Residential exposure to air pollution, access to neighborhood greenspace and their association with hair cortisol concentrations in the second and third trimester of pregnancy

Background: Exposure to air pollution during pregnancy has been associated with adverse pregnancy outcomes in studies worldwide, other studies have described benecial effects of residential greenspace on pregnancy outcomes. The biological mechanisms that underlie these associations are incompletely understood. Recent studies have shown that a biological stress response, with release of cortisol, may underlie associations between air pollution and health effects. The available research on air pollution exposure in relation to biological stress during pregnancy is still scarce. Methods: We explored associations between residential exposure to air pollution, access to neighborhood greenspace and hair cortisol concentrations in a prospective pregnancy cohort study. We modelled participants’ residential air pollutant concentrations (particulate matter with an aerodynamic diameter ≤ 2.5 μm, nitrogen dioxide (NO 2 ), black carbon (BC)), assessed residential distance to a major road and access to a neighborhood greenspace. Hair cortisol concentrations, reecting cortisol secretion over a period of 3 months prior to sampling, were determined at the end of the second (n = 133) and third pregnancy trimester (n = 81). Results: Three-month mean residential NO 2 and BC concentrations were positively associated with third pregnancy trimester hair cortisol concentrations. The residential distance to a major road was negatively associated with second and third trimester hair cortisol concentrations. Access to a greenspace of 10 hectares or more within 800 meters travel distance signicantly moderated the association between residential proximity to a major road and second trimester hair cortisol concentrations. At an average residential distance of 304 meters from a major road, mean second trimester HCC were estimated 22% lower for mothers with access to a neighborhood greenspace (3.71 (95% CI: 3.24, 4.24) pg/mg hair) compared to mothers without access (4.22 (95% CI: 3.26, 5.47) pg/mg hair). The moderation tended towards signicance in the third trimester (p < 0.10). Conclusions: Increased residential exposure to air pollution and closer proximity to a major road are associated with an increased biological stress response in the second and third trimester of pregnancy, access to neighborhood greenspace may moderate the association.

impact of maternal exposure to air pollution on birth outcomes is of major public health importance, considering the ubiquitous nature of air pollution in urban settings [10]. The biological pathways that relate maternal exposure to air pollution to adverse pregnancy outcomes however, remain incompletely understood. Recent experimental animal research has shown that a neuroendocrine stress response is among the early biological responses triggered by exposure to ne particulate matter with an aerodynamic diameter ≤ 2.5 μm (PM 2.5 ) and exposure to nitrogen dioxide (NO 2 ) [11]. The biological stress response includes activation of the hypothalamic-pituitary-adrenal (HPA) axis and release of glucocorticoid stress hormones, with the glucocorticoid cortisol as its main downstream effector in humans [12]. The relevance of these experimental observations to humans has been con rmed in a few recent studies [13,14]. To date, most human studies have assessed short-term variations in cortisol levels in relation to air pollution exposure, using blood and saliva as a matrix. Longer-term cortisol concentrations are di cult to evaluate using blood and saliva, due to circadian variations in cortisol secretion and the need for multiple sampling [15]. Repeated sampling increases discomfort for study participants. Hair however, is a suitable matrix for the assessment of longer-term cortisol concentrations [16]. As cortisol is incorporated into growing hair, hair cortisol concentrations (HCC) retrospectively re ect cortisol secretion over a period of several months [17]. Importantly, chronic activation of the maternal HPA axis during pregnancy is associated with gestational hypertensive disorders, intrauterine growth restriction and developmental programming of disease susceptibility [18,19]. With regard to birth outcomes, a signi cant negative association between maternal HCC in the 2 nd pregnancy trimester and gestational age at delivery has been reported [20].
Interestingly, a growing body of research suggests a bene cial relationship between residential access to greenspaces (publicly accessible vegetation, urban parks, forests) and pregnancy outcomes for both mothers and babies [21]. Residential exposure to air pollution and greenspace occur simultaneously, therefore, the adverse effects of air pollution may, to some degree, be moderated by the bene cial effects of residential access to greenspace [22].
