Introduction Air pollution is a suspected developmental neurotoxicant [1]. In animals, inhalation of diesel exhaust and ultrafine particles results in elevated cytokine expression and oxidative stress in the brain [2,3] and altered animal behavior [4,5]. In children, exposure to traffic-related air pollutants during pregnancy or infancy, when the brain neocortex rapidly develops, has been related to cognitive delays [6–8]. Children spend a large proportion of their day at school, including the period when daily traffic pollution peaks. Many schools are located in close proximity to busy roads, which increases the level of traffic-related air pollution in schools and impairs children’s respiratory health [9]. There is currently very little evidence on the role of traffic-related pollution in schools on cognitive function [10]. Though the brain develops steadily during prenatal and early postnatal periods, resulting in the most vulnerable window [1], high cognitive executive functions essential for learning [11] develop significantly from 6 to 10 y of age [12]. The brain regions related to executive functions such as working memory and attention—largely the prefrontal cortex and the striatum [13]—have shown inflammatory responses after traffic-related air pollution exposure [2,14]. We aimed to assess the relationship between long-term exposure to traffic-related air pollutants at school and cognitive development measurements in primary school children within the BREATHE (Brain Development and Air Pollution Ultrafine Particles in School Children) project. Methods Funding The research leading to these results received funding from the European Research Council under ERC Grant Agreement number 268479 for the BREATHE project. Design Forty schools in Barcelona (Catalonia, Spain) were selected based on modeled traffic-related nitrogen dioxide (NO2) values [15]. Low- and high-NO2 schools were paired by socioeconomic vulnerability index and type of school (i.e., public/private). A total of 39 schools agreed to participate and were included in the study (Fig. 1). Participating schools were similar to the remaining schools in Barcelona in terms of socioeconomic vulnerability index (0.46 versus 0.50, Kruskal-Wallis test, p = 0.57) and NO2 levels (51.5 versus 50.9 μg/m3, p = 0.72). 10.1371/journal.pmed.1001792.g001 Fig 1 Map of Barcelona and the schools by high or low air pollution by design. Black dots indicate the locations of schools with high air pollution, and white dots indicate the locations of schools with low air pollution, based on NO2 levels. All school children (n = 5,019) without special needs in grades 2 through 4 (7–10 y of age) were invited to participate, and families of 2,897 (59%) children agreed. All children had been in the school for more than 6 mo (and 98% more than 1 y) before the beginning of the study. All parents or guardians signed the informed consent form approved by the Clinical Research Ethical Committee (No. 2010/41221/I) of the Institut Hospital del Mar d’Investigacions Mèdiques–Parc de Salut Mar, Barcelona, Spain. Outcomes: Cognitive Development Cognitive development was assessed through long-term change in working memory and attention. From January 2012 to March 2013, children were evaluated every 3 mo over four repeated visits, using computerized tests in series lasting approximately 40 min in length. We selected working memory and attention functions because they grow steadily during preadolescence [12,16]. The computerized tests chosen (the n-back task on working memory [12] and the attentional network test [ANT] [17]) have been validated with brain imaging [13,17] and in the general population [18]. Groups of 10–20 children were assessed together, wearing ear protectors, and were supervised by one trained examiner per 3–4 children. For the n-back test, we examined different n-back loads (up to three back) and stimuli (colors, numbers, letters, and words). For analysis here, we selected two-back and three-back loads for number and word stimuli as they showed a clear age-dependent slope in the four measurements and had little learning effect. Numbers and words activate different brain areas. The two-back test predicts general mental abilities (hereafter called working memory), while the three-back test also predicts superior functions such as fluid intelligence (hereafter called superior working memory) [19]. All sets of n-back tests started