Trimester-Specific Air Pollutant Exposure During Pregnancy and Infant Neurodevelopment at One Year: Insights into the Role of Inflammation and Oxidative Stress

Abstract

Prenatal exposure to air pollution is a major public health concern due to its potential to impair fetal brain development. This study examined whether maternal inflammatory and oxidative stress biomarkers mediate the association between trimester-specific air pollutant exposure during pregnancy and infant neurodevelopment at one year. We analyzed 87 mother–infant pairs from the OBESO perinatal cohort in Mexico City. Trimester-specific exposure to CO, PM₁₀, PM_2.5, SO₂, and O₃ was estimated using residential geolocation. Biomarkers were measured in the first and third trimesters by protocol, and intra-pregnancy change was calculated as Δ(3T–1T) for cytokines (IL-1β, IL-6, TNFα) and oxidative stress markers (malondialdehyde (MDA), protein carbonyls (PC), and total antioxidant capacity (TAC). Infant neurodevelopment at 12 months was assessed using Bayley-III. Exploratory mediation analyses were conducted, adjusting for gestational age at birth, pre-eclampsia, gestational diabetes, fetal growth restriction, marital status, mode of delivery, and infant sex; bootstrapping was applied to obtain robust estimates. Third-trimester CO exposure was associated with poorer receptive language (coef = 0.754, p = 0.02). PM_2.5 exposure showed direct effects on expressive language in the first (coef = 0.01, p = 0.04) and third trimesters (coef = 0.007, p = 0.015) in models including IL-1β. Third-trimester O₃ and SO₂ exposures were linked to lower expressive scores in models including TNFα (coef = 0.007, p = 0.02), MDA (coef = 0.008, p = 0.04), and PC (coef = 0.007, 95% p = 0.04). Meanwhile PM₁₀ exposure was associated with socio-emotional outcomes in models with IL-6 and TAC (coef = 0.003, p = 0.04). These findings indicate that maternal inflammation and oxidative stress biomarkers did not mediate the associations between prenatal air pollution exposure and infant neurodevelopment, and this study cannot elucidate their specific biological role in neurodevelopment.

1. Introduction

According to the World Health Organization, more than 90% of the global population is exposed to air pollution levels exceeding recommended limits, contributing to an estimated seven million premature deaths yearly [1]. Beyond its profound health impact, air pollution imposes a substantial economic burden [2], particularly in low- and middle-income countries [3]. Densely populated urban areas with high levels of industrial activity are among the most affected [2].

The primary health consequences of air pollution include ischemic heart disease, chronic obstructive pulmonary disease, stroke, lung cancer, and respiratory infections [3]. Vulnerable populations, such as children, pregnant women, older adults, and individuals with pre-existing conditions, are disproportionately affected [4]. A growing body of evidence suggests that prenatal exposure to ambient air pollution may disrupt fetal development, with adverse effects that can persist into childhood and beyond [5].

Airborne pollutants are inhaled and processed in the lungs, where they activate immune cells, such as alveolar macrophages and neutrophils, triggering systemic low-grade inflammation by releasing cytokines such as TNF-α, IL-6, and IL-1β [6,7]. Many pollutants, particularly ultrafine particles, can cross the alveolar–capillary barrier and enter maternal circulation, where they may reach the placenta and stimulate local immune responses with further cytokine production [6,7,8]. While these molecules play essential roles in pregnancy, their excessive levels have been linked to complications, such as pre-eclampsia and gestational diabetes [9], and may disrupt fetal brain development by impairing synaptic plasticity and neuronal maturation in cortical and hippocampal regions, which are critical for cognition, emotion, and language [8,10,11,12].

In parallel, exposure to PM₁₀, PM_2.5, NO₂, and O₃ has been associated with oxidative stress, characterized by increased lipid peroxidation (malondialdehyde-MDA), protein damage (protein carbonyls-PC), and reduced total antioxidant capacity (TAC) [13]. Given that the fetal antioxidant system is intrinsically limited due to high metabolic demands and immature detoxification mechanisms, such an imbalance can cause oxidative damage to vulnerable neural structures in the developing brain [14,15,16,17].

Building on this evidence, it has been hypothesized that maternal systemic inflammation and oxidative stress may mediate the relationship between prenatal exposure to air pollutants and offspring neurodevelopment [18]. However, no studies have formally assessed these biological pathways using mediation analysis. [18,19]. Mediation models enable the decomposition of total effects into direct and indirect components, offering insight into potential mechanisms and improving causal inference in observational studies [20,21]. In studies examining the impact of air pollution during pregnancy and its association with neurodevelopment, preterm birth has been explored as a mediator of academic performance [22], while biological mediators such as mitochondrial function [22] and gut microbiome composition [23] have been investigated in individuals diagnosed with autism spectrum disorder. These findings highlight the importance of mediation analyses in advancing research in this field.