To our knowledge, no research is available on residential air pollution exposure or access to neighborhood greenspace in relation to longer-term biological stress during pregnancy. Accordingly, the main aim of this study was to explore associations between residential exposure to air pollution and road tra c and maternal hair cortisol concentrations in a pregnancy cohort. In addition, we aimed to explore whether residential access to neighborhood greenspace could moderate associations between residential air pollution exposure and maternal biological stress during pregnancy.

Study population and design
This study was conducted in the framework of IPANEMA (Impact of Particulate Matter on Mothers and Babies in Antwerp), a prospective pregnancy cohort study of the Antwerp University Hospital (UZA) in collaboration with the Flemish Institute for Technological Research (VITO) and the University of Antwerp (UA). Pregnant women were recruited between April 2015 and January 2018 at the UZA prenatal clinic by a midwife or obstetrician at a gestational age of 12 to 14 weeks. The inclusion criteria were: a singleton pregnancy; the ability to ll out extensive Dutch questionnaires; delivery planned in the Antwerp University Hospital. All participating mothers gave written informed consent. The study protocol was approved by the ethical committee of the University of Antwerp (14/40/411) and registered under number NCT02592005 at clinicaltrials.gov. Health-related information on mothers and babies was extracted from the hospital records and questionnaires that participants completed at enrolment, during pregnancy and after delivery. These questionnaires provided detailed information on participants' socio-demographic and lifestyle characteristics. A detailed protocol of the IPANEMA study can be found elsewhere [23].

Hair sample collection and cortisol measurement
At the end of second trimester of pregnancy in-hospital consultation and shortly after delivery, a strand of hair of at least 2mm thick was bound together with a cotton thread and cut close to the scalp from the posterior vertex region of the head. This area of the scalp exhibits the lowest intra-individual variability in HCC [24]. Hair samples were stored in paper envelopes at room temperature until analysis. When protected from ultraviolet light, cortisol concentrations in hair samples remain stable at room temperature for several years [25]. Cortisol concentrations were determined from the 3 cm of hair closest to the scalp. Based on an average hair growth of 1 cm per month, this length represents cortisol secretion in a 3-month period, a trimester, prior to sampling [26]. There is a wide consensus that the rst 5-6 cm of hair nearest to a person's scalp can reliably re ect HPA activity [27]. Analysis was performed at the Institute of Public Aliquots of 100 mL 20 ng/mL isotope labeled cortisol (cortisol-D 4 ) were added as internal standard, together with 0.9 mL methanol. Samples were incubated in the dark at 25°C while whirl mixed at 2000 revolutions per minute for 5 days and subsequently centrifuged at 3000g for 5 minutes. 20 μL of the supernatant was injected onto a High-Performance Liquid Chromatography (HPLC) column. HPLC was performed using an Accella 1250 pump (Thermo Scienti c, San Jose, CA) and a PAL autosampler (CTC analytics, Zwingen, Switzerland). The analytical column was a Kinetex C18 column, 100 x 4.6 mm (2.6 μm) equipped with a 2 x 4 mm C18 SecurityGuard column (Phenomenex, Torrance, CA). Isocratic elution was performed with a mobile phase system consisting of methanol and 0.1 M formic acid (80:20) at a ow rate of 400 μL/min for 6 min. After the peaks were eluted, a wash procedure was performed before the next samples was injected onto the column. The triple quadrupole mass spectrometer utilized was a TSQ Vantage (Thermo Scienti c, San Jose, CA). The calibration curve and calculation of the sample concentration were based on the area ratio of the analyte/isotope labeled internal standard. Quality control samples were included in each series of samples. The limit of quanti cation (LOQ) for cortisol was 1.0 pg/mg hair. The intra-day repeatability coe cient of variation was 8.7% and the inter-day reproducibility coe cient of variation was 9.5% Residential exposure assessment Assessment of all residential exposure variables was based on the participants' geocoded home address.
