1.1 Time series evidence

The time-series evidence on NO₂ has increased since the 2005 global update of the WHO air quality guidelines. Since 2004 (the cut-off date for studies included in the last WHO review), 125 new peer-reviewed ecological time-series studies on NO₂ have been published up to April 2011. These were identified using the Air Pollution Epidemiology Database (APED).³ Table 5 shows the total number of studies, according to health outcomes examined and the number of multicity studies available.

Table 5.

Summary of ecological time-series studies of NO₂ in APED.

The new studies were conducted mainly in the WHO Western Pacific Region (which includes China), Europe, the United States and Canada. The majority used 24-hour average concentrations of NO₂, measured mainly at urban background locations. A few studies published after the April 2011 cut-off for APED were included in the review, as they contain information on issues of relevance to the Question. These are not reflected in Table 5, but are discussed in the text which follows. A list of all time-series references considered in this project is available upon request. Only a selection of these references is discussed in the text below.

Given the crucial issue of understanding whether NO₂ has direct adverse effects on health, emphasis has been placed on time-series studies that investigated whether associations of NO₂ are robust to adjustment of concentrations of particles (and other pollutants). More than 65 time-series studies of NO₂ that used two-pollutant and/or multipollutant statistical models are available – only a proportion of these included adjustment for a metric of PM, and these are discussed in the sections that follow.

Mortality

In the 2005 global update of the WHO air quality guidelines, WHO concluded that daily concentrations of NO₂ are associated significantly with increased daily all-cause, cardiovascular and respiratory mortality within the range of concentrations studied; however, it also noted the reductions in the overall effect estimates in an important meta-analysis, following adjustment for PM (Stieb, Judek & Burnett, 2002, 2003). Since then, comprehensive reviews of the time-series literature on NO₂ have emerged with similar conclusions – that is, the short-term associations of NO₂ with mortality are suggestive of a direct effect, but there is some uncertainty about the causal nature of these associations (CARB, 2007; EPA, 2008b). In addition to these, a peer-reviewed research report of a comprehensive systematic review and meta-analysis of single pollutant model estimates (Anderson et al., 2007) reported that increases in NO₂ concentrations (per 10 µg/m³, 24-hour averages) are associated with increases in all-cause mortality: 0.49% (95% CI: 0.38–0.60%) in all ages and 0.86% (95% CI: 0.50–1.22%) for those older than 65 years of age. Results for maximum 1-hour average concentrations of NO₂ were lower: 0.09% (95% CI: -0.01–0.20%) and 0.15% (95% CI: 0.03–0.26%) in all-ages and for those older than 65 years of age, respectively. Increases in daily mortality, for all ages, for cardiorespiratory mortality (0.18% (95% CI: 0.08–0.27%), 24-hour average); cardiovascular mortality (0.34% (95% CI: 0.19–0.48%), maximum 1-hour average, 1.17% (95% CI: 0.82–1.53%), 24-hour average); and respiratory mortality (0.45% (95% CI: 0.21–0.69%), maximum 1-hour average, 1.76% (95% CI: 1.35–2.17%), 24-hour average) were also reported with NO₂. Anderson et al. (2007) also compared multipollutant model estimates for NO₂ from multicity studies and reported that consistent positive estimates for mortality (and hospital admissions) were found before and after adjustment for co-pollutants, with the size and precision of the estimates not being substantially reduced after such adjustment. The authors also concluded that these findings suggested that the short-term associations between NO₂ and health outcomes were unlikely to be confounded by other pollutant measures.

Twenty-four time-series studies of mortality, which used two-pollutant and/or multipollutant models for NO₂, have been published since the 2005 global update of the WHO air quality guidelines (Burnett et al., 2004; Dales et al., 2004; Kan, Jia & Chen, 2004a,b; Zeka & Schwartz, 2004; Simpson et al., 2005a; Díaz, Linares & Tobías, 2006; Samoli et al., 2006; Brook et al., 2007; Qian et al., 2007, 2010; Yamazaki et al., 2007; Chen et al., 2008; Hu et al., 2008; Ren Y et al., 2008c; Wong et al., 2008; Breitner et al., 2009; Chen et al., 2010a; López-Villarrubia et al., 2010; Park, Hong & Kim, 2011; Chiusolo et al., 2011; Chen et al., 2012a,b; Faustini et al., 2012; Chen et al., 2013).⁴ All of these papers included adjustments for a metric of PM; 17 of the 24 papers reported positive, though not always statistically significant, short-term associations of NO₂ with mortality for a range of diagnoses and age groups, after adjustment for a PM metric.

Multicity studies (of the aforementioned 24 studies), which included adjustment for particles in two-pollutant models, show robust short-term associations of NO₂ with increased all-cause, cardiovascular and respiratory mortality (Table 6), though some evidence of confounding (by black smoke) of the NO₂ association with respiratory mortality was identified in the European study, APHEA-2.⁵ Mainly PM₁₀ was used when controlling for particles in these multicity studies. In contrast, the large American multicity study, NMMAPS, did not find such associations between NO₂ and daily mortality: only a small and non-significant association of NO₂ with all-cause mortality was reported after adjustment in a two-pollutant model with PM₁₀; no association was found in a multipollutant model, which included PM₁₀, carbon monoxide, SO₂ and ozone (Zeka & Schwartz, 2004; see Table 6). Previous NMMAPS analyses of a subset of the 90 cities in the United States in this study found approximately a 0.7% (95% CI: 0.3–1.2%)⁶ increase in all-cause mortality per 45.12 µg/m³ (24 ppb)⁷ NO₂ at lag 1 (other lagged model results showed that the strongest association between NO₂ and all-cause mortality was identified at lag 1) (HEI, 2003). Following adjustment for PM₁₀ or other pollutants (O₃ and SO₂), the central estimates either increased or were unchanged (based on a plot of the estimates), though they lost statistical significance (HEI, 2003). The reason for the difference between NMMAPS and the other multicity studies is unclear. We note that the method used by Zeka & Schwartz (2004) differed from those used in the other studies, as it sought to deal with measurement error. The shape of the concentration–response function in the multicity studies was often assumed to be linear. Samoli et al. (2006) reported that their assumption of a linear relationship between maximum 1-hour concentrations of NO₂ and mortality was based on other results from APHEA-2 (Samoli et al., 2003), which suggested that it was appropriate to make such an assumption. Where tested, the relationship between NO₂ and all-cause mortality did in fact appear to be linear (Wong et al., 2008; Chen et al., 2012b).

Table 6.

Summary of two-pollutant model results for NO₂ from multicity time-series and case-crossover studies of mortality.

1.2 Panel studies

The short-term effects of NO₂ on respiratory health in children with asthma were recently reviewed (Weinmayr et al., 2010). The review is based on 36 panel studies published 1992–2006, of which 14 are from the Pollution Effects on Asthmatic Children in Europe (PEACE) project, all from 1998. In the meta-analysis of these studies, NO₂ showed statistically significant associations with asthma symptoms when considering all possible lags, but not when similar lags (0, 1 or 0–1) were evaluated from each study. The association with cough was statistically significant, but only when the PEACE studies were not considered (The PEACE study was negative. On the one hand it is the only multicentre study with a uniform protocol; on the other hand, concerns have been raised that the results were influenced by an influenza epidemic during the relatively short observation period (2 months)). The estimated effect on peak expiratory flow was statistically insignificant.

A PubMed search identified 11 new articles: 6 from the Americas (Sarnat et al., 2012; O’Connor et al., 2008; Escamilla-Nuñez et al., 2008; Liu et al., 2009; Castro et al., 2009; Dales et al., 2009), 2 from Europe (Andersen et al., 2008b; Coneus & Spiess, 2012), and 3 from Asia (Min et al., 2008; Yamazaki et al., 2011; Ma et al., 2008). All, except two, studies investigated school-aged asthmatic and/or symptomatic children, while the remaining two considered infants and toddlers (Andersen et al. 2008b; Coneus & Spiess, 2012). Most of the studies observed positive associations between short-term exposure to NO₂ (or nitrogen oxides) for different lags and respiratory symptoms, such as coughing and wheezing (O’Connor et al., 2008; Escamilla-Nuñez et al., 2008; Andersen et al., 2008b), as well as for exhaled nitric oxide (Sarnat et al., 2012) and also for pulmonary function decrease (O’Connor et al., 2008; Liu et al., 2009; Castro et al., 2009; Dales et al. 2009; Min et al., 2008; Yamazaki et al., 2011; Ma et al., 2008) in children, although not all of them were statistically significant (Dales et al., 2009). One article reported numerical data only for PM (Min et al., 2008).

The largest of these studies included 861 children with asthma from seven inner-city communities in the United States (O’Connor et al., 2008). A difference of 20 ppb in the 5-day average concentrations of NO₂ was associated with an odds ratio (OR) of 1.23 (95% CI: 1.05–1.44), for having a peak expiratory flow less than 90% of personal best. Similar risk elevations were found for decreased FEV1, cough, night-time asthma, slow play and school absenteeism. Multipollutant models showed independent effects for NO₂ after adjustment for ozone and PM_2.5 for most of these end-points.

In addition, some panel studies were published earlier, but were not included in the meta-analysis mentioned above (Svendsen et al., 2007; Delfino et al., 2006; Trenga et al., 2006; Moshammer et al., 2006; Mar et al., 2005; Rabinovitch et al., 2004; Ranzi et al., 2004; Boezen et al., 1999). These studies are not reviewed here, but are cited to illustrate, combined with the points made earlier, that a new combined analysis would contribute to a more precise quantitative estimation of the effects of short-term fluctuations in outdoor NO₂ concentrations on changes in pulmonary function and respiratory symptoms in asthmatic children, given the increased number of studies available. This might be performed by enlarging the existing review (Weinmayr et al., 2010) with the recently published panel studies. A new combined analysis could also consider panel studies performed on indoor exposure and respiratory effects – for example, Marks et al. (2010).

This section has concentrated on panel studies of respiratory health in children. There are also studies on chronic obstructive pulmonary disease in adults (Peacock et al. (2011) found effects of NO₂ weakened by control for PM₁₀) and on cardiovascular end-points. Panel studies with end-points relevant to mechanistic questions are discussed in the toxicology section.

Of particular interest is a recent study conducted in the Netherlands that attempted to isolate which component of the ambient aerosol could be related to various acute response endpoints in subjects exposed at five separate microenvironments, of contrasting source profile: an underground train station, two traffic sites, a farm and an urban background site. To date, only the results of the associations with respiratory end-points have been published, but this data suggested that interquartile increases in particle number concentration and NO₂ were related to decreased lung function (FVC and FEV1) – associations that were insensitive to adjustment for other gaseous pollutants and an extensive range of PM metrics, including black carbon, particle number concentration, oxidative potential and transition metals (Strak et al., 2012).

1.3 Chamber studies

Since 2005, the literature on human exposure to NO₂ has undergone a number of systematic reviews, as part of the EPA integrated scientific assessment for oxides of nitrogen (EPA, 2008b) and later by Hesterberg et al. (2009) and Goodman J et al. (2009). The human chamber study evidence was also reviewed by the California Environmental Protection Agency Air Resources Board (CARB, 2007) in their assessment of the California NO₂ standard. Since the publication of these reviews, limited NO₂ chamber studies that address lung function and airway inflammation have been performed, and two studies that address cardiovascular end-points have been published (Langrish et al., 2010; Scaife et al., 2012). Therefore, the conclusions that arise from these reviews remain valid for consideration of the current NO₂ standards.

The human chamber studies generally only address a single pollutant, but this can be helpful when it is unclear whether the health effects reported in the so-called real world are related to the pollutant itself, or a mixture of pollutants for which NO₂ serves as a surrogate. This is particularly true for NO₂, where there is an apparent mismatch between the human chamber exposure data (with mixed evidence of inflammation or impaired lung function between 0.26 ppm and 0.60 ppm) and the population-based epidemiology (with effects reported at lower concentrations within the ambient range). The results from the chamber studies are therefore useful in determining whether NO₂ should be considered toxic to the population in its own right or should be considered a tracer for local source emissions – that is, traffic in urban areas.

Human clinical studies generally examine healthy subjects, or patients (asthmatics, chronic obstructive pulmonary disease patients, allergic rhinitis patients, and patients undergoing rehabilitation following cardiac events) with relatively stable symptoms exposed to a single NO₂ concentration for durations of 1–6 hours. Results are generally compared against a control air exposure, with a range of end-points examined, including lung function, exhaled gases, airway inflammation (either by bronchoscopy or by induced sputum) and airway hyperresponsiveness. The results observed from these acute exposures are usually transient and, though often statistically significant, are seldom clinically so. Healthy subjects have been exposed acutely to NO₂ concentrations that range from 0.3 ppm to 2.0 ppm (1–4 hours); sensitive subgroups, including asthmatics and chronic obstructive pulmonary disease patients have been exposed to 0.26–1.00 ppm. Above 1 ppm, clear evidence of inflammation has been observed in healthy subjects in a number of studies (Helleday et al., 1995; Devlin et al., 1999; Pathmanathan et al., 2003; Blomberg et al., 1999), but the picture is less established at concentrations between 0.2 ppm and 1.0 ppm (Vagaggini et al., 1996; Jörres et al., 1995; Gong et al., 2005; Frampton et al., 2002; Riedl et al., 2012), partially because of inconsistent responses in the varying end-points examined.

