Associations of maternal blood biomarkers of prenatal APAP exposure with placental gene expression and child attention deficit hyperactivity disorder

Abstract

Despite evidence linking prenatal acetaminophen (APAP) exposure and adverse neurodevelopment in humans and animals, over half of pregnant women in most populations use APAP. Prior studies could be biased by inaccurate self-reported APAP use, and the molecular mechanisms linking prenatal APAP with adverse neurodevelopment are unknown. We estimated associations between maternal plasma biomarkers of APAP exposure, child attention deficit hyperactivity disorder (ADHD), and placental gene expression in 307 African-American mother-child pairs. Overall, detection of APAP in 2^nd trimester plasma was associated with higher odds for child ADHD diagnosis (Odds Ratio (OR)=3.15 [95%CI: 1.20–8.29]). Prenatal APAP exposure and ADHD were associated with placental upregulation of immune system pathways in females, and downregulation of oxidative phosphorylation in both sexes. In females only, prenatal APAP was associated with 5.22% higher odds (0.0456%−13.1%) of ADHD statistically mediated through increased immunoglobulin heavy constant gamma 1 (IGHG1) expression. These results highlight placental molecular mechanisms that may underlie developmental toxicity of prenatal APAP exposure.

An estimated 41%−70% of pregnant people report taking N-acetyl-p-aminophenol (APAP, otherwise known as acetaminophen and paracetamol) in American, European, and Asian populations.^1–9 The United States Food and Drug Administration (FDA) and European Medicines Agency (EMA) have traditionally viewed APAP as posing minimal risks to the fetus when used as directed during pregnancy.^10,11 However, accumulating evidence from human and animal model studies show links between prenatal APAP exposure and adverse reproductive and neurodevelopmental outcomes (reviewed by^12,13). In particular, observational studies have shown associations of prenatal exposure with higher risks for child attention deficit hyperactivity disorder (ADHD) and autism spectrum disorders (ASD).^{2–5,8,14–19} Reviewing the literature, the FDA and EMA determined that the weight of evidence is inconclusive, and have called for more rigorous studies and better safety data to inform their recommendations on how pain medicines are used during pregnancy.^10,11 Although definitive evidence could be obtained from a randomized clinical trial, ethical standards preclude the inclusion of pregnant women when harm to the offspring can be plausibly inferred.²⁰ Thus, well-designed observational studies that specifically address limitations cited by the FDA, EMA, and other medical agencies are critical for informing future regulatory action.

Many human observational studies have shown associations of prenatal APAP with child ADHD,^{2–5,8,14–19} yet there are some important limitations, most notably the reliance on maternal self-reported drug use.²¹ While two studies have observed associations between ADHD and drug concentrations measured in cord blood² and meconium¹⁴ more studies assessing prenatal APAP exposure using biomarkers are needed. An additional major knowledge gap is lack of information about APAP’s mechanism(s) of action at the molecular level. Epidemiologic studies have examined alterations in functional brain connectivity¹⁴ and cord-blood DNA methylation^9,22–24 as potential links between prenatal APAP and neurodevelopment, but more work is needed. For instance, functional brain connectivity and DNA methylation patterns cannot be readily utilized to study known reduction-oxidation (redox) mechanisms of APAP toxicity. Oxidative stress from glutathione depletion is the cause of APAP induced hepatoxicity and liver failure following overdose,^25–30 but whether this mechanism also mediates prenatal APAP-neurodevelopment associations remain understudied. Omics technologies can potentially bridge these gaps, providing a means to assess APAP exposure biomarkers and explore molecular mechanisms.

Metabolomics have been used to assess APAP use^2,9,31–33 and can mitigate self-report limitations. Because hundreds of medications contain APAP, individuals may unknowingly ingest it. For instance, one study found APAP analytes in serum from 20.2% of women reporting no intake.³¹ Follow-up investigation found that 28 of 157 (17.8%) reporting no intake took pain or cold medications that may contain APAP. While addressing self-report limitations, metabolomic biomarkers may reflect a narrow exposure window owing to APAP’s <3 hour blood half-life.³⁴

Transcriptomics may provide functional insights into redox biology and other pathways that may be perturbed by fetal/intrauterine APAP exposure. For instance, RNA-sequencing has shown dysregulated gene expression in human embryonic stem cells exposed to APAP during neuronal differentiation,³⁵ and gene expression changes in the brains and placentas of rodents following developmental APAP exposure.^36,37 However, transcriptomic mechanisms of prenatal APAP toxicity have not been studied in humans. Neurotoxicity of prenatal APAP exposure could be mediated via perturbations to the placenta, which regulates nutrient, waste, and gas exchange between the mother and fetus, providing a functional link between the maternal environment and fetal programming.^38,39

To address these gaps, here we: 1) examined associations of maternal plasma APAP metabolites during pregnancy with child ADHD and ADHD-related outcomes among 307 mother-child pairs, and 2) evaluated associations of plasma APAP metabolites with differential gene expression and transcriptomic pathway perturbations in the placenta. We additionally investigated sex-specific relationships based on prior reports of effect modification by sex of prenatal APAP-neurodevelopment associations.^2,3,8,36

Results

Descriptive Data

This study took place in the Conditions Affecting Neurocognitive Development and Learning in Early Childhood (CANDLE) cohort in Memphis, Tennessee.^40,41 A previous analysis, which investigated the connections between prenatal metabolomics profiles and obesity risk in African Americans,⁴² randomly selected 450 mothers who self-identified as Black/African American from this cohort for untargeted metabolomics analysis of second trimester (mean [range] gestational age of 22.8 [15.3–28.0] weeks) plasma (Figure 1). We evaluated ADHD-related outcomes in a follow-up visit at age 8–10 years for 307 children among this subsample. Mothers were on average 24.8 years of age during pregnancy, and the majority were not exposed to tobacco (87.3%) or alcohol (93.5%) during pregnancy, self-identified as Not Hispanic/Latino (98.0%) and delivered vaginally (61.6%) (Table 1). Children were on average 9.1 years of age at neurodevelopmental assessment, approximately half were female (51.1%), and most did not have immediate family members with mental health disorders (82.4%).