This study evaluated whether maternal systemic inflammation and oxidative stress mediate the association between trimester-specific prenatal exposure to air pollutants and neurodevelopmental outcomes at 12 months of age. Specifically, we examined infant performance in receptive language, expressive language, and socio-emotional domains. To our knowledge, no prior study has jointly evaluated inflammatory cytokines and oxidative stress markers as biological pathways linking prenatal air pollution exposure with infant neurodevelopment. By integrating trimester-specific pollutant exposures with dynamic maternal molecular responses, this study addresses a critical gap in the literature and provides a novel perspective on early-life environmental vulnerability.

2. Materials and Method

2.1. Study Population

This study included mother–child pairs from the OBESO perinatal cohort, an ongoing prospective birth cohort based at the Instituto Nacional de Perinatología (INPer) in Mexico City. The cohort investigates the effects of maternal obesity and environmental exposures on child development. The study was approved by the Internal Review Board of the Instituto Nacional de Perinatología (3300-11402-01-575-17; 14 January 2024), and all participants provided written informed consent before enrollment.

Pregnant women receiving prenatal care at INPer were recruited between 2017 and 2019 at the first trimester (11.0–13.6 weeks of gestation, confirmed by ultrasound) and were followed prospectively with visits scheduled in the second (18–22.6 weeks of gestation) and third trimesters (28–34.6 weeks of gestation).

Eligibility criteria included the following: singleton pregnancy, maternal age ≥ 18 years, no pre-existing conditions (e.g., diabetes mellitus, cardiovascular, infectious, autoimmune, renal, or hepatic disease), absence of major congenital malformations, and residence within the Valley of Mexico (Mexico City and the State of Mexico). The present analysis included 87 mother–infant pairs with complete data on prenatal air pollution exposure, maternal inflammatory and stress biomarkers, and infant neurodevelopment at 12 months (Figure 1).

Maternal baseline sociodemographic and clinical characteristics—including age, socioeconomic status, intelligence quotient (IQ), pregestational body mass index BMI (kg/m²), and the presence of hypothyroidism—were assessed, and the occurrence of pre-eclampsia during pregnancy was recorded.

2.2. Air Pollution Exposure Assessment

Trimester-specific exposure to five air pollutants—carbon monoxide (CO), ozone (O₃), sulfur dioxide (SO₂), particulate matter ≤ 2.5 (PM_2.5), and particulate matter ≤ 10 μm (PM₁₀)—was estimated using daily data (January 2017 to December 2019) from the Mexico City’s official air quality monitoring network, the Red Automática de Monitoreo Atmosférico (RAMA) (http://www.aire.cdmx.gob.mx/default.php?opc=%27aKBhnmI=%27&opcion=Zg==, accessed on 20 June 2024), which includes 44 monitoring stations in the Valley of Mexico (21 in Mexico City and 13 in the State of Mexico). This approach has been implemented before for estimating ambient pollutant exposure in perinatal cohorts [13,24], as well as in non-pregnancy-related studies [25]. To estimate individual exposure to ambient air pollutants during pregnancy, the nearest air quality monitoring station was identified for each participant based on her residential location. Residential addresses were geocoded using Universal Transverse Mercator (UTM) coordinates. Geographic distances between each residence and the available monitoring stations were calculated using Geographic Information Systems (GIS).

To identify the most appropriate monitor representing each participant’s exposure, a weighted minimum distance approach was applied. This method considered not only the linear distance between residence and monitors but also the completeness of monitoring data during the gestational period. The weighted distance Dp for each monitor was calculated as follows:

$$\mathit{D}\mathit{p}={\displaystyle \frac{\mathit{D}}{\mathit{W}}}$$

where D represents the Euclidean distance (km) between residence and monitoring station, and W is a weighting factor based on the proportion of valid daily pollutant data available for the monitor during the relevant exposure window.

2.3. Maternal Inflammatory and Oxidative Stress Biomarkers

Fasting maternal blood samples were collected in red-top Vacutainer tubes (Becton-Dickinson, Franklin Lakes, NJ, USA) during the first and third trimesters of pregnancy and processed for serum separation within one hour of collection. Serum samples were stored at −70 °C until analysis.

Inflammatory cytokines (IL-6, TNF-α, IL-1β, and IL-8) were quantified using Duo-Set sandwich ELISA kits (R&D Systems, Minneapolis, MN, USA), following the manufacturer’s protocols. Detection limits were 10, 0.3, 0.5, and 0.4 pg/mL, respectively. Oxidative stress was evaluated by measuring serum malondialdehyde (MDA) using the 1-methyl-2-phenylindole method [22], protein carbonyls (PC) by the dinitrophenylhydrazine method [23], and total antioxidant capacity (TAC) using CUPRAC method [24]. All assays were run in duplicate, and intra-assay coefficients of variation were <10%.

Changes in biomarker concentrations were calculated as the arithmetic difference between third- and first-trimester levels to reduce potential confounding by pre-pregnancy exposures.

