Associations between oxidative stress biomarkers during pregnancy and infant cognition at 7.5 months

Abstract

Oxidative stress has been identified as an important biological pathway leading to neurodevelopmental delay. However, studies assessing the effects of oxidative stress on cognitive outcomes during infancy, a critical period of neurodevelopment, are limited. Our analysis included a subset of those enrolled in the Illinois Kids Development Study (IKIDS; N=144). Four oxidative stress biomarkers (8-isoprostane-PGF_2α, 2,3-dinor-5,6-dihydro-8-iso-PGF_2α, 2,3-dinor-8-iso-PGF_2α, and prostaglandin-F_2α) were measured in 2^nd and 3^rd trimester urine and were averaged. Infant cognition was measured using a visual recognition memory task consisting of five blocks, each with one familiarization trial (two identical stimuli) and two test trials (one familiar and one novel stimulus). Outcomes measured included average run duration (a measure of information processing speed), novelty preference (a measure of recognition memory), time to reach familiarization and shift rate (measures of attention). Linear regression was used to estimate associations between individual oxidative stress biomarkers and each outcome. Increasing 8-isoprostane-PGF_2α, 2,3-dinor-8-iso-PGF_2α, and prostaglandin-F_2α were associated with a decrease in novelty preference (β=−0.02, 95% confidence interval [CI]=−0.03, 0.00; β=−0.02, 95% CI=−0.04, 0.00; β=−0.01, 95% CI=−0.02, 0.00, respectively), as well as a modest increase in shift rate. These findings suggest that oxidative stress may be associated with poorer recognition memory in early infancy.

Introduction

Prenatal exposures to chemical and non-chemical stressors have been associated with impaired neurodevelopment and cognitive abilities.^2–4 Despite these well-established associations, the underlying biologic mechanisms remain largely unknown. Oxidative stress, defined as the imbalance between the reactive oxygen species (ROS) and antioxidant systems in the body, has been proposed as a key mechanism linking chemical and non-chemical stress exposures to cognitive development. For example, numerous epidemiologic studies have shown that prenatal exposure to environmental chemicals, including phthalates, metals, and per- and polyfluoroalkyl substances, are associated with an increase in multiple oxidative stress biomarkers during pregnancy.^5–8 Experiences of stressful life events and socioeconomic deprivation have also been linked to elevated oxidative stress biomarkers in pregnant and non-pregnant populations.^9,10 Exposure to psychosocial stress and environmental toxicants during pregnancy has also been associated with a reduction in telomere length.^11–13 It is hypothesized that telomere shortening, which has been linked to cognitive delay,¹⁴ occurs as a result of an increase in ROS and oxidative stress.¹⁵

Animal and preliminary epidemiologic studies support the notion that oxidative stress is on the causal pathway to poor neurodevelopment.¹⁹ Specifically, cross-sectional human studies find that oxidative stress levels, commonly measured by the biomarker 8-isoprostane-prostaglandin-PGF_2α (8-iso PGF_2α), are elevated among those with Autism Spectrum Disorder (ASD) and Attention Deficit Hyperactivity Disorder (ADHD) relative to neurotypical controls.^20–24 Lower prenatal total antioxidant status has also been associated with worse neuro-motor outcomes in offspring at one year of age.²⁵ Further, a 2020 study observed that prenatal levels of 8-iso-PGF_2α were associated with greater social impairment and that higher prostaglandin-F_2α (PGF_2α) levels were associated with worse externalizing problems at age 4,²⁶ indicating potential direct effects of oxidative stress on the developing brain. Prior work also finds that maternal oxidative stress biomarkers are moderately to strongly correlated with newborn oxidative stress measured in cord blood,^27,28 supporting the notion that maternal oxidative stress in utero is of importance to the developing fetus.

