Is There Evidence for Synergy Among Air Pollutants in Causing Health Effects?

Background Environmental air pollutants are inhaled as complex mixtures, but the long dominant focus of monitoring and research on individual pollutants has provided modest insight into pollutant interactions that may be important to health. Trends toward managing multiple pollutants to maximize aggregate health gains place increasing value on knowing whether the effects of combinations of pollutants are greater than the sum of the effects of individual pollutants (synergy). Objective We reviewed selected published literature to determine whether synergistic effects of combinations of pollutants on health outcomes have actually been demonstrated. Methods and results We reviewed 36 laboratory studies of combinations of ozone with other pollutants that were reported in the recent U.S. Environmental Protection Agency Ozone Criteria Document. We examined original reports to determine whether the experimental design tested for synergy and whether synergy was demonstrated. Fourteen studies demonstrated synergism, although synergistic, additive, and antagonistic effects were sometimes observed among different outcomes or at different times after exposure. Conclusions Synergisms involving O3 have been demonstrated by laboratory studies of humans and animals. We conclude that the plausibility of synergisms among environmental pollutants has been established, although comparisons are limited, and most involved exposure concentrations much higher than typical of environmental pollutants. Epidemiologic research has limited ability to address the issue explicitly.


Commentary
Air pollution exists as a complex mixture that varies spatially and temporally in its charac teristics. In the United States and other countries, major air pollutants are regulated individually, although when present in a mixture, their toxicity could differ from that found in investigations directed at individual pollutants. As recommended by the National Research Council (NRC 2004a), there is increasing interest in managing environmen tal air quality using multipollutant strategies aimed at reducing the aggregate health bur den of air pollution. Therefore, from both the public health and regulatory perspectives, the potential for synergy among mixture components is a particular concern. A fair and increasingly frequent question is whether there are any documented examples of syner gies among air pollutants. That question was the focus of our review and this commentary.
Definitions and strategies. Synergy is strictly defined as occurring if the effect of the combined exposure is greater than the sum of the effects of the two or more individual components of the mixture (see definitions in Appendix 1). The term is used loosely and sometimes applied to any effect of a combi nation of pollutants that is greater than the effect of one of the components alone. This circumstance is not an instance of synergy as defined within the public health community unless the effects are greater than additive. The term also extends to an effect caused by a combination of pollutants but not caused by exposure to the individual pollutants, absent exposure to the others. This applied defini tion makes clear the need to have evidence on both the individual and the combined effects of any combination of pollutants to evaluate the presence of synergism.
The term "interaction" also requires a careful definition. As defined in statistical modeling, interaction refers to the inter dependence of the effects of two or more variables. Product terms of the potentially interacting variables are inserted into ana lytical models to test for the presence of interaction, which may be synergistic or antagonistic. The statistical tests for inter action have low statistical power, and con sequently the joint effects of several factors may be only imprecisely characterized, unless there is a strong interaction or abundant data. Often, the term "interaction" is used loosely to refer to synergism or antagonism, although more accurately, interactions can be characterized as synergistic or antagonistic, depending on the direction of the combined effect. Epidemiologists also use the term "effect modification" to refer to interdepen dence of the effects of two or more variables (Last 2000).
Synergism between two environmental pollutants might occur through a variety of mechanisms. First, the two pollutants might act at the same or different steps in the same mechanistic pathway; second, the presence of one might influence ability to mitigate the action of the other; and third, the presence of one might influence the dose of the other.
There are the possibilities that the existence of synergism could be dose dependent, that the same combined exposure might be syn ergistic for one effect and not for others, or that the same effect may be synergistic in some tissues and not in others, as illustrated by examples presented later.
Laboratory and epidemiologic studies have demonstrated effects of combined or sequential exposures that were greater than the effects of either exposure given singly, but the combined effects were often simply addi tive. Other such studies have demonstrated combined effects that were less than additive. A strict quantitative test of synergy requires measurement of the effects of each com ponent and of the combined components administered under identical conditions. For environmental exposures studied with epide miologic methods, the corresponding condi tions would include sufficiently comparable populations, measurements of effect, coex posures to other pollutants, and other modi fying factors. Such identical conditions are impossible to achieve in the strictest sense, but may be approximated to varying degrees in epidemiologic studies through careful design and analysis.