2.4. Neurodevelopment Assessment

Infant neurodevelopment was assessed at 12 months of age using the Bayley Scales of Infant Development, Third Edition (Bayley-III), administered by trained psychologists blinded to exposure data and independent from the environmental assessment team. For this study, we focused on three domains: receptive language, expressive language, and socio-emotional development. Raw scores were converted to age-standardized scaled scores (mean = 10, SD = 3) according to Bayley-III guidelines. Impaired neurodevelopment was defined as a scaled score of 7 or less (1 SD below the mean) in one or more of the assessed neurodevelopmental domains.

2.5. Statistical Analysis

Descriptive statistics were used to characterize the study population. Bivariate analyses were performed to assess associations between sociodemographic and clinical variables, pollutant exposures, and biomarker levels. Normality of continuous variables was assessed using the Kolmogorov–Smirnov test. Group differences were analyzed using t-tests for normally distributed continuous variables, Mann–Whitney U tests for non-normally distributed variables, and chi-square tests for categorical variables. All statistical analyses were conducted using R version 4.4.1 in RStudio (version 2024.09.1 + 394, “Cranberry Hibiscus”).

2.6. Mediation Analysis

Exploratory mediation analyses were conducted to test whether changes in maternal inflammatory and oxidative stress biomarkers mediated the association between trimester-specific pollutant exposures and infant neurodevelopmental outcomes, as illustrated in Figure 2. Separate models were constructed for each pair of pollutants (trimester-specific mean) and mediators (arithmetic difference Δ3T–1T). As a consequence, in all cases, references to the second trimester correspond exclusively to pollutant exposures, which were estimated separately for each trimester (T1, T2, and T3); inflammatory and oxidative stress biomarkers were consistently modeled as the difference between third- and first-trimester levels (Δ3T–1T). These analyses followed a classical mediation framework and were not intended to infer causal or effect modification pathways.

All models were adjusted for relevant clinical and sociodemographic covariates, including gestational age at birth, pre-eclampsia, gestational diabetes, fetal growth restriction, maternal educational level, socioeconomic status, marital status, mode of delivery, and infant sex. Mediation analyses were performed in R using the mediation package (version 4.5.0), with additional support from the lmtest (version 0.9-40) and sandwich (version 3.1-1) packages to obtain heteroskedasticity-consistent standard errors. To ensure robust estimation of the mediation effects, nonparametric bootstrapping with 100 simulations was conducted for each model, allowing for the construction of 95% confidence intervals for both the indirect (mediated) and direct pathways.

3. Results

This study included 87 mother–infant pairs from the OBESO perinatal cohort at INPer in Mexico City who met the established inclusion criteria and had complete data on neurodevelopmental assessment at 12 months of age. Based on their performance in the Bayley-III scales, infants were classified as having either normal or altered neurodevelopment across three domains: receptive language (Normal = 51, Altered = 36), expressive language (Normal = 47, Altered = 40), and socio-emotional development (Normal = 66, Altered = 21). These classifications were used to define comparison groups for subsequent analyses. The mean maternal age was 29.9 years, and the average pregestational BMI (pBMI) was 27.46 kg/m². No significant differences between groups were found in demographic or clinical characteristics (

Table 1. Maternal sociodemographic and clinical characteristics by neurodevelopmental domain at 12 months.

