Oxidative Damage to DNA and Lipids as Biomarkers of Exposure to Air Pollution

1Exposure to particulate matter (PM) from combustion processes contributes substantially to cardiovascular and pulmonary ill health and premature mortality globally. PM represents highly complex mixtures with large variations in size, chemical composition, shape, surface, reactivity, and charge, in both time and space, due to variable sources, atmospheric chemical reactions, and meteorological conditions. Nevertheless, for exposure assessment for epidemiological association with health outcomes and regulation, PM is usually only considered as mass defined by size cutoff at 2.5 and 10 μm in aerodynamic diameter (PM2.5 and PM10, respectively). These measures are often little affected by ultrafine particles (UFPs) from, for example, diesel engine emission because of their low mass, although UFPs are thought to have important health effects due to their high alveolar deposition, small size (≤ 0.1 μm in aerodynamic diameter), large surface area, and potential to translocate. Moreover, in epidemiological studies, exposure levels are often assigned crudely and groupwise according to air monitoring data and sources near the residence, because the modest individual risks require large numbers and long observation times to assess. Personal exposure can be assessed by portable monitors or carefully registered time–activity patterns in well-defined microenvironments of exposure in small numbers of subjects. However, the internal dose of PM also depends on breathing patterns and airway deposition, and cardiovascular and other systemic effects of PM require further translocation of PM constituents or signaling molecules or cells from the airways. Oxidative stress with inflammation is thought to be central in the mechanisms of action for both the pulmonary and extrapulmonary health effects of PM. Thus, biomarkers of oxidative stress should serve as proxy measures of the true internal exposure to PM to compare potential health impacts of different sources in both small controlled exposure settings and large population approaches. Oxidative modification of DNA and lipids are particularly relevant for cancer and cardiovascular disease where oxidative stress in the circulation is important. Experimental studies in animals and cell cultures have consistently shown that combustion-related PM induces oxidative stress and DNA damage in relevant organs and cells. In our experience the effects on biomarkers of oxidized DNA and lipids are observed within a lag period < 24 hr after the exposure to PM.

2A number of studies of PM exposure in humans have applied biomarkers of oxidative damage to DNA and lipids in the blood compartment or in terms of products excreted in urine or exhaled breath condensate (EBC), as outlined in Table 1. The biomarkers of oxidatively damaged DNA include 8-oxo-7,8-dihydroguanine (8-oxoGua) or the corresponding deoxynucleoside 8-oxo-7,8-dihydro-2′-deoxyguanosine (8-oxodG) measured in DNA and urine, the exocyclic M1 adduct to guanine (M1dG), and lesions detected as sites in DNA sensitive to formamidopyrimidine DNA glycosylase (FPG) and endonuclease III (ENDOIII). The biomarkers of lipid peroxidation (LPO) products include conjugated dienes (CDs), lipid hydroperoxides, malondialdehyde (MDA), thiobarbituric acid–reactive substances (TBARS), and F2-isoprostanes measured in EBC, plasma, serum, or urine. However, a systematic approach is required to evaluate the validity of their use as biomarkers of biological effective dose in this context. We undertook a systematic review of the published studies to assess the extent and consistency of associations between exposure to combustion-related PM and the biomarkers of oxidative damage to DNA and lipids.

3The publications were identified by searches in the PubMed, EMBASE, and Web of Science databases, as well as reference lists in the identified papers [see Supplemental Material for the search strategy (doi:10.1289/ehp.0901725)]. We included studies that investigated measures of effects of environmental air pollution exposure. This encompassed studies of subjects who had been exposed at work to environmental air pollution (e.g., policemen exposed to traffic exhaust). We excluded studies on occupational exposures to air pollution, such as metal smelting or incineration, because these are characterized by exposure to other air pollution components than those found in urban air. We searched for publications with reported data on LPO products and oxidatively damaged DNA by means of the biomarkers outlined in Table 1 in airways, the blood compartment, and urine. Isolated leukocytes, lymphocytes, or mononuclear blood cells are all referred to as white blood cells (WBCs). We used the term “oxidized nucleobases” for the urinary excretion of 8-oxodG and 8-oxoGua because they are not measured in DNA. Tables 2–5 summarize details of the included studies. The results from some of the studies have been reported in multiple publications; we discuss these as studies rather than as individual publications because they originated from the same investigation.