In assessing synergism among pollutants using epidemiologic approaches, the mini mum information needed is estimation of exposure or dose for the two or more pollut ants under investigation. Most often, syner gism is assessed using multivariable models that include terms representing the effects of the individual pollutants and one or more interaction terms that represent potential Background: Environmental air pollutants are inhaled as complex mixtures, but the long dominant focus of monitoring and research on individual pollutants has provided modest insight into pollutant interactions that may be important to health. Trends toward managing multiple pollutants to maximize aggregate health gains place increasing value on knowing whether the effects of combinations of pollutants are greater than the sum of the effects of individual pollutants (synergy). oBjective: We reviewed selected published literature to determine whether synergistic effects of combinations of pollutants on health outcomes have actually been demonstrated. Methods and results: We reviewed 36 laboratory studies of combinations of ozone with other pollutants that were reported in the recent U.S. Environmental Protection Agency Ozone Criteria Document. We examined original reports to determine whether the experimental design tested for synergy and whether synergy was demonstrated. Fourteen studies demonstrated synergism, although synergistic, additive, and antagonistic effects were sometimes observed among different outcomes or at different times after exposure. conclusions: Synergisms involving O 3 have been demonstrated by laboratory studies of humans and animals. We conclude that the plausibility of synergisms among environmental pollutants has been established, although comparisons are limited, and most involved exposure concentrations much higher than typical of environmental pollutants. Epidemiologic research has limited ability to address the issue explicitly.  (Greenland 1983;Rothman et al. 2008). A model for two pollutants would be as follows: where Y is the outcome, x 1 and x 2 are two specific pollutants, β 1 and β 2 estimate the pollutantspecific effects of x 1 and x 2 , and β 3 estimates the joint effect of x 1 and x 2 , and ε is the error term. In the presence of inter action (synergism is positive interaction), β 3 is not zero, the value indicating no interaction, and the effect of x 1 is x 1 (β 1 + β 3 x 1 x 2 ) and that of x 2 is x 2 (β 2 + β 3 x 1 x 2 ). In the example of smoking, radon exposure, and lung cancer, Y might represent risk for lung cancer death, x 1 the cumulative amount smoked, and x 2 the cumulative exposure to radon. For three pol lutants, there are three twoway interaction terms (for x 1 x 2 , x 1 x 3 , and x 2 x 3 ) as well as a threeway interaction term for x 1 x 2 x 3 . As the number of potentially interacting pollutants increases, the number of possible interaction terms increases progressively. The coefficients are estimated by the algorithm used to fit the model to the data-for example, least squares or maximum likelihood. Even for two exposures, statistical power is limited for detecting interactions, and the value of β 3 is estimated with limited preci sion, unless the data are substantial or there is particularly strong synergism (Greenland 1983;Rothman et al. 2008). If inter actions among three or more pollutants are of interest, power is likely to be extremely limited, given the number of terms in the model. Interpretation of models directed at interaction also needs to consider error in the measurement of the pollutants and the consequences of errors that are differential among the pollutants. A further consider ation is the scale on which the interaction is assumed to take place: additive or multiplica tive. Typically, additive models are used for continuous outcome measures, such as lung function level, whereas multiplicative models are used for risk for events, such as prob ability of dying. Interpretation of interaction terms is scale dependent, although for public health purposes, epidemiologists are in con sensus that positive departure from additivity constitutes synergism (Rothman et al. 2008).
Although the scale of interaction should be based in biological understanding, there is not a specific correspondence between bio logical interaction and statistical interaction in the modeling of epidemiologic data.