with colors as a training phase to ensure the participant’s understanding. The n-back parameter analyzed was d prime (d′), a measure of detection subtracting the normalized false alarm rate from the hit rate: (Z hit rate − Z false alarm rate) × 100. A higher d′ indicates more accurate test performance. Among the ANT measures, we chose hit reaction time standard error (HRT-SE) (standard error of reaction time for correct responses)—a measure of response speed consistency throughout the test [20]—since it showed very little learning effect and the clearest growth during the 1-y study period among all the ANT measurements. A higher HRT-SE indicates highly variable reactions related to inattentiveness. Exposures: Direct Measurements of Traffic-Related School Air Pollution Each pair of schools was measured simultaneously twice during 1-wk periods separated by 6 mo, in the warm and cold periods of the year 2012. Indoor air in a single classroom and outdoor air in the courtyard were measured simultaneously. The pollutants measured during class time in schools were real-time concentrations of black carbon (BC) and ultrafine particle number (UFP; 10–700 nm in this study) concentration, measured using the MicroAeth AE51 (AethLabs) and DiSCmini (Matter Aerosol) meters, respectively, and 8-h (09:00 to 17:00 h) particulate matter 0.30). Correlations between modeled BC and NO2 at home and measured EC and NO2 at school were weak (r = 0.27, p 0.1 in the mixed effects linear models), and the detrimental associations occurred in all the groups. Given that development was significantly lower in grade 4 for all tasks, we repeated the analyses stratifying by grade, and the results were homogeneous. Moreover, in order to control for the “summer learning loss” phenomenon occurring between the two academic years, we excluded tests done in the second academic year that did not result in a notable change in our observed associations. Furthermore, we excluded the first exam, to prevent a potential practice effect, and the association, if anything, became stronger for working memory and superior working memory (S3 Table). Finally, sequential exclusion of school pairs one by one from the models did not change the results, suggesting that exceptional influential cases were not affecting the results. 10.1371/journal.pmed.1001792.t008 Table 8 Stratified analyses of adjusted 12-mo change in cognitive development by school air pollution exposure (high/low group or interquartile range increase) in 2,715 children and 10,112 tests from 39 schools. Cognitive Outcome By Sex By Maternal Education By ADHD By High/Low Air Pollution By Type of School Boys (n = 1,357) Girls (n = 1,358) High (n = 1,590) Low–Middle (n = 1,125) No (n = 2,409) Yes (n = 275) High (n = 1,358) Low (n = 1,357) Public (n = 931) Private (n = 1,784) Working memory (two-back numbers, d′) High/low −13 (−23, −4.2)* −6.1 (−15, 2.6) −15 (−23, −6.4)* −3.2 (−13, 6.7) −7.7 (−14, −0.97)* −26 (−45, −6.7)* − − −0.15 (−12, 11) −14 (−22, −6.4)* EC outdoor −6.4 (−12, −0.75)* −1.3 (−6.7, 4.0) −10 (−15, −5.1)* 4 (−2.2, 10) −1.9 (−6.0, 2.3) −17 (−29, −5.6)* 1.2 (−4.6, 6.9) −6.9 (−16, 2.4) 3.9 (−3.0, 11) −8.0 (−13, −3.1)* EC indoor −8.9 (−15, −2.8)* −3.2 (−9.1, 2.8) −10 (−16, −4.7)* −0.64 (−7.5, 6.2) −3.5 (−8.0, 1.1) −22 (−35, −8.5)* −2.7 (−8.8, 3.5) −6.6 (−18, 5.0) −0.53 (−11, 10) −7.1 (−12, −2.3)* Superior working memory (three-back numbers, d′) High/low −10 (−18, −3.0)* −1.9 (−8.8, 5.0) −7.5 (−14, −0.74)* −3.7 (−11, 4.0) −5.2 (−11, 0.14) −12 (−26, 3.0) − − −2.1 (−11, 7.1) −7.3 (−13, −1.2)* EC outdoor −9.6 (−14, −5.1)* 1.2 (−3.1, 5.5) −6.7 (−11, −2.6)* −1.2 (−6.0, 3.6) −3.3 (−6.7, 0.03) −11 (−19, −1.8)* −3.1 (−7.8, 1.5) −4.8 (−12, 2.5) −1.8 (−7.3, 3.7) −5.5 (−9.4, −1.6)* EC indoor −10 (−15, −5.4)* −0.85 (−5.6, 3.9) −8.9 (−13, −4.5)* −1.4 (−6.7, 3.9) −4.7 (−8.4, −1.1)* −11 (−20, −0.95)* −5.7 (−11, −0.71)* −4.2 (−13, 4.9) −4.9 (−13, 3.4) −5.7 (−9.6, −1.9)* Inattentiveness (HRT-SE, milliseconds) High/low 8.1 (1.8, 15)* 1.4 (−4.9, 7.8) 9.0 (3.1, 15)* −0.93 (−8.2, 6.3) 5.5 (0.69, 10)* 3.6 (−11, 18) − − 1.1 (−7.1, 9.2) 7.9 (2.4, 13)* EC outdoor 5.8 (1.9, 9.6)* 1.8 (−2.2, 5.7) 5.2 (1.7, 8.7)* 1.4 (−3.0, 5.9) 2.3 (−0.63, 5.2) 13 (4.9, 22)* 4.7 (0.72, 8.8)* −2 (−8.6, 4.5) 3.6 (−1.1, 8.3) 4.5 (1.0, 8.0)* EC indoor 5.2 (1.0, 9.4)* 2.0 (−2.3, 6.4) 4.6 (0.84, 8.4)* 1.9 (−3.1, 6.8) 1.9 (−1.3, 5.2) 16 (7.0, 26)* 3.9 (−0.47, 8.2) −2 (−10, 6.2) 5.3 (−2.0, 13) 4.2 (0.74, 