A number of studies have suggested that NO₂ can augment allergen-induced inflammation in asthmatics (Barck et al., 2002, 2005) following short (15–30 minutes) consecutive day exposures at relatively low concentrations (0.26 ppm). However other studies at comparable NO₂ concentrations, but at a higher total dose (0.4 ppm for 3 hours), have failed to demonstrate a similar enhancement of airway inflammation or a reduced house-dust-mite allergen provocation dose in allergic asthmatics (Witten et al., 2005). Similarly, Vagaggini et al. (1996) were unable to demonstrate NO₂-induced upper airway inflammation in mild asthmatics, by nasal lavage and induced sputum 2 hours after an exposure to 0.3 ppm NO₂ for 1 hour. The evidence for inflammation in asthmatics exposed to NO₂ concentrations near environmental concentrations is therefore ambiguous.

Few studies have reported acute lung function changes near the current guideline concentration value (Bauer et al., 1986; Jörres et al., 1995) in the absence of a specific or nonspecific challenge (reviewed in Hesterberg et al., 2009; EPA, 2008b). In those studies that reported decrements, exposure concentrations were high (more than 1 ppm) and the magnitude of the observed changes was unlikely to be of clinical significance (Blomberg et al., 1999).

The reported effects of NO₂ on nonspecific and specific airway hyperresponsiveness have been reviewed in detail in the EPA integrated science assessment (2008b), as well as by Goodman J et al. (2009) and Hesterberg et al. (2009). In healthy individuals, small increases in nonspecific airway hyperresponsiveness have been reported following short-term (1–3 hour) exposure to NO₂ in the range 1.5–2.0 ppm (Mohsenin, 1987; Frampton et al., 1989). In asthmatics, the data on nonspecific hyperresponsiveness to NO₂ suggests a sensitizing effect between 0.2 ppm and 0.6 ppm, though the responses, where significant, were small and unlikely to be of clinical significance (Bylin et al., 1985; Mohsenin, 1987; Strand et al., 1996). A limited number of studies have evaluated airway hyperresponsiveness in asthmatics at multiple NO₂ concentrations, and the results of these studies do not support a clear concentration–response relationship between 0.1 ppm and 0.5ppm (Bylin et al., 1988; Roger et al., 1990; Tunnicliffe, Burge & Ayres, 1994). It should be noted that, in all of the studies cited, a range of responses were observed; and in many studies, subgroups of responders were identified. Given that chamber studies by their very nature rely on typically small numbers of healthy volunteers, or clinically stable patients, it is likely that they may underestimate the responses of sensitive subgroups within the population, especially in relation to disease severity.

Langrish et al. (2010) failed to demonstrate vascular dysfunction (vascular vasomotor or fibrinolytic function) in healthy volunteers exposed to 4 ppm NO₂ or filtered air for 1 hour. This is an important paper, as this result contrasts markedly with the group’s previous findings using diesel exhaust containing high concentrations of NO₂ (Mills et al., 2007). A subsequent paper, examining the vascular and prothrombic effects of diesel exhaust (300 µg/m³ for 1 hour), with and without inclusion of a particle trap, demonstrated that a reduction of particle number and mass concentration, in the absence of changes in nitrogen oxides, was associated with reduced adverse cardiovascular outcomes (Lucking et al., 2011). This was interpreted as strongly supporting the view that fine particles, and not NO₂, were driving the previously reported cardiovascular effects, especially as NO₂ concentrations increased almost five-fold with the particle trap. Further evidence, suggesting the absence of an acute cardiovascular effect of NO₂, was reported by Scaife et al. (2012). In this study they found no significant changes in heart variability parameters in 18 heart bypass and myocardial infarction patients following exposure to 400 ppb NO₂ for 1 hour.

1.4 Toxicological studies on short-term exposure (hours to days)

The WHO Regional Office for Europe indoor air quality guideline (2010) noted that acute exposures (hours) in the range of 0.04–1.0 ppm were rarely observed to cause effects in animals. The California Environmental Protection Agency Air Resources Board (CARB, 2007) noted acute effects (hours) in rats and mice at 0.2–0.8 ppm NO₂ (increased mast cells, quoted from Hayashi & Kohno (1985) in CARB, 2007)*¹¹ and increased synthesis of the carcinogen dimethylnitrosamine (Iqbal, Dahl & Epstein, 1981)* at 0.2 ppm. Also, there were: effects on liver detoxification enzymes at 0.25 ppm (Miller et al., 1980)*; effects on macrophages at 0.3–0.5 ppm (e.g. Robison et al., 1993)*; and increased bronchiolar proliferation at 0.8 ppm (Barth et al., 1994)*. The EPA (2008b) also highlighted the study by Barth et al. (1994). The effects at 0.2–0.25 ppm are harder to interpret (the mast cells may not have been degranulating, dimethylnitrosamine synthesis relied on an additional administered precursor and the effect on liver enzymes may not be adverse), but it is clear that the effects are adverse above 0.3 ppm. Tables in these reports quote other studies that report acute no effect levels from 0.4 ppm to 0.8 ppm.

A literature search, and literature abstracts already held from 2008 onwards (the literature cut-off for WHO Regional Office for Europe (2010)),¹² did not indicate any animal studies with short-term exposure to NO₂ alone that would change the CARB (2007) or EPA (2008b) view. Urea selective catalytic reduction-treated diesel engine emissions (0.78 ppm NO₂ and dilutions) were generally less toxic to rat lungs than conventional diesel engine emissions (0.31 ppm NO₂ and dilutions), for various endpoints over durations of 1, 3 or 7 days. However, there were differences in the nature of the oxidative stress produced with either increases in 8-hydroxy-2-deoxyguanosine with conventional diesel engine emissions, or increases in haemoxygenase-1 mRNA expression with urea selective catalytic reduction-treated diesel engine emissions (Tsukue et al., 2010). The latter suggests oxidative stress related to NO₂, but is not conclusive, given the mixture of constituents present.

Using exposures lasting from hours to days, recent mechanistic animal studies show: protein S-glutathionylation in the lung at 25 ppm NO₂ (Aesif et al., 2009); a decrease in aggregating activity of surfactant-protein D at 10 ppm or 20 ppm (Matalon et al., 2009); and increases in markers of oxidative stress, endothelial dysfunction, inflammation and apoptosis in the hearts of rats from exposures of 2.7 ppm to 10.6 ppm (Li H et al., 2011). Zhu et al. (2012) found that 2.6 ppm NO₂ delayed recovery from stroke (slowed reduction in infarct volume) in a rat stroke model and increased behavioural deficits. Dose-related endothelial and inflammatory responses were also found in the range 2.6–10.6 ppm. Channell et al. (2012) found that plasma from healthy adults exposed to 0.5 ppm NO₂ for 2 hours was able to activate cultured primary human coronary endothelial cells. This suggested that a circulating factor was mediating, at least in part, the endothelial response that may underly the cardiovascular epidemiological findings. One study in mice (Alberg et al., 2011) found no increased sensitization to intranasal ovalbumin at 5 ppm or 25 ppm NO₂ when diesel exhaust particles did show an increase, whereas another study in mice (Hodgkins et al., 2010) at 10 ppm NO₂ showed sensitization to inhaled ovalbumin due to increased antigen uptake by antigen-presenting cells. A study in mice suggested that effects in the lung due to 20 ppm NO₂ for 10 days are worse with a small increase in vitamin C dose than with a large increase (Zhang et al., 2010), perhaps due to vitamin C increasing NO₂ absorption into the epithelial lining fluid (Enami, Hoffmann & Colussi, 2009). All these concentrations (apart from Channell et al., 2012) are far in excess of the ambient concentrations linked to health effects in population studies, and the studies are not designed to show whether the mechanisms extend down to lower concentrations – that is, the mechanisms may or may not be relevant.

Recent studies indicate that the NO₂ radical can be directly involved in nitration of tyrosines (Surmeli et al., 2010, for example) and also that it causes a nitration-dependent cis-trans-isomerization to trans-arachidonic acid (linked to microvascular injury), described by the authors as a characteristic process for NO₂ (Balazy & Chemtob, 2008). This is not just relevant to the lung. While NO₂ itself is not absorbed systemically, as it is likely to react first, its reaction products, nitrite and nitrate ions, are found in the blood after NO₂ inhalation (Saul & Archer, 1983). Nitrite and nitrate anions in the blood are now regarded as carriers of nitric oxide (Lundberg, Weitzberg & Gladwin, 2008; Lundberg & Weitzberg, 2010; Weitzberg, Hezel & Lundberg, 2010).

The cycle whereby nitrate is excreted into the saliva and reduced to nitrite by oral bacteria and then converted to nitric oxide in the stomach or in the blood and tissues, after absorption of nitrite, may have a physiological role (to ensure nitric oxide production for vasodilation under hypoxic conditions when oxygen dependent nitric oxide synthases may fail), but may also have adverse consequences (Panesar, 2008). Nitrite can also lead to the NO₂ radical via peroxidase catalysed oxidation with hydrogen peroxide, via formation of nitrous acid, which dissociates to nitric oxide and NO₂, via formation of peroxynitrite from nitric oxide and superoxide and subsequent breakdown to the NO₂ and hydroxyl radicals and, less commonly, direct oxidation of nitric oxide (d’Ischia et al., 2011; Signorelli et al., 2011). In other words, there is an indirect transfer of inhaled NO₂ to the NO₂ radical in tissues via reactive intermediates.

There is a great deal of literature on reactive nitrogen species (Abello et al., 2009; d’Ischia et al., 2011; Sugiura & Ichinose, 2011), including in literature on atherosclerosis (Upmacis, 2008). The review by Upmacis (2008) describes the occurrence of protein nitrotyrosine in atherosclerotic plaques and also in the bloodstream and describes an emerging view that this is a risk factor for cardiovascular disease. A major proportion of the nitrotyrosine is thought to come via nitric oxide from inducible nitric oxide synthase (induced under inflammatory conditions), but the source of the remainder is unknown. Nitrite and nitrate in the blood from NO₂ inhalation provides a potential link with this literature. Human beings exposed to 0.1 ppm NO₂ by inhalation have been inferred to form about 3.6 mg nitrite per day, more than the dietary intake of nitrite (Saul & Archer, 1983). More work is needed, however, to understand how significant any contribution of NO₂ inhalation to systemic nitrative stress might be in quantitative terms.

Although this is a section on toxicology, some epidemiological studies are considered here as they concentrate on mechanisms. They provide a potential link between the mechanisms in animal studies discussed above and mechanisms operating in human beings, although they lose the advantage of confident specificity for NO₂. These studies have mostly addressed cardiovascular end-points: no effects on blood coagulation and inflammatory markers (Zuurbier et al., 2011b; Steinvil et al., 2008) or a non-significant increase (significant for sP-selectin) (Delfino et al., 2008); increases in brachial artery diameter and flow mediated dilatation (perhaps actually due to nitric oxide) (Williams et al., 2012); adverse effects on heart rate variability in heart disease patients (Zanobetti et al., 2010) and in cyclists (Weichenthal et al., 2011) (independent of PM metrics); a non-significant increase in the QT interval in electrocardiograms (significant in diabetics) (Baja et al., 2010); and an increase in lipoprotein-associated phospholipase A2 (linked to inflammation in atherosclerotic plaques) (Brüske et al., 2011).

2.1 Long-term epidemiological studies

Since the publication of the 2005 global update of the WHO air quality guidelines, a large number of new studies on long-term effects related to NO₂ and other traffic pollutants have been published. To prepare a response to the question on new evidence, systematic literature searches were performed for original articles published 2004 to April 2012 on long-term studies (cohort studies, cross-sectional studies, case-control studies) that investigated outdoor NO₂ exposure and mortality or diagnosed diseases or lung function. We did not look at other physiological markers and at birth outcomes, as it is not clear whether they are really consequences of long-term exposure, and there were too few studies meeting the criteria. The searches in the literature databases PubMed, ISI Web of Science and Ludok resulted in 160 publications: on mortality or specific death causes (n = 37); on lung function (n = 34); on incidence or prevalence of cardiovascular diseases (n = 19); on diabetes mellitus (n = 6); on asthma (n = 46); and on bronchitis (n = 18).

In addition, two review reports from the California Environmental Protection Agency Air Resources Board and the EPA published in 2007 and 2008, respectively, have been reviewed.