Geographical Information System (GIS) analyses were carried out using ESRI ArcGIS software version 10.4 (Environmental Systems Research Institute, Redlands, California, USA). The residential degree of urbanization was assessed according to the Eurostat de nition that classi es local administrative units as cities, towns, suburbs or rural areas based on a combination of geographical contiguity and population density, applied to 1 km² population grid cells [29]. We assessed residential exposure to ne particulate matter (PM 2.5 ), nitrogen dioxide (NO 2 ) and black carbon (BC), primary constituents of tra crelated air pollution. Residential exposure to PM 2.5 , NO 2 and BC was modelled using a spatial temporal interpolation method. In Flanders, atmospheric pollutants are continuously measured by a network of automatic monitoring stations by the Flemish Environment Agency. The Belgian Interregional Environment Agency (IRCEL, Intergewestelijke Cel voor het Leefmilieu) uses these measurements together with information on land cover to interpolate the air pollutant concentrations on a 4x4 km² resolution [30]. These background results are combined with a bi-gaussian dispersion model based on emissions from point sources and line sources, the Immission Frequency Distribution Model (IFDM). The combined RIO-IFDM model chain produces daily averaged pollutant concentrations in Belgium on a high resolution receptor grid [31]. We calculated mean air pollutant concentrations at the residential address over a 3month period before sampling, similar to the period of cortisol accumulation in the hair samples, and over a 1-year period before sampling. Residential proximity to major roads is often used as a surrogate measure of long-term exposure to tra c-related air pollution [32]. We calculated the straight-line distance from each residence to the nearest major road. Major roads included international motorways (E-roads) and the network of large national and local roads of Belgium (N-roads).
Residential access to a neighborhood greenspace was based on the 2016 version of the land-use map of Flanders, which maps land cover types, i.e. natural vegetated land cover and urban greenery, in 10x10 m² raster cells [33]. Green cells were clustered to assess the area and public accessibility of greenspace in the maternal residential surroundings. Access to a small neighborhood greenspace (NHGS) was de ned as access to at least 0.2 hectares (ha) of greenspace within a travel distance of 400 meters (m) from residence, access to a large greenspace was de ned as access to at least 10 ha of greenspace within a travel distance of 800 m from residence. In the large greenspace typology, small water bodies are included when surrounded by > 50% greenspace, agricultural land is included when surrounded by > 30% greenspace. More technical background information on the green typology indicators can be found elsewhere [34].

Potential covariates of hair cortisol concentrations
Possible covariates of HCC were identi ed based on available data within the IPANEMA cohort and on existing literature [35]. Tested covariates included maternal age, parity, maternal socioeconomic status (SES) de ned as the highest educational attainment of the mother and categorized as low/intermediate/high, pre-existing chronic diseases (diabetes, asthma, cardiovascular disease), pre-pregnancy body mass index (BMI), gestational week at sampling, season of sampling, smoking and alcohol consumption before pregnancy, systemic use of glucocorticoids and daily hair washing. We assessed maternal ethnic background as European/non-European country of birth since hair growth rate may be in uenced by ethnicity [36]. We additionally tested variables that may have a link with both residential environment and biological stress, i.e. neighborhood SES and residential exposure to noise. A systematic review in the World Health Organization (WHO) European Region showed that lower neighborhood SES is usually linked with higher levels of air pollutants [37]. Independent from higher levels of exposure, deprived mothers may have a higher vulnerability, leading to more pronounced adverse health effects of a given environmental exposure [38]. The Area Deprivation Index (ADI) is a yearly calculated indicator for neighborhood SES on a sub-municipality level in Flanders (Statistics Flanders, n.d.). Deprivation is recorded by the Flemish Child and Family Government Agency (www.kindengezin.be). Selection criteria for deprivation are the family's monthly income, the parents' educational attainment, the children's development, the parents' employment situation, housing and health. If a family ful ls three or more criteria, it is considered to be underprivileged (OECD, 2000). The index of year X (%), i.e. 2017, considers all children born in year X, X-1 and X-2 that live in deprived households in a given area in Flanders, divided by the total number of children born in the area during the same period. The ADI of the participants' neighborhood was subsequently categorized into tertiles representing low, intermediate and high area deprivation across the range of ADI among all participants.