To date, few studies have examined neurodevelopment during early infancy, a critical developmental period for early attention, memory, and information processing. To address this knowledge gap, we leveraged the Illinois Kids Development Study (IKIDS) to evaluate the relationship between prenatal oxidative stress biomarkers and infant cognition at 7.5 months using a visual recognition memory (VRM) task.¹ We have previously shown that the VRM task is sensitive to in utero chemical and non-chemical stress exposures.^11,29 We included four F₂-isoprostanes (F₂-IsoPs) which are widely considered to be the “gold standard” indicators of lipid peroxidation, and are positively correlated with one another.³⁰ We quantified 8-iso-PGF_2α, as it is the most widely studied F₂-IsoPs in epidemiologic studies. We also included 2,3-dinor-5,6-dihydro-8-iso-PGF_2α and 2,3-dinor-8-iso-PGF_2α, two of the major metabolites 8-iso-PGF_2α, as they may be more sensitive than the parent compound when measured in urine.³¹ We also included PGF_2α, as elevated levels are associated with respiratory distress in early life,³² suggesting that PGF_2α may play a role in the development of brain injury.³³ We hypothesized that elevated levels of oxidative stress would be negatively associated with infant cognition.

Methods

Study Population

This analysis included 144 mother-child pairs enrolled in the Illinois Kids Development Study (IKIDS). This subset included those for whom information on maternal urinary oxidative stress biomarker concentrations were available and the infant had completed the VRM task at 7.5 months (Figure S1). Information on recruitment and the study population has been described in detail elsewhere.³⁴ Briefly, pregnant persons were eligible for inclusion in IKIDS if they were between 18 and 40 years of age, able to communicate in English, not pregnant with multiples, considered a low-risk pregnancy (e.g., the pregnancy had not been classified as high-risk by their doctor for a reason other than advanced maternal age), and resided within 30 minutes of the University of Illinois campus. A self-reported interview questionnaire was used to ascertain information regarding maternal education, maternal age, maternal race/ethnicity, and marital status. Information on pre-pregnancy body mass index (BMI; kg/m²), parity, infant sex and gestational age at delivery were obtained via a combination of self-report and medical record abstraction. All participants provided written, informed consent prior to enrollment and the Institutional Review Board at the University of Illinois Urbana-Champaign (09498) approved this study.

Measurement of urinary oxidative stress biomarkers

Urine samples were collected at up to two timepoints during pregnancy (mean 17.2 weeks gestation and 23.5 weeks gestation) and were frozen at −80 °C prior to analysis of oxidative stress biomarkers at the Eicosanoid Core Laboratory at Vanderbilt University Medical Center. Using analytic techniques previously described in detail,^8,35 we quantified urinary levels of four oxidative stress biomarkers using liquid chromatography–mass spectrometry. These compounds included 8-iso-PGF_2α, its two major metabolites, 2,3-dinor-5,6-dihydro-8-iso-PGF_2α, 2,3-dinor-8-iso-PGF_2α, and PGF_2α. Values below the limit of detection (LOD) were imputed using the LOD divided by the square root of 2.³⁶ We accounted for urinary dilution by correcting the oxidative stress biomarkers with specific gravity using the equation: ${\mathrm{O}\mathrm{X}}_{\mathrm{c}}=\mathrm{O}\mathrm{X}\left(\frac{\left({\mathrm{S}\mathrm{p}\mathrm{G}}_{\mathrm{m}\mathrm{e}\mathrm{d}\mathrm{i}\mathrm{a}\mathrm{n}}-1\right)}{(\mathrm{S}\mathrm{p}\mathrm{G}-1)}\right)$. In this equation, SpG_median is the median specific gravity in the IKIDS cohort (1.015), SpG is the specific gravity level, OX is the uncorrected oxidative stress biomarker level and OX_c is the specific gravity-corrected oxidative stress biomarker level. For participants with repeated measures of oxidative stress biomarkers available, we calculated the geometric average to obtain a more stable estimate of oxidative stress across gestation. If only one measure was available, we used only that measure. The geometric averaged oxidative stress biomarkers were used as our primary exposures in downstream analyses.