1. Review of the California Ambient Air Quality Standard for NO₂, 2007. The following recommendation was released for a NO₂ annual-average standard: establish a new annual average standard for NO₂ of 56 µg/m³ (0.030 ppm), not to be exceeded. It was based on evidence from epidemiological studies showing that long-term exposures (that is, one or more years) to NO₂ may lead to changes in lung function growth in children, symptoms in asthmatic children, and preterm birth (CARB, 2007).
2. EPA integrated science assessment for oxides of nitrogen – health criteria, 2008. This document rated the epidemiological and toxicological evidence, examining the effect of long-term exposure to NO₂ on respiratory morbidity, as suggestive, but not sufficient to infer a causal relationship. The document also rated the evidence on other morbidity or mortality as inadequate to infer the presence or absence of a causal relationship. It found the relationship between long-term exposure to NO₂ and mortality to be inconsistent. Furthermore, when associations were noted, they were not specific to NO₂, but also implicated PM and other traffic indicators.

Long-term studies generally investigate geographic differences in NO₂ exposure for a period of several years. The spatial distribution of NO₂ shows a high variation, as it varies mainly with traffic – the most important NO₂ source in European countries. Therefore, individual assessment of exposure is much more important than the assessment of more homogeneously distributed PM_2.5 or (often) PM₁₀. So, the spatial resolution of the exposure assessment is crucial. In contrast to particle mass, black smoke (or black carbon) often shows a similar spatial distribution to NO₂. The answer to the question about whether long-term effects are due to NO₂ per se should ideally be based on studies that include NO₂ together with a particle mass indicator (on the one hand) and an indicator such as black carbon or ultrafine particles (on the other hand).

Some of the newly published studies were ecological in nature (lacking individual covariates) or did not give numerical results for NO₂; and some had been superseded by more recent analyses with more data and longer follow up. These studies were excluded. As an indicator of long-term traffic exhaust, 31 of the studies modelled exposure to NO₂ or nitrogen oxides and did not include a particle measure (14 of them investigating asthma). In these cases, it is not possible to attribute the resulting effects to NO₂ per se, so they are not discussed in detail here. A total of 81 studies analysed NO₂ and a particle measure in parallel, most of them investigating mortality, asthma, or lung function.

Some of these studies will be highlighted below, according to the following criteria:

• cohort studies with individual follow-up, as their validity is higher than cross-sectional studies or studies based on so-called ecological units; and
• studies with a contemporary evaluation of a PM fraction, preferably PM_2.5 or black carbon.

Meta-analyses that include studies with and without a contemporary evaluation of a PM fraction and with a comparison with PM are also highlighted. A tabulation of the full set of studies is available on request.

The following discussion focuses on respiratory health in children and long-term mortality studies, as these are the outcomes with the largest number of investigations and with results relevant to possible standard settings.

Summary of mortality-related to long-term exposure

A total of 18 cohort studies and two case-control studies published since 2004 provide RR estimates for associations of natural mortality, cardiovascular and respiratory mortality or lung cancer with NO₂ and PM. With the exception of studies conducted in China, most investigations were conducted in areas where the average NO₂ levels were below 40 µg/m³. Not all of the studies presented correlations between NO₂ and other pollutants, but those that did indicated moderate to high correlations; in European studies, the correlation is typically greater than 0.80. We focus mainly on the four cohort studies with multipollutant models and the five European cohort studies that analysed NO₂ and a particle metric – but without estimating two-pollutant models – giving details also in Tables 7 and 8.

Table 7.

Cohort studies with multipollutant models for NO₂ and a particle measure and mortality risk.

Table 8.

Cohort studies from Europe on long-term mortality risk and NO₂ and PM without multipollutant models.

The following European studies are relevant to the evaluation.¹⁴

Recently, the results from a cohort of more than a million adults in Rome were published (Cesaroni et al., 2013). Long-term exposures to both NO₂ and PM_2.5 were associated with increased risks of nonaccidental mortality. The strongest association was found for ischaemic heart diseases, but also mortality for cardiovascular diseases and lung cancer were significantly associated with both pollutants. Respiratory mortality was marginally associated with NO₂ and not associated with PM_2.5. The cohort was built on administrative data and lacked information on such behavioural risk factors as smoking (smoking was available on a subset of the cohort and no confounding from this factor was suggested in a sensitivity analysis). The models were therefore adjusted for pre-existing conditions that share lifestyle risks (diabetes, chronic obstructive pulmonary disease and hypertensive heart disease) and for individual and small-area socioeconomic position. The average exposure estimated for individual address for the follow-up (the mean follow-up duration was 8.3 years) was 43.6 µg/m³ (range: 13.0–75.2 µg/m³) for NO₂ and 23.0 µg/m³ (range: 7.2–32.1 µg/m³) for PM_2.5. The functional form of the association showed no evidence of deviation from linearity for non-accidental mortality, cardiovascular mortality, respiratory mortality and lung cancer, but it showed some deviation from linearity for the association between NO₂ and ischaemic heart disease mortality. In a two-pollutant model, the estimated effect of NO₂ on non-accidental mortality was independent of PM_2.5.

In a French study (Filleul et al., 2005) a comparison of mortality over a period of 25 years between 18 areas, including both NO₂ and black smoke, revealed positive and statistically significant effects of NO₂ – concentrations averaged between 12 µg/m³ and 32 µg/m³ over a period of 3 years for natural, cardiopulmonary and lung cancer mortality. The effects of black smoke were lower than those observed for NO₂. Six of the twenty-four areas with a NO₂ average from 36 µg/m³ to 61 µg/m³ were excluded from the analyses, as the monitors were not considered representative of the population’s exposure (assessed with a high nitrogen oxide-to-NO₂ ratio, indicating localized traffic sources). In Germany, Heinrich et al. (2013) followed the vital status of women with baseline NO₂ and PM₁₀ exposure values for more than 18 years (SALIA study). Positive effects were found for all-cause and cardiopulmonary mortality in relation to NO₂ (median exposure: 41 µg/m³ in the year before baseline investigation, based on the annual average of the next monitoring station)) and for all-cause and cardiopulmonary mortality and for lung cancer mortality in relation to PM₁₀. The association between cardiopulmonary mortality and PM₁₀ was reduced in this extended follow-up period, during which PM₁₀ concentrations (but not NO₂ concentrations) were lower, compared with an earlier analysis, after 12 years (Gehring et al., 2006).

In contrast to the analyses of Gehring et al. (2006) and Heinrich et al. (2013), which investigated all-cause and cardiopulmonary mortality, the analyses of the SALIA-study data by Schikowski et al. (2007) investigated cardiovascular mortality risk in relation to NO₂ and PM₁₀, and a possible effect modification in women with impaired lung function or pre-existent respiratory diseases. The significantly elevated mortality risk from cardiovascular causes in relation to NO₂ (median exposure: 46 µg/m³) was not higher in the susceptible subgroup than in the whole sample. In an analysis of all inhabitants of Oslo, Norway, Naess et al. (2007) reported that the RR estimates for cardiovascular diseases, respiratory diseases and lung cancer were very similar for NO₂ and PM_2.5 when evaluated across the quartiles of the two pollutants. The Norwegian study also investigated the functional form of the association. The authors give separate results for the two age groups of 51–70 years and 71–90 years. In the younger age group, they observed a threshold for the effect on overall mortality of about 40 µg NO₂/m³ due to the results for cardiovascular and lung cancer mortality. Thresholds were also observed for PM_2.5 (about 14 µg/m³) and PM₁₀ (about 19 µg/m³). In the older age group, the increase in overall mortality risk was linear in the interval 20–60 µg NO₂/m³. For chronic obstructive pulmonary disease, a linear effect was seen for both age groups. The large Dutch study by Beelen et al. (2008a) evaluated several pollutants, including NO₂, PM_2.5 and black smoke. The effect sizes for natural, cardiovascular, and respiratory mortality were greater for NO₂ (population average: 36.9 µg/m³; SD: 8.2 µg/m³) than for PM_2.5 (population average: 28.3 µg/m³) and were similar to those for black smoke. The RRs per 30 µg NO₂/m³ were 1.08 (95% CI: 1.00–1.16) for natural mortality, 1.07 (95% CI: 0.94–1.21) for cardiovascular mortality and 1.37 (95% CI: 1.00–1.87) for respiratory mortality; the RRs per 10 µg PM_2.5/m³ were 1.06 (95% CI: 0.97–1.16) for natural mortality, 1.04 (95% CI: 0.90–1.21) for cardiovascular mortality and 1.07 (95% CI: 0.75–1.52) for respiratory mortality. The spatial correlation between different pollutants was high.

For cardiopulmonary mortality, the large American Cancer Society cohort study found no or a marginally significant association with NO₂ (a 1% increase per 18.8 µg NO₂/m³ in 1980 (95% CI: 0–2%)) and a marginally increased risk of ischaemic heart disease mortality (a 2% increase (95% CI: 0–3%) at a median exposure of 49 µg NO₂/m³). The study compared mortality and pollution between populated regions and was not designed to investigate smaller scale variations in pollution (Krewski et al., 2009). By contrast, a cohort study in Toronto, using land use regression to predict NO₂ at the residential address and an interpolation method to predict PM_2.5, did find a large effect on natural and cardiovascular mortality for NO₂, but not for PM_2.5 (Jerrett et al., 2009b). Therefore, the authors modelled NO₂ together with traffic proximity (and not with PM_2.5), and the effects for NO₂ were slightly weakened, but still significant (from 1.40 (95% CI: 1.05–1.86) to 1.39 (95% CI: 1.05–1.85) per interquartile range of 7.6 µg NO₂/m³). The median of individually assessed NO₂ exposure was 43.05 µg NO₂/m³ (25^th percentile: 39.1 µg NO₂/m³; 75^th percentile: 38.46 µg NO₂/m³).

A new analysis of the California data from American Cancer Society Cancer Prevention Study II (Jerrett et al., 2011) used several different methods to model exposures to air pollution with high spatial resolution. It found, in a cohort of more than 76 000 adults with 20 432 deaths in the follow up period from 1982 to 2000, consistent associations of PM_2.5 with mortality from cardiovascular causes, comparable with the findings from the national American Cancer Society Cancer Prevention Study II. However, the strongest associations were found for NO₂. Individual exposure data were derived from a land use regression model of NO₂ that predicted local variations in the exposure of participants in the years 1988–2002 (mean of individual level exposure 12.3 ppb or 23.1 µg/m³). The authors stated that NO₂ is generally thought to represent traffic sources and concluded that combustion-source air pollution was associated with premature death. The analyses are published as a report for the California Environmental Protection Agency Air Resources Board and are not (yet) published in a peer reviewed journal. The report does not give effect estimates with confidence intervals for the two-pollutant models.

Only a few of the studies reviewed conducted a formal multipollutant analysis designed to differentiate the specific effects of the pollutants. The large registry cohort study in Rome found associations of NO₂ with non-accidental mortality, independent of PM_2.5 and also independent of traffic density or distance to the next main road (Cesaroni et al., 2013). The analysis of the above-mentioned California data of the American Cancer Society Cancer Prevention Study II observed that, in two-pollutant models of NO₂ with PM_2.5, both pollutants were associated with significantly elevated effects on mortality from cardiovascular disease and ischaemic heart disease. NO₂ was also independently associated with an elevated risk for premature death from all causes and lung cancer (Jerrett et al., 2011), but the report did not provide the adjusted quantitative effect estimates.

Hart et al. (2011) examined the association of ambient residential exposure to PM₁₀, PM_2.5, NO₂, and SO₂ with mortality in 53 814 men in the United States trucking industry. One of the unique features of this study was the ability to model the mortality effects of exposures to multiple pollutants. In the multipollutant models, adverse effects were predominantly seen with exposures to NO₂ and SO₂, and they were reduced for PM₁₀ and PM_2.5. A weakness of the study was that there was no information on other risk factors for mortality, such as cigarette smoking. A subsample with information on smoking showed some correlation with pollution, and a recalculation, adjusting for smoking on that basis, suggested that the hazard ratio would be reduced in size but still remain – that is, it did not affect their qualitative conclusions. Similar findings are available from a Canadian cohort study (Gan et al., 2011), where the effects on mortality were stronger for black carbon and NO₂ than for other pollutants; the study used a multipollutant model (with NO₂, PM_2.5, and black carbon in the same model) that attenuated the NO₂ effect, but did not reduce it to null (RR = 1.03, 95% CI: 0.99–1.07). It should be emphasized that the effect estimates for black carbon were robust against adjustment for NO₂, whereas the effect estimate for NO₂ was reduced to non-significance after adjustment for black carbon.

Overall, the findings suggest that traffic and other sources of fossil fuel combustion are important pollution sources that result in greater overall lung cancer, cardiovascular disease and respiratory disease mortality in the cohorts. In contrast to the cardiovascular mortality risk found to be associated with long-term exposure to NO₂ in the above studies, such an association was not found consistently in studies on the incidence or prevalence of cardiovascular disease. Only seven studies with cross-sectional, case-control or cohort designs investigated the relationship of these outcomes with NO₂ and particles in parallel. Five of those did not find any association or just a small insignificantly higher risk; two recent publications found an increased risk for ischaemic heart disease and for heart failure, but not for other cardiovascular diagnoses (Beckerman et al., 2012; Atkinson et al., 2013).