Residential proximity to major roads may also lead to elevated noise levels [41]. Residential noise exposure levels were assessed using the Flemish strategic noise map of 2016, which includes major road infrastructure as de ned in the EU-guideline 2002/49/EG [42]. The strategic noise map expresses noise levels in L den , the average sound level over a 24 hour period with a penalty of 5 dB added for evening hours and a penalty of 10 dB added for nighttime hours [43]. The WHO guideline for average noise exposure produced by road tra c is set at 53 decibels (dB) L den , road tra c noise above this level has been associated with adverse health effects, including adverse birth outcomes [44]. Noise exposure was therefore evaluated binary as exposure to a noise level ≥ 53 dB L den .

Statistical analysis
Statistical analysis was performed using SPSS Statistics (version 25; IBM, Armonk, NY, USA) and R (version 2018; R Foundation for Statistical Computing, Vienna, Austria). Descriptive statistics provide an overview of study population characteristics, residential exposure characteristics and geometric mean HCC concentration with 95% con dence interval. Air pollution variables were logarithmically transformed (ln-scale) because of skewed distributions, distance to major roads was logarithmically transformed to re ect the non-linear distance decay of tra c-related exposure to air pollutants [45].
Spearman rank correlations between residential exposures variables were assessed, since correlations of 0.9 or higher between exposure variables indicate strongly connected exposures that cannot be disentangled [46]. The outcome variable HCC was logarithmically transformed to obtain a normal distribution. For HCC below the LOQ of 1 pg/mg hair, a random imputation from a log-normal probability distribution was performed where the mean was allowed to depend on observed values for hair cortisone concentrations that were measured simultaneously with cortisol, since both glucocorticoids were highly correlated (p < 0.01, Pearson's r = 0.711 for 2 nd trimester cortisol and cortisone, p < 0.01, Pearson's r = 0.758 for 3 rd trimester cortisol and cortisone). Linear regression models were used to analyze associations between 3-month mean air pollutant concentrations (PM 2.5 , NO 2 , BC) , distance to major roads and access to greenspace as a predictor and 2 nd and 3 rd trimester Hair Cortisol Concentrations as an outcome. Given the limited number of study participants, we decided not to adjust for a set of a priori selected covariates. The nal regression models were only adjusted for signi cant covariates (p < 0.05). All assumptions of linear regression were checked. To quantify the association, the estimated change in HCC (β) with 95% con dence interval (95% CI) is presented for an increase in exposure from the 25 th to the 75 th percentile.
Effect modi cation by access the neighborhood greenspace was assessed by adding the interaction term of exposure to air pollution or distance to major roads and access to greenspace into the regression model. The level of signi cance for estimates was set at p < 0.05.
We conducted several sensitivity analyses to evaluate our results. We tested 1-year mean air pollutant concentrations in relation to 2 nd and 3 rd trimester HCC to con rm the robustness of 3-month mean results. We additionally adjusted our 2 nd trimester models for frequency of hair washing, a signi cant determinant of 2 nd trimester HCC, independent of biological stress.

Results
Hair samples for cortisol analysis were provided by 152 participants. Characteristics of the study population are described in Table 1. We excluded 3 participants due to inexplicable high HCC values (> 3 times the interquartile range above the third quartile). As a result, 149 pregnant women were included in this study, of which 133 women donated a sample at the end of the 2 nd trimester (week 26 ± 1.6) and 81 women shortly after delivery (week 39 ± 1.6), 65 women donated a sample twice. Almost half of the 149 mothers (48%) was aged between 26 and 30 years, 61% of participants were primigravid. Most of the study participants were of European origin (75%, 21% data missing), enjoyed higher education (57%, 23% data missing) and were employed prior to their pregnancy (72%, 23.5% data missing).
Residential characteristics are described in Table 2. Study participants lived in cities (38%), towns and suburbs (62%) in Flanders, none of the participants lived in a rural area. The mean ADI of our study population was 16.4% (95% CI: 14.6, 18.1) whereas the mean 2017 ADI for the study region Antwerp was 17.6% (Statistics Flanders, n.d.). We tested the signi cance of the association between ADI as an arealevel SES indicator and maternal educational attainment as a personal SES-indicator. We did not observe a signi cant association between neighborhood SES and personal SES (Spearman rank r = -0.074, p = 0.404). A small neighborhood greenspace was accessible for 94% of participants, 76% had residential access to a large greenspace. Three-month geometric mean PM 2.5 was 11.61 (95% CI: 11.06, 12.21) µg/m 3 and 11.55 (95% CI: 10.95, 12.18) µg/m 3 for 2 nd trimester and 3 rd trimester sampling respectively.