In the laboratory, evaluating synergism between pollutants A and B, for example, would require four exposure scenarios: a) con trol, b) A alone, c) B alone, and d) A + B (each at the same concentration as when given singly). Either the same or different groups of subjects might receive each treat ment, depending on the nature of the treat ment, persistence of effects, and meas ure ment methods.
Background knowledge. The Health Effects Institute convened a working group during 1990-1992 to review the subject (Samet and Speizer 1993a), and information in the resulting papers described challenges and strategies for evaluating complex mix tures by epidemiologic (Dockery 1993;Samet and Speizer 1993b;Weiss 1993) and labora tory studies (Mauderly 1993;McDonnell 1993). The NRC and other organizations have convened numerous committees on the related topic of complex mixtures. Beyond subsequent reports of several laboratory stud ies examining interactions among pollutants at high doses, it is not clear that the under standing of synergistic interactions among environmental pollutants has advanced sub stantially beyond the 1993 reports.
Synergisms among occupational inhala tion exposures are known; perhaps the best documented example is the synergy between radon progeny and cigarette smoking in pro ducing lung cancer in underground miners (NRC 1988(NRC , 1999. In this example, the presence of synergism was shown by a pooled analysis of data that contained information from cohorts of underground miners on both exposure to radon progeny and smoking. Statistical models were used to estimate the degree of synergism, which could be deter mined with reasonable precision because substantial data were available. Wellstudied examples of potential synergism between smoking and other disease risk factors include asbestos and lung cancer [Agency for Toxic Substances and Disease Registry (ATSDR) 2006; Selikof et al. 1968] and oral contra ceptives and pulmonary thromboembolism (Lidegaard 1999;Petitti et al 1996).
A review by the ATSDR of the toxico logic and epidemiologic literature on effects of mixtures (ATSDR 2004) found some, but scant, evidence for synergisms from exposures to multiple chemical agents. In addition to examples in which each component caused measurable effects, examples were cited in which combined exposures caused measurable effects that were not observed when the com ponents were administered individually at the same doses. Among the combined exposures reviewed by ATSDR, the effects were more commonly less than additive, rather than greater than additive.
The air pollution literature yields similarly mixed results. For example, Chen et al. (1991) observed greater than additive effects on the pulmonary function of guinea pigs from expo sure to sulfuric acid (H 2 SO 4 ), acidcoated zinc oxide particles, and ozone. Kleinman et al. (2000) observed less than additive effects on the proliferation of the respiratory epithe lium of rats exposed to O 3 , carbon particles, and ammonium bisulfate given singly and combined. Anderson and Avol (1992) did not find significant synergy between the effects of carbon particles and H 2 SO 4 on the respi ratory function of experimentally exposed humans; however, the combination elicited effects in some subjects having no response to either carbon or acid alone, suggesting pos sible synergy. Jakab and Hemenway (1994) found that the effects of inhaled O 3 and car bon black were synergistic in causing lung inflammation and suppressing phagocytosis by alveolar macrophages recovered from rats in bronchoalveolar lavage fluid (BALF). The effects of carbon black and O 3 were syner gistic only when inhaled together, not when administered sequentially. Creutzenberg et al. (1995) found the combined effect of inhaled O 3 and instilled carbon black on the uptake of particles by cells collected from exposed rats in BALF to be less than additive, but the effect on the migration of cells in response to a chemoattractant was synergistic. The latter two studies demonstrated that effects of com bined and sequential exposures might differ and that the same exposure may be synergistic for some, but not all, outcomes.