Characteristic		Receptive Language			Expressive Language			Socioemotional
		Normal (n = 36)	Altered (n = 51)	p Value	Normal (n = 47)	Altered (n = 40)	p Value	Normal (n = 66)	Altered (n = 21)	p Value
Mothers											Total (n = 87)
Maternal age (Years)		30 ± 5	30 ± 5	0.38	29 ± 5	31 ± 5	0.28	30 ± 5	30 ± 6	0.82	29.94 ± 5.2
p-BMI (kg/m2)		26.23 ± 5.3	28.23 ± 4.53	0.06	26.9 ± 5.81	28.13 ± 4.38	0.26	27.27 ± 5.44	28.08 ± 4.46	0.5	27.46 ± 5.19
Maternal education at enrollment
Less than high school		13 (14.9%)	11 (12.6%)		14 (16.1%)	10 (%)		4 (4.6%)	20 (%)		24 (27.6%)
Some high school/high school graduate		9 (10.3%)	16 (18.4%)		14 (16.1%)	11 (12.6%)		8 (9.2%)	17 (%)		25 (28.7%)
More than high school		14 (16.1%)	24 (27.6%)	0.24	19 (21.8%)	19 (21.8%)	0.79	9 (10.3%)	29 (%)	0.45	38 (43.7%)
Marital status
Single		4 (4.6%)	8 (9.2%)		6 (6.9%)	6 (6.9%)		4 (4.6%)	8 (9.2%)		12 (13.8%)
Married		15 (17.2%)	23 (26.4%)		16 (18.4%)	22 (25.3%)		9 (10.3%)	29 (%)		38 (43.7%)
Free union		17 (19.5%)	20 (23.0%)	0.88	25 (%)	12 (13.8%)	0.08	8 (9.2%)	29 (%)	0.71	37 (42.5%)
Pre-eclampsia
	Yes	4 (4.6%)	6 (6.9%)		6 (6.9%)	4 (4.6%)		1 (1.1%)	9 (10.3%)		10 (11.5%)
	No	32 (36.7%)	45 (51.72%)	0.8	41 (47.2%)	36 (41.4%)	0.94	57 (65.5%)	20 (23%)	0.47	77 (88.5%)
Gestational diabetes
	Yes	4 (4.6%)	4 (4.6%)		2 (2.3%)	6 (6.9%)		1 (1.1%)	7 (%)		8 (9.2%)
	No	32 (36.7%)	47 (54%)	0.9	45 (51.7%)	34 (39.1%)	0.175	59 (67.8%)	20 (23%)	0.71	79 (90.8%)
Gestational growth restriction
		0 (0%)	5 (5.7%)		3 (3.4%)	2 (2.3%)		2 (2.3%)	3 (3.4%)		5 (5.7%)
		36 (41.4%)	46 (52.9%)	0.85	44 (50.6%)	38 (43.7%)	0.99	63 (72.4%)	19 (21.8%)	0.94	82 (94.3%)
Infants
Gestational age at birth (weeks)		38.71 ± 1.57	38.47 ± 1.64	0.45	38.72 ± 1.54	38.52 ± 1.77	0.79	38.52 ± 1.64	38.72 ± 1.54	0.99	38.57 ± 1.61
Sex
Female		16 (18.4%)	26 (29.8%)		19 (21.8%)	23 (26.4%)		9 (10.3%)	33 (37.9%)		42 (48.3%)
Male		20 (23.0%)	25 (28.7%)	0.5	21 (24.1%)	24 (27.6%)	0.65	12 (13.8%)	33 (37.9%)	0.46	45 (51.7%)
Born from cesarean section		16 (18.4%)	27 (31.0%)	0.28	17 (19.5%)	26 (29.8%)	0.32	34 (39.1%)	9 (10.3%)	0.79	43 (49.4%)
Apgar 1 min		7.82 ± 0.55	7.62 ± 1.61	0.49	7.33 ± 1.48	7.68 ± 1.53	0.66	7.83 ± 1.49	7.33 ± 1.83	0.4	7.7 ± 1.58
Apgar 5 min		9 ± 0.001	8.96 ± 0.28	0.06	8.98 ± 0.26	8.97 ± 0.16	0.09	8.98 ± 0.22	8.95 ± 0.22	0.08	8.98 ± 0.22

Data are presented as means ± standard deviation and counts. Comparison of means by Mann–Whitney U test and chi square. pBMI: pregestational body mass index.

).

Mean trimester-specific pollutant exposures were compared across neurodevelopmental outcome groups (

Table 2. Trimester-specific average pollutant exposure by neurodevelopmental domain at 12 months.

Pollutant	Receptive Language			Expressive Language			Socioemotional
	Altered	Normal	p Value	Altered	Normal	p Value	Altered	Normal	p Value
First trimester
PM10 (µg/m3)	35.47 ± 12.75	31.13 ± 11.8	0.37	35.04 ± 12.62	32.52 ± 12.39	0.8	34.47 ± 15.68	33.42 ± 11.42	0.57
PM2.5 (µg/m3)	15.87 ± 7.82	13.15 ± 8.5	0.25	16.17 ± 6.95	13.54 ± 8.98	0.22	16.24 ± 8.87	14.27 ± 7.95	0.55
CO (ppb)	0.48 ± 0.18	0.46 ± 0.18	0.88	0.47 ± 0.17	0.48 ± 0.19	0.9	0.46 ± 0.191	0.48 ± 0.18	0.3
SO2 (ppb)	3.97 ± 2.02	3.67 ± 1.39	0.57	4.01 ± 1.99	3.72 ± 1.59	0.77	3.84 ± 2.1	3.85 ± 1.69	0.73
O3 (ppb)	25.26 ± 8.29	27.1 ± 6.7	0.13	26.86 ± 9.08	25.31 ± 6.27	0.79	23.58 ± 7.98	26.8 ± 7.48	0.02
Second trimester
PM10 (µg/m3)	31.79 ± 10.96	27.61 ± 10.98	0.04	31.16 ± 12.08	29.13 ± 10.23	0.62	32.52 ± 12.05	29.28 ± 10.76	0.28
PM2.5 (µg/m3)	14.83 ± 7.8	11.83 ± 7.72	0.12	15.32 ± 7.05	12.12 ± 8.28	0.05	14.19 ± 8.51	13.41 ± 7.7	0.64
CO (ppb)	0.42 ± 0.14	0.44 ± 0.13	0.5	0.44 ± 0.15	0.43 ± 0.13	0.77	0.43 ± 0.14	0.43 ± 0.13	0.9
SO2 (ppb)	3.47 ± 2	3.28 ± 1.58	0.71	3.62 ± 2.23	3.21 ± 1.41	0.7	3.87 ± 2.56	3.24 ± 1.52	0.47
O3 (ppb)	26.85 ± 7.73	27.35 ± 7.49	0.79	27.95 ± 8.68	26.31 ± 6.54	0.48	26.68 ± 8	27.18 ± 7.53	0.9
Third trimester
PM10 (µg/m3)	27.6 ± 9.9	24.84 ± 9.24	0.17	26.86 ± 9.95	26.12 ± 9.53	0.55	30.41 ± 9.74	25.2 ± 9.39	0.01
PM2.5 (µg/m3)	12.29 ± 7.43	11.06 ± 7.57	0.47	12.58 ± 6.82	11.1 ± 7.99	0.43	10.41 ± 6.96	12.22 ± 7.63	0.22
CO (ppb)	0.43 ± 0.11	0.38 ± 0.11	0.05	0.39 ± 0.1	0.41 ± 0.12	0.38	0.39 ± 0.1	0.4 ± 0.12	0.6
SO2 (ppb)	3.22 ± 1.59	3.53 ± 2.07	0.9	3.39 ± 1.72	3.3 ± 1.89	0.77	3.6 ± 2.18	3.27 ± 1.67	0.78
O3 (ppb)	25.29 ± 6.05	24.75 ± 6.04	0.41	26.46 ± 6.56	23.89 ± 5.3	0.06	25.96 ± 4.96	24.78 ± 6.32	0.3