4We stratified the studies into three broad categories: controlled exposures, panel studies, and cross-sectional studies. Studies with controlled exposure to air pollution PM are the most robust type of design as either crossover studies or parallel groups of subjects exposed to air pollution constituents or filtered air. In panel studies, samples are collected from the same individuals at different times of the year in order to exploit contrasts in exposure due to temporal changes. This study design minimizes the influence of interindividual variation because the subjects are their own controls. However, the design is vulnerable to confounding because other factors such as diet and sunlight show temporal (e.g., seasonal) variation, which can affect the value of the biomarker as shown, for instance, for DNA damage in WBCs detected by the comet assay. In addition, the quality of the panel study depends on the exposure characterization; personal exposure characterization shows a closer association with biomarker levels than does exposure assessed from stationary monitoring stations. Cross-sectional studies have a less controlled design than do the panel studies because the exposure gradient is obtained by collecting samples from subjects from different geographical areas or occupations. The cross-sectional studies can have optimal exposure characterization, but confounding can be a problem because individual factors such as lifestyle, including diet, may influence the biomarker and covary with the exposure. This problem typically arises, for instance, when policemen and office personal or subjects from rural and urban areas are being compared.

5We critically analyzed the studies with special focus on suboptimal exposure assessment and the quality of biomarkers. We referred to these problems as potential measurement error in the exposure assessment and biomarkers because they can be regarded as systematic errors that may affect the validity. Figure 1 outlines the relationship between the measurement error in exposure assessment and biomarker. We should emphasize that specific studies included in our analysis may have other sources of bias, including selection bias and small numbers of observations, which can affect the estimated effect size in a particular study.

6The primary exposure assessment in our analysis is the mass concentration of particles as PM2.5 or PM10 or the number concentration of UFPs. The exposure characterization encompasses data obtained from stationary monitor stations and personal monitors. We regarded personal exposure to PM as the optimal exposure assessment; for studies without PM measurements we regarded data on ambient gasses [nitrogen oxides (NOx), ozone (O3), or sulfur dioxide], polycyclic aromatic hydrocarbons (PAHs), and benzene as indirect estimates of PM exposure with greater potential for error. Similarly, the urinary excretion of metabolites of PAHs [e.g., 1-hydroxypyrene (1-HOP)] and benzene [S-phenylmercapturic acid (S-PMA) and trans,trans-muconic acid (tt-MA)] generated by their biotransformation are potentially biased estimates of the ambient concentration of PM, although they may be important biomarkers of internal dose of the parent compound. The measured PM showed the strongest association with the biomarkers of oxidized DNA and lipids in studies that measured the exposure as both PM and nitrogen dioxide (NO2), 1-HOP, S-PMA, and tt-MA. We expect that the measurement error will be nondifferential, which usually biases effect estimates toward the null because they tend to obscure contrasts between the exposed and unexposed and those with or without the outcome of interest.

7The potential for biomarker measurement error originates from unspecific measurements or analytic flaws due to poor assay conditions. For instance, suboptimal assay procedures used to detect oxidatively damaged DNA may cause spurious oxidation that increases the apparent level of DNA damage. The level of 8-oxodG in DNA from unexposed mammals is approximately 1 lesion/106 dG (deoxyguanosine); the,; ) recommended that publications that report levels of 8-oxodG above a threshold of 5 lesions/106 dG should be interpreted with caution. The comet assay detects DNA damage by migration of DNA in agarose gels, and the end points are usually reported as extent of migration, although they can be transformed to lesions per unaltered nucleotides by calibration with ionizing radiation. The level of oxidatively damaged DNA measured by the comet assay in WBCs of humans is < 1 lesion/106 nucleotides. Oxidatively damaged DNA, nucleobases, and LPO products can be measured by antibody-based methods, but artificially high background levels can occur because of unspecific binding of the antibodies to other biomolecules. The simple assay of TBARS and CDs has been seriously criticized and is not recommended for in vivo detection of LPO products, whereas improved methods using high-performance liquid chromatography (HPLC) purification steps are more reliable assays of TBARS. We classified biomarkers with suboptimal biochemical analysis as follows: (1) simple spectrophotometric measurement of TBARS without a prepurification step; (2) simple assays for CDs and lipid hydroperoxides, (3) levels of 8-oxodG exceeding a threshold of 5 lesions/106 dG in the unexposed group, and (4) detection of oxidatively damaged DNA, nucleobases, or lipids by antibody-based methods without prepurification steps. We expect that the biomarker measurement error will result in reduced effect estimates for both nonspecific biomarkers and assays having low sensitivity or a high limit of detection.