7.6)* Difference (95% CI) in the 12-mo change, adjusted for age, sex, maternal education, residential neighborhood socioeconomic status, and air pollution exposure at home; school and individual as nested random effects. *p < 0.05. Discussion This large study with repeated and objective measures demonstrated that cognitive development is reduced in children exposed to higher levels of traffic-related air pollutants at school. This association was consistent for working memory, superior working memory, and inattentiveness, and robust to several sensitivity analyses. The association was observed both when the exposure was treated as high/low traffic-related air pollution and when using specific pollutants including outdoor and indoor EC, NO2, and UFP, which are largely traffic-related [21,22]. Changes in the developmental trajectory could resemble those suggested for the adverse impact of urban air pollution on lung function development [29]. Mechanisms of air-pollution-induced neurotoxicity have been explored [30]. The findings provide strong support for air pollution being a developmental neurotoxicant and point towards the primary school age as a particularly vulnerable time window for executive function development. A strength of this study is the longitudinal ascertainment of executive function trajectories that specifically develop during school age and the direct measures of air pollution. A concern, however, is potential residual confounding by socio-demographic characteristics, although in European cities, the relationship between proximity to traffic and economically disadvantaged areas is not always evident [31]. In the city of Barcelona, the highest air pollution was observed in the “Eixample,” a wealthy central area of the city where most of our schools with high traffic were selected [23]. We paired by design high- and low-traffic schools by socioeconomic characteristics and type of school, and although there was an inverse relation between school pollution and socioeconomic vulnerability index, such differences between schools after matching became small. In addition to the association of cognitive parameters observed with high- compared to low-exposed schools, we also observed a consistent association of cognitive parameters with specific pollutants whose relation with socio-demographics was weak and in some cases nonexistent. Furthermore, cognitive development was unrelated to social determinants in our study, in contrast to cognitive function at baseline. Besides, the associations remained in the stratified analyses (e.g., for type of school or high-/low-polluted area) and after additional adjustment (e.g., for commuting, educational quality, or smoking at home), contradicting a potential residual confounding explanation. Other potential limitations are the potential misclassification error of the UFP exposures. Seasonalized measures of UFP showed the lowest correlation among the pollutants between the first and the second campaign and weaker associations with the cognitive parameters (e.g., −4.0 [95% CI −8.6 to 0.49] for indoor UFP and working memory) than non-seasonalized UFP, which is probably because of its large geographical and temporal instability due to constant and rapid secondary formation [22]. In contrast, EC and NO2 showed very similar associations with cognitive parameters using both seasonalized and non-seasonalized measures. Another potential limitation is non-response. A total of 182 out of the initial 2,897 children (6%) were excluded because of incomplete data on individual variables. When these children were included in the analysis in models that did not require the complete dataset (i.e., a model not adjusted for maternal education), results were identical. Another level of non-response refers to children (41%) from families that did not want to be part of the study, although they were invited. This non-response affects representativeness rather than internal validity, given that the participation rate per school was unrelated to the school social gradient and that adjustment for participation rate did not change the results. Based on the results from one school, participants had less neuropsychological problems than non-participants, which likely made them less susceptible to air pollution effects. Therefore, any effect observed in the present study would likely be a conservative estimate for extrapolation to the entire population. A