In summary, the European cohort studies provide evidence that the NO₂ effects on natural and cause-specific mortality are similar to, if not larger than, those estimated for PM. The recent registry cohort study from Italy (Cesaroni et al., 2013) and the American (Jerrett et al., 2011; Hart et al., 2011) and Canadian (Gan et al., 2011) studies have attempted multipollutant models, and they provide support for an effect of NO₂ independent from particle mass metrics. In three of these mortality studies with multipollutant models, the major fraction of the populations studied was exposed to NO₂ levels lower than 40 µg/m³; in one of them, nearly all participants were exposed to levels lower than 40 µg/m³ (Jerrett et al., 2011). Four of the six European analyses were centred around 40 µg NO₂/m³. In the French study, areas with (possibly non-representative) monitor averages above 32 µg NO₂/m³ were excluded. The study by Naess et al. (2007) looked at non-linear exposure–response functions and found a possible threshold around 40 µg/m³ for NO₂ and also a threshold for particles in the age group of 51–70-year-old people, especially for cardiovascular mortality. In contrast, they found no thresholds in the age group of 71–90-year-old people for all cause and cardiovascular mortality and none for chronic obstructive pulmonary disease mortality in either age group. The Italian study found some evidence for non-linearity in the association between NO₂ and ischaemic heart mortality, but not for cardiovascular or non-accidental mortality. All other investigators applied linear exposure–response functions.

As the long-term mortality studies have all included populations exposed in part to annual average NO₂ concentrations of well below the current WHO air quality guidelines of 40 µg/m³, or even been conducted over a range almost entirely below the air quality guidelines, it would be wise to consider whether the guideline should be lowered at the next revision of the guidelines.

2.2 Sub-chronic and chronic toxicological studies

Animal studies

Since the 2005 global update of the WHO air quality guidelines, the toxicological evidence up to about 2007 has been described in several reports (EPA, 2008b; CARB, 2007; COMEAP, 2009b; WHO Rgional Office for Europe, 2010). Table 9 summarizes the lowest effect concentrations at the lower end of the dose range, as highlighted in the conclusions sections of the various reports. Such studies as Mercer et al. (1995), highlighted by the EPA, that included spikes of higher concentrations have been excluded. It can be seen that there are now four studies showing effects below 0.34 ppm, including older studies not previously quoted by WHO, marked with an asterisk. Tabacova, Nikiforov & Balabaeva (1985) did not describe maternal toxicity,*¹⁵ so it cannot be determined whether the neurobehavioural posture and gait changes are a direct toxic effect of NO₂ on the offspring or an indirect effect, via toxicity to the mother. The study has not been replicated by other studies in this dose range, although similar effects have been found at much higher doses. The effects in the studies by Zhu et al. (2012) (reviewed in more detail below) and Takano et al. (2004)* are on risk factors for ischaemia and atherosclerosis, rather than actual occurrence of ischaemia and atherosclerosis themselves. Also, the change in endothelin-1 (Zhu et al., 2012) was in an unexpected direction, a direction that has not yet been replicated in another study. These three studies should not be dismissed, but there is an element of uncertainty in using them to describe a firm lowest effect concentration.

Table 9.

Lowest effect concentrations highlighted by WHO, EPA and CARB at low end of dose range^a.

Short-term exposure

In comparing the epidemiological and the toxicological and chamber study evidence for the concentrations at which effects occur, it is important to realize that, in the epidemiological studies, the daily variability at the background sites also reflects the daily variability at hot spots, so that the actual concentrations inducing health effects may be higher than those measured at the background site. Kerbside concentrations can be regularly in the range 380–560 µg/m³ (200–300 ppb) (1-hour average) at some polluted sites (London Air, 2013), and a wider range of sites often exceed the 1-hour limit value of 200 µg/m³ (EEA, 2012). This is within (or approaching) the range of concentrations that result in small effects in the chamber studies (380–1160 µg/m³; 200–600 ppb). Of course, it depends how long people are in those locations but, in addition to such activities as queuing at bus stops, in-vehicle exposure can be similar to the outdoor concentrations in heavy traffic (Chan & Chung, 2003), and car journey durations in congested cities can be considerable. Personal exposures to NO₂ reflect the sum of the microenvironment concentrations that an individual moves through over a determined time interval and are, therefore, strongly influenced by high concentration environments.

Concern has been expressed in some studies that ambient concentrations of NO₂ are not correlated with personal exposures to NO₂ and are, actually, better correlated with personal exposure to PM_2.5 (e.g. Sarnat et al., 2001). However, a recent systematic review addressed the issue of the relationship between personal and ambient NO₂ exposures, identifying significant correlations between these two exposures estimates, though the strength of the association varied across studies (Meng et al., 2012). Studies not quoted by Meng et al. (2012) also varied, both finding (Rijnders et al., 2001) and not finding (Bellander, Wichmann & Lind, 2012) significant correlations.

It is difficult to extrapolate quantitatively from animal toxicological studies. Modelling (with many assumptions) suggests about a tenfold higher local NO₂ concentration in the bronchi of humans than in those of rats, at the same concentration in the air that is being inhaled, and vice versa (much lower) in the alveoli (Tsujino, Kawakami & Kaneko, 2005). Several of the respiratory effects in animals occur in the bronchi, and it is plausible that some of the epidemiological results in human beings (such as asthma admissions) are primarily the result of effects on the bronchi. There may, therefore, be some evidence that supports application of the tenfold safety factors used in traditional toxicology for extrapolation from animals (strictly rats) to human beings. This, together with the distribution of personal exposures and the wide range of susceptibility in the human population, suggests that the epidemiological findings are not necessarily incompatible with the toxicological evidence on NO₂ itself. There are too few studies that examine NO₂ and particles from defined sources in the same experimental system, to directly compare the toxicological importance of these two pollutants, and their relative toxicological importance may vary by end-point.

The apparent mismatch between the time-series evidence and the lack of apparent responses in the chamber studies at background concentrations may be a consequence of only a small proportion of the population responding at particular times, an effect that could be picked up only in the much larger samples used in time-series studies. Specifically, more sensitive groups, such as severe asthmatics, are not studied in chamber studies because of the risks involved. Thus, the lack of robust effects and/or the lack of evidence at lower doses in chamber studies is insufficient to rule out the reported associations with NO₂ in the time-series studies found at the concentrations present in the wider environment. Considering the presence of more sensitive subgroups in the population together with the higher concentrations at microenvironments (such as the kerbsides described above) could explain some of the apparent mismatch, a point also made by Frampton & Greaves (2009).

While the preceding discussion supports the causality of the short-term respiratory effects, the issue is more uncertain for the short-term cardiovascular effects. Positive associations robust to adjustment for PM mass metrics are found in many time-series studies for cardiovascular mortality. The new evidence continues to suggest positive associations between NO₂ and cardiovascular hospital admissions, but findings are mixed in terms of robustness of the associations after adjustment for co-pollutants. It is therefore difficult to comment further on the nature of the relationship between NO₂ and cardiovascular hospital admissions. The results of panel studies on markers of cardiovascular risk (discussed in the toxicology section) found some interesting results in some cases, but no effects in others. Of the few chamber studies available, most did not find effects on cardiovascular end-points. The few short-term animal studies available did find effects on markers of systemic oxidative stress, inflammation and endothelial dysfunction at above ambient concentrations, but without a defined no-effect level. One study found that plasma from volunteers exposed to NO₂ at 0.5 ppm had an effect on coronary epithelial cells in vitro. Although slightly longer term (3 months), the study of modern diesel emissions, more highly dominated by NO₂, did not find evidence of cardiovascular effects in healthy young animals. There are theoretical mechanisms by which NO₂ inhalation could cause increased nitrative stress in the diseased heart. The overall picture is mainly one of an absence of a sufficient volume of evidence to resolve the mixed results found in the studies available, so it adds uncertainty, rather than challenging the causality of the epidemiological associations.

The current short-term guideline has the advantage of being set on the basis of chamber studies that use NO₂ itself. However, there is now a large body of time-series and panel evidence that, in contrast to other pollutants, has not been used so far in determining the level of a short-term guideline. It is recommended that this evidence be included in future considerations of a short-term guideline for NO₂.

Long-term exposure

Another challenging issue is whether the effects of long-term exposure are due to NO₂ per se.

Human chamber studies are not suitable for investigating long-term exposures. While animal toxicological studies do provide evidence of long-term effects, the degree to which this applies at ambient concentrations is less clear. In studies that compare long- and short-term exposures, the effects of long-term exposures occur at lower concentrations (0.25 ppm for clear adverse effects and some possible effects below this (Table 9)). However, while this concentration can be found at kerbsides, it is less likely that people are exposed to these concentrations several hours a day long-term (the long-term toxicological studies were often not based on exposure for 24 hours a day) than it is for people to be exposed for the shorter durations needed for the short-term effects. The toxicological evidence now includes a few studies that show cardiovascular effects, but the evidence is too limited for firm conclusions. It is much harder to judge the robustness to adjustment in the long-term epidemiological studies than it is in the short-term studies, because the spatial correlations between NO₂ and other pollutants are often high. However, there are a few studies that do suggest effects independent of particle mass metrics (including studies on cardiovascular mortality). The existence of effects of short-term exposure provides some plausibility for the effects of long-term exposure – particularly, for respiratory effects.

In summary, there is also some support, to a lesser degree than for the short-term, for a long-term effect of NO₂ per se. Again, NO₂ may also be capturing the effects of other traffic-related pollutants.

The current long-term guideline developed historically from one based on indoor studies. Now, there are enough outdoor air pollution studies (those described here plus pre-2004 studies) on respiratory effects and on all-cause mortality to consider using them to help set a new long-term guideline, with appropriate caveats about uncertainties. These studies included exposures to concentrations above and below, or all below, the current guideline and, where studied, linear relationships without a threshold have been found for at least some outcomes. This suggests that consideration should be given to lowering the guideline.

How to reflect these uncertainties in regulations is a matter beyond the scope of WHO, but consideration could be given to flagging the guideline to emphasize uncertainty.

Research gaps will be highlighted later in answers to Question C9, but it should be emphasized – particularly for NO₂ – where much of the important chamber and toxicological evidence comes from 20–30 years ago. Information on mechanisms is crucial and needs to take advantage of the full range of modern experimental techniques, including systems biology, to capture interacting causal pathways.

3.1 NO₂ per se

When both PM metrics and NO₂ measures are used in health impact assessments, some of the effects attributed to each of them may overlap. So, estimates from two-pollutant models should be used to attempt to avoid this. Unfortunately, very few long-term studies provide estimates from two-pollutant models (see Question C2), often because NO₂ and PM were too closely correlated. In addition, papers may not perform two-pollutant models when one of the pollutants did not show a significant association.

Studies that examine NO₂ using exposures based on a few monitoring sites as a broad surrogate for personal exposure may have lower correlations with primary particle metrics. But this advantage may be offset by the exposure measure being less representative of individual personal exposure (measurement error) and, therefore, showing weaker effects in two-pollutant models, independent of a possible so-called true association. Correlations with PM mass metrics may be higher in such studies. This illustrates the point that the spatial scale of the study can influence the interpretation of the concentration–response function to be used in the health impact assessment.

Black carbon, elemental carbon or black smoke, and carbon monoxide, being alternative measures to capture traffic exhaust effects, share with NO₂ the aspect of traffic proximity and association with emissions from combustion. They are often more closely associated with NO₂ than with particulate mass. It is self-evident that the results of an impact assessment for those indicators should not be added with the results for NO₂, if single-pollutant models are being used.

The health outcomes associated most consistently with long-term exposure to NO₂ are mortality, respiratory symptoms or asthma, and lung function. Cardiovascular outcomes are more uncertain – not because there is a large body of evidence showing a lack of effects, but because there are relatively few studies and there is less support from mechanistic evidence. The degree to which these associations are a consequence of NO₂ per se was discussed in Question C2, and the points relevant to choices of concentration-response functions are highlighted below for each end-point.

Children’s respiratory symptoms and/or asthma symptoms. As discussed in Question C2, there are several cohort studies that show the effects of long-term exposure to NO₂ on the respiratory condition of children. Several of these effects were related to lung function, which is a difficult end-point to use in a cost–benefit analysis, as it does not necessarily correspond exactly with symptoms and is therefore hard to give a value in monetary terms (The end-point could be used in cost–effectiveness analysis). Nonetheless, the finding in some studies of an effect on lung function that is stable to adjustment for other pollutants provides some support for the studies that examine respiratory symptoms. Publications from the Southern California Children’s Health Study include a paper on bronchitic symptoms in asthmatic children; the paper used multipollutant models for a variety of PM metrics, as well as NO₂ and ozone (McConnell et al., 2003). While the effects of NO₂ and organic carbon were difficult to separate, associations with NO₂ were found both across communities and across different years within communities, whereas associations with organic carbon were only robust in the latter case. An adjusted coefficient from this study could be used, although the end-point of bronchitic symptoms in asthmatics, as a result of year-to-year variations in annual average NO₂, is rather unusual, and there are an insufficient number of studies with adjusted coefficients to perform a meta-analysis. It is difficult to judge whether to use this coefficient for a central analysis. For the evidence currently available, it is the least uncertain of the possible concentration–response functions, as it has an adjusted coefficient and is for a respiratory outcome. Although it is only one study, it is supported in a general way by epidemiological long-term exposure studies that show effects on lung function. For now, before further studies are available, we suggest the use of the adjusted coefficient in a central analysis (item d(i)) above, with acknowledgement of the greater degree of uncertainty (see the summary in section 4 below), compared with other central analysis effects.