8The studies differ considerably in design, and the results are reported in ways and units that preclude direct comparison of the estimated effect size in the studies. Thus, we have estimated the effect of exposure on biomarkers as standardized mean differences with 95% confidence intervals (CIs) between exposed subjects and referents in a random effects meta-analysis by means of Review Manager (RevMan; version 5.0; Nordic Cochrane Centre, Cochrane Collaboration 2008, København Ø, Denmark). The standardized mean difference is the difference in means of groups divided by the pooled standard deviation (SD). It is a measurement of estimated effect, which can be used in a meta-analysis when all studies assess the same outcome (level of oxidized biomolecules in this analysis), but it is measured in a variety of ways with different scales. We assessed the estimated effect size in a random model meta-analysis because it incorporates heterogeneity among studies. The heterogeneity between studies was analyzed by tau squared, chi squared, and I2 tests; tests for subgroup differences were carried out using the chi-square test. We obtained means, SDs, and the number of subjects from the studies, or we calculated these values from data reported in the publications (see Tables 2–5). The variance was reported in different ways in the original publications; thus, our estimates of 95% CIs may be biased, but the central estimates (means) should not be. We calculated the mean and SD from regression analyses for studies that modeled continuous data and defined the exposure gradient as equal to either the interquartile range or a 10 μg/m3 increase in PM. The interquartile range in PM2.5 was in the same range as a 10-μg/m3 increase in PM2.5 in the studies carried out in Windsor, Ontario, Canada, whereas the interquartile range (66.5 μg/m3 measured as PM10) measured by in Iran was substantially higher, which could reflect different sources of exposure or PM fraction. We calculated overall means and SD for studies that included more than one group of exposed subjects or investigated the same subjects under different exposure scenarios. We assumed that the interquartile ranges would be equal to the SDs found in studies that used nonparametric analyses and reported variation as ranges.

9Table 2 summarizes studies on the association between air pollution exposure and oxidized DNA and lipids in controlled exposure studies. The number of subjects in these studies is rather low (3–41 subjects; mean ± SD, 18 ± 12), which may be because carrying out controlled exposure on a large number of subjects is demanding. A study of air pollution exposure to UFPs in persons bicycling for approximately 90 min in a laboratory or on traffic-intense streets reported that the level of FPG sites in WBCs was associated with the number concentration of UFPs. A subsequent investigation by the same group had a similar correlation between particulate fractions with median particle sizes of 23 nm and 57 nm (consistent with semivolatile organic compounds from diesel exhaust and carbonaceous soot emissions into the air of a busy street) and the level of FPG sites in WBCs. A subsequent study in elderly subjects showed a statistically non-significant decrease in urinary excretion of the F2-isoprostane 8-iso-prostaglandin-F2 after a period of home air filtration. In another study of elderly patients with coronary heart disease, reported that inhalation of concentrated air pollution particles (CAPs) was associated with increased concentration of 8-isoprostanes in EBC. This finding is in keeping with that found by ) who observed that healthy young subjects had elevated level of MDA in EBC after intensive exercise at a location with high-traffic intensity compared with the same type of exercise at a location with less traffic.