Data are presented as means ± standard deviation. Comparison of means by Mann–Whitney U test.

). Infants with altered receptive language had significantly higher second-trimester exposure to PM₁₀ (31.79 ± 10.96 µg/m³ vs. 27.61 ± 10.98 µg/m³, p = 0.04) and third-trimester exposure to CO (0.43 ± 0.11 ppb vs. 0.38 ± 0.11 ppb, p = 0.05). In addition, infants with altered expressive language also had greater second-trimester exposure to PM_2.5 (15.32 ± 7.05 µg/m³ vs. 12.12 ± 8.28 µg/m³, p = 0.05). Infants with impaired socio-emotional skills had significantly higher exposure to PM₁₀ in the third trimester (30.41 ± 9.74 µg/m³ vs. 25.2 ± 9.39 µg/m³, p = 0.01).

Inflammatory and oxidative stress biomarkers were measured in maternal serum during the first and third trimesters of pregnancy (

Table 3. Maternal inflammatory and oxidative stress biomarkers during pregnancy by neurodevelopmental outcome domain at 12 months.

	Receptive Language			Expressive Language			Socioemotional
Molecule	Altered	Normal	p Value	Altered	Normal	p Value	Altered	Normal	p Value
First trimester
IL-1β (pg/mL)	91.13 ± 272.6	42.96 ± 156.6	0.1	104.6 ± 305.7	42.72 ± 139.6	0.13	122.9 ± 404.7	54.75 ± 140.0	0.6
IL-6 (pg/mL)	153.2 ± 361.6	134.7 ± 367.31	0.45	120.5 ± 218.1	166.9 ± 451.2	0.44	97.16 ± 184.9	160.9 ± 402.2	0.83
TNFα (pg/mL)	216.6 ± 372.5	194.0 ± 693.3	0.33	223.8 ± 395.8	193.3 ± 619.3	0.54	204.1 ± 373.5	208.3 ± 568.2	0.79
MDA (nmol MDA/mg dry weight)	0.04 ± 0.04	0.04 ± 0.02	0.99	0.05 ± 0.05	0.04 ± 0.02	0.64	0.04 ± 0.02	0.04 ± 0.04	0.83
PC (nmol PC/mg of protein)	6.97 ± 2.77	7.63 ± 4.17	0.86	6.83 ± 2.81	7.59 ± 3.84	0.38	5.98 ± 2.05	7.64 ± 3.66	0.08
TAC (nmol trolox equivalent)	0.08 ± 0.02	0.09 ± 0.01	0.09	0.08 ± 0.02	0.08 ± 0.01	0.62	0.09 ± 0.02	0.08 ± 0.02	0.23
Third trimester
IL-1β (pg/mL)	109.3 ± 297.6	51.87 ± 155.1	0.28	117.5 ± 328.2	58.3 ± 152.2	0.23	127.8 ± 357.7	72.1 ± 204.9	0.64
IL-6 (pg/mL)	168.8 ± 346.8	176.2 ± 486.5	0.3	125.0 ± 220.9	211.7 ± 515.7	0.55	132.6 ± 259.4	184.3 ± 445.7	0.99
TNFα (pg/mL)	229.5 ± 404.6	182.8 ± 588.3	0.6	228 ± 428.9	195.0 ± 534.6	0.68	237.2 ± 410.5	201.6 ± 510.7	0.68
MDA (nmol MDA/mg dry weight)	0.05 ± 0.06	0.06 ± 0.07	0.99	0.05 ± 0.06	0.06 ± 0.06	0.67	0.05 ± 0.05	0.05 ± 0.07	0.58
PC (nmol PC/mg of protein)	9.21 ± 6.46	9.04 ± 8.26	0.52	10.14 ± 8.93	8.28 ± 5.3	0.61	9.57 ± 6.1	9 ± 7.57	0.36
TAC (nmol trolox equivalent)	0.07 ± 0.02	0.08 ± 0.03	0.54	0.07 ± 0.02	0.07 ± 0.02	0.56	0.07 ± 0.02	0.08 ± 0.02	0.29
Difference Third–First trimester
Δ IL-1β	18.18 ± 139.3	8.91 ± 47.75	0.89	12.88 ± 142.6	15.59 ± 7.22	0.80	4.85 ± 91.13	17.36 ± 116.6	0.89
Δ IL-6	15.63 ± 249.1	41.41 ± 176.5	0.44	4.56 ± 65.7	44.8 ± 295.06	0.69	35.5 ± 224.4	23.37 ± 221.8	0.57
Δ TNFα	12.85 ± 215.7	−11.19 ± 138.3	0.92	4.22 ± 127.3	1.78 ± 227.3	0.8	33.07 ± 234.7	−6.7 ± 170.16	0.62
Δ MDA	0.01 ± 0.05	0.02 ± 0.06	0.25	0 ± 0.04	0.02 ± 0.06	0.44	0.01 ± 0.05	0.01 ± 0.05	0.5
Δ PC	2.24 ± 5.97	1.41 ± 8.13	0.39	3.32 ± 8.09	0.69 ± 5.53	0.16	3.6 ± 5.33	1.36 ± 7.3	0.3
Δ TAC	−0.01 ± 0.03	−0.01 ± 0.03	0.86	−0.01 ± 0.03	−0.01 ± 0.03	0.51	−0.02 ± 0.03	−0.01 ± 0.03	0.6