third limitation relates to the lack of measurements in preceding periods. However, all children had been in their school for more than 6 mo before the beginning of the study, and when we limited the study to children with more than 2 y in the school (94% of the children), associations remained the same. We interpreted these associations as chronic effects (i.e., due to exposures longer than 6 mo) since it is unlikely that the geographical pattern of air pollution occurring during the study period had changed in the last 2 y. Finally, indoor assessment was limited to a single classroom. This is not a problem for the indoor assessment of pollutants such as EC, given the high correlation between outdoor and indoor levels and similar coefficients for the association with cognition between outdoor and indoor exposures. However, it could be a problem for school noise since the correlation between outdoor and indoor noise was strongly dependent on the street orientation of the classroom (ranging from 0.07 for classrooms facing away from the street to 0.70 for classrooms facing the street). However, residual confounding by noise was unlikely given the weak correlation between the pollutants and noise measured in the same classrooms, and the robustness of the coefficients for the different pollutants after adjusting for noise and for the interaction between noise and age. This study addresses the role of traffic air pollution in schools on cognitive development. Previous studies on the effects of polluted air at schools were a study in two schools in Quanzhou (China) on attention disorders [10], two studies on aircraft noise that secondarily assessed the association between NO2 and cognitive function [32,33], and an ecological study in Michigan (US) on industrial pollution and school failure [34]. Other studies in children have evaluated the effect of maternal personal air pollution exposure or maternal/child exposure at home [35]. We found here an association between traffic-related air pollution exposure at school and cognitive development during primary school age, independent of residential air pollution and beyond the effects related to home exposures in early life found by previous studies. Total cumulative exposure in school, home, and commuting and the different time windows of exposure are not addressed here, but the continuous monitoring of BC and physical activity with personal samplers in 54 of our children showed that exposure at school was significantly higher than at home and did not change by commuting mode. This higher exposure level at school could be attributed to peaks of pollution occurring during school time, and higher inhaled dose during school time due to exercise and physical activity at schools. Besides, the fact that children at schools in the most polluted area traveled a shorter distance from home suggests a shorter commute, which could explain the lack of confounding after adjusting for commuting distance and mode. We could not disentangle the time frame of the exposures occurring under the long-term school exposure measured here. However, in the case of inattentiveness, in contrast to what was seen for working memory, the association at baseline was larger than at follow-up. Given that inattentiveness develops earlier than working memory [12], this finding could suggest that the adverse effect of air pollution could have preceded the study period, and that the lower improvement in scores may be associated with previous exposures, too. Among the individual traffic-related pollutants, we found an adverse association between EC and child cognitive development. EC comes almost exclusively from diesel vehicles in Barcelona, with an ambient air mode of around 30–40 nm, in the UFP range [22]. EC and traffic-derived metals were an important fraction of indoor and outdoor quasi-ultrafine particles (PM0.25) in our study schools [36]. We observed a high penetration of EC into the classrooms (indoor/outdoor ratio 94%) and similar associations of indoor and outdoor EC with cognitive development. Although the indoor/outdoor ratio was weaker (70%) for UFP, we also found cognitive associations with UFP. These findings remained after adjustment for traffic noise at school, pointing towards UFP as a neurotoxic traffic component, which is coherent with the numerous and consistent findings in animal studies that UFP may cause disruption of the blood–brain