There are meta-analyses available for NO₂ and asthma incidence (Anderson, Favarato & Atkinson, 2013b) within communities (positive) and for NO₂ and asthma prevalence across communities (no associations) (Anderson, Favarato & Atkinson, 2013a). There are also several studies of NO₂ and asthma prevalence within communities (most positive) that would be suitable for a meta-analysis. Annual prevalence would be preferable for quantification, particularly as the asthma incidence meta-analysis was intended only for hazard assessment and, thus, did not control for the length of the follow-up period. As correlations were so close, multipollutant models were not available. Nonetheless, these studies could be used to provide coefficients for a sensitivity analysis that compares the possibilities of the effect being entirely due to black smoke, entirely due to NO₂, or (more likely) somewhere in between (item d(iii) above).

Mortality. There is now a good body of evidence on NO₂ and all-cause mortality in within-community cohort studies. Unfortunately, there is no recent meta-analysis, and only a few of the studies could apply two-pollutant models (see Question C2). A large cohort study investigating multipollutant models with NO₂ (Jerrett et al., 2011) did not yet provide exact estimates with confidence intervals for the two-pollutant model NO₂–PM_2.5. However, the recent large Rome Longitudinal Study (Cesaroni et al., 2013) provided effect estimates for NO₂-related natural mortality, while adjusting for PM_2.5 in a two pollutant model. In addition, the Canadian study (Gan et al., 2011), which investigated the association of coronary heart disease mortality with the three pollutants PM_2.5, black carbon and NO₂ together, could (in theory) serve as a cautious approach in sensitivity analysis for calculating independent effects of NO₂ from particulate mass and from traffic soot. NO₂ had correlation coefficients with PM_2.5 and black carbon of 0.47 and 0.39, respectively. It should be noted, however, that this study used a population free of cardiovascular disease at baseline, rather than the general population, so general population baseline rates would not apply. Results from more (including large) cohort studies with results on NO₂ in multipollutant models are expected in the near future, together with comprehensive meta-analyses. The additional uncertainty about other supporting evidence for cardiovascular effects would need to be acknowledged. The best option would be to use an adjusted all-cause mortality estimate from the studies expected in sensitivity analysis (item d(ii) above), due to uncertainties about the causality of the cardiovascular component of all-cause mortality. Alternatively, a meta-analysis of single-pollutant models could be used in sensitivity analysis (d(iii)) to compare (but not add) the possible effects, if the relationship is due to NO₂, compared with it being due to PM.

3.2 Capturing the effects of traffic pollution and/or small-scale variations in pollution

NO₂, particularly primary NO₂ from traffic, shows a much larger spatial variation than does PM_2.5. With regard to health impact assessments, PM_2.5 is not optimally suited to capture long-term effects from small-scale spatial pollution variations (Beelen et al., 2008a; Bemis, Torous & Dertinger, 2012). A health impact assessment that intends to estimate the whole burden of air pollution will miss these effects, if it relies only on PM_2.5. The additional burden of small-scale variations in pollution can be estimated by using NO₂, as it has been found to be associated with effects not always captured by using PM mass (Jerrett et al., 2009b), in studies where individual exposure to pollution was estimated with more spatially exact methods (such as land use regression models). Especially in Europe, NO₂ has therefore been used to investigate the effect from traffic pollution (Brunekreef, 2007; Cesaroni et al., 2013).

A health impact assessment intended to estimate only the long-term health effects of traffic pollution could rely on single-pollutant model concentration–response functions for NO₂, alternatively or as a sensitivity analysis to an evaluation of effects associated with PM mass. In that case, estimates from one-pollutant models could be used. The respective results can be compared, but not added. The caveat is that NO₂ might not be acting as a marker for traffic in the same way it did at the time of the study − for example, due to the use of particle traps increasing emissions of primary NO₂, (see Question C2).

Children’s respiratory symptoms and/or asthma symptoms. The McConnell et al. (2003) study (described in section 3.1, on NO₂ per se) is less obviously suitable for a concentration–response function that represents pollution varying on a fine spatial scale, such as traffic pollution, as the study is based on a combination of cross area comparisons and yearly variations. The yearly variation within community component (which may be driven more by local pollution sources) has, however, been used in health impact assessments of local pollution changes (Perez et al., 2009b; Perez et al., 2012; Brandt et al., 2012).

The meta-analyses available for NO₂ and asthma incidence (Anderson, Favarato & Atkinson, 2013b) within communities (positive) – or (better (see 3.1)) the studies of NO₂ and asthma prevalence within communities (most positive) that have the potential for meta-analysis – are suitable for quantifying the effect of traffic pollution or fine scale varying pollution, in general (that is, d(iv)), where other traffic pollutants are not quantified. To quantify the effect of traffic pollution or fine scale varying pollution, it does not matter that the correlations are too close to perform multipollutant models, as NO₂ is being used as an indicator (with an acknowledgement of the caveats).

Mortality. Similarly, a meta-analysis of single-pollutant models of long-term exposure to NO₂ and all-cause mortality could be used for the same (item d(iv) above) scenarios. For the reasons given before, we propose concentrating on all cause mortality. If desired, however, there would be less uncertainty in using a cardiovascular mortality concentration–response function than there would in using a NO₂ per se calculation. This is not only because there is no need to distinguish between NO₂ and PM (or at least traffic PM) in the epidemiology studies, but also because the toxicology that supports an effect of PM on cardiovascular outcomes also supports the epidemiology on a cardiovascular effect of traffic pollution, for which NO₂ is being used as a marker.

Present WHO air quality guidelines

Exposure to arsenic occurs in inorganic and organic forms, and in most cases oral intake predominates. The critical effect of inhalation of inorganic arsenic is considered to be lung cancer. The 2005 global update of the WHO air quality guidelines (WHO Regional Office for Europe, 2006) used the unit risk of 1.5 × 10^-3, based on the risk of lung cancer (WHO Regional Office for Europe, 2000). Thus, a lifetime exposure to 6.6 ng/m³ would cause an excess risk of 10^-5. For genotoxic carcinogens, such as arsenic, the 2005 global update of the WHO air quality guidelines did not present any guideline level. The evaluation was based mainly on three occupational smelter cohorts (Tacoma and Montana in the United States, and Ronnskar in Sweden). An updated pooled analysis of these cohorts was extrapolated and transformed into unit risk of 1.5 × 10^-3 used by the guidelines (Viren & Silvers, 1994). The EU target value for annual average arsenic in PM₁₀ is 6 ng/m³ (EU, 2005). Typical ambient arsenic concentrations in England are about 1 ng/m³ (EPAQS, 2009). Inhalation is a minor part of total exposure.

Later reviews

The United States Agency for Toxic Substances and Disease Registry reviewed the health risks of arsenic (ATSDR, 2007), but the Agency provides minimal risk levels for non-cancer end-points and did not publish any guideline for inorganic arsenic. There is a draft integrated risk information system (IRIS database) document from the EPA (2010b), and it does not propose any guideline limit (inhalation reference concentration, Rfc).

A United Kingdom expert panel evaluated the health risk of inhalation of inorganic arsenic (EPAQS, 2009). As a starting point, it used the midpoint of estimated cumulative exposure (125 µg/m³ multiplied by the number of years) in the lowest stratum in the Swedish smelter study with a significant increase in lung cancer. For a 40-year working life, this gave a lowest-observed-adverse-effect level (LOAEL) of 3 µg/m³. It was divided: by 10, to obtain a presumed no-observed-adverse-effect level (NOAEL); by 10, to obtain a longer exposure time for the general population; and by 10, to obtain the possible susceptible groups. With a factor of 1000, a guideline of 3 ng/m³ was proposed for the PM₁₀ fraction, as an annual average.

Recent studies

We found three studies published after 2005 that are relevant to the risk assessment for inhaled arsenic. The first, by Jones et al. (2007), found no significant association between cumulative arsenic exposure and lung cancer in a United Kingdom smelter, but did find significant associations when less weight was given to exposures that occurred long before the outcome. A statistically significant increase in the RR of lung cancer was found in the stratum with a mean (weighted) cumulative exposure to arsenic of about 1.5 mg/m³ multiplied by the number of years. This is higher than the LOAEL in the Swedish smelter study cited above, but the point estimates of strata with lower exposure are not incompatible with risk estimates in previous reviews.

The second study, by Lubin et al. (2008), reanalysed the Montana cohort and found a higher RR at high-intensity exposure to arsenic (for example, 0.6 mg/m³ for 5 years) compared with low-intensity exposure (for example, 0.3 mg/m³ for 10 years) at the same level of cumulative exposure. This makes sense in view of possible limitations in demethylation and/or detoxification ability. If it is true also for low-level exposure, arsenic in ambient air would be less risky than indicated in previous estimates made from occupational exposure. However, the conclusions are not generally accepted, and the results cannot be directly transformed into a unit risk estimate.

The third study, by Smith AH et al. (2009), compared the RRs of lung cancer of inhaling inorganic arsenic at the Tacoma smelter with the risk at ingestion of inorganic arsenic in drinking water in Chile, using arsenic concentrations in urine. They found that the excess RR of lung cancer versus urinary arsenic level was similar in the two cases. This would indicate that the absorbed dose is carrying the risk, independent of whether it was inhaled or ingested. This is significant. If cancer risk estimates for arsenic in drinking-water (NAS, 2001), calculated as a function of arsenic concentrations in urine, could be transformed into air levels of arsenic, this would be an alternative way of estimating the cancer risk for the general population at low-level arsenic concentrations in ambient air. Using assumptions on absorbed arsenic by inhalation and ingestion, the United States National Academy of Sciences (NAS) estimate of the excess absolute risk of lung cancer would transform into a unit risk of 1 x 10^-3, very similar to the 1.5 x 10^-3 estimate calculated from completely different data. However, one consequence of this notion is that we must then also consider urinary bladder cancer. It is generally accepted (NAS, 2001) that intake of arsenic in drinking water also increases the risk. The NAS estimates for the United States population for ingesting 10 µg arsenic per day from water for this site (based on Taiwanese data) is 12–23 per 10 000 (lifetime risk).

Evaluation

In the WHO air quality guidelines, a unit risk of 1.5 x 10^-3 was proposed, based on extrapolations from cumulative exposure to arsenic in smelter cohorts. In the past decade, studies have suggested that the true unit risk could be lower or higher than that. Another technique, applying an unsafety factor of 1000 to the LOAEL in one of the smelter cohorts resulted in a guideline value of 3 ng/m³. In summary, the new evidence is insufficient to have an impact on the current EU target value.

Present WHO air quality guidelines

The main human exposure sources of cadmium are diet (higher uptake at low iron stores, making women usually more exposed than men) and smoking. The most well-known health effects of cadmium are kidney damage and toxic effects on bone tissue (osteomalacia and osteoporosis). Cadmium has been classified by the International Agency for Research on Cancer as a Group 1 human carcinogen, mainly due to the increased risk of lung cancer from occupational exposure to cadmium. In the WHO air quality guidelines (WHO Regional Office for Europe, 2000), the data behind the classification of cadmium as carcinogenic was considered to be complicated to interpret due to concomitant exposure to arsenic. Therefore, no unit risk of lung cancer, based on these studies, could be derived. The 2000 WHO Regional Office for Europe air quality guidelines noted that average kidney cadmium levels in Europe are very close to the critical level for renal effects. A further increase in dietary intake of cadmium, due to accumulation of cadmium in agricultural soils, must be prevented. Therefore, a guideline value of 5 ng/m³ was set for cadmium in air (WHO Regional Office for Europe, 2000). This was also applied as an EU target value (EU, 2005). Present levels of cadmium in air are 0.1–1 ng/m³ in rural areas, 1–10 ng/m³ in urban areas, and higher than 10 ng/m³ in some industrial areas (WHO Regional Office for Europe, 2007). Inhalation is a minor part of total exposure, but ambient levels are important for deposition in soil and, thereby, dietary intake.

Later reviews

WHO working group on long-range transboundary air pollution. A WHO Regional Office for Europe working group published the document Health risks of heavy metals from long-range transboundary air pollution (LRTAP) (WHO Regional Office for Europe, 2007). Emissions, depositions, air levels, and health risks were reviewed, including the impact of other factors (such as fertilizers and sewage). Information on the effects of low-level exposure to cadmium on markers of renal function and bone was updated. The working group mentioned two critical effects at low-level exposure to cadmium: excretion of low molecular weight proteins, due to tubular cell damage, and also osteoporosis. Studies on cadmium balance in topsoil in Europe indicated that the amount of input exceeds that of removal. The working group noted that several European studies in the late 1990s and the beginning of the 2000s showed effects on kidney and/or bone at environmental exposure levels for urinary cadmium as low as 0.5–2.0 µg/g creatinine (µg/gC). The working group proposed a LOAEL of 2 µg/gC. The evidence of lung cancer from inhalation of cadmium was considered to be rather weak. The margin of safety for adverse effects on the kidney and bone is very narrow for the European population and non-existent for sensitive subgroups, such as women with low iron stores. Since food represents more than 90% of the cadmium intake in non-smokers, and no decline has been shown, further efforts should be made to reduce cadmium emissions.