Data are presented as means ± standard deviation. Comparison of means by Mann–Whitney U test.

). For each biomarker, the arithmetic difference between third- and first-trimester concentrations was calculated to reduce the influence of pre-pregnancy baseline levels. Bivariate analyses showed no significant differences in these biomarkers’ changes between infants with normal and altered scores across any neurodevelopmental domains.

Mediation models were used to assess whether inflammatory and oxidative stress markers mediated associations between prenatal pollutant exposure and neurodevelopmental outcomes (Figure 3, Figure 4 and Figure 5). The results presented correspond to models in which a significant mediation effect was observed, while all evaluated models are available in the Supplementary materials (Supplementary Tables S1–S3).

For receptive language, no biomarker significantly mediated the associations. However, third-trimester CO exposure showed a direct effect when we evaluated Δ IL-1β (coef = 0.754, 95% CI: 0.12 to 0.93 p = 0.02) and Δ TNFα (coef = 0.75, 95% CI: 0.056 to 0.92, p = 0.04). It was similar in total effects Δ IL-1β (coef = 0.75, 95% CI: 0.12 to 0.93 p = 0.02), Δ TNFα (coef =0.75, 95% CI: 0.11 to 0.92, p = 0.02), and Δ IL-6 (coef = 0.0001, 95% CI: 0.032 to 0.093, p = 0.02). Interestingly, second-trimester PM₁₀ exposure became significant in the mediation model (direct effect: coef = 0.008, 95% CI: 0.0003 to 0.014, p = 0.04), despite being non-significant in the direct model, indicating a suppression effect.

In expressive language models, PM_2.5 exposure in the first (coef = 0.01, 95% CI: 0.002 to 0.016, p = 0.04) and third trimesters (coef = 0.007, 95% CI: 0.008 to 0.01, p = 0.015) showed significant direct associations when IL-1β was included as a mediator, although IL-1β itself was not statistically significant. PM_2.5 exposure in the second trimester showed significant total effects in models including TNFα (coef = 0.012, 95% CI: 0.0007 to 0.018, p = 0.04) and TAC (coef = 0.013, 95% CI: 0.0003 to 0.018, p = 0.04), although no significant indirect effects were detected.

PM_2.5 in the third trimester had a significant total effect (coef = 0.007, 95% CI: 0.008 to 0.015, p = 0.015) in models including IL-6 as a mediator, but neither direct nor indirect effects were significant. Third-trimester exposure to O₃ was associated with poorer expressive language scores in models including TNFα (total and direct effects: coef = 0.007, 95% CI: 0.0004 to 0.009, p = 0.02) and MDA (coef = 0.008, 95% CI: 0.002 to 0.009, p = 0.04). TAC also showed a significant direct effect (coef = 0.008, 95% CI: 0.006 to 0.011, p = 0.0001), without mediation. SO₂ exposure was associated with significant total effects when PC were included as mediators (coef = 0.007, 95% CI: 0.002 to 0.001, p = 0.04), although no significant indirect effect emerged.

For socio-emotional development, third-trimester PM₁₀ exposure was significantly associated with adverse outcomes in models including IL-6 (total and direct effects: coef = 0.003, 95% CI: 0.004 to 0.005, p = 0.04) and TAC (total effect: coef = 0.003, 95% CI: 0.0002 to 0.005, p = 0.04; direct effect: p = 0.02), although no significant mediation was observed.