Assessment of Visual Recognition Memory (VRM)

Infant cognition was assessed at approximately 7.5 months of age using an automated infrared eye-tracking VRM task. This task has been previously described in detail,¹ and is based on work by Rose et al. showing that outcome measures from a nonautomated version of this VRM task are predictive of cognitive outcomes later in life.^37–39 The assessment was completed at the lab in a dark room, where infants were seated on a parent’s lap in front of a large high definition television screen that displays the stimuli. Infant looking behavior was recorded using an EyeLink 1000 Plus infrared eye tracker. The area surrounding the infant and their caregiver was surrounded by black curtains to minimize distractions. Prior to the task, infants’ gaze was drawn to three different points on the screen using animated audiovisual clips to calibrate the eye tracker. Calibration was validated by repeating this process to ensure the infant looks at the same three locations. Before each trial, infants’ gaze was drawn to the center of the screen to ensure that looking was not biased by where infants happened to be looking when the trial started. Caregivers were instructed to look down at their infant’s head, so as to not influence their infant’s looking behavior during the task.

The task consisted of five blocks of three trials each with each block consisting of one familiarization trial followed by two test trials. In the familiarization trial, infants saw two identical black-and white photographs of human faces side by side until they accumulated 20 seconds looking at them. Approximately half of the infants were familiarized to stimulus set 1 and the other half of infants were familiarized to stimulus set 2, such that the novel faces in set 1 were the familiar faces in set 2 and vice versa (Figure S2). In each test trial, the familiar stimulus was paired with the novel stimulus and shown until the infant accumulated a total of 1 second looking at the stimuli, plus 5 additional seconds, regardless of looking behavior. All test trials were counterbalanced so that half of the infants saw the novel stimuli on the right side first and the other half of the infants saw the novel stimuli on the left side first.

In the familiarization trials, we quantified shift rate (the number of times infants shift their gaze between the left and right stimuli) as a measure of visual attention, as infants who shift more are attending to and comparing the two stimuli. Average run duration (length of time infants spent looking at stimuli before looking away) was included as a measure of information processing speed, as faster processors tend to have shorter average run duration. Time to reach familiarization (time to reach familiarization criterion) was included as an additional measure of visual attention, as more time off-task will lead to longer time to meet the 20 second looking time criterion. In the test trials, novelty preference (the proportion of time spent looking at the novel face) was used as an indicator of recognition memory, with more time spent looking at the novel over the familiar face being indicative of a better memory of the familiar stimulus. Data were processed using the SR Research DataViewer software (SR Research Ltd., Mississauga, Ontario, Canada) and all cognitive measures were averaged across trials.

Statistical analysis

We examined the distribution of sociodemographic characteristics and VRM outcomes using frequencies, counts, means, and standard deviations (SDs). The distribution of oxidative stress biomarkers was similarly assessed using geometric means, geometric SDs, and selected percentiles. We used a series of linear regression models to examine the association between oxidative stress biomarkers and VRM outcomes. Each biomarker – outcome combination was assessed separately using individual models, in which the oxidative stress biomarker was treated as a natural log transformed, continuous measure. We first examined these associations in models that were unadjusted for covariates, followed by models that were minimally adjusted for infant sex, stimulus set, and corrected infant age at assessment. Covariates included in fully adjusted models additionally included maternal age, maternal education (a proxy for socioeconomic status), and pre-pregnancy BMI. These covariates were chosen via a Directed Acyclic Graph (DAG; Figure S3), a tool commonly applied in epidemiology research that is used to identify the minimal set of covariates to be retained in adjusted models. Our DAG was supplemented by associations between exposures and outcomes in our study population.^34,35 Smoking status was not included as a confounder as only one participant reported smoking during pregnancy.

We performed several sensitivity analyses to ensure the robustness of our findings. First, we examined effect modification by infant sex and oxidative stress using an interaction term for sex*oxidative stress, as well as in models stratified by infant sex. We similarly assessed effect modification by stimulus set using the same methods. Next, we examined the association between oxidative stress biomarkers and VRM outcomes in models stratified by both infant sex and stimulus set (four combinations total), as we have previously observed three-way-interactions between exposure, set and sex with the VRM task.¹¹ We then included gestational age at delivery as an additional covariate in our fully adjusted models to account for the possible influence of preterm birth. To assess windows of susceptbility during pregnancy, we examined the association between time-point specific oxidative stress biomarkers and VRM outcomes. We separately adjusted for averaged gestational age at sample collection in models that included averaged oxidative stress biomarkers as the exposure. Lastly, we examined the association between oxidative stress and VRM outcomes in models that were restricted those who completed just the first two blocks of the VRM task (N=163), in an effort to increase our statistical power.