10Controlled exposure to wood smoke containing very high mass concentration of particles has been associated with increased levels of LPO products in serum, urine, and EBC. However, observed no association between exposure and FPG sites in WBCs and suggested that this result may be due to increased DNA repair activity of oxidized nucleobases because urinary excretion of 8-oxoGua and WBC expression levels of the 8-oxoguanine DNA glycosylase (OGG1) base excision repair enzyme, which removes 8-oxoGua from DNA, were increased after exposure to wood smoke but not after exposure of the same subjects to clean air. Increased urinary excretion of 8-oxoGua was also observed in a study where subjects were exposed to exhaust on a traffic-intense street for 4 hr, but 2 hr of exposure to a high concentration of diesel exhaust was not associated with urinary excretion of 8-oxodG or F2-isoprostanes in subjects with metabolic syndrome.

11Table 3 summarizes studies on the association between air pollution exposure and oxidized DNA and lipids in panel studies. These studies involved multiple measurements over time, and the number of subjects in these studies was higher than the number of subjects in controlled exposure studies (2–182 subjects; mean ± SD, 62 ± 59). Several panel studies showed that concurrent air pollution exposure was associated with elevated levels of LPO products in EBC and plasma, as well as elevated levels of 8-oxodG in plasma. Subjects without doctor-diagnosed cardiovascular diseases or who did not take medication for diabetes showed an association between outdoor levels of PM2.5 and TBARS in plasma, although the analysis of all subjects in the study only indicated statistically nonsignificant associations between personal exposure to PM2.5 and TBARS or 8-isoprostanes in plasma. Personal exposure to PM2.5 was associated with increased levels of 8-oxodG in WBCs of students living in the center of Copenhagen, whereas the exposure was only associated with the MDA levels of women and not with the level of FPG sites in WBCs in any group. Interestingly,, observed no correlation between 8-oxodG in WBCs and the background mass concentration of PM2.5 measured at stationary monitoring stations, suggesting that a relatively clean urban air may provide too little contrast in the long-range transported fractions of PM to be a reliable indicator of traffic-generated exposure. Moreover, they found significant association between the biomarkers and personal exposure to NO2 supporting the key role of PM. This is in keeping with observations from a controlled exposure study with constant NO2 exposure, which showed a strong effect of change in PM exposure on DNA oxidation.

12The design of the cross-sectional studies can be grouped into two main categories. The first category is characterized by studies that achieved the exposure contrast by studying subjects in occupations with different ambient air pollution levels (Table 4). The other category is characterized by studies of subjects, often with comparable occupations or ages, from geographical areas with different ambient air pollution levels (Table 5). The cross-sectional studies have generally included more subjects than the controlled exposure studies and panel studies. The number of subjects in the cross-sectional studies on different occupations has been in the range of 31–356 subjects (mean ± SD = 109 ± 95 subjects), whereas the studies that have contrasted exposure in different geographical areas have used even higher number of subjects (43–894 subjects; mean ± SD = 222 ± 234).

13Using job titles as the basis for stratification of exposure, studies showed that subjects in occupations with high exposure to traffic emissions had higher levels of FPG sites and 8-oxodG and M1dG in WBCs than did referents. However, the latter two studies reported levels of 8-oxodG that were above the threshold of 5 lesions/106 dG, suggesting the potential for spurious oxidation of the samples. Another study showed higher levels of FPG and ENDOIII sites in WBCs of traffic emission exposed policemen compared with other policemen working indoor during a month when air pollution was relatively high (i.e., January) but no association during a month with low air pollution exposure (i.e., September). In contrast, found no difference in serum lipid hydroperoxides concentration when comparing traffic officers with healthy indoor workers; the primary purpose of their study appears to have been to compare lipid hydroperoxide levels between subjects that received phytochemicals and subjects that did not, and they evaluated associations with air pollution in a secondary analysis. Evidence of null associations in studies that evaluate air pollution as a secondary exposure suggests the possibly of a general trend toward publication bias favoring studies that report positive associations when air pollution is the primary exposure of interest, but the findings of the study may also have been due to the use of a nonspecific spectro-photometric assay for the detection of LPO products, which can bias the estimated effect toward null. Studies using urinary biomarkers have also shown increased levels of 8-oxodG and 15-F2t-isoprostanes in subjects exposed to high concentrations of traffic-vehicle exhausts.