Figure 1: Study Inclusion Flowchart.

Study flowchart for inclusion of participants shows analytic sample sizes remaining from the implementation of each exclusion criterion. Participants come from Conditions Affecting Neurocognitive Development and Learning in Early Childhood (CANDLE) birth cohort.

Table 1:

Descriptive statistics of the study sample

	No Prenatal APAP (N=245)	Prenatal APAP (N=62)	Total (N=307)	P-value1
Child ADHD Diagnosis (n (%))				0.035
No	224 (91.4%)	51 (82.3%)	275 (89.6%)
Yes	21 (8.6%)	11 (17.7%)	32 (10.4%)
Prescription medications for ADHD (n (%))				0.034
No	232 (94.7%)	54 (87.1%)	286 (93.2%)
Yes	13 (5.3%)	8 (12.9%)	21 (6.8%)
CBCL ADHD DSM5-oriented scale (raw score)				0.511
N-Miss	2	0	2
Mean (SD)	3.403 (3.056)	3.694 (3.272)	3.462 (3.098)
Range	0.000 – 14.000	0.000 – 13.000	0.000 – 14.000
CBCL externalizing (raw score)				0.297
N-Miss	2	0	2
Mean (SD)	5.284 (6.222)	6.226 (6.745)	5.475 (6.332)
Range	0.000 – 47.000	0.000 – 28.000	0.000 – 47.000
NIH Flanker (uncorrected standard score)				0.633
N-Miss	24	2	26
Mean (SD)	92.905 (14.179)	91.917 (14.211)	92.694 (14.166)
Range	68.000 – 146.000	65.000 – 116.000	65.000 – 146.000
Digit Span (scaled score)				0.725
N-Miss	36	5	41
Mean (SD)	8.416 (2.797)	8.561 (2.584)	8.447 (2.748)
Range	1.000 – 17.000	4.000 – 16.000	1.000 – 17.000
Child sex (n (%))				0.841
Female	126 (51.4%)	31 (50.0%)	157 (51.1%)
Male	119 (48.6%)	31 (50.0%)	150 (48.9%)
Maternal age (years)				0.074
Mean (SD)	24.535 (5.219)	25.871 (5.345)	24.805 (5.264)
Range	16.000 – 40.000	16.000 – 36.000	16.000 – 40.000
Pre-pregnancy BMI (kg/m 2 )				0.266
N-Miss	1	1	2
Mean (SD)	29.016 (8.672)	27.689 (6.717)	28.751 (8.325)
Range	15.000 – 62.000	18.000 – 45.000	15.000 – 62.000
Maternal tobacco (n (%))				0.029
No	219 (89.4%)	49 (79.0%)	268 (87.3%)
Yes	26 (10.6%)	13 (21.0%)	39 (12.7%)
Maternal alcohol (n (%))				0.982
No	229 (93.5%)	58 (93.5%)	287 (93.5%)
Yes	16 (6.5%)	<5 (<7%)	20 (6.5%)
Prenatal stressful life experiences (number)				0.469
Mean (SD)	1.959 (1.971)	1.758 (1.870)	1.919 (1.949)
Range	0.000 – 9.000	0.000 – 8.000	0.000 – 9.000
Gravidity (number of pregnancies)				0.403
Mean (SD)	2.755 (1.803)	2.968 (1.708)	2.798 (1.784)
Range	1.000 – 9.000	1.000 – 7.000	1.000 – 9.000
Maternal ethnicity (n (%))				0.418
Not Hispanic/Latino	241 (98.4%)	60 (96.8%)	301 (98.0%)
Hispanic/Latino	<5 (2%)	<5 (<7%)	6 (2.0%)
Maternal education (n (%))				0.117
High School completion	144 (58.8%)	36 (58.1%)	180 (58.6%)
<High School	40 (16.3%)	9 (14.5%)	49 (16.0%)
Graduated college or technical school	56 (22.9%)	12 (19.4%)	68 (22.1%)
Some graduate work or graduate/professional degree	5 (2.0%)	5 (8.1%)	10 (3.3%)
Family income adjusted (USD)				0.598
N-Miss	22	5	27
Mean (SD)	24642 (21687)	26379 (24138)	24995 (22173)
Range	2493 – 83437	2545 – 82637	2493 – 83437
Neighborhood deprivation index (standardized score)				0.378
N-Miss	1	0	1
Mean (SD)	0.719 (0.727)	0.626 (0.800)	0.700 (0.742)
Range	−0.866 – 2.804	−0.879 – 2.804	−0.879 – 2.804
Mental health disorders in immediate family (n (%))				0.358
N-Miss	10	2	12
No	196 (83.4%)	47 (78.3%)	243 (82.4%)
Yes	39 (16.6%)	13 (21.7%)	52 (17.6%)
NSAIDs during pregnancy (n (%))				0.141
N-Miss	6	0	6
No	235 (98.3%)	59 (95.2%)	294 (97.7%)
Yes	<5 (2%)	<5 (<7%)	7 (2.3%)
Antibiotics during pregnancy (n (%))				0.134
N-Miss	6	0	6
No	207 (86.6%)	58 (93.5%)	265 (88.0%)
Yes	32 (13.4%)	<5 (<7%)	36 (12.0%)
Delivery method (n (%))				0.961
Vaginal	151 (61.6%)	38 (61.3%)	189 (61.6%)
C-section	94 (38.4%)	24 (38.7%)	118 (38.4%)
Labor type (n (%))				0.299
Spontaneous, augmented	87 (35.5%)	16 (25.8%)	103 (33.6%)
Spontaneous	60 (24.5%)	18 (29.0%)	78 (25.4%)
Induced	54 (22.0%)	19 (30.6%)	73 (23.8%)
No Labor	44 (18.0%)	9 (14.5%)	53 (17.3%)
Birthweight (g)				0.680
Mean (SD)	3086.498 (601.996)	3121.935 (612.947)	3093.655 (603.382)
Range	780.000 – 4640.000	757.000 – 4252.000	757.000 – 4640.000
Gestational age at birth (weeks)				0.342
Mean (SD)	38.610 (2.405)	38.276 (2.698)	38.543 (2.466)
Range	26.571 – 41.857	25.857 – 41.286	25.857 – 41.857
Child age at study visit (years)				0.820
Mean (SD)	9.096 (0.824)	9.123 (0.919)	9.102 (0.843)
Range	8.010 – 10.980	8.010 – 10.970	8.010 – 10.980

¹Two-sided P-values calculated from analysis of variance and chi-square goodness of fit tests for continuous and categorical variables, respectively. P-values are not adjusted for multiple comparisons.