Results

There were 144 participants included in our analytic sample. The average maternal age at delivery was 31 years (SD=3.8) and the average pre-pregnancy BMI was 26 kg/m2 (SD=6.4). The majority of our participants were well-educated, with 53.5% having a graduate degree (Table 1). Most participants self-identified as White (83.3%), with few participants identifying as Black, Asian/Pacific Islander, Hispanic, or Multi-Racial. Relative to the larger IKIDS cohort, our analytic sample included a slightly higher percentage of participants who were married or living with a partner (94.4% versus 87.3%) and delivered preterm (9.7% versus 4.4%). No other differences between our analytic sample and the larger cohort were observed (Table 1). When examining the distributions of VRM outcomes in our analytic sample, we observed that the mean of novelty preference, time to reach familiarization, shift rate, and average run duration were 57% (SD=6.7), 50 seconds (SD=2.1), 0.43 looks between stimuli (SD=0.16), and 4.4 seconds (SD=2.6), respectively (Table 1). Across the averaged oxidative stress biomarkers, the geometric mean was highest for 2,3-dinor-8-iso-PGF_2α (geometric mean= 4.43 ng/mL [geometric SD= 1.77]) and lowest for 2,3-dinor-5,6-dihydro-8-iso-PGF_2α (geometric mean= 0.56 ng/mL [geometric SD= 4.69]) (Table 2).

Table 1.

Distribution of demographic characteristics and visual recognition memory outcomes in the Illinois Kids Development Study.

	Analytic Sample (N=144)	Full Cohort (N=563)
Maternal Age at Delivery (years)
Mean (SD)	31 (3.8)	30 (4.2)
Pre-pregnancy Body Mass Index (kg/m2)
Mean (SD)	26 (6.4)	27 (6.8)
Missing	1 (0.7%)	4 (0.7%)
Maternal Education
Less than College Degree	15 (10.4%)	111 (19.7%)
College Degree	52 (36.1%)	197 (35.0%)
Graduate Degree	77 (53.5%)	255 (45.3%)
Maternal Race/Ethnicity
Asian/Pacific Islander	7 (4.9%)	31 (5.5%)
Black	<5	31 (5.5%)
Hispanic	<5	15 (2.7%)
Other/Multi-Racial	10 (6.9%)	35 (6.2%)
White	120 (83.3%)	451 (80.1%)
Infant Sex
Male	64 (44.4%)	254 (45.1%)
Female	80 (55.6%)	271 (48.1%)
Missing	0 (0.0%)	38 (6.7%)
Parity
1+ Births	86 (59.7%)	338 (60.0%)
No Prior Births	58 (40.3%)	225 (40.0%)
Marital Status
Married/Living Together	136 (94.4%)	492 (87.4%)
Single	8 (5.6%)	71 (12.6%)
Preterm Birth
No	130 (90.3%)	500 (88.8%)
Yes	14 (9.7%)	25 (4.4%)
Missing	0 (0.0%)	38 (6.7%)
Gestational Age (weeks)
Mean (SD)	39 (1.8)	39 (1.5)
Missing	0 (0.0%)	38 (6.7%)
Novelty Preference (%)
Mean (SD)	0.57 (0.07)	0.57 (0.07)
Missing	0 (0.0%)	303 (53.8%)
Time to Reach Familiarization (seconds)
Mean (SD)	50 (21)	52 (21)
Missing	0 (0.0%)	303 (53.8%)
Shift Rate – Faces
Mean (SD)	0.43 (0.16)	0.39 (0.18)
Missing	0 (0.0%)	304 (54.0%)
Average Run Duration - Faces (seconds)
Mean (SD)	4.4 (2.6)	4.4 (2.4)
Missing	0 (0.0%)	303 (53.8%)

Note: Full cohort includes those who delivered prior to July 2020.

Table 2.

Distributions of averaged urinary oxidative stress biomarkers corrected with specific gravity in the Illinois Kids Development Study (N=144).