14Cross-sectional studies of subjects living, working, or going to school in locations with different ambient air pollution levels encompass investigations of subjects with predefined age groups, such as children, adolescents, adults, or elderly. Studies of children living in areas with different levels of exposure have shown positive associations with 8-oxodG levels in nasal cells, 8-oxodG in WBCs, LPO products in plasma, and urinary excretion of 8-oxodG. Studies of air pollution exposure in adults have provided more mixed results: benzene as a marker of urban air pollution exposure was associated with urinary excretion of S-PMA and 8-oxodG in WBCs, but not with ENDOIII and FPG sites in WBCs or urinary excretion of 8-oxodG. Other studies of urinary excretion of 8-oxodG have shown positive associations, but studies of LPO products in plasma have indicated both positive associations with oxidative damage and no apparent effects.

15The qualitative assessment in the preceding sections indicates that most of the reports showed associations between air pollution exposure and oxidatively damaged DNA, nucleobases, and lipids. Most studies measured the biomarkers in surrogate tissue cells such as WBCs or noncellular bodily fluids such as plasma, urine, and EBC. The data on biomarkers of the airways mainly encompass measurements of LPO products in EBC, whereas only one study examined 8-oxodG in nasal cells. Figure 2 shows study-specific and overall estimates of effect for exposure to PM on oxidatively damaged DNA, nucleobases, and lipids in the airways, blood, and urine. In general there is considerable heterogeneity between the studies and between subgroups; methodological diversity between studies may explain the heterogeneity. In addition, the categorization of the exposure does not take into account that exposure gradients most likely differ between the studies. It is not possible to calculate a standardized exposure unit (e.g., effect per 10-μg/m3 increase PM2.5) because the studies reported exposure measurements in different fractions of PM, or they contained no information about the concentration of PM. The overall standardized mean differences between exposed subjects and nonexposed referents for the oxidized DNA and for LPO products in the blood were 0.53 (95% CI, 0.29–0.76) and 0.73 (95% CI, 0.18–1.28), respectively. In the urine the estimated effect size by PM exposure versus nonexposed referents for oxidatively damaged DNA and nucleobases and for LPO products was 0.52 (95% CI, 0.22–0.82) and 0.49 (95% CI, 0.01–0.97), respectively. This suggests that exposure to PM is associated with comparable increases in oxidized DNA and lipids, although it should be emphasized that the heterogeneity between subgroups might mask real differences between the biomarkers. The effect on DNA damage in the airways is presently difficult to assess because there was only one study of oxidized DNA. The estimated effect on LPO products in EBC was 0.64 (95% CI, 0.07–1.21), which is comparable to overall standardized mean differences in LPO products in the blood and urine with PM exposure. This finding suggests that LPO products in plasma and urine are suitable biomarkers of biologically effective PM dose reflecting oxidative stress in the airways.

