confounding factors that may have influenced HRV measures, as we adjusted our models for age, BMI, mean arterial pressure, fasting blood glucose, cigarette smoking, alcohol consumption, room temperature, outdoor apparent temperature, season, and use of beta-blockers, calcium channel blockers, and ACE inhibitors. Therefore, chances that the observed associations reflected bias due to confounders are minimized.

29Our results can be generalized only to an aged population that consists of older males who are almost all white. The effect on women and children as well as different ethnic groups should be addressed in future studies, particularly in relation to the exposure of different population groups to PM2.5 with varying geographic location, occupation, socioeconomic status, and behavioral characteristics. Other health outcomes including respiratory responses may also be affected by responses to ROS in an interaction with PM2.5 exposure. Our findings provide new information to guide research on the breadth of the effect of PM2.5 exposure.