There have also been studies of interac tions between air pollutants and other agents, and these, too, have produced mixed results. For example, Spannhake et al. (2002) found a greater than additive release of the proin flammatory cytokine interleukin8 from the BEAS2B human airway epithelial cell line treated with rhinovirus and exposed to either O 3 or nitrogen dioxide. Harrod et al. (2003) found less than additive effects on broncho alveolar lavage cell counts, proinflammatory cytokines, and Clara cell secretory protein levels in mice exposed to respiratory syncytial virus and diesel emissions. One could also consider many of the studies of the effects of pollutant exposure on airway reactivity as tests of synergy, if they include exposures to the pollutant and the airway agonist alone as well as in combination. For example, several of the 29 studies listed in

Review of Combined Exposure Studies Cited in 2006 U.S. EPA Ozone Criteria Document
The above studies indicate that synergies involving environmental air pollutants have been demonstrated in the laboratory and that O 3 has been a component of synergistic combinations of pollutants. We turned to the 2006 U.S. EPA Criteria Document for O 3 (U.S. EPA 2006) to examine in detail a larger set of studies of combined exposures involving O 3 . This subset of the air pollution literature is a potentially informative body of evidence because it has been assembled in a policydirected document. O 3 is a common pollutant that has been studied extensively in population and laboratory settings, and the three tables encompass a sizable number of studies addressing the combined effects of O 3 and other pollutants. The original publica tions were examined to determine whether the study designs provided tests of synergy and, if so, whether a positive departure from additivity was demonstrated. The 14 studies we found to demonstrate synergies are listed in Table 1.
We did not attempt a more comprehen sive review of the air pollution literature, although such a review would be useful. People are certainly exposed to a wide spec trum of additional pollutant combinations, and the effects of some internally complex pollutant classes such as particulate matter (PM) may involve synergisms among com ponents. We found that our limited review was sufficient to solidly confirm the answer to the question of whether synergies among pol lutants found in the environment had been demonstrated.
Laboratory studies of humans exposed to combinations including O 3 . We examined 19 papers cited in Table AX614 (Ozone Mixed with Other Pollutants) of the 2006 U.S. EPA Criteria Document. We found that the designs of 13 of 19 studies demonstrat ing significant effects were adequate to test for synergy. Only the following two studies ( Table 1) demonstrated effects of combined exposures to O 3 and another pollutant that were greater than the sum of the effects of the individual pollutants. The other 11 studies testing for synergy did not demonstrate effects that were greater than additive. Horvath et al. (1986) exposed young, nonsmoking women for 2 hr during intermit tent exercise to 485 ppb O 3 , to 2 ppb peroxy acetyl nitrate (PAN), and to the combination of O 3 and PAN. Measurements included heart rate, breathing pattern, lung volumes, forced exhalation, and O 2 -CO 2 exchange. PAN alone caused some symptoms but either no or very small, insignificant effects among measured variables. O 3 caused more reports of symptoms and significant effects on several measured variables. The combined exposure was additive with respect to total reported symptoms but greater than additive for most affected variables. For example, forced vital capacity near the end of exposure was reduced 2% by PAN, 23% by O 3 , and 31% by the combination, indicating a combined effect 24% greater than additive.
DrechslerParks (1995) exposed healthy older adults (five men and one woman, 56-85 years of age, completed the proto col) for 2 hr during intermittent exercise to 450 ppb O 3 , 600 ppb NO 2 , or to O 3 and NO 2 combined. Measurements included respiration, heart rate, cardiac output, and stroke volume. Cardiac output was increased 1% by O 3 , reduced 5% by NO 2 , and reduced 14% by the combination. Stroke volume was reduced 2% by O 3 , reduced 7% by NO 2 , and reduced 12% by the combination. These results suggest modest synergy.
Laboratory studies of animals exposed to combinations including O 3 . We exam ined 17 papers listed as reporting synergis tic effects in Tables AX517 (Interactions  of Ozone EPA 2006). We found that only 14 of 17 stud ies were designed to test for synergism, and that only the 12 listed in Table 1 actually demonstrated synergism by reporting effects of combined exposures that were greater than the sum of effects of the individual exposures.
This discrepancy between the authors' con clusions and our review reflects the common tendency to use the term "synergism" loosely. The 12 studies demonstrating synergism for at least one measured outcome are described briefly here.