	Geometric Mean (Geometric SD)	5%	25%	50%	75%	95%
8-iso-PGF2α	1.12 (1.9)	0.37	0.82	1.18	1.61	2.64
2,3-dinor-5,6-dihydro-8-iso-PGF2α	0.56 (4.96)	0.03	0.18	0.76	1.95	6.03
2,3-dinor-8-iso-PGF2α	4.43 (1.77)	2.24	3.4	4.73	6.18	9.6
PGF2α	2 (2.48)	0.41	1.21	2.35	3.72	6.4

In fully adjusted models, we observed that a natural log unit increase in 8-iso-PGF_2α, 2,3-dinor-8-iso-PGF_2α, and PGF_2α was associated with a non-significant increase in shift rate (β= 0.02, 95% confidence interval [CI]= −0.02, 0.07; β= 0.02, 95% CI= −0.03, 0.07, and β= 0.03, 95% CI= −0.01, 0.06, respectively) (Figure 1; Table S1). When novelty preference was the outcome of interest, we found that a natural log unit increase in 8-iso-PGF_2α, 2,3-dinor-8-iso-PGF_2α, and PGF_2α was associated with a reduction in novelty preference, reflecting poorer memory. Increasing PGF_2α was also associated with a non-significant decrease in average run duration (β= −0.34, 95% CI= −0.86, 0.19), and a non-significant increase in time to reach familiarization (β= 3, 95% CI= −1.22, 7.22) (Figure 1; Table S1). In contrast, a natural log unit increase in 2,3-dinor-5,6-dihydro-8-iso-PGF_2α was associated with a non-significant increase in average run duration (β= 0.22, 95% CI= −0.07, 0.51), and a non-significant decrease in time to reach familiarization (β= −2.31, 95% CI= −4.61, −0.01) (Figure 1; Table S1). Associations were similar in unadjusted models and in models minimally adjusted for infant sex, stimulus set, and infant age at assessment (Table S1).

Figure 1.

Adjusted linear regression estimates and 95% confidence intervals for the association between averaged urinary specific gravity corrected oxidative stress biomarkers and visual recognition memory outcomes in the Illinois Kids Development Study (N=143).

Note: Model are adjusted for infant sex, stimulus set, infant age at assessment, maternal age, maternal education, and pre-pregnancy BMI.

When stratifying by both infant sex and stimulus set, we observed that the positive association between 2,3-dinor-5,6-dihydro-8-iso-PGF_2α and average run duration was driven by males who saw set 2 (Figure 2; Table S2). We similarly found that increasing 2,3-dinor-5,6-dihydro-8-iso-PGF_2α and 2,3-dinor-8-iso-PGF_2α were both associated with a decrease in novelty preference only among males who saw set 2 (β= −0.02, 95% CI= −0.03, 0; β= −0.03, 95% CI= −0.06, −0.01, respectively). No other notable differences across strata of sex and set were observed (Figure 2; Table S2).

Figure 2.

Adjusted linear regression estimates and 95% confidence intervals for the association between averaged urinary specific gravity corrected oxidative stress biomarkers and visual recognition memory outcomes in the Illinois Kids Development Study, stratified by infant sex and stimulus set.

Note: Model are adjusted for infant age at assessment, maternal age, maternal education, and pre-pregnancy BMI.

In our sensitivity analyses stratified solely by infant sex, we observed an inverse association between 8-iso-PGF_2α, 2,3-dinor-8-iso-PGF_2α, and average run duration among males only, although p-values indicated that the interaction was non-significant (p-interaction= 0.24 and 0.36, respectively) (Table S3). We also observed an inverse association between 8-iso-PGF_2α and time to reach familiarization among females only, however p-values for interaction similarly showed non-signifance (p-interaction= 0.53) (Table S3). When stratifying by stimulus set only, a natural log unit increase in 8-iso-PGF_2α, 2,3-dinor-8-iso-PGF_2α, and PGF_2α was associated with an increase in average run duration only among those who saw set 2 (Table S4). In contrast, an increase in 8-iso-PGF_2α was associated with a reduction in novelty preference only among those who saw set 1 (p-interaction= 0.14 (Table S4). When visit specific oxidative stress biomarkers were included as the exposure, we observed a positive association with shift rate when 8-iso-PGF_2α was measured at the 17 weeks. A natural log unit increase in 8-iso-PGF_2α, 2,3-dinor-8-iso-PGF_2α, and PGF_2α was also associated with a significant reduction in novelty preference at 17 weeks only (Table S5). No major changes in effect estimates were observed in models additionally adjusted for gestational age at delivery (Table S6) or averaged gestational age at sample collection (Table S7). When restricting to VRM outcomes in block 1 and block 2, the association between 8-iso-PGF_2α and average run duration and time to reach familiarization, as well the associations between all oxidative stress biomarkers and novelty preference were attenuated reactive to our overall results (Table S8).