Spearman rank correlations of residential exposure characteristics are presented in Table 3 (2 nd trimester study population) and Table 4 (3 rd trimester study population).
We observed strong positive correlations between 3-month mean air pollutant concentrations (r ranged from 0.61 to 0.89). Distance to major roads was negatively correlated with NO 2 and BC concentrations (r ranged from -0.24 and -0.37), but not with PM 2.5 concentrations.
Access to a neighborhood greenspace did not signi cantly correlate with air pollutants and distance to major roads in the 2 nd trimester. In the 3 rd trimester study population, we did nd weak negative correlations between access to a large neighborhood greenspace and air pollutants (r ranged from -0.28 to -0.31) and a weak positive association of access to a large neighborhood greenspace with distance to major road (r = 0.26). The ADI was weakly positively correlated with NO 2 , BC and noise exposure above the WHO guideline (r ranged from 0.21 to 0.46) and negatively correlated with distance to major roads and access to a large neighborhood greenspace (r ranged from -0.18 to -0.25).
Season of sampling and daily hair washing were identi ed as signi cant covariates of 2 nd trimester HCC, no signi cant covariates were identi ed for 3 rd trimester HCC (see Table S1 for details). None of the participants reported the systemic use of glucocorticoids. Residential noise exposure above the WHO guideline (≥53 dB L den ) was not signi cantly associated with 2 nd or 3 rd trimester HCC (p = 0.871, p = 0.190 respectively). Nor did we nd signi cant associations between the ADI and 2 nd or 3 rd trimester HCC (p = 0.661, p = 0.388 resp.).
Results of the associations between air pollution exposure, access to neighborhood greenspace and maternal biological stress are presented in Table 5. We found a signi cant negative association between 3-month mean PM 2.5 concentrations and 2 nd trimester HCC in the unadjusted model (p = 0.009), the association did not remain signi cant after adjustment for season of sampling (p = 0.357). In the 3 rd trimester, 3-month mean PM 2.5 concentrations were not signi cantly associated with HCC.
We did not observe signi cant associations between 3-month mean NO 2 and BC concentrations and 2 nd trimester HCC. We did observe a signi cant positive association between 3-month mean NO 2 concentrations and 3 rd trimester HCC (p = 0.016). For an increase of 3-month mean residential NO 2 concentrations from 18.35 µg/m 3 (p25) to 30 µg/m 3 (p75), an increase of 3 rd trimester HCC with a factor 1.42 (95% CI: 1.07, 1.88) was estimated. We also observed a signi cant positive association between 3month mean BC concentrations and 3 rd trimester HCC (p = 0.032). For an increase of 3-month mean residential BC concentrations from 0.84 µg/m 3 (p25) to 1.48 µg/m 3 (p75), an increase of 3 rd trimester HCC with a factor 1.37 (95% CI: 1.03, 1.82) was estimated. Residential 3-month mean NO 2 concentrations explained 6.2% of the variations in 3 rd trimester HCC, 3-month mean BC concentrations explained 4.6%.