Two studies examined interactions between O 3 and acid aerosols. Kimmel et al. (1997) exposed SpragueDawley rats 4 hr/ day for 2 days to O 3 at 600 ppb, to fine (300 nm) or ultrafine (60 nm) H 2 SO 4 aero sol at 500 µg/m 3 , or to the combinations of O 3 and each of the acid aerosols and per formed morphometric measurements of cell proliferation and damage to alveolar septa. The volume percentage of markedly damaged alveolar tissue was only slightly increased by either acid exposure but markedly increased by O 3 . The combined effects were greater than additive. A similar pattern was observed for cell proliferation, but the degree of syn ergy was less pronounced. Sindhu et al. (1998) exposed Fischer344 (F344) rats 4 hr/day, 3 days/week for 40 weeks to O 3 at 150 ppb, nitric acid (HNO 3 ) at 50 µg/m 3 , or the combination, and meas ured the lung content of polyamines. O 3 , but not HNO 3 , increased lung putrescine, and the combined effect was 2fold greater than the O 3 effect. Both exposures also increased lung spermidine and spermine contents, but the combined effects were less than additive.
Two studies examined interactions between O 3 and aerosols of resuspended PM collected in Ottawa, Ontario, Canada. Vincent et al. (1997) exposed F344 rats for 4 hr to either 800 ppb O 3 ; 5,000 or 48,000 µg/m 3 of resuspended urban PM from Ottawa (EHC93); or to the combina tion of O 3 and each concentration of par ticles. They evaluated cell proliferation in the lung parenchyma and bronchiolar region by cell labeling. Only O 3 increased cell labeling when given alone. The combined effects were synergistic for labeling in the bronchioles but  Adamson et al. (1999) exposed F344 rats for 4 hr to either 800 ppb O 3 , 57,000 µg/m 3 PM, or to the combination and evaluated cell proliferation in the whole lung and alveolar duct region by cell labeling. As in the former study, only O 3 increased cell labeling when given alone, and the combined effects were approximately additive for wholelung label ing; however, the effect was clearly synergis tic (approximately 4fold) for labeling in the alveolar duct region.
Two studies examined interactions between O 3 and cigarette smoke, a complex mixture encountered in indoor and outdoor environ ments. Wu et al. (1997) exposed guinea pigs to a few breaths (two puffs) of diluted cigarette smoke followed by exposure to 1,500 ppb O 3 for 1 hr and then measured dynamic lung com pliance and resistance. The effects of smoke and O 3 alone were compared with the effects of the sequential combination. Compliance was reduced by both exposures, and the com bined effect was slightly greater than additive. Resistance was increased only by smoke when exposures were given singly, but the increase due to combined exposure was approximately 10fold greater than the effects of smoke alone. Yu et al. (2002) exposed B6C3F1 mice 6 hr/ day for 3 days to aged, diluted sidestream ciga rette smoke at 30,000 µg PM/m 3 and then to 500 ppb O 3 for 24 hr, or to the two pollutants alone, followed by measurement of centriacinar cell proliferation and markers of inflammation in BALF. Both exposures increased BALF neu trophils, and the combined effect was decid edly greater than additive. Neither exposure increased BALF lymphocytes, but the com bined exposure elicited a nearly 2fold increase over control levels. The effects on BALF tumor necrosis factor alpha were slightly synergistic, and the effects on BALF interleukin8 and cell proliferation were additive.
Five studies examined interactions between O 3 and endotoxin. Although endotoxin is not a commonly measured air pollutant, it is ubiq uitous in the environment and a frequent com ponent of environmental PM. Johnston et al. (2002) exposed C57BL6J mice for 24 hr to O 3 at 1,000 ppb, to aerosolized endotoxin at an esti mated lung dose of 37.5 EU, or to O 3 followed by endotoxin, and measured responses in BALF at 4 and 24 hr. At 4 hr, there were slightly syn ergistic increases in interleukin1β and inter leukin6, an additive effect on interleukin1Ra, and less than additive effects on interleukin1α and macrophage inhibitory factor. However, the effects on all of the markers were slightly to markedly synergistic at 3 days.