Discussion

In the present study, we examined associations between oxidative stress biomarkers, averaged across the second and third trimesters of pregnancy, with infant cognition outcomes assessed using an automated VRM task. We observed that increasing levels of 8-iso-PGF_2α, 2,3-dinor-8-iso-PGF_2α, and PGF_2α were associated with a reduction in novelty preference, reflecting poorer visual recognition memory. These negative associations were present primarily among males who were tested with stimulus set 2. We similarly observed that 2,3-dinor-8-iso-PGF_2α was positively associated with average run duration—suggesting slower (i.e., poorer) information processing speed—only among males who saw stimulus set 2. Taken together, our results suggest that oxidative stress biomarkers, particularly 2,3-dinor-8-iso-PGF_2α, is associated with negative effects on information processing and recognition memory in males who saw stimulus set 2.

In our analysis, we observed that the majority of our oxidative stress biomarkers were associated with a reduction in novelty preference, reflecting worse recognition memory. This inverse association was only statistically significant for 2,3-dinor-8-iso-PGF_2α and marginally significant for 8-iso-PGF_2α, where the significant effects for both biomarkers persisted only among males who saw stimulus set 2. This may suggest that stimulus set 2 is a more challenging test to male infants, as opposed to stimulus set 1. While these results were obtained with small sample size due to the stratification by set and sex, they are still in line with several other studies from our group showing significant results in set 2 but not in set 1, as well as other studies finding sex effects with one or the other sex impacted negatively. Specifically, in a larger analysis of over 200 infants in the IKID cohort, our prior work shows that novelty preference scores are higher (i.e., memory is better) among infants who were tested with stimulus set 1 when compared to those tested with stimulus set 2;¹ a finding that suggests that stimulus set 2 is harder for infants to remember than stimulus set 1. This bias may also impact infants’ looking behavior during familiarization, as the familiar stimuli in set 1 are less engaging to infants than in set 2. Thus, infants tend to look away from the stimuli more often during familiarization in set 1 than in set 2. While these breaks in looking at the familiar stimuli tend to generate shorter run durations in set 1 relative to set 2, this may also suggest that infants are having to work harder to study the stimulus during the familiarization phase (as indicated by the observed increase in shift rate), which is followed by a decrease in novelty preference scores during the test phase. Taken together, this suggests that when examining infants’ performance using an automated VRM task, set 2 may be a more sensitive test of these subtle associations because set 2 displays stimuli that do not favor infants’ natural biases. Similarly, our observed sex differences are also consistent with the literature.⁴⁰ In the context of our study, our finding that males were uniquely susceptible to the negative effects of oxidative stress may be reflective of underlying critical windows of susceptbility during gestation. Notably, we also observed that the associations between oxidative stress and reduced novelty preference were stronger when restricted to samples obtained during early pregnancy (16–18 weeks gestation), which is the timeframe in which testosterone levels in male fetuses are thought to be peaking.⁴¹ Taken together, this may suggest that oxidative stress exposure in utero may have a more determinantal effect on male developing fetuses, as opposed to females.