We detected one or more APAP analytes in the second trimester plasma metabolomes of 62 mothers (20.2%, Table 1). Pair-wise Pearson correlations between individual APAP analytes were generally high (all P<0.05, Supplemental Figure 1A). The lowest correlations, ranging from 0.41 to 0.75, were between 3-(methylthio)APAP sulfate and other analytes. Pair-wise correlations among all other APAP analytes ranged from 0.68 to 0.97. Among individuals with each analyte detected, distributions of analyte concentrations were approximately normal (Supplemental Figure 1B).

Prenatal APAP and child ADHD

At child age 8–10 years, parents reported ADHD diagnoses in 11 children (17.7%, 5 male) prenatally exposed to APAP and 21 children (8.6%, 15 male) who were not exposed, and ADHD medication use in 8 children (12.9%, 4 male) prenatally exposed to APAP and 13 children (5.3%, 10 male) who were not exposed (Table 1, Supplemental Table 1). Children with parent-reported ADHD diagnoses had higher Child Behavior Checklist (CBCL) DSM5-oriented ADHD subscale and externalizing syndrome scores, and lower NIH Toolbox Flanker Inhibitory Control and Attention Test (NIH Flanker) and Wechsler Intelligence Scale for Children Digit Span scores (Supplemental Figure 2).

Adjusting for potential confounders and precision variables (see Figure 2 for conceptual model), detection of APAP in maternal 2^nd trimester plasma was associated with higher odds of ADHD diagnosis (Odds Ratio (OR) = 3.15 [95% CI: 1.20–8.29]), and higher odds of taking ADHD medications (OR = 5.20 [1.48–18.28]) compared with no APAP exposure (Figure 3, Supplemental Tables 1,2). Furthermore, there was a suggestive but non-significant association of prenatal acetaminophen with 1.53 [−0.16–3.22] higher CBCL externalizing score. Prenatal APAP exposure was not associated with CBCL DSM5-oriented ADHD scale (β = 0.51 [−0.38 – 1.4]), NIH Flanker (β = −1.87 [−6.07 – 2.33]) or Digit Span (β = 0.09 [−0.73 – 0.91]) scores (Figure 3, Supplemental Table 2).

Figure 2: Conceptual Model.

Conceptual model of confounders, mediators, precision variables, and effect modifiers in the relationship between prenatal APAP exposure and child ADHD-related outcomes (A). Primary models adjusted for confounders and precision variables, while sensitivity models additionally adjusted for birthweight and gestational age. Effect modification was tested by including APAP by child sex interaction terms. Prenatal NSAID and antibiotics, which were included as covariates in the primary models, were conceptualized as instrumental variables for APAP indications (B).

¹ Potential mediators were only included as adjustment variables in a sensitivity analysis.

Figure 3: Associations of Prenatal APAP Exposure with Child ADHD.

Associations of prenatal APAP with maternal report of child ADHD diagnosis and ADHD prescription medications (N = 307) (A), CBCL DSM5-oriented ADHD scale and externalizing scale (N = 305) (B), and task-based NIH Flanker (N = 281) and Digit Span (N = 251) scores (C). Non-interaction models (indicated as “Both” in legend) included child sex as a covariate with no sex interaction terms. Sex-specific model estimates were obtained by adding prenatal APAP by child sex interaction product terms to the models. Models were re-run with male and female sex set as the reference level to obtain sex specific estimates for prenatal APAP. All models adjusted for maternal age, education, ethnicity, pre-pregnancy BMI, prenatal alcohol, prenatal tobacco, prenatal stressful life events, gravidity, delivery method, labor type, household income, neighborhood deprivation index, child age at the study visit, mental health disorders in the child’s immediate family, and maternal NSAID and antibiotic use (as instruments for APAP indications). Points indicate odds ratios from logistic regressions and beta coefficients from linear regressions, while error bars depict 95% confidence intervals.

Sex-specific analyses suggested more pronounced associations among females versus males, but estimates were imprecise and APAP by sex interactions were non-significant (Figure 3, Supplemental Tables 1,2). For example, prenatal APAP exposure was associated with higher odds for child ADHD diagnosis in females (OR = 6.16 [1.58–24.05]) while there was a positive, but non-significant association in males (OR = 1.7 [0.44 – 6.61]; interaction P = 0.18). Similarly, the odds of ADHD medication use were elevated in both females (OR = 10.26 [1.71–61.68]) and males (OR = 2.84 [0.52–15.51]), although the male result was non-significant (interaction P = 0.29). Prenatal acetaminophen exposure was associated with 2.60 [0.19–5.02] higher CBCL externalizing scores in females, while no association was observed in males (β = 0.41 [−1.77 – 2.59]; interaction P=0.17). We observed a suggestive but non-significant association of prenatal acetaminophen with 1.03 [−0.24–2.30] higher CBCL DSM5-oriented ADHD scores in females, but no association in males (β = −0.02 [−1.15 – 1.11]; interaction P=0.21). Prenatal APAP was not associated with NIH Flanker or Digit Span scores in the analyses testing effect modification by sex (Figure 3, Supplemental Table 2). Results were not appreciably different when additionally adjusting for birthweight and gestational age, or in unadjusted models (Supplemental Table 2). There was no evidence for prenatal APAP by tobacco (P = 0.49) or alcohol (P = 0.99) interactions on child ADHD.