barrier, microglial activation, and brain inflammation [14]. Evidence points towards chronic microglial stimulation and altered innate immune response and inflammation as the key neurotoxic mechanisms of UFP [14,29,37]. UFP has been shown to cause microglial inflammation following either brain UFP deposition or systemic inflammation originating in UFP-exposed organs such as the lungs [36]. Microglial stimulation affects neurons, and UFP has been shown to decrease neuronal glutamatergic function and disrupt synapses [38]. Similarly, airborne metals have been shown to alter dopamine function [39]. The underlying brain mechanisms are beyond the present study, but the observation of associations with executive functions, the lack of confounding by ADHD or behavior, and the association among children without ADHD suggests a general brain dysfunction. Boys appeared more susceptible to air pollution, although both boys and girls showed an adverse association of school air pollution with cognitive development. Although results could be due to chance, in animals, males were more susceptible to airborne metals than females, which may be because of sex-specific altered dopamine function [39]. The possible higher vulnerability of children with ADHD could also indicate abnormalities related to dopamine [40]. Stratification by maternal education or type of school showed a larger association among students with the most educated mothers and those from private schools. This resembles what has been observed with other hazards for neurodevelopment such as genetic effects [41], presumably because fewer adverse factors play a role among students with educated mothers or in private schools, thus causing less interference with the factors under study. The observed association between air pollution and cognitive development was strong. For example, an increase from the first to the fourth quartile in indoor EC resulted in a 13.0% reduction in the growth of working memory. In contrast, the association at baseline was smaller (1.9%). Part of this larger association during primary school may be a matter of bigger magnitude of exposure to traffic pollution in schools, but it could indicate that some executive functions are particularly vulnerable during primary school age, as has also been seen for lead [42]. The long-term effect probably occurs over the period of maximum development of these executive functions, resulting in a notable cumulative effect by the end of this period in preadolescence. The observed association was consistent across cognitive measurements, though it was more evident for superior working memory, which is a good predictor of learning achievement [19]. Impairment of high cognitive functions has severe consequences for school achievement [11]. Thus, reduced cognitive development in children attending the most polluted schools might result in a disadvantage in mental capital, which may have a long-lasting life course effect. Overall, we have shown that children attending schools with higher levels of exposure to traffic-related air pollutants had a smaller growth in cognitive development over time, suggesting that traffic-related air pollution in schools negatively affects cognitive development. This may have consequences for learning, school achievement, and behavior. With regard to air pollution regulation, the present study shows that the developing brain may be vulnerable to certain traffic-related air pollutants. Supporting Information S1 Table Crude difference (and 95% CI) in cognitive development at baseline and 12-mo change by school air pollution exposure (high versus low or interquartile range increase) in 2,715 children and 10,112 tests from 39 schools. (DOCX) Click here for additional data file. S2 Table Difference (and 95% CI) in cognitive development at baseline and 12-mo change by school air pollution exposure (high versus low or interquartile range increase) in 2,715 children and 10,112 tests from 39 schools, after further adjustment for high/low area, commuting, and smoking at home. (DOCX) Click here for additional data file. S3 Table Difference (and 95% CI) in cognitive development (12-mo change) by school air pollution exposure (high/low group or interquartile range increase) in 2,715 children and 10,112 tests, after excluding some child-visits. (DOCX) Click here for additional data file. S1 Text STROBE checklist. (DOCX) Click here for additional data file.