Residential distance to a major road was negatively associated with second trimester HCC (p = 0.016 unadjusted model, p = 0.11 adjusted model). For an increase in distance to a major road from 143 m We tested whether access to a neighborhood greenspace moderated the associations between air pollution exposure, proximity to major roads and maternal biological stress. We found no signi cant interaction between access to a small or large neighborhood greenspace and air pollution constituents in relation to 2 nd or 3 rd trimester HCC (see Table S2 for details). We did observe a signi cant interaction between access to a large neighborhood greenspace (NHGS) and distance to major roads in relation to 2 nd trimester HCC in both the unadjusted model, as presented in Figure 1, and the model adjusted for season of sampling (p = 0.021, p = 0.034 resp.). At an average residential distance of 304 meters from a major road, estimated mean 2 nd trimester HCC were 22% lower for mothers with access to a large neighborhood greenspace (3.71 (95% CI: 3.24, 4.24) pg/mg hair) compared to mothers without access  Figure 1 Interaction between distance to a major road and access to a large neighborhood greenspace in relation to 2nd trimester HCC The interaction between access to a large neighborhood greenspace (NHGS) and distance to major roads in relation to 3 rd trimester HCC tended towards signi cance (p = 0.073), as presented in Figure 2. At an average residential distance of 264 m from a major road, the model estimated 19% lower mean 3 rd trimester HCC for mothers with access to a large neighborhood greenspace (5.46 (95% CI: 4.30, 6.95) pg/mg hair) compared to mothers without access (6.75 (95%CI: 4.46, 10.22) pg/mg hair). The interaction model explained 7.9% of the variations in 3 rd trimester HCC. Figure 2 Interaction between distance to a major road and access to a large neighborhood greenspace in relation to 3rd trimester HCC In a sensitivity analysis, we evaluated the signi cance of associations between 1-year mean PM 2.5 , NO 2 and BC concentrations and HCC to re ect the participants' longer-term residential exposure to air pollution. Results are presented in Table S3. Extending the exposure period did not change our results, we found signi cant positive associations between 1-year mean NO 2 concentrations and 3 rd trimester HCC ( p = 0.013) and between 1-year mean BC concentrations and 3 rd trimester HCC (p = 0.046). The robustness of our results was also evaluated by additional adjustment of our 2 nd trimester model with daily hair washing, a signi cant but external, non-biological determinant of HCC. Our results, presented in Table S4 and S5, remained robust. Residential distance to a major road, adjusted for season of sampling and daily hair washing was signi cantly associated with 2 nd trimester HCC (n = 103, p = 0.006, R 2 = 17.4% ). In accordance, access to a large greenspace signi cantly moderated the association between distance to major roads and 2 nd trimester HCC in the model adjusted for season of sampling and daily hair washing (n = 103, p = 0.001, R 2 = 22.1%). The model estimated 17.66% lower mean 2 nd trimester HCC for participants with access to large NH greenspace compared to participants without access to large NH greenspace (

Discussion
This study provides new insights in the relation between residential exposure to air pollution and road tra c and hair cortisol as a biomarker for longer-term biological stress during pregnancy. We observed signi cant positive associations between residential 3-month mean NO 2 and BC concentrations and maternal biological stress in the 3 rd pregnancy trimester. It should be noted that NO 2 exposure levels were strongly correlated with BC exposure levels (r=0.89), making it impossible to disentangle the effects of both pollutants. In urban settings, road tra c is the principal source of NO 2 and BC in ambient air [47]. We also observed signi cant associations between residential proximity to major roads and maternal biological stress in the 3 rd pregnancy trimester. As previously reported, residential proximity to major roads and maternal biological stress in the 2 nd pregnancy trimester were signi cantly associated [48].
The difference in signi cant associations between tra c-related exposures and 2 nd and 3 rd trimester HCC may be due to the difference in study population between both trimesters or to the increase in circulating cortisol concentrations towards the end of pregnancy, which is a normal biological process [20]. Our observations are in line with recent human studies that reported associations between air pollutants and short-term variations in cortisol secretion. In a panel study among 43 students in Shanghai, residential exposure to PM 2.5 was associated with higher serum cortisol levels [13]. In a cross-sectional analysis of 1793 adults, residential NO 2 exposure was associated with higher wake-up salivary cortisol [14]. To our knowledge, only one epidemiological study has examined the association between personal air pollution exposure and HCC; the study, including Belgian schoolchildren and adolescents, did not nd a signi cant relationship [49]. Pregnancy however, is a vulnerable period for both mother and fetus [50]. Several mechanisms potentially underlie the association between air pollution and biological stress during pregnancy. Air pollutants may induce oxidative stress and low grade in ammation [51]. Depending on size and chemical composition, inhaled air pollution constituents may translocate from the lungs to the systemic circulation or migrate via olfactory transport to the brain and directly interact with brain tissues including the hypothalamus [11]. During pregnancy, oxidative stress is known to be higher than in the non-pregnant state; residential exposure to air pollutants and road tra c may further amplify the level of maternal oxidative stress [52]. Oxidative stress may in turn lead to low grade in ammation and HPA axis activation, resulting in a marked increase in the secretion of cortisol into the circulation [53]. In addition to indirect activation of the HPA axis by systemic low grade in ammation, low grade in ammation in the brain may directly activate the hypothalamus [54]. Flanders, the IPANEMA study region, is characterized by a dense road network and high emissions from tra c [55]. The fraction of the Flemish population, living and working in close proximity to tra c, is high and access to neighborhood greenspace is typically limited. Interestingly, n our urban and suburban pregnancy cohort, access to signi cantly moderated the association between residential proximity to tra c and maternal biological stress in the 2 nd pregnancy trimester. Bene cial relationships between residential access to greenspace and hair cortisol concentrations have been described in previous studies [56,57]; whereas other studies have reported a bene cial impact of surrounding greenness on fetal growth and birth weight [58-61]. Neighborhood greenspace may improve health by relieving psychophysiological stress, supporting physical activity, increasing social contacts and by reducing exposure to air pollution, noise and excessive heat [62-65]. In our study, we found a weak inverse correlation between access to a large greenspace and residential air pollutant concentrations in the 3 rd pregnancy trimester, but not in the 2 nd trimester. This suggest a moderating effect of residential access to a large greenspace, independent of the effect on air pollution exposure levels.