Placental transcriptomics of prenatal APAP and child ADHD

The 174-participant subsample with placental RNA-sequencing data had a higher (average) birthweight and a higher rate of induced labor but was otherwise comparable to participants without RNA-sequencing data (Supplemental Table 3). Among 13,961 transcripts analyzed here, no single transcripts were differentially expressed in association with prenatal APAP exposure or with child ADHD (false discovery rate (FDR)<0.10). In prenatal APAP by sex interaction models, no transcripts were associated with APAP exposure in males. In females, however, prenatal APAP exposure was associated with higher placental expression of interleukin 1 receptor antagonist (IL1RN) by 212% [94%−404%] (FDR = 0.0944; interaction P = 0.0004). In child ADHD by sex interaction models, no transcripts were associated with ADHD diagnosis in males. In females, however, placental expression of immunoglobulin heavy constant gamma 1 (IGHG1) was 360% [143%−768%] higher among children who developed ADHD compared to children without parent-reported ADHD diagnoses (FDR = 0.0809; interaction P = 0.014).

We performed causal mediation analysis to examine the roles of placental IL1RN and IGHG1 expression as potential links between prenatal APAP and child ADHD among females. Although IL1RN was associated with prenatal APAP exposure in females (see above), there was little evidence that IL1RN was differentially expressed according to ADHD diagnosis (P= 0.664). Furthermore, the average causal mediation effect (ACME) was non-significant in mediation analyses (ACME = −4.11% [−9.99%−0.789%]). In addition to higher IGHG1 expression being associated with ADHD diagnosis in females (Figure 4A), there was a suggestive association of prenatal APAP exposure in females with higher IGHG1 expression by 57.9% [−9.58%−94.4%] (P = 0.110; Figure 4B). In causal mediation analysis, prenatal APAP exposure was associated with 5.22% [0.0456%−13.1%] higher odds of ADHD diagnosis among females statically mediated through increased placental IGHG1 expression (Figure 4C). In males, neither ADHD nor APAP were associated with IGHG1 expression, and there was no evidence for mediation (Figure 4D–F).

Figure 4: Placental IGHG1 Expression Links Prenatal APAP with Child ADHD in Females but not Males.

Boxplots depict median (center line), 25^th percentile (minima), and 75^th percentile (maxima), while whiskers extend to the highest and lowest values that are within 1.5 times the inter-quartile range of the box. One point per individual (N = 174) depicts placental IGHG1 expression according to child ADHD diagnosis (A) and prenatal APAP exposure (B) among females. Diamonds indicate mean gene expression. Unadjusted and false discovery rate (FDR) adjusted two-sided p-values from moderated t-tests from limma-voom differential expression analysis shown. Mediation by placental IGHG1 expression of the association between prenatal APAP exposure and child ADHD among females shows mediation effect estimates and 95% confidence intervals from causal mediation analysis (C). IGHG1 expression according to child ADHD diagnosis (D) and prenatal APAP exposure (E), and mediation analysis (F) in males.

Functional enrichment analyses using competitive gene set testing revealed several Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways that were differentially regulated (FDR<0.05) according to both APAP exposure and ADHD diagnosis (Figure 5). The most abundant differentially regulated pathways belonged to the KEGG signal transduction and immune system groups. APAP exposure was associated with immune system pathways in different directions according to fetal sex, with 11 pathways upregulated in females, and two pathways downregulated in males.

Figure 5: KEGG Pathways Associated with Prenatal APAP Exposure and Child ADHD.

T-statistics (Log2FCs divided by standard error) from the differential expression analyses were input into gene set enrichment analyses for Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways using the generally applicable gene set enrichment (GAGE) and Correlation Adjusted MEan RAnk (CAMERA) methods. Gene set enrichment P-Values (larger among the two methods) and mean gene T-statistics shown for pathways with FDR<0.05 in both CAMERA and GAGE methods (N = 174).

Several KEGG pathways were regulated in the same direction in association with both prenatal APAP and child ADHD, particularly among females. Three of the immune system KEGG pathways upregulated in association with APAP in females were also positively associated with child ADHD in females, namely, antigen processing and presentation, intestinal immune network for IgA production, and NOD-like receptor signaling (Figure 5). Three other pathways – JAK-STAT signaling, cell adhesion molecules (CAMs) and phagosome – were similarly upregulated in association with both prenatal APAP and child ADHD among females. There was only one pathway in males that was associated with APAP and ADHD in the same direction: oxidative phosphorylation (Figure 5).

Discussion

In this pregnancy cohort subsample of 307 mother-child dyads self-identifying as Black/African American in the urban South, prenatal APAP exposure was associated with ADHD-related outcomes and perturbed placental gene expression pathways. Detection of APAP analytes in 2^nd trimester plasma was associated with higher odds for child ADHD diagnosis and prescriptions for ADHD medications. Although estimates were imprecise and the sample for computing sex-specific associations was small, ORs for ADHD diagnosis and prescriptions for ADHD medications were larger in females compared to males, and prenatal APAP was significantly associated with higher CBCL externalizing symptoms in females but not in males or in the pooled analysis. These results align with prior studies that found stronger prenatal APAP-ADHD associations in females,^2,3,8 but our small sample size warrants replication of sex-specific associations in better powered studies. Placental transcriptomic signatures associated with prenatal APAP were also sex specific. These findings support well-known associations of prenatal APAP with adverse child neurodevelopment^{2–5,8,12,14–19} and point to potential mechanisms related to placental gene expression.