16The focus of our analysis was on the estimated effect of exposure to PM on biomarkers of oxidized DNA, nucleobases, and lipids. A number of studies have not measured personal exposure to PM, which may be because of their focus on other air pollution constituents or lack of resources or because the study was too large to do personal exposure measurements. Ideally, all studies should have included personal measurements of PM, and the analysis should have included mutual adjustment for coexposures to other air pollutants. We assumed that PM is the most important contributor to oxidative stress among the air pollutants and that personal exposure can be estimated from the personal exposure to the other pollutants or ambient PM levels, although with potential bias. Of the studies identified in our meta-analysis, only has estimated personal exposure levels using mathematical modeling of data from stationary monitoring stations. We have categorized the studies according to whether or not they characterized exposure using personal versus ambient PM measurements. Based on the categorization of the studies according to the potential measurement error in Tables 2–5, the estimated effect size is 0.55 (95% CI, 0.19–0.90), 0.66 (95% CI, 0.37–0.95), and 0.65 (95% CI, 0.34–0.96) for studies categorized as having no potential measurement error, potential measurement error in either biomarker analysis or exposure assessment, and potential measurement error in both biomarker analysis or exposure assessment, respectively [see Supplemental Material for the Forest plots (doi:10.1289/ehp.0901725)]. The effect size is essentially identical in the three groups, which could be caused by opposite acting effects of potential measurement errors (regression toward null effect) and uncontrolled confounding factors (increased effect size) in panel and cross-sectional studies. The percentages of studies that did not use personal measurements to assess PM exposure were 8% (1 of 12), 36% (4 of 11), and 100% (29 of 29) in the studies categorized as controlled exposures, panel studies, and cross-sectional studies, respectively (χ2 = 36.7, p < 0.001). The use of error-prone biomarker assays was less likely in the controlled exposure studies (31%; 4 of 13) than in the panel studies (64%; 7 of 11) and cross-sectional studies (59%; 17 of 29), although no differences were found between the distributions (χ2 = 3.4, p > 0.05). Overall, the quality of the studies, in terms of the likelihood of exposure and outcome measurement error, appears highest in the controlled studies and lowest in the cross-sectional studies. Both of these measurement errors may bias the effect estimate toward null, whereas uncontrolled confounding factors in panel studies and cross-sectional studies most likely increased the estimated effect size. In addition, we emphasize that the studies that measured personal PM exposure and used more accurate biomarker assays mainly investigated the effect of air pollution particles in realistic urban air concentrations, although one study used high concentrations of wood smoke particles. Collectively, associations between PM exposure and biomarkers of oxidative stress estimated by studies likely to have more accurate exposure and outcome measurements cannot be explained by exposure to excessive concentrations of PM. Yet, the controlled exposure studies in our meta-analysis had few numbers of observations, which might reduce the precision with which effects have been estimated. In addition, controlled exposure studies may be more prone to selection bias due to random errors in selection and have limited generalizability because they are restricted to participants with specific characteristics.

17Our analysis shows that exposure to combustion particles is consistently associated with elevated levels of oxidatively damaged DNA and nucleobases and LPO products in human blood cells, plasma, urine, and EBC. The association is seen across studies with optimum designs, including controlled or personal exposure assessment and biomarkers with low potential measurement error due to indirect exposure assessment and/or use of biomarkers prone to artifacts. Still, we emphasize that the identified studies are inhomogeneous in design and quality of biomarkers, which weakens our conclusions about specific exposure–effect relationships for particulate air pollution. In addition, few numbers of subjects, especially in the controlled exposure studies, is a minor limitation that might affect the generalizability of our meta-analysis results.

18Our critical analysis indicates that the range in exposure to realistic ambient concentrations of combustion particles is associated with a 50% increase in the level of oxidatively damaged DNA, nucleobases, and lipids, supporting the notion that they are suitable biomarkers of the biologically effective dose of PM. However, we caution against the use of suboptimal biomarkers; ideal biomarkers of oxidative damage should detect a major part of the total ongoing oxidative damage in vivo, have small assay variation, have smaller intraindividual than interindividual variation, not be confounded by diet, be stable on storage, and have the same level obtained in target and surrogate tissue. They should also have predictive value of risk of disease, which can be firmly assessed only in prospective studies. The biomarkers of urinary excretion of 8-oxodG and TBARS in plasma are among the few biomarkers that have been studied in prospective cohort studies; they have predictive value regarding development of lung cancer and cardiovascular diseases, respectively. Further development of oxidized DNA and LPO products as biomarkers of biological effective dose of air pollution exposure should focus on the most reliable and well-validated assays, including assays for the measurement of isoprostanes and techniques that consistently measure low background levels of oxidatively damaged DNA and nucleobases. The relevant biomarkers with low potential of measurement error are constantly developed to increase assay capacity toward high throughput, for instance, the comet assay and urinary excretion of 8-oxodG (Henriksen and Poulsen 2009; ). This development will allow the use of these biomarkers of exposure to PM in large-scale population studies.