Our findings build on prior studies linking elevated oxidative stress to neurobehavioral changes in children that seem for the most part negative. Specifically, a 2022 review postulated that ASD is caused, in part, by an increase in ROS and elevated oxidative stress.⁴² Additionally, cross-sectional studies have found that oxidative stress biomarkers in children are elevated among those with ADHD relative to neurotypical controls.²⁴ Finally, prior work within The Infant Development and the Environment Study (TIDES), has shown that although 3^rd trimester levels of 8-iso-PGF_2α were associated with an increase in total scores on the social responsiveness scale, PGF_2α was positively associated with externalizing problems, as measured by the Behavioral Assessment for Children (BASC).²⁶

Oxidative stress may represent one mechanism linking environmental chemical and non-chemical stress exposures to adverse maternal and child health outcomes. Previously within the IKIDS cohort, we observed that 8-iso-PGF_2α was elevated in response to prenatal exposure to perfluoro octane sulfonic acid (PFOS),⁸ a chemical within the per- and polyfluoroalkyl substances (PFAS) class. While PFAS was not strongly associated with cognitive outcome variables in the VRM task in the IKIDS study population,⁴³ other epidemiologic studies have found that prenatal exposures to PFAS were associated with alternations in memory in early childhood.^44,45 Similarly, studies have consistently found that prenatal exposures to phthalates, a group of chemicals commonly found in plastics, are associated with elevated levels of 8-iso-PGF_2α and its major metabolites.^6,7,46 Members of our study team have shown that prenatal phthalate exposure during pregnancy was associated with increased run duration in the VRM task in the IKIDS infant cohort which is indicative of slower information processing.²⁹ In contrast, prenatal stress was associated with longer times to familiarization in the VRM task suggesting poorer visual attention.¹¹ In the present study, we observed that oxidative stress was associated with lower novelty preference which indicates poorer recognition memory. Taken together, these last set of findings suggest that the VRM task is sensitive to a variety of prenatal factors and has been able to identify different patterns of effect for a variety of exposures.

A novel aspect of our study was that we used an automated VRM task to measure infant cognition. Importantly, this measure tracks looking behavior using an infrared eye tracker, limiting potential outcome misclassification relative to studies which assess neurodevelopment using parent self-report. However, we note that our outcome measures were assessed in early infancy, which may not be predictive of future cognitive outcomes. Nonetheless, previous studies have demonstrated that both nonautomated looking time^37–39 and eye tracking methods^47,48 are reliable measures of foundational cognitive processes (i.e., memory, attention, processing speed) that are also predictive of cognitive function later in childhood. Additionally, we included multiple F₂-IsoPs as biomarkers of oxidative stress. These measures are unaffected by lipids in the diet and are stable throughout the day in spot urine samples.⁴⁹ Our primary analysis focused on average oxidative stress biomarker concentrations. While this allowed us to create a more stable estimate across pregnancy, we acknowledge that this limited our ability to assess critical windows of susceptibility. Our study was also limited by a relatively small sample size, particularly for stratified analyses, which limits our statistical power to test for stimulus set and sex differences. Additionally, our analysis plan involved running many statistical tests, and in that sense our findings may be due to chance as we did not correct for multiple comparisons. However, we note that correcting for multiple comparisons is not necessary in epidemiologic studies as it increases the type II error.⁵⁰ Lastly, our study was comprised of predominately White participants who were highly educated, which in combination with our limited sample size, limits our external generalizability to other groups.

Conclusions

Within the longitudinal IKIDS cohort, we observed that prenatal oxidative stress biomarkers were associated with decreased novelty preference suggesting poorer recognition memory. Our results suggest that increased concentrations of both 8-iso-PGF_2α, 2,3-dinor-8-iso-PGF_2α, and PGF_2α levels were associated with a decrease in recognition memory, and the former was also associated with slower information processing. Future studies, which focus on socioeconomically diverse study populations and include additional measures of oxidative stress at different gestational weeks, as well as in the placenta, cord blood, and children, are needed to confirm the findings from our study. Additional studies examining oxidative stress as a mediator linking chemical and non-chemical stressors to neurodevelopment are also warranted.

Data availability statement:

The datasets for this manuscript are not publicly available because, per the National Institutes of Health (NIH)-approved Environmental influences on Child Health Outcomes (ECHO) Data Sharing Policy, the entirety of the ECHO-wide cohort data has not yet been made available to the public for review/analysis. Requests to access the datasets should be directed to the ECHO Data Analysis Center, ECHO-DAC@rti.org.