The literature on epidemiologic associations between prenatal APAP and adverse neurodevelopment is large.^{2–5,8,14–19} A recent meta-analysis of 7 studies encompassing 195,957 participants reported an adjusted pooled risk ratio of 1.34 for ADHD.⁴³ Our OR could overestimate the association. Alternatively, it may better approximate the association of frequent APAP use rather than any intake during pregnancy with ADHD. Owing to the <3 hour plasma half-life of APAP,³⁴ our measurement may only reflect use within a few days before the blood draw, and could disproportionately capture frequent APAP users over sporadic users. Consistent with this hypothesis, the literature reports a stepwise increase in ADHD risks with increasing frequency of use.^3,5 Furthermore, the only two other studies of prenatal APAP biomarkers (rather than self-report) and ADHD also reported larger odds ratios ranging from 2.26–2.86.^2,14

While recent studies based in the population medical registers of Norway⁴⁴ and Sweden⁴⁵ have shown null associations between prenatal APAP exposure and child ADHD via sibling-matched analyses, several limitations in these studies limit their conclusions. The Norwegian study observed a two-fold increase in the risk of ADHD associated with exposure to APAP for 29 days or more during pregnancy. Following sibling control, this association was nullified at the within-family level. Limiting this analysis, only 34 sibling pairs were discordant for this long-term APAP exposure and ADHD. Statistically, only sibling pairs discordant in both exposure and outcome contribute to the estimated “within-pair” association.⁴⁶ A study in the same Norwegian cohort with a continuous outcome, which facilitated a larger sample size of doubly discordant sibling pairs, found significant associations between prenatal APAP exposure and adverse neurodevelopment in the sibling-controlled analysis.¹⁶ In the Swedish study, midwives asked pregnant individuals to report medications without specifically inquiring about APAP or other over-the-counter analgesics/antipyretics, which likely explains why the prevalence of APAP exposure during pregnancy was so severely underestimated at 7.5%.⁴⁵ The global literature (reviewed by^12,13), including biomarker and self-report measures in Sweden during the same time period as the national registry study,^47,48 suggests that approximately 50% of pregnant individuals use APAP at some time during their pregnancy. Open-ended questions about medication use are liable to underestimate APAP, possibly owing to forgetfulness, the misconception that only prescribed drugs should be reported, or failure to recognize that an over-the-counter medication like APAP is still a drug.⁴⁹ Supporting the likelihood of underreporting, a validation study in the Netherlands found a small 0.57 specificity for self-reported acetaminophen use in baseline web-based questionnaires, where only 18% of pregnant individuals reported acetaminophen use, compared with 6-week paper-based medication diaries, where 32% (of the same participants) reported use.⁵⁰ The most likely consequence of exposure misclassification is bias toward the null, and this type of bias more strongly impacts the “within-pair” estimate from sibling control analyses compared to the overall population estimate.⁵¹ Finally, rather than being confounders, factors that are shared within families, including genetic variants, could instead interact with acetaminophen exposure. Sibling control models adjust for genetic and familial factors as if they were confounders and are thus biased in the presence of effect measure modification by these shared factors.⁵² A recent simulation study has shown that both under-reporting of acetaminophen and adjustment for such interacting variables can bias associations with neurodevelopment toward the null.⁵³ Nevertheless, confounding by shared genetic or environmental factors remains a possibility. If such confounding is partially responsible for the link between prenatal APAP and adverse neurodevelopment, it will be important to identify these confounders, which could be future targets for harm reduction.

Citing limitations in the literature, including the fact that APAP “use was self-reported (e.g., potential for recall bias)” in prior studies,²¹ the FDA, EMA, American College of Obstetricians and Gynecologists, Society of Obstetricians and Gynecologists of Canada, and Society for Maternal-Fetal Medicine do not support adjusting recommendations for APAP intake in pregnancy.^{12,21,54–57} In addition to our biomarker-based study, two other studies have measured biomarkers of prenatal APAP in cord blood² and meconium¹⁴ and shown associations with child ADHD.

This is the first study in humans to evaluate RNA-sequencing changes associated with prenatal APAP exposure. We identified altered placental IGHG1 expression as a possible mechanism linking prenatal APAP exposure to child ADHD in females. This gene encodes the constant heavy chain of Immunoglobulin Gamma 1 (IgG1), the most abundant antibody in humans.⁵⁸ The placenta links the maternal environment with fetal programming,^38,39 and we present evidence that APAP neurotoxicity could be statistically mediated via perturbations to this interface. However, causality cannot be inferred from these data, and could be explored in future experimental studies. Additionally, the higher odds of ADHD associated with higher IGHG1 expression was small compared with the overall association between prenatal APAP and ADHD, suggesting that a wider set of mechanisms may be involved. APAP easily crosses the placenta^59,60 and blood-brain barrier,⁶¹ so it is also possible that APAP induces neurotoxicity directly in the fetal brain.

There are remarkable similarities between the transcriptomic results here and in rodent studies. For instance, prenatal APAP up-regulated inflammatory and immune-related genes in the rat placenta.³⁷ Strikingly, prenatal APAP exposure was associated with increased placental IL1RN expression in this population, and Il1rn was also upregulated by APAP in the rat placenta.³⁷ Two immune system pathways upregulated by prenatal APAP in females here – antigen processing and presentation and hematopoietic cell lineage – were also upregulated in the brain following developmental APAP exposure in mice.³⁶ Antigen processing and presentation was also positively associated with child ADHD in females. CAMs, which were upregulated in this study in association with both prenatal APAP and child ADHD in females, were also upregulated in the brains of mice following developmental APAP exposure.³⁶ Finally, oxidative phosphorylation, which was negatively associated with prenatal APAP in both sexes, and negatively associated with ADHD in males here, was also downregulated by developmental APAP treatment in the brains of male but not female mice.³⁶

Several gene expression changes associated with both prenatal APAP and ADHD can modulate neurodevelopment. Signaling factors traditionally regarded as elements of the immune system are increasingly acknowledged for their roles in brain development.^62,63 For instance, cytokines are expressed in developing neurons, microglia, and astrocytes, and their regulation affects neurogenesis and neuronal proliferation.⁶² Additionally, pregnancy requires maternal suppression of an immune response to the allogenic fetus, and APAP exposure may disrupt such fetal immune tolerance.⁶⁴ Moreover, studies show associations of elevated inflammatory and immune responses during pregnancy with offspring neurodevelopmental disorders.^65–67 Beyond the immune response, oxidative phosphorylation was downregulated in association with APAP and child ADHD. Impaired oxidative phosphorylation is expected, as APAP disrupts the electron transport chain.^28–30 Oxidative phosphorylation is the canonical primary energy source within human cells, including neurons.⁶⁸ Deficits in this pathway can disrupt normal neuronal function and brain development, and contribute to neurodevelopmental disorder pathology.^68–70 The placenta has immense energy requirements, and disruptions to placental modulation of energy generation may underlie adverse outcomes.⁷¹ This work supports immune/inflammatory activation and impaired energy metabolism as potential mechanisms linking prenatal APAP with neurodevelopment.