The United States Agency for Toxic Substances and Disease Registry reviewed the health risks of cadmium and recommended a minimal risk level for this hazardous substance (ATSDR, 2012). As a point of departure, the Agency selected the lower confidence limit for cadmium concentrations in urine, resulting in a 10% increase of the excretion of beta-2-microglobulin in one of the European studies – 0.5 µg/gC; it chose the study that found effects at the lowest cadmium concentration in urine. Based on a toxicokinetic model, long-term inhalation of 0.1 µg/m³, combined with the average dietary cadmium intake in the United States population, would result in cadmium concentrations in urine of 0.5 µg/gC. An uncertainty factor of 9 was applied to the cadmium concentration in air, and thus a minimal risk level of 10 ng/m³ was set.

The European Food Safety Authority (EFSA) reviewed cadmium in food (CONTAM, 2009). In a meta-analysis, it found the lower limit of the benchmark dose for a 5% increased prevalence of “elevated” beta-2-microglobulin to be 4 µg/gC for the cadmium concentration in urine. After adjusting for the interindividual variability of the cadmium concentration in urine, a critical concentration of cadmium in urine of 1 µg/gC was derived. EFSA also reviewed data on cadmium effects on bone: “studies summarized indicate a range of U-Cd [cadmium concentrations in urine] for possible effects on bone, starting from 0.5 µg/gC, which is similar to the levels at which kidney damage occurs”. EFSA also reviewed studies on cancer, including those based on environmental exposure (lung cancer: Nawrot et al., 2010; endometrial cancer: Akesson, Julin & Wolk, 2008). The data on hormone-related cancers were considered to need confirmation from other studies.

The Joint WHO/FAO Expert Committee on Food Additives (JECFA) used essentially the same studies on beta-2-microglobulin as did EFSA, but it used a slightly different statistical modelling technique and came to the conclusion that the point estimate for the break point for elevated beta-2-microglobulin was 5.2 µg/gC for cadmium concentrations in urine (JECFA, 2011). This was then transformed to a dietary intake of 0.8 µg/kg/day (lower confidence limit). The effects on bone were not considered.

The International Agency for Research on Cancer recently updated their evaluation of cadmium and cadmium compounds (IARC, 2012). The Agency found epidemiological support for lung cancer in humans from inhalation of cadmium and also found sufficient evidence of lung cancer in animals. Therefore, cadmium and cadmium compounds are carcinogenic to humans (Group 1), mainly based on the increased risk of lung cancer.

Recent studies

The above-mentioned reviews include literature up to 2006–2008. Thereafter, the published studies of major interest are of two types.

1. Some studies indicate that the associations between low-level exposure to cadmium and excretion of low molecular weight proteins shown in several other studies may not be due to cadmium toxicity. Instead, co-excretion of cadmium and proteins is more likely to be caused by physiological factors, such as varying reabsorption of cadmium and proteins in renal proximal tubules (Chaumont et al., 2012; Akerstrom et al., 2013).
2. Other published studies showed effects on bone at low-level exposure to environmental cadmium (Gallagher, Kovach & Meliker, 2008; Schutte et al., 2008; Wu, Magnus & Hentz, 2010; Nawrot et al., 2010; Thomas et al., 2011; Engström et al., 2011, 2012), although some did not find positive associations (Rignell-Hydbom et al., 2009; Trzcinka-Ochocka et al., 2010).

Evaluation

Research performed in the new millennium has indicated adverse effects of long-term dietary cadmium on kidney and bone at cadmium concentrations in urine commonly seen in most European countries – about 1 µg/gC. The WHO Regional Office for Europe air quality guidelines for 2000 is still valid; further increase of cadmium in agricultural soils must be prevented. The cadmium input in European agricultural soils is larger than the output, suggesting that the cadmium intake will not decrease. Overall, deposition from cadmium in air contributes typically about half of the cadmium input to soils. Present levels of cadmium in air are too high to obtain a cadmium balance in soils (WHO Regional Office for Europe, 2007). This should be taken into account when deciding whether the WHO air quality guidelines should be reconsidered.

3. Mercury (Hg)

4. Lead

5. Nickel

Short-term exposure and mortality

Since the publication of the 2005 global update of the WHO air quality guidelines, Anderson et al. published a peer-reviewed research report that contains meta-analyses of time-series studies, based on studies up until 2006 (Anderson et al., 2007). In single-city studies, positive and statistically significant associations with SO₂ were generally found for all-cause, cardiovascular, cardiac and respiratory mortality in single-pollutant models. While the majority of the single-city studies are from before 2004, there are a few post-2004 studies included in some of the summary estimates described below (Jerrett et al., 2004, Penttinen, Tiittanen & Pekkanen, 2004). Most estimates showed significant heterogeneity, except those for cardiorespiratory mortality and mortality from lower respiratory infections, but there was also significant heterogeneity for the main mortality end-points for PM₁₀. Adjustment for publication bias, where shown (all except cardiorespiratory, cardiac and stroke mortality), reduced the size of the summary estimates, but they remained positive and statistically significant. The size of the estimates was only slightly lower than those for PM₁₀.

There was substantial overlap between the multicity studies reviewed for the 2005 global update of the WHO air quality guidelines (WHO Regional Office for Europe, 2006) and for the Anderson et al. report (Anderson et al., 2007), but the latter covered more studies (none published after 2004). Again there were positive, statistically significant associations in single-pollutant models with all-cause, cardiovascular, cardiac and respiratory mortality, but those associations tested in multipollutant models were reduced by control for other pollutants, often substantially, and often lost statistical significance. Multipollutant models in single-city studies were not reviewed in this report.

The EPA view, published in 2008 (EPA, 2008a), covered much of the same material, but also included the Public Health and Air Pollution in Asia literature review (HEI, 2004), which found that a meta-analysis of time-series studies of SO₂ and mortality in Asia gave similar results to those of European multicity studies. Overall, the EPA considered that there was suggestive evidence of a causal relationship (consistent in single-pollutant models, more uncertain after adjustment for co-pollutants). The Hong Kong Intervention study (Hedley et al., 2002) was noted for the effect of a change in SO₂ on mortality, in the absence of a change in PM₁₀; also noted was that this change in sulfur in fuel may have included changes in heavy metal content (reported reference: Hedley et al. (2006); now available in journal form: Hedley et al. (2008)).

The Hong Kong Intervention has now been studied in more detail (Wong et al., 2012). The decrease in SO₂ concentrations between the pre- and post-intervention periods was accompanied by decreases in NO₂ and increases in ozone. There were also decreases in metals (aluminium, iron, manganese, nickel, vanadium, lead and zinc) associated with particles, of which the decreases in nickel and vanadium were most consistent. While nickel and vanadium were associated with mortality in general terms, it was not possible to show a clear link between changes in their concentrations being associated with the intervention and with changes in their effect on mortality after (compared with before) the intervention (although there was a decline). In this more detailed study, SO₂ only showed a decrease in the excess risk of respiratory mortality (not statistically significant), but also showed an increase in the excess risk of all-cause (not statistically significant) and cardiovascular mortality after (compared with before) the intervention. In the overall study (irrespective of the intervention), the effects of SO₂ on mortality were not stable to adjustment for nickel and vanadium. In summary, while the intervention was still beneficial, a more detailed study suggests that identifying the responsible pollutant is difficult.

APED, used to prepare the review by Anderson et al. (2007), has now been updated to 2009 for papers on SO₂. The database uses a comprehensive literature searching strategy, with sifting for study quality criteria. The database indicates seven new studies (Cakmak, Dales & Vidal, 2007; Filleul et al., 2006a; Kowalska et al., 2008; Lee, Son & Cho, 2007b; Tsai et al., 2003; Wong et al., 2008; Yang et al., 2004), for all-cause mortality, all ages, all-year and 24-hour average SO₂, that had quantitative estimates and did not overlap with other studies (see Table 11). The literature search identified a further two studies meeting these criteria (Rajarathnam et al., 2011; Rabczenko et al., 2005) and a further paper has been identified since (Chen et al., 2012c). Most of the studies showed positive associations (8 of 10) of which five were statistically significant. The range of the central estimates (-1.4–3% per 10 µg/m³) were all within the range of the 144 previous estimates (-4.63–6.4% per 10 µg/m³). Although two Asian study estimates (Wong et al., 2008, Lee, Son & Cho, 2007b) were higher than the current summary estimate, it is unlikely that this would change an updated summary estimate much, given the large number of studies already in the meta-analysis.

Table 11.

Selected quantitative estimates for SO₂ and mortality.

In a study in Santiago, Chile, the SO₂ estimate generally remained stable after adjustment for PM₁₀, ozone and carbon monoxide (Cakmak, Dales & Vidal, 2007), and a negative association in Delhi, India, was unchanged by adjustment for PM₁₀, with and without NO₂ (Rajarathnam et al., 2011). The HEI report collating four Asian city studies (Wong et al., 2010b) found the estimate remained stable after adjustment for PM₁₀ and ozone, but not for NO₂, with the exception of Bangkok, where the estimate was reduced by both PM₁₀ and NO₂. In the study of 17 Chinese cities (Chen et al., 2012c), the estimate was reduced, but remained positive and statistically significant when controlled for PM₁₀; the reduction on adjustment for NO₂ was substantial, and the estimate lost significance. There were too few multipollutant-model studies in the earlier Asian city meta-analysis to draw an overall conclusion (HEI, 2004). This does not change the previous view (Anderson et al., 2007; WHO Regional Office for Europe, 2006) that estimates for all-cause mortality could be sensitive to adjustment for other pollutants. It is possible that the short sharp peaks of SO₂ mean it is more subject to measurement error, which can affect multipollutant models. However, Zeka & Schwartz (2004) found that, using a statistical technique for multipollutant adjustment less subject to measurement error, the SO₂ estimate was small and was not statistically significant.

The report on four Asian cities (Wong et al., 2010b) examined the shape of the concentration–response function with mixed results (2 of 4 linear), but none of them examined the shape after control for other pollutants.

Short-term exposure and respiratory hospital admissions

Anderson et al. (2007) found a positive and statistically significant association with respiratory hospital admissions, all ages, all year of 1.51% (95% CI: 0.84–2.18%) per 10 µg/m³ SO₂ that showed evidence of heterogeneity. The association remained positive and statistically significant after adjustment for some publication bias. Multicity studies gave a similar result. There were no multicity studies that examined multipollutant models, and the single-city study multipollutant results were not reviewed. A further three studies have been added to APED since (Leem et al., 1998; Jayaraman & Nidhi, 2008; Chang, Hsia & Chen, 2002), another two meeting the criteria (Wong et al., 2010a; Cakmak, Dales & Judek, 2006b) were identified in the literature search, and another one later (Chen et al., 2010b). All six found positive associations with SO₂ (two non-significant) that ranged from 0.13% to 8.2% compared with -0.5–22.5% for the previous estimates. Where there was adjustment for other pollutants (Cakmak, Dales & Judek, 2006b; Jayaraman & Nidhi, 2008; Leem et al., 1998), the estimates were reduced and remained significant in only one study (Cakmak, Dales & Judek, 2006b).

Given the chamber study findings, asthma admissions will be described here. Results for other respiratory diagnoses are described in Anderson et al. (2007). The relationship with asthma admissions was significant in children in single and multicity studies and was robust to adjustment in multipollutant models in the multicity studies (unexamined in single-city studies) (Anderson et al., 2007). Further studies published since 2006 on asthma admissions in children are also positive and statistically significant in two studies (Lee, Wong & Lau, 2006; Samoli et al., 2011a), with a range from 1.3% to 6% per 10 µg/m³ SO₂ compared with 0.8% to 9% in Anderson et al. (2007). In Samoli et al. (2011a), although control for PM₁₀ resulted in a loss of significance, it also resulted in a smaller reduction (20%) in the coefficient than when PM₁₀ was controlled for SO₂ (30% reduction). The SO₂ coefficient was stable to adjustment for NO₂ and ozone. The coefficient was said to be non-significant (direction not given) in a study in Hong Kong (Ko et al., 2007). Results in other age groups or all ages were more mixed (Anderson et al., 2007; Bell, Levy & Lin, 2008; Ko et al., 2007; Tsai et al., 2006a; Yang et al., 2007).