4. Discussion

This study conducted exploratory mediation analyses to evaluate whether changes in maternal oxidative stress and inflammatory biomarker concentrations between the first and third trimesters of pregnancy could explain the association between prenatal exposure to air pollutants and neurodevelopmental outcomes at one year of age. Several airborne pollutants—specifically CO, PM_2.5, PM₁₀, O₃, and sulfur dioxide SO₂—were independently associated with alterations in infant neurodevelopment, particularly in receptive and expressive language domains. Despite these consistent direct associations, no statistically significant indirect effects were detected through the evaluated biomarkers, suggesting that maternal systemic inflammation and oxidative stress do not mediate the observed relationships.

In the receptive language domain, we observed that third-trimester exposure to CO was significantly associated with poorer outcomes, independent of maternal levels of IL-1β, IL-6, or TNFα. The significant bivariate association suggests a direct relationship potentially involving biological mechanisms not captured by the measured mediators. Mechanistically, CO has been proposed to induce fetal neuroinflammation and early maternal immune activation [25]. In murine models, prenatal CO exposure has been linked to impairments in spatial memory and social behavior [26], as well as disrupted neuronal migration in the fetal cerebral cortex [27]. Although evidence on CO’s effect on maternal inflammation is limited, studies have shown that CO exposure increases the expression of SOD-1, SOD-2, HO-1, iNOS, and nitrotyrosine [28], contributing to systemic hypoxia due to its 240-fold greater binding affinity for hemoglobin compared to oxygen, with consequent cardiovascular and neurological effects [29,30] Complementing our findings, a cohort study in rural Guatemala found that higher third-trimester CO exposure—measured in households using wood-burning cookstoves—was associated with poorer cognitive outcomes in children aged 6 to 7 years, particularly in memory, visuospatial abilities, and fine motor coordination [31]. While the affected domains differ from those assessed in our study, these results may reflect a shared developmental vulnerability to prenatal CO exposure. Early deficits in receptive language, such as those we observed at 12 months, could signal broader neurodevelopmental disturbances that manifest later in childhood across multiple domains.

Second-trimester PM₁₀ exposure was also linked to receptive language impairment, particularly in models that included PC, suggesting that higher PM₁₀ levels and insufficient oxidative stress defenses, may adversely affect early brain development. Third-trimester PM₁₀ exposure was linked to socio-emotional impairments, aligning with previous studies reporting PM₁₀—related alterations in brain volume and [32] white matter microstructure during childhood and adolescence [32,33].

In the expressive language domain, PM_2.5 exposure during all trimesters was consistently associated with adverse outcomes, although only second-trimester exposure reached statistical significance in bivariate analyses. In mediation models, total effects were significant in several cases, but no clear mediating or direct pathways emerged. These results support prior evidence that PM_2.5 is one of the most neurotoxic airborne pollutants, implicated in deficits in cognition, learning, memory, executive function, adaptive behavior, and attention [34,35,36]. In addition, several studies have identified the second trimester as a sensitive window during which exposure to PM_2.5 is associated with adverse neurodevelopmental outcomes. Exposure during mid-pregnancy (approximately 20–24 weeks of gestation) has been linked to lower language scores at age two [37], and exposure around 17–18 weeks of gestation has been associated with reduced cognitive performance (MDI scores). This period coincides with critical stages of synaptogenesis and neuronal migration in brain regions involved in language development [38] These associations are likely mediated by their impact on brain development, including thickness and volume of the thalamus [39], size of the hippocampus and corpus callosum [40], impaired white matter development [40,41], and disruptions in synaptic pruning, myelination, and interconnectivity [39,42].

The total effect of third-trimester SO₂ exposure was observed in the model evaluating PC as a mediator. However, studies assessing SO₂ exposure remain inconclusive regarding its association with neurodevelopmental outcomes. The findings most comparable to ours are from a study conducted in China by Yu et al., which reported associations with lower motor and language scores [43]. In contrast, three cohort studies in children with autism spectrum disorder did not identify significant associations with neurodevelopmental outcomes [44,45,46]. These results suggest that SO₂ exposure may exert subtle effects on neurodevelopment without leading to significant alterations.

O₃ exposure during the third trimester was associated with poorer expressive language scores, particularly in models that included TAC as a covariate. This suggests that both increased O₃ exposure and a decrease in antioxidant defenses are strongly associated with adverse outcomes. While the role of O₃ in neurodevelopment has received less attention, [41] prenatal exposure has been linked to intellectual disability and social integration difficulties [36,47]. Recent evidence suggests it may disrupt functional connectivity in brain regions involved in emotion and social cognition, such as the amygdala and hippocampus [42].