One group reported three different stud ies using identical sequences of exposure to O 3 followed by intranasal instillation of endotoxin, including groups receiving the individual treatments. Fanucchi et al. (1998) exposed F344 rats 8 hr/day for 3 days to 500 ppb O 3 followed by two daily intrana sal instillations of endotoxin and evaluated nasal tissues by immunohistochemistry and RNA analysis at 3 hr or 6 days. Endotoxin alone had little effect on the volume density of AB/PASstaining mucosubstances, but O 3 increased staining. The combined effect was somewhat synergistic at 6 hr and strikingly synergistic (> 6fold) at 3 days after exposure. The effect on expression of the rMuc-5AC gene was approximately additive at 6 hr, but synergistic (3fold) at 3 days. Using the same exposure protocol in two subsequent studies of F344 rats, Wagner et al. (2001aWagner et al. ( , 2001b demonstrated the repeatability of the syner gistic effect on nasal mucosubstances. In another study by the same group, Wagner et al. (2003) exposed F344 rats 8 hr/ day for 2 days to 1,000 ppb O 3 , preceded each day by intranasal instillation of 2 or 20 µg endotoxin, and included groups receiv ing the individual treatments. Three days later, they examined effects in the lung by analysis of BALF and changes in airway epithelium. The exposures produced synergistic increases in BALF neutrophils, mucin glycoprotein, and elastase, and also in intraepithelial muco substances and epithelial cell density in distal airways. Increases in BALF lymphocytes and macrophages were approximately additive, and the effect on epithelial cell density in proximal airways was less than additive.
One study examined interactions between O 3 and ovalbumin instilled intranasally. Wagner et al. (2002) exposed Brown Norway rats to 500 ppb O 3 for 1 or 3 days and instilled 50 µL 1% ovalbumin after each O 3 exposure. At 24 hr after the last treatment, they examined inflammatory and epithelial cell populations, the volume densities of intra epithelial mucosubstances, and cell prolifera tion rates in nasal epithelium. The effect of O 3 and antigen on epithelial mucosubstances was synergistic in the maxilloturbinates but less than additive in the septum. The effect on eosinophil influx was also synergistic in the maxilloturbinates but less than additive in the septum. The combined effects were less than additive for neutrophil influx, cell labeling, and epithelial cell density.

Summary, Discussion, and Conclusions
Our examination of studies cited in the 2006 O 3 Criteria Document (U.S. EPA 2006) con firmed that synergisms between O 3 and other pollutants have been demonstrated in labora tory studies involving humans and animals. Fourteen studies among the 13 human and 14 animal studies testing for synergy demon strated greater than additive effects for one or more outcomes. The copollutants in these studies included urban PM, cigarette smoke, H 2 SO 4 , HNO 3 , NO 2 , PAN, endotoxin, and antigen. These copollutants are all plausible classes of environmental air contaminants (accepting the neoantigen ovalbumin as a model for environmental proteinaceous anti gens). The additional studies not contained in the 2006 O 3 Criteria Document, but cited above as examples of synergy involving O 3 , included carbon black, virus, and airway ago nists. This limited review, therefore, identi fied diverse examples of synergies between O 3 and other air pollutants, as well as syner gies involving combinations of H 2 SO 4 with carbon particles, and diesel emissions with virus. Of course, environmental exposures involve much more complex mixtures than those typically used in the laboratory. The studies we reviewed involved only a very few of the myriad pollutants encountered in the environment in different combinations. Only the studies including tobacco smoke or die sel emissions approached a realistic level of complexity, and only Kleinman et al. (2000) used a factorial design to examine the single and combined effects of more than two treat ments (they included three).