The added value of prospective cohort studies such as IPANEMA, is the possibility to provide more insight into early pathophysiological mechanisms, triggered by air pollution exposure, in real-world settings. The urban and suburban character of the IPANEMA cohort made it possible to go deeper into tra c-related air pollution exposure, notwithstanding the low number of participants. Residential exposure to air pollution was estimated using a high spatial resolution model, residential mobility of participants was considered. In addition to maternal tra c and air pollution exposure, we took access to neighborhood greenspaces into account. Evaluation of only one residential environmental exposure i.e. air pollution, ignoring potential interaction with other jointly occurring exposures i.e. access to greenspaces, could lead to an inaccurate estimate of the true effect of exposures [66]. We measured hair cortisol concentrations, a novel method in epidemiological studies to retrospectively determine longer-term biological stress in a noninvasive and reliable way. Hair samples were collected according to a strict protocol by trained midwives at the in-hospital consultation to avoid interindividual differences in hair collection. Some limitations of the study need to be addressed. In this study, we had a limited number of participants and did not have the same study population in the second and third pregnancy trimester, leading to differences in residential exposures. We had a considerable percentage of missing questionnaire-based data. Future prospective cohort studies, ideally including a larger number of participants from pre-conception onwards, should enhance efforts to collect questionnaires from all participants, including relevant information on time spent in residential greenspaces, physical activity, wellbeing and health. The exposure assessment was limited to residential surroundings, we did not consider air pollution exposure while commuting and working. IPANEMA participants were mostly of a higher socioeconomic status, it was therefore not possible to explore increased vulnerability to environmental exposure in participants with lower SES.
Moreover, the neighborhood SES indicator in this study did not re ect participants with lower socioeconomic status. In literature, the pattern of air pollution is often described as U-shaped , although the most deprived areas have the highest levels of poor air quality, the least deprived areas also experience higher levels of air pollutants than some other social groups [37]. In our cohort, the mean ADI was 18.1% for lower educated mothers, 15.3% for medium education mothers and, as described in literature, we observed a slightly higher ADI of 15.6% for higher educated women. Future studies should enhance efforts to include participants of all SES.

Conclusions
This study observed signi cant positive associations between residential exposure to tra c-related air pollution during pregnancy and longer-term biological stress in the 2 nd and 3 rd trimester of pregnancy. In the 2 nd trimester of pregnancy, the association was signi cantly moderated by residential access to a large neighborhood greenspace. Air pollution and urban spatial planning are in the center of public debate in Flanders. Because of the ubiquitous nature of tra c-related air pollution and the adverse pregnancy outcomes that have been associated with elevated maternal biological stress for both mothers and babies, even a small increase in maternal biological stress may be of public health interest. Our research, if con rmed in future studies, may provide guidance towards a more sustainable urban planning and support environmental health protection for both pregnant women and their babies.      Figure 1 Interaction between distance to a major road and access to a large neighborhood greenspace in relation to 2nd trimester HCC