Anderson et al. (2007) reported a summary estimate per 10 µg/m³ SO₂ of a 2.65% (95% CI: 0.39–4.96%) increase in asthma emergency room visits in children, all year. The summary estimate was not significant in adults and the outcome was not examined in multicity studies. The database has been partially updated for emergency room visits, identifying four studies (Jalaludin et al., 2008; Ito, Thurston & Silverman, 2007; Szyszkowicz, 2008; Villeneuve et al., 2007). The latter two found no associations, but the first two found positive and statistically significant associations in children and adults, respectively. In Jalaludin et al. (2008), the association was reduced, but remained significant, on adjustment for other pollutants; the association, however, was not robust to adjustment for NO₂ in Ito, Thurston & Silverman (2007) (only the summer association was examined). The literature search identified a few other studies of asthma emergency room visits, notably a large multicity study in Canada (Stieb et al., 2009) that found a non-significant negative association. This study was for all ages. As the Anderson meta-analysis found a greater effect in children, it would have been interesting to see the analysis split by age group. A study in Toronto did find a significant relationship in children across both sexes and in the top and bottom quintiles of socioeconomic position (Burra et al., 2009). A positive and statistically significant association was also found for SO₂ and emergency room visits for wheeze in children aged 0–2 years in a multicity study in Italy (Orazzo et al., 2009).

A preliminary study from Taiwan, Province of China, used daily variations in spatially modelled pollutants as the exposure metric (Modelling may be more uncertain for SO₂ than for other pollutants, as SO₂ concentrations are characterized more by sharp peaks, which are harder to model). The study found a negative association that was not statistically significant in both single and multipollutant models for asthma emergency room visits in all age groups (Chan et al., 2009). Overall, the conclusions are unclear for SO₂ and emergency room visits for asthma, as there were mixed results in multipollutant models in the recent studies, but the larger number of earlier studies suggests the association is robust (EPA, 2008a).

Short-term exposure and cardiovascular admissions

Anderson et al. (2007) reported a summary estimate of 0.96% (95% CI: 0.13–1.79%) per 10 µg/m³ for cardiovascular admissions and 24-hour average SO₂ for all ages, all year across 5 studies and a summary estimate of 2.26% (95% CI: 1.30–3.22%) for cardiac admissions across 12 studies. A similar result was found for cardiac admissions for a multicity study in eight Italian cities and a lower one in seven European cities. The multicity studies did not include a multipollutant model and the report did not collate multipollutant model estimates for single-city studies. A further multicity study from Spain meeting the same criteria has been published since (Ballester et al., 2006) and reported an estimate of 1.33% (95% CI: 0.21–2.46%). This was not stable to adjustment for carbon monoxide, but was to other pollutants. A study in Shanghai found a positive and statistically significant association with cardiovascular admissions that was stable to adjustment for PM₁₀, but was reduced to some degree and lost statistical significance on adjustment for NO₂ (Chen et al., 2010b). The study also reported a positive and statistically significant estimate for cardiac admissions as did the study by Wong et al. (2010a). The latter two estimates were not examined in multipollutant models.

Short-term respiratory effects

With a variety of assumptions, modelling of gas transport in theoretical airway models predicted that the local concentration of SO₂ in the upper airways of human beings would be 3–4 times higher than in rats or dogs (Tsujino, Kawakami & Kaneko, 2005). Exposure to 50 ppm SO₂ 1 hour a day for 3 days, followed by ovalbumin, exaggerated chronic allergic airway inflammation and subepithelial fibrosis (Cai et al., 2008). Another study exposed rats to 2 ppm SO₂ for 1 hour a day for 7 days prior to or without sensitization with ovalbumin. Prior exposure to SO₂ increased the mRNA and protein expression of epidermal growth factor (EGF), epidermal growth factor receptor (EGFR) and cyclooxygenase-2 (COX-2), markers of regulation of mucus hypersecretion, and airway repair and inflammation (Li, Meng & Xie, 2008). A study that developed an animal model for chronic obstructive pulmonary disease found that 5, 10 and 20 ppm SO₂ for 3 days resulted in no change in basal mucus secretory activity in trachea preparations and a decrease (not an increase) at 40 ppm and 80 ppm. No changes were found in acetylcholine stimulated secretory activity. Single cell necrosis and loss of cilia occurred at concentrations of 10 ppm and higher (Wagner et al., 2006). SO₂ inhalation at l20 ppm for 6 hours a day for 7 days caused significant increases in the proto-oncogenes c-fos and c-jun mRNA and in protein levels in the lungs of rats (Qin & Meng, 2006a) and caused a further increase in the presence of benzo[a]pyrene. It was hypothesized that this might explain the co-carcinogenicity of SO₂ and benzo[a]pyrene in hamsters, although the EPA view was that SO₂ was not a clear co-carcinogen.

There have also been studies of lung cells in vitro. Human bronchial epithelial (BEP2D) cells were treated with a range of concentrations (0.0001–1 mM) of the SO₂ derivatives sodium bisulfite (NaHSO₃) and sodium sulfite (Na₂SO₃) in a 1:3 ratio for various durations up to 24 hours (Li, Meng & Xie, 2007). Expression of EGF, EGFR, intercellular adhesion molecule 1 (ICAM-1) and COX-2 mRNA and protein showed a dose-dependent increase that was greatest at 30 minutes. It was suggested that this would result in mucin overproduction and inflammation, if occurring in vivo. The same dose protocol (with the addition of a 2 mM dose of the SO₂ derivatives) for 4 hours was applied to the same cell line, leading to mRNA and protein overexpression of c-fos, c-jun and c-myc at all doses, with other changes in proto-oncogenes and tumour suppressor genes at 0.1–2.0 mM (Qin & Meng, 2009).

Another study, in A549 cells, found significantly reduced cell viability in an air–liquid interface culture, with concentrations from 10 ppm to 200 ppm SO₂ (Bakand, Winder & Hayes, 2007). In the same cell line, sodium sulfite (a derivative of SO₂) at 1000–2500 µM was shown to enhance interleukin 8 (IL-8) release. IL-8 is a chemical signal involved in neutrophil recruitment and activation. IL-8 release was inhibited by a selection of asthma drugs (Yang et al., 2009). Sulfite oxidation by a mammalian peroxidase–hydrogen peroxide system, resulting in the highly reactive sulfate radical (SO₄^.-), has been shown for the first time in an experiment using human myeloperoxidase and human neutrophils and (bi)sulfite anions at 20–100 µM (Ranguelova et al., 2012). Ranguelova et al noted that healthy individuals have a mean serum concentration of 5 µM, but it can be raised in disease.

Short-term systemic and cardiovascular effects

A group from China published a series of papers on the systemic effects of SO₂. Meng & Liu, 2007, showed morphological changes in various organs in mice after inhalation of 10.6 ppm SO₂ and above for 4 hours a day for 7 days. Qin & Meng (2006b) showed a dose-related reduction in activities and mRNA levels of the detoxifying enzymes CYP2B1/2 and CYP2E1 in the lungs, and CYP2B1/2 (but not CYP2E1) in the livers of rats treated with 5.3–21.2 ppm SO₂ for 6 hours a day for 7 days. Protein oxidative damage and DNA-protein cross-links were increased in the lungs, liver and heart (in that order) in mice exposed to 5.3–21.2 ppm SO₂ for 6 hours a day for 7 days (Xie, Fan & Meng, 2007). Unsurprisingly, 21.2 ppm SO₂ for 4 hours a day for 10 days led to oxidative stress in the livers and brains of the mice. This oxidative stress was ameliorated by moderate, but not high, levels of vitamin C and by salicylic acid (Zhao H et al., 2008).

Other than the study at high doses by Xie, Fan & Meng (2007) mentioned above, no other animal studies that reported cardiovascular toxicity were picked up in the search. A review of amino acids as regulators of gaseous signalling (Li et al., 2009) notes that endogenous SO₂ (derived from cysteine) can activate guanylyl cyclase and thus elicit a variety of responses, including relaxation of vascular smooth muscle cells. A review on the same subject that is more specific to SO₂ is available (Chen S et al., 2011). In a section of the Chen et al. review, on older and newer studies of the effects of exogenous SO₂, the authors highlighted a study by Nie & Meng (2007) showing that the SO₂ derivatives NaHSO₃ and Na₂SO₃, in a 1:3 ratio at a concentration range of 5–100 µM, inhibited the sodium/calcium exchanger current in rat myocytes and that SO₂ derivative concentrations above 10 µM increased intracellular myocyte free calcium. Nie & Meng (2007) also showed that SO₂ and its derivatives can lower blood pressure in rats. The Chen et al. review also mentioned that SO₂ and its derivatives can act as vasoconstrictors or vasodilators, depending on concentration.

The section of the Chen S et al. (2011) review on endogenous SO₂ notes that a decrease in SO₂ can protect against vascular structural remodelling in spontaneously hypertensive rats and can protect against pulmonary hypertension in rats with hypoxic pulmonary hypertension. An increase in endogenous SO₂ protected against pulmonary hypertension in monocrotaline-induced hypertension (by inhibiting smooth muscle cell proliferation), but mediated myocardial ischaemia reperfusion injury. The concentrations for these studies are not discussed in the review. The authors concluded that endogenous SO₂ is involved in regulation of cardiovascular function and that disturbances in this regulation can be found in disease.

Reducing tailpipe soot may lead in specific cases to an increase NO₂ and ultrafine particles

Reducing the emission of soot particles from motor vehicles has in some cases significantly altered the chemical composition and particle size distributions in specific urban environments (Keuken et al., 2012; Herner et al., 2011). When filter traps were applied without catalysts (urea selective catalytic reduction), NO₂ concentrations increased locally at traffic sites, even though a total reduction in concentrations of nitrogen oxides was observed. The regulatory relevance of this shift is clear, although the real health impact of this change in the emission is still under discussion (Keuken et al., 2012). In these specific cases, mass-related emission of PM was significantly reduced with the change of the combustion conditions and the use of a particle filter without urea selective catalytic reduction. The use of particle traps significantly removed the larger particles and, hence, led to formation of a nucleation mode with a significantly increased particle number concentration of about 10 nm at end of the tailpipe (Herner et al., 2011), and a shift of the mode led to a particle number concentration of about 10–30 nm at some distance –for example, that of a pedestrian on a street (Harrison, Beddows & Dall’Osto, 2011; Casati et al., 2007).

Changing reactivity over time or with increasing distance from the source

Particle reactivity is believed to be an important characteristic with direct influence on the hazard potential. Particle reactivity is a general term that includes, for example, the particle intrinsic formation potential of reactive oxygen species and the ability to catalyse redox reactions. Its possible importance for health is currently being discussed (Shiraiwa, Selzle & Pöschl, 2012). Initial studies on spatial and temporal variations of particle reactivity parameters have been conducted (Boogaard et al., 2012a; Künzli et al., 2006). Still missing is the link between particle reactivity and other particle characteristics, the actual mechanisms triggered, and the ageing and/or changes of particle reactivity during atmospheric transport due to chemical reactions. Understanding how particle surface properties are altered during atmospheric transport, physically and chemically, may be one key in linking particle characteristics and emissions to health effects.

Secondary organic aerosol

Although laboratory experiments have shown that organic compounds in both gasoline fuel and diesel engine exhaust can form secondary organic aerosols, the fractional contribution from gasoline and diesel exhaust emissions to ambient secondary organic aerosols in urban environments is poorly understood. Recently, Bahreini et al. (2012) demonstrated that, in Los Angeles, the contribution from diesel emissions to secondary organic aerosol formation is very low and that gasoline emissions dominate diesel exhaust emissions in forming secondary organic aerosol mass. Chamber studies performed in Europe seem to confirm this hypothesis. In a very recent paper, Gentner et al. (2012) reported that diesel exhaust is seven times more efficient at forming aerosol than gasoline exhaust, that both sources are important for air quality and, depending on a region’s fuel use, that diesel is responsible for 65–90% of the vehicle-derived secondary organic aerosols. These conflicting results have important implications for air quality policy, but at present large uncertainties exist.

Verma et al. (2009b) have shown for Los Angeles in summer that both primary and secondary particles possess high redox activity; however, photochemical transformations of primary emissions with atmospheric ageing enhance the toxicological potency of primary particles, by generating oxidative stress and leading to subsequent cell damage.

Studies by Biswas et al. (2009) – using direct exhaust PM emissions from heavy duty vehicles, with and without emission abatement technologies implemented – suggest that the semivolatile fraction of particles are far more oxidative than solid (carbon) particles. It is also possible, in our opinion, that the secondary organic aerosols formed from the condensation of previously volatilized PM are highly oxidative.

Ozone and organics

Reactions of ozone with certain organic molecules that occur indoors at certain concentrations can produce short-lived products that are highly irritating, relative to the reaction precursors, and may also have long-term health effects. Known products of indoor ozone reactions include such compounds as formaldehyde, acetaldehyde and other organic acids. Some of these compounds are known to cause ill health in human beings (Weschler, 2006). The EPA Building Assessment Survey and Evaluation study data was analysed for associations between ambient ozone concentrations and building-related symptom prevalence (Apte, Buchanan & Mendell, 2008). Ambient ozone correlated with indoor concentrations of some aldehydes, a pattern suggesting the occurrence of indoor ozone chemistry. Apte, Buchanan & Mendell (2008) hypothesized that ozone-initiated indoor reactions play an important role in indoor air quality and building occupant health. They also hypothesized that ozone carried along into buildings from the outdoor air is involved in increasing the frequency and the range of upper and lower respiratory, mucosal, and neurological symptoms by as much as a factor of 2 when ambient ozone levels increase from those found in low-ozone regions to those typical of high-ozone regions.