Differences in dose and dose pattern are key caveats in extrapolating these laboratory findings to environmental exposures. Nearly all of the laboratory studies involved high concen trations that are not reflective of typical envi ronmental exposures. The studies of humans involved 2hr exposures to 450-485 ppb O 3 and concentrations of PAN and NO 2 much higher than ambient levels in the United States. The O 3 exposures of animals ranged from 150 to 1,500 ppb, and only the study by Sindhu et al. (1998) included exposures longer than 3 days. In that study, the exposure of rats 4 hr/day, 3 days/week for 40 weeks to 150 ppb O 3 ± 50 µg/m 3 HNO 3 used concentrations of O 3 only twice the current 1hr U.S. National Ambient Air Quality Standard (U.S. EPA 2008), but a concentration of HNO 3 many fold higher than is typical of the environment. The publications provide no indication as to whether any of these research groups have explored the dose-response relationships of their published findings down to environmental exposure levels. Unfortunately, the limitation posed by the range of exposure concentrations is a widespread issue in interpreting toxicology studies of air pollution and not only studies of combined exposures.
Existing information yields little indica tion of whether synergies observed in animals might also occur in humans, in animals of other species, in animals of other strains of the same species, or in subjects of different ages. Interspecies, interstrain, and agerelated differences exist in the uptake and metabo lism of inhaled materials and in the suscep tibility and sensitivity to adverse effects. The fact that more animal than human studies in our review demonstrated synergies can not be interpreted to mean that synergies are less likely to occur in humans. Our limited survey revealed no studies in which different species, strains, or ages were exposed to the same combinations of pollutants. Like the dose issue, the lack of systematic comparisons among research models is a prevalent limita tion throughout the air pollution literature.
The current evidence for synergism is restricted primarily to subclinical responses. The outcomes for which synergism was demonstrated in the studies described above encompassed a diverse spectrum of responses, including indicators of activated gene expression (e.g., Muc5AC), chemopro tective responses (e.g., lung polyamines), proinflammatory cytokines (e.g., interleu kins), frank inflammation (e.g., BALF neu trophils), alterations of cellular populations and functions (e.g., epithelial proliferation, increased mucosubstances), and alterations of organlevel function (e.g., cardiac out put, forced vital capacity). Although these effects reflect perturbations of response path ways that can contribute to health outcomes observed in population studies, only the last (lung function) has been commonly mea sured in the population studies undergird ing healthbased air quality regulations. It is possible that some synergisms occurring at intermediate steps in pathogenic pathways may not be manifest in outcomes at the clini cal level. For many examples of synergism we identified, it is uncertain whether the syner gism would be manifest at clinical and public health scales. Effects at subclinical levels have provided supportive evidence for associations between air pollutants and health, but air quality regulations have been based generally on mortality and morbidity outcomes that are a consequence of the integrated actions of myriad molecular and cellular biological responses. The interpretation of the impli cations of synergistic effects on subclinical homeostatic and pathologic responses raises the possibly insoluble dilemma of defining at what level measured responses constitute adverse effects (American Thoracic Society 2000). The present information offers only indirect evidence for synergies among pol lutants in morbidity or mortality caused or exacerbated by air pollutants.
Ideally, hypotheses related to synergism would be tested under realworld conditions in epidemiologic studies. However, several bar riers weaken the epidemiologic approach to investigating synergism. First, the exposures to the multiple pollutants of concern need to be estimated; this may prove difficult because of the complexity of estimating exposures to pol lutants as they vary spatially and temporally. Inaccuracies in exposure assessment will tend to decrease the sensitivity of a study for esti mating the degree of synergy. Second, unless a high degree of synergism is anticipated, study populations of substantial size are needed to characterize the combined effects of multiple agents because of the limited statistical power of the analytic methods used to assess synergism. It is impractical to test for synergisms in the popu lation as precisely as the laboratory permits. The composition of the environmental air pollution mixture varies in location and time, but the spatial-temporal variation generally does not provide the degree of exposure contrasts needed to compare effects of multiple pollutants alone and in combination. The effects of single pollut ants are generally inferred from such variations, but characterization of quantitative relationships is far more imprecise for the effects of combina tions of pollutants than for the effects of single pollutants. The precision of exposure estimates also poses a limitation. Accurate measures of personal exposures have only been linked to individual outcomes in panel studies of limited scope, and even then, no study has measured the full range of pollutant species to which the subjects were exposed. The weight of air pollu tion epidemiology necessarily rests largely on estimates of exposure derived from monitor ing data mandated by air quality regulations; thus, substantial data are available for only a few pollutants and pollutant classes.