Overlapping results between this study and the rodent model studies cited above have important implications for causal inference and clinical translation. Confounding is minimal in well-controlled randomized animal model studies, making these overlapping genes and pathways strong candidates as causal agents in the link between prenatal APAP exposure and neurotoxicity. To specifically address confounding by indication, animals are treated with APAP following random assignment, not in response to indications for use. Moreover, many human studies find that adjusting for indications such as headache, pain, or fever has no impact on associations between prenatal APAP and child ADHD.^72,73 Regarding translation, increased confidence in clinical relevance is warranted for genes and pathways uncovered in animal model studies that were also associated with prenatal APAP here.

Prenatal APAP was associated with ADHD diagnosis in both sexes and externalizing symptoms in females, but not with task related measures that are often impaired in children with ADHD. We did not ask caregivers to withhold ADHD medications, and did not collect data on medication use on the day of neurodevelopmental assessment. ADHD medications could have moderately improved scores and possibly attenuated associations, although children with ADHD still had poorer performance on task related outcomes. These task related outcomes could capture different ADHD-related domains from ADHD diagnosis and CBCL scores. For instance, prenatal APAP could not affect visual selective attention and working memory while still being associated with other aspects of ADHD. Finally, confounding by maternal health behaviors related to maternal medication use and ADHD diagnosis/CBCL warrants discussion,^74,75 as task based outcomes would not be subject to such confounding. However, whether such confounding would attenuate or inflate associations of APAP with ADHD is unclear.^74,76,77

This work should be considered in the context of several limitations. First, we only included Black/African American women in the urban South, limiting generalizability, but also enhancing data for a group traditionally underrepresented in research. Second, our sample size was small, particularly for RNA-sequencing, which only included 174 placentas. Third, we measured expression using bulk RNA-sequencing. Single-cell transcriptomics could additionally capture toxicant-associated changes in the relative abundance and/or impairment of specific cell types.⁷⁸ Fourth, we used a one-time measurement of metabolites in maternal blood during pregnancy to characterize APAP exposure. A single time point measurement is more likely to capture frequent compared with sporadic users, and APAP use during one time point could correlate with use at other time points in pregnancy. Thus, our results should not be interpreted as pinpointing the second trimester as a window of high susceptibility. The literature suggests that the third trimester may be a more relevant window of APAP susceptibility in relation to neurodevelopment.^12,13 This could be a consequence of the peak brain growth spurt that occurs in the third trimester,⁷⁹ or because APAP used by the mother during the immediate pre-partum period may be left to newborns with underdeveloped capacities to metabolize it.⁸⁰ Additionally, the prevalence of individuals with APAP biomarkers detected in our study likely underestimates the prevalence of APAP use throughout all of pregnancy. Finally, while we adjusted for multiple potential sources of confounding, unobserved confounding is always possible in observational epidemiologic studies.

Conclusions

The links between biomarkers of prenatal APAP exposure and child ADHD presented here are consistent with the prior epidemiologic literature based on self-report and address concerns about recall bias. Gene expression changes, including upregulation of the IGHG1 gene and KEGG immune system pathways, and downregulation of oxidative phosphorylation, could be mechanistic links to ADHD. Furthermore, the alignment of identified genes and pathways with rodent studies enhances confidence in their potential as mechanistic agents. However, omics studies should ultimately be considered as hypothesis generating. Additionally, owing to small sample sizes, sex-specific associations with ADHD-related outcomes and transcriptomic analyses warrant replication.

Methods

Study population

The Conditions Affecting Neurocognitive Development and Learning in Early Childhood (CANDLE) prospective pregnancy cohort of mother–child dyads in Shelby County, Tennessee (Memphis area) enrolled 1,503 women aged 16–40 with relatively healthy singleton pregnancies in their second trimester between 2006 and 2011 resulting in 1,455 live births.^40,41 Exclusion criteria included an existing chronic disease requiring medication (e.g., hypertension, diabetes, and sickle cell disease), known pregnancy complications (e.g., complete placenta previa and oligohydramnios), or plans to deliver at a nonparticipating hospital. The Institutional Review Boards of the University of Tennessee Health Science Center and the University of Washington granted approval for the study. Participants aged 18 years or older provided informed consent, while assent and consent from legally authorized representatives were obtained for those under 18 years prior to enrollment. Participants were compensated with up to $850 (USD) across all study visits from pregnancy through child age 8–10 years.

Exposure

A previous analysis, which investigated the connections between prenatal metabolomics profiles and obesity risk in African Americans,⁴² randomly selected 450 out of 953 mothers who self-identified as Black/African American from the CANDLE study for untargeted metabolomics analysis. We collected non-fasting maternal plasma samples in the second trimester and stored them at −80 °C before conducting metabolomics analysis using the Metabolon Discovery HD4TM Platform (Metabolon Inc., Morrisville, NC), which includes four previously described⁴² ultra-high-performance liquid chromatography (UPLC)-mass spectrometry methods. Briefly, all methods were conducted using a Waters ACQUITY UPLC system paired with a Thermo Scientific Q-Exactive high-resolution/accurate mass spectrometer. The system was equipped with a heated electrospray ionization (HESI-II) source and an Orbitrap mass analyzer set to operate at a mass resolution of 35,000. Sample extracts were dried and reconstituted in solvents tailored for each of the four analytical methods, which contained standards at fixed concentrations to ensure injection and chromatographic consistency. Two methods were analyzed using acidic positive ion conditions. One of these methods was optimized for more hydrophobic compounds, where sample extracts were gradient eluted from a C18 column (Waters UPLC BEH C18–2.1×100 mm, 1.7 μm) using water and methanol, containing 0.05% perfluoropentanoic acid (PFPA) and 0.1% formic acid (FA). The other method was chromatographically optimized for more hydrophobic compounds, where the extract was gradient eluted from the same C18 column operated at a higher organic content using methanol, acetonitrile, water, 0.05% PFPA and 0.01% FA. The third method utilized basic negative ion conditions on a separate, dedicated C18 column. Gradient elution was performed using methanol and water with 6.5mM Ammonium Bicarbonate at pH 8. For the fourth method, negative ionization was applied following elution through an HILIC column (Waters UPLC BEH Amide, 2.1 × 150 mm, 1.7 μm), with a gradient of water and acetonitrile containing 10 mM ammonium formate at pH 10.8. The scan range varied between methods, covering 70–1000 m/z.