Secondary inorganic aerosols

In regions with high photochemical activity, the reduction of PM mass pollution and the possible increase in frequency of droughts may lead to an increase in midday nucleation episodes with a consequent increase in levels of secondary nano-size or ultrafine particles. Thus, in highly polluted atmospheres, the secondary PM mass grows by condensation on pre-existent particles; however, in cleaner conditions, and especially under high insolation and low relative humidity, new formation on nanoparticles (nucleation) from gaseous precursors may dominate the condensation sink in urban areas (Reche et al., 2011). The nucleation starts from the oxidation of SO₂ and the subsequent interaction with ammonia, and these nanoparticles immediately grow – probably by condensation of volatile organic compounds on the nucleated particles (Kulmala & Kerminen, 2008).

Another key atmospheric component is urban ammonia. This is emitted mostly by traffic and other fugitive sources, such as city waste containers and sewage, and is also emitted from animal farming. Ammonia is an alkaline gas and, when emitted in a high NO₂ scenario, may enhance the formation of ammonium nitrate (a major component of PM_2.5). Furthermore, the levels of ammonium nitrate may also increase due to the marked decrease of SO₂ emissions that yielded a marked decrease of ammonium sulfate levels across Europe. This is expected to occur because sulfuric acid is more reactive with ammonia, and most of ammonia is consumed by sulfuric acid; when sulfuric acid decreases and more ammonia is available, more ammonium nitrate can be formed from nitric acid and ammonia.

Inorganic aerosols and metals

Thus, pure ammonium sulfate particles are rare in the atmosphere, and they usually occur as a coating on (or are coated by) other substances. They can be formed from SO₂ emissions being converted photochemically into sulfuric acid. This acid coats the outside of other particles, such as metal oxide particles, which can come from the same power plant, from brake wear of cars and trucks, from metal processing, and so on. Alternatively, they may adsorb metal particles on their surface. By internal mixing, the surface components diffuse towards the core of the particle, leading to reactions between acidic sulfates and metals, converting insoluble (and hence weakly toxic) metal oxides into soluble metals. This is critical because transition metals can catalytically induce the production of highly reactive oxygenating compounds in the lung and elsewhere in the body. For example, Ghio et al. (1999) reported that soluble iron concentrations correlate with sulfate concentrations in particle filters and that the ability of soluble extracts from the particles to generate damaging oxidants is directly proportional to the sulfate concentrations. More recently, Rubasinghege et al. (2010) simulated the transformation of non-bioavailable iron to dissolved and (hence) bioavailable iron in atmospheric iron particles in the presence of acids, in both light and dark conditions. The presence of sulfuric acid on the particles results in a dramatic increase in the bioavailable iron.

Metals are not the only case where the presence of sulfates can change the toxicity of other particle components. Popovicheva et al. (2011) showed that the extent of water uptake and modification of elemental carbon particles depended on the sulfate content of the particles. Also, Li W et al. (2011) reported that sulfate aided the ageing of freshly emitted soot particles, which occurred within 200–400 m of major roads.

Wu et al. (2007) examined the effect of ammonium sulfate aerosol on the photochemical reactions of toluene (mostly from cars) and nitrogen oxides to form secondary organic particles. They found that the sulfate particles reduced the time to reach maximum concentrations of secondary organic aerosols, and also increased the total aerosol yield from toluene. That is, in the presence of sulfates, more gaseous emissions from mobile sources will be converted into particles.

Ozone and NO₂

It is clear that ozone precursors make an important hemispheric (external to the EU) contribution, but because the high ozone levels are recorded mostly in rural areas, policy pressure on PM and nitrogen oxides is much higher than on ozone. The way of abating ozone levels by local measures that focus on local ozone precursors is very complex. This is because the relationship between ozone and volatile organic compounds and between ozone and NO₂ is not linear, so that specific cases of reducing NO₂ without compensating for the decrease of volatile organic compounds, or vice versa, may result in ineffective results, or even in an increase in ozone. On the other hand, biogenic volatile organic compounds may also be involved in the process. It is expected, however, that measures applied to reducing nitrogen oxides and industrial volatile organic compounds and also climate measures to abate methane (an ozone precursor) may contribute to abating ambient ozone. But it is also true that it is very difficult to quantify the impact of such abatement.

Interaction due to ultraviolet and/or changes in volatile organic compound and/or particulate organic carbon composition

It is well known that photochemical reactions by, for example, ultraviolet and visible radiation have a significant impact on the gaseous and particulate chemical composition of the atmosphere. One of the components currently believed to be relevant to health is organic carbon, either in the gas phase or the particulate phase. Jimenez et al. (2009) present an overview of the evolution of organic aerosols in the atmosphere. Studies closer to the source – for example, investigations of the photo-oxidation of organic compounds from motor vehicle emissions – also show significant changes (Miracolo et al., 2010). Any changes in the composition of organic carbon compounds in the atmosphere are of great importance, since their relevance to health (independent of gaseous or particle phase) has a huge spectrum, from no health effects to high toxicity. To understand the changes on a small scales – for example, near sources and close to the public – as well as the changes occurring during transport, they have to be monitored more closely.

PM components and ozone

Some evidence suggests that ambient concentrations of ozone can increase the biological potency of particles. Ozonized diesel exhaust particles may play a role in inducing lung responses to ambient PM (Madden et al., 2000). Similar findings have been observed in clinical studies (Bosson et al., 2008). In terms of vascular and cardiac impairment in rats inhaling ozone and diesel exhaust particles, Kodavanti et al. (2011) reported that the joint effect of exposure to ozone (0.803 mg/m³) and diesel exhaust particles (2.2 mg/m³) was less prominent than exposure to either substance alone. An explanation for this might be found in the duration of the exposure protocol (1 day a week for 16 weeks) that may have led to adaptive responses known to occur for ozone. Also, concentrations as low as 0.423 mg/m³ ozone increased the toxic response of mixtures of carbon and ammonium sulfate particles in rats – including histopathological markers of lung injury, bronchoalveolar lung fluid proteins, and measures of the function of the lung’s innate immunological defences – whereas these effects were not observed with ozone alone (Kleinman et al., 2003).

PM components and nitrogen oxides

As NO₂ can cause nitrative stress (and production of nitric oxide via nitrite) and PM can cause oxidative stress (and production of superoxide radicals), the combination of exposure to both might increase production of peroxynitrite over and above either pollutant alone. Peroxynitrite is recognized as a key intermediate, with the potential to affect protein function (Gunaydin & Houk, 2009). It is formed from the reaction of nitric oxide and the superoxide radical. NO₂, as is well known, can lead to increased levels of circulating nitrite and nitrate. Nitrite can be reduced back to nitric oxide and the NO₂ radical in remote tissues (Lundberg, Weitzberg & Gladwin, 2008). Overall, co-exposure to PM and NO₂ may lead to simple additive effects in the lung, and reduction of either one of these components may therefore lead to lower effect estimates in epidemiological studies of the other component. Although the literature on this is very sparse, it seems unlikely that the reduction of PM has a major effect on the health impacts of nitrogen oxides.

PM and other gases and/or vapours

Some papers show that the oxidative potential and toxicity of soot decreases up to 75% after heating and loosing the external organic carbon shell (Biswas et al., 2009). Also, volatile organic compounds are able to form PM (Robinson et al., 2007; Jimenez et al., 2009) In addition, studies of human beings and diesel engine exhaust and clean carbon particles (Mills et al., 2011) strongly suggest that organic chemical compounds on the surface of the carbon particles are responsible for immediate cardiovascular responses. Interestingly, reducing the amount of soot can lead to an increase in NO₂, but the net effect is still reduced adverse effects on health (Lucking et al., 2011).

PM: carbon and iron

Animals exposed to soot particles at a concentration of 250 µg/m³ and to iron alone at a concentration of 57 µg/m³ had no adverse respiratory effects, but a synergistic interaction between soot and iron particles, in the combined exposure, was identified with strong inflammatory responses (Zhou et al., 2003).

Ozone and nitrogen oxides

Very little is known about the effects of co-exposures of ozone and NO₂. Decades ago, Mautz et al. (1988) were able to show synergistic toxic responses in mice exposed to these pollutants. Since ozone and NO₂ will form nitric acid vapour and nitrate radicals, the synergistic effects were explained by chemical interactions (Mautz et al., 1988). Various chemical compounds, as well as allergenic proteins, are efficiently oxygenated and nitrated upon exposure to ozone and NO₂, which leads to an enhancement of their toxicity and allergenicity (Shiraiwa, Selzle & Pöschl, 2012). Oxidative stress has been postulated as the underlying mechanism for adverse health outcomes, suggesting a rather unifying and standard response.

Interactions with aeroallergens

The implications of co-exposures to air pollutants and aeroallergens are a rather complex issue, with both antagonistic and synergistic effects observed, depending on the sequence and levels of exposures. Eggleston (2009) concluded that “the same environmental exposures that may cause increased symptoms at one point in time may be protective when the exposure occurs earlier or at high enough levels”.

PM. Co-exposures to aeroallergens (such as grass pollen) and particles result in synergistic effects that lead to much stronger allergic responses in experimental animals and human beings than the sum of the responses to each of these constituents (D’Amato et al., 2005; Steerenberg et al., 2003; Diaz-Sanchez et al., 1999). However, this is different from studies in which the effects of particles have been studied on already allergic subjects. For example, controlled exposures, lasting 2 hours with intermittent exercise, to diluted diesel engine exhaust at a particle mass concentration of 100 µg/m³ did not evoke clear and consistent lower-airway or systemic immunological or inflammatory responses in mildly asthmatic subjects, with or without accompanying challenge with cat allergen (Riedl et al., 2012). Likewise, these diesel engine exhaust exposures did not significantly increase nonspecific or allergen-specific bronchial reactivity. A few isolated statistically significant or near-significant changes were observed during and after exposure to diesel engine exhaust, including increases in nonspecific symptoms (such as headache and nausea) suggestive of subtle, rapid-onset systemic effects. From several inhalation studies with PM_2.5 in rat and mice models for allergic asthma, it was concluded that allergic inflammation and other effects can be enhanced by PM exposures for all size ranges of PM₁₀ (Kleinman et al., 2007; Li N et al., 2010; Heidenfelder et al., 2009). However, such an interaction points more towards an enhancement of an existing disease and not to a synergy due to co-exposures.

In summary, ambient particles can, indeed, act as a carrier and adjuvant of aeroallergens, and reductions of PM may therefore result in stronger effects than those based on the concentration–response functions of PM alone, depending on the nature of the co-exposures.

NO₂. Alberg et al (2011) were not able to detect an impact of NO₂ on allergy induced by ovalbumin, whereas diesel exhaust particles were shown to be a potent adjuvant. In contrast, Bevelander et al. (2007) and Hodgkins et al. (2010) did find a potentiating effect of NO₂ within the concentration range used by Alberg et al. (2011) (5–25 ppm) when NO₂ exposure occurred immediately beforehand and ovalbumin exposure was by inhalation. Recent in vitro findings suggested that levels of SO₂ and NO₂ below current EU health based exposure standards can exacerbate pollen allergy on susceptible subjects (Sousa et al., 2012). Moreover, exposure to NO₂ significantly enhanced lung inflammation and airway reactivity in animals that were treated with ovalbumin (Layachi et al., 2012). So, there are mixed results in human clinical studies in which NO₂ exposure preceded allergen exposure.

Changing fuel composition

There is conflicting evidence about the extent to which biodiesel exhaust emissions present a lower risk to human health when compared with petroleum diesel emissions (Swanson, Madden & Ghio, 2007). German studies have shown significantly increased mutagenic effects, by a factor of 10, of particle extracts from rapeseed oil in comparison with fossil diesel fuel; and the gaseous phase caused even stronger mutagenicity (Bünger et al., 2007). Biodiesel (rapeseed oil methyl ester) has been shown to have a four times higher cytotoxicity than conventional fossil diesel under idling conditions, while no differences were observed for the transient state (Bünger et al., 2000). This was particularly evident for mixtures of rapeseed and fossil diesel, suggesting that a mixture can lead to more harmful particulate emissions. So far, the opposite was found by others: no differences for cytotoxicity with vehicle emissions under idling conditions (Jalava et al., 2010).

Results indicate an elevated mortality risk from short-term exposure to ultrafine particles, highlighting the potential importance of locally produced particles. In an epidemiological study in Erfurt, Germany, decreases in RRs for short-term associations of air pollution were calculated as pollution concentrations decreased and control measures were implemented. However, the mass concentration changes did not explain the variation in the coefficients for NO₂, carbon monoxide, ultrafine particles, and ozone (Breitner et al., 2009). The control measures included restructuring of the eastern bloc industries, a changed car fleet (the number of cars with catalytic converters increased over time, as did the number of cars in general) and complete fuel replacement and an exchange from brown coal to natural gas in power plants and in domestic heating (Acker et al., 1998).