In this review we identified a need for more rigorous reporting of findings of studies that consider synergism. The term is widely and loosely used. Claims of synergism should be accompanied by the needed quantitative assessment of the evidence and preferably by an assessment of the precision with which synergism has been demonstrated. Our deter mination of whether synergism was dem onstrated in the studies we examined was semiquantitative and often based on graphs rather than numeric data. Reports of statisti cal significance were limited to differences between effects of combined and sham expo sures or between combined and single expo sures; the statistical significance with which combined exposures had greater than additive effects was not reported.
The results of this review indicate that synergies among environmental air pol lutants are plausible and that experimental approaches may document their existence. It is both plausible and widely accepted that few, if any, effects of air pollution are attributable to single pollutants exclusively, although both epidemiologic and laboratory findings indi cate that single pollutants or pollutant classes can dominate certain effects. Nonetheless, research strategies have been driven largely by singlepollutant, singlesource regulatory frameworks, and thus have focused on detect ing and confirming the causality of single pol lutants or specific complex source emissions treated as a single exposure material. Very lim ited emphasis has been given to apportioning effects among the full spectrum of pollutants or evaluating pollutant interactions. A shift of the emphasis of air pollution health research toward a more comprehensive, forward thinking, multipollutant perspective would be timely in view of the increasing trend toward multipollutant regulatory strategies. Despite the limitations of laboratory and epidemio logic research tools, both approaches could plausibly be directed more toward a better understanding of the roles of a much broader spectrum of air contaminants, and thus their sources, to the health impacts associated statis tically with indices of air pollution. Although it is clearly impossible to study large numbers of pollutants using a full factorial study design, there are multiple strategies for disentangling the contributions of multiple pollutants in the laboratory (Mauderly 1993). The evaluation of synergies and antagonisms among pollutants, including their dose-response relationships, is a necessary foundation for progressing toward a multipollutant air quality management framework, as noted by the NRC Committee on Air Quality Management in the United States (NRC 2004b).
In summary, our selective review con firmed that synergisms (greater than additive effects) among air pollutants in causing mea surable biological effects have been demon strated in laboratory studies of humans and animals. The limited sample of studies encom passed by this review served to answer this question, and it might be expected that a more comprehensive review of the air pollution lit erature would reveal additional evidence. Our review also identified evidence for additive and less than additive responses to combined expo sures to multiple pollutants, sometimes among Appendix I. Potential interactions among pollutants.
Additivity: effect of the combination equals the sum of individual effects. Synergism: effect of the combination is greater than the sum of individual effects. Antagonism: effect of the combination is less than the sum of individual effects. Inhibition: a component having no effect reduces the effect of another component. Potentiation: a component having no effect increases the effect of another component. Masking: two components have opposite, cancelling effects such that no effect is observed from the combination.
"Effect" means the observed expression of the par ticular health outcome in question. A combination of pollutants could have different interactions for different outcomes. The interaction could occur at any level of biological pathway from exposure to expression of the outcome (U.S. EPA 2000).
volume 117 | number 1 | January 2009 • Environmental Health Perspectives different outcome measures in studies in which certain outcomes revealed synergisms, and sometimes among the same outcomes at different times after exposure. To our knowl edge, there has been no systematic review of all literature on the health effects of air pollution for evidence of pollutant interactions; thus, we have limited insight into the overall balance of the effects of the many possible combinations of pollutants. Although little, if any, evidence of synergisms has been developed at common environmental exposure levels, the possibility of both synergisms and antagonisms needs to be considered. In the absence of relevant evidence to the contrary, the assumption of additive effects appropriately remains the default for regulatory risk assessment purposes. However, considering the likelihood that both synergisms and antagonisms result from envi ronmental exposures and their importance to accurate assessments of risk, evaluations of pollutant interactions by both epidemiologic and laboratory research approaches will be critical for developing a stronger foundation for multipollutant air quality management.