Among 949 metabolites with known structural identity (named biochemicals) identified, we included the APAP parent compound and known metabolites of APAP in the present study. Analytes included unmetabolized APAP, APAP glucuronide, APAP sulfate, 2-hydroxyAPAP sulfate, 2-methoxyAPAP glucuronide, 2-methoxyAPAP sulfate, and 3-(methylthio)APAP sulfate. We dichotomized prenatal APAP exposure, categorizing individuals with one or more analytes detected in plasma as exposed, and individuals with no analytes detected as unexposed. Analyte distributions were plotted and pair-wise Pearson correlations among analytes were computed.

We chose not to model individual analytes or continuous analyte concentrations because the relative concentrations of individual analytes are mostly related to the time elapsed between APAP intake and analysis.³² Furthermore, distributions of APAP analyte concentrations were approximately normal and not truncated, indicating that non-detects likely reflect individuals not using APAP, rather than individuals using APAP but with low blood concentrations that were left censored (Supplemental Figure 1B).

Outcomes

We included multiple distinct outcomes, evaluated at child age 8–10 years, to capture different domains of ADHD-related symptoms and behaviors. These included 1) parent report of child ADHD diagnosis and their child taking prescription ADHD medications within the past month, 2) CBCL (completed by parents) for the DSM5-oriented ADHD subscale and externalizing subscale, and 3) task-based behaviors relevant to ADHD, namely the NIH Toolbox Flanker Inhibitory Control and Attention Test (NIH Flanker) and the Wechsler Intelligence Scale for Children Digit Span (Digit Span). In the computer-based NIH Flanker task, children indicate the left-right orientation of a central arrow over 20 trials while inhibiting attention to flanking arrows oriented in the same or opposite direction as the central arrow. Higher scores indicate better inhibitory control and visual selective attention. We used NIH Flanker uncorrected standard scores for all analyses. The Digit Span subtest measures auditory working memory. The examiner reads a sequence of numbers, and the child verbally recalls the numbers in the same order (Forward task), reverse order (Backward task), and ascending order (Sequencing task). We used Digit Span scaled scores for all analyses. We ensured that examiners were thoroughly trained on the administration of task-based outcomes prior to data collection. Administrations were video-recorded throughout the study and reviewed for quality control. We excluded scores that were invalid or missing due to child behavior problems, examiner error, or interfering events during testing (Figure 1).

Funded through the National Institutes of Health sponsored Environmental influences on Child Health Outcomes (ECHO) program, the ECHO prenatal and early childhood pathways to health (ECHO-PATHWAYS) consortium generated placental RNA-sequencing data on a total of 794 CANDLE participants (previously described in^81–83), including 174 placentas among our study sample. Briefly, within 15 min of delivery, a piece of placental villous tissue in the shape of a rectangular prism with approximate dimensions of 2×0.5×0.5cm was dissected from the placental parenchyma and cut into four ~0.5-cm cubes. The tissue cubes were placed in a 50-mL tube with 20mL of RNAlater (ThermoFisher, Waltham, MA; Catalog Number [CN]: AM7020) and refrigerated at 4°C overnight (≥8h but≤24h). Each tissue cube was transferred to an individual 1.8-mL cryovial containing fresh RNAlater. The cryovials were stored at −80°C, and the fetal villous tissue was manually dissected and cleared of maternal decidua. Following dissection, the fetal samples were placed into RNAlater and stored at −80°C.

RNA was isolated using the AllPrep DNA/RNA/miRNA Universal Kit (Qiagen, Germantown, MD; CN: 80224) according to the manufacturer’s recommended protocol. Integrity of RNA was determined with an Agilent 2100 Bioanalyzer instrument (Agilent, Santa Clara, CA). All RNA samples had RNA Integrity Numbers (RIN) >7. RNA sequencing was performed at the University of Washington Northwest Genomics Center. Total RNA was poly-A enriched to remove ribosomal RNA, and cDNA libraries were prepared from 1 μg of total RNA using the TruSeq Stranded mRNA kit (Illumina, San Diego, CA; CN: 20020595) and the Sciclone NGSx Workstation (Perkin Elmer, Waltham, MA). Each library was uniquely barcoded and subsequently amplified using a total of 13 cycles of PCR. Library concentrations were quantified using Qubit Quant-it dsDNA High Sensitivity assay fluorometric quantitation (Life Technologies, Carlsbad, CA; CN: Q32853). Average fragment size and overall quality were evaluated with the DNA1000 assay (Agilent, Santa Clara, CA; CN: NC1758992) on an Agilent 2100 Bioanalyzer. Each library was sequenced to an approximate depth of 30 million reads on an Illumina HiSeq sequencer. RNA sequencing quality control was performed using both the FASTX-toolkit (v0.0.13) and FastQC (v0.11.2).⁸⁴ Transcript abundances were estimated by aligning to the GRCh38 transcriptome (Gencode v33) using Kallisto,⁸⁵ then collapsed to the gene level using the Bioconductor tximport package, scaling to the average transcript length.⁸⁶ We retained only protein-coding genes, processed pseudogenes, and lncRNAs, and removed genes with average log counts per million (CPM)<0, resulting in a final sample of 13,961 transcripts. Log-CPM values were normalized to library size using the weighted trimmed mean of M-values.⁸⁷