Despite the lack of statistically significant mediation by inflammatory or oxidative stress biomarkers, our findings suggest a biological interaction between prenatal pollutant exposure and maternal physiological responses that may influence fetal neurodevelopment. This is consistent with experimental data showing that pollutants activate inflammatory pathways (e.g., IL-1β, IL-6, TNF-α, CRP), oxidative stress responses (e.g., MDA, 8-OHdG, hypoxanthine), and innate immune signaling such as the NLRP3 inflammasome [36,48]. While the selected biomarkers may not have captured the full extent of these responses, their inclusion modified effect estimates in some models, indicating a possible modulatory role [48,49].

We observed no significant differences in maternal biomarker concentrations between infants with normal vs. altered neurodevelopment, nor did these biomarkers mediate the associations in formal models. This may reflect the involvement of alternative biological pathways such as epigenetic regulation, hormonal signaling, or placental function.

A major strength of this study is the simultaneous integration and evaluation of maternal inflammatory and oxidative stress biomarkers, providing a broader and more biologically comprehensive perspective on maternal physiological responses to environmental exposures during pregnancy. By collecting biological data at two key gestational time points (first and third trimesters), we were able to assess dynamic changes in these molecular pathways across pregnancy, rather than relying on single time-point measurements. Additionally, using a well-characterized perinatal cohort with prospective follow-up and standardized neurodevelopmental assessments at 12 months strengthens the internal validity of our findings. The trimester-specific estimation of pollutant exposures further enhances temporal resolution in exposure assessment. Finally, the application of exploratory mediation models enabled the investigation of potential biological pathways linking environmental exposures to early neurodevelopment, moving beyond traditional exposure–outcome associations and offering a mechanistic framework to guide future research and targeted prevention strategies.

However, several limitations should be acknowledged. Our biomarker panel was restricted to a subset of markers of systemic inflammation and oxidative stress, which may not fully capture the complexity of immune and redox pathways involved in prenatal environmental exposures. Additional biomarkers—such as CRP, glutathione, or 8-OHdG—could provide a more comprehensive picture of the mechanisms at play. Biomarker concentrations were only available for the first and third trimesters, and changes were estimated as a simple difference between these time points; this approach may overlook non-linear or fluctuating patterns across gestation, particularly during the second trimester. Pollutant exposure was derived from trimester-specific arithmetic means from fixed-site monitoring stations, which may not reflect individual-level variation related to mobility, indoor exposure, and microenvironmental conditions, potentially leading to exposure misclassification. However, recent studies have adopted similar methodologies for estimating ambient pollutant exposure in perinatal cohorts [13,24], as well as in non-pregnancy-related studies [25]. Although we adjusted for key demographic and clinical covariates, residual confounding cannot be ruled out. The relatively small sample size (n = 87) limited statistical power to detect indirect effects in mediation models and increased the likelihood of Type II error. Finally, although multiple associations were tested, no formal correction for multiple comparisons was applied, given the exploratory nature of the study. These considerations underscore the need for cautious interpretation and highlight the importance of replication in larger, longitudinal studies.

These findings highlight the importance of considering gestational timing and pollutant-specific effects when evaluating neurodevelopmental risk. It is important to note that the mediation models applied in this study are exploratory and do not imply causal relationships; instead, they aim to identify potential biological pathways and generate hypotheses for future research. Fetal brain vulnerability to air pollutants involves complex biological processes, potentially including disruptions in maternal oxidative stress responses. Future research should incorporate broader panel biomarkers to better understand the pathways through which prenatal air pollution exposure contributes to adverse neurodevelopmental outcomes. In addition to oxidative stress and inflammation, future studies should explore alternative or complementary mechanisms—such as epigenetic regulation, endocrine disruption, and placental response pathways—to provide a more comprehensive understanding of fetal vulnerability. A deeper understanding of these mechanisms could inform targeted public health interventions and help design clinical strategies to identify at-risk pregnancies and implement early developmental surveillance or preventive care.

5. Conclusions

Although exploratory mediation models did not yield statistically significant indirect effects, our findings indicate pollutant- and trimester-specific associations with early neurodevelopment, particularly in receptive, expressive, and socio-emotional domains. Including maternal inflammatory and oxidative stress biomarkers in the models modified the effect estimates. However, they could not be linked to the causal pathway between prenatal pollutant exposure and neurodevelopmental outcomes. Given the modest sample size and reliance on biomarker changes between the first and third trimesters, the study may have been underpowered to detect subtle mediation effects, and results should be interpreted with caution.

Our findings highlight the importance of considering gestational timing, pollutant-specific mechanisms, and maternal physiological responses in future research, and underscore the need for larger, longitudinal studies with more comprehensive biomarker panels to clarify the biological pathways through which prenatal air pollution exposure may affect child neurodevelopment.

Importantly, this study is novel in integrating trimester-specific exposures with maternal inflammatory and oxidative stress biomarkers, and in jointly evaluating their role as potential mediators or modulators of the neurodevelopmental effects of prenatal air pollution. While exploratory, this approach provides new insights into maternal molecular responses during pregnancy and their potential relevance for fetal programming, underscoring the value of replication in larger cohorts.