Covariates

We collected covariate data on maternal characteristics via questionnaires and medical record abstraction.⁴⁰ We adjusted all models for precision variables, defined as variables that could affect child neurodevelopment but have no clear casual effect on maternal APAP use, and confounders, defined as potential causes of both maternal APAP use and child neurodevelopment. These adjustment variables included maternal age, education, ethnicity, pre-pregnancy body mass index (BMI), stressful life events during pregnancy, tobacco and alcohol use during pregnancy, non-steroidal anti-inflammatory drug (NSAID) use during pregnancy, antibiotic use during pregnancy, household income adjusted for region and inflation, neighborhood deprivation index (which includes indicators related to poverty, unemployment, and home ownership derived from the US Census)⁸⁸, delivery method, labor type, maternal report of parent or sibling mental health disorders (yes versus no), and child sex assigned at birth (see Figure 2 for conceptual model). Covariates were modeled continuously or categorically according to the variable groupings in Table 1, with the largest group for each variable set as the reference level.

Mothers self-reported alcohol use during pregnancy, while we included in the positive tobacco exposure group all individuals with maternal urine cotinine above 200 ng/mL,⁸⁹ as well as individuals who were below this cut-off but self-reported tobacco use during pregnancy. We assessed exposures to prenatal stressful life events using a survey adapted from the Centers for Disease Control and Prevention Pregnancy Risk Assessment Monitoring System.⁹⁰ Women responded yes or no to 14 statements covering multiple types of stressful life events they experienced during pregnancy, such as those relating to relationship problems, housing or financial issues, legal problems, and illness or death of a loved one. We additionally adjusted models examining neurodevelopmental outcomes for child age at the study visit, and placental transcriptomics outcome models for RNA-sequencing batch. We further included birthweight and gestational age at birth as covariates in sensitivity models as they may be predictors of the outcomes, but could also be on the pathway between exposure and outcome.

Statistical analysis

We calculated descriptive statistics for covariates and neurodevelopmental outcomes and performed univariate tests for associations between prenatal APAP exposure and each variable via analysis of variance and chi-square goodness of fit tests for continuous and categorical variables, respectively.

We utilized logistic regression to model associations of prenatal APAP exposure with binary outcomes (child ADHD and ADHD medications), and coefficients were exponentiated into odds ratios. We utilized linear regressions with robust standard errors to model CBCL outcomes. We modeled normally distributed NIH Flanker and Digit Span scores with linear regressions. We present unadjusted models, and models adjusted for the primary set of covariates discussed above.

We analyzed these data using sex as a covariate, and in separate models we tested interactions by including APAP by sex product terms. We imputed missing data using multivariate imputation by chained equations (MICE) via the ‘mice’ R package.⁹¹ We implemented MICE separately for males and females to create 100 datasets, including all covariates and outcomes in the predictor matrix. We combined the sex-stratified multiple imputations, ran models described above on all 100 datasets, and pooled results using Rubins’ rules.⁹²

Tobacco and alcohol exposure could alter APAP susceptibility by affecting liver and/or cytochrome p450 metabolism. In two separate sensitivity analyses, we tested interactions by including an APAP by tobacco or APAP by alcohol product term in the fully adjusted models for ADHD.

We performed whole-transcriptome analysis via RNA-sequencing for 174 placentas among our study sample. We constructed linear models for each outcome transcript using the RNAseqCovarImpute package,⁹³ which implements MICE⁹¹ for missing covariates and the limma-voom pipeline for differential expression analysis.⁹⁴ RNAseqCovarImpute accommodates high dimensional expression data by binning genes into smaller groups, creating separate multiply imputed datasets and differential expression models within each bin, and pooling results with Rubin’s rules. This method integrates with the limma-voom differential expression pipeline to fit weighted linear models for each gene that take into account individual-level precision weights based on the mean-variance trend.⁹⁴ Results are further moderated using the limma empirical Bayes procedure in which gene-wise variances are squeezed towards a global gene variance prior.⁹⁵

We obtained log₂-adjusted fold-changes (Log2FCs) associated with 1) prenatal APAP exposure versus no exposure, and 2) ADHD diagnosis versus no ADHD. In addition to RNA-sequencing batch, we included all covariates described above, with the exception that prenatal APAP models did not include child age. Using the formula (2^Log2FC-1)*100%, we transformed Log2FCs into percent gene expression changes. We report transcripts associated with prenatal APAP or ADHD at FDR<0.10.

We included transcripts associated with prenatal APAP or ADHD at FDR<0.10 in causal mediation analyses using the ‘mediation’ R package.⁹⁶ Because we only identified female sex-specific associations, we conducted sex-moderated mediation analyses. We specified the mediator model as “Transcript Expression ~ APAP*Sex + Covariates” and the outcome model as “ADHD ~ APAP*Sex + Transcript Expression + Covariates.” We performed moderated mediation with the ‘mediate’ function using the Quasi-Bayesian approximation with 1,000 simulations, and female sex set as a pre-treatment covariate to return female conditional estimates. After performing each mediation analysis on 100 multiply imputed datasets, we combined the results using the ‘amelidiate’ function to stack all simulations across all imputed datasets and determine estimates and 95% confidence intervals.

We input t-statistics (Log2FCs divided by standard error) from the differential expression analyses into gene set enrichment analyses for Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways using the generally applicable gene set enrichment (GAGE)⁹⁷ and Correlation Adjusted MEan RAnk (CAMERA)⁹⁸ methods. We excluded KEGG human disease pathways, and only reported pathways with FDR<0.05 in both CAMERA and GAGE methods by plotting their gene set enrichment P-Values (larger among the two methods) and mean gene t-statistics.

Data availability

The RNA-sequencing data utilized in this study are available at the database of Genotypes and Phenotypes (dbGaP) study accession phs003619.v1.p1. The covariate data utilized for this study contain protected health information and are not publicly available, but de-identified data may be available on request, subject to approval by the internal review board and under a formal data use agreement. Contact the corresponding author for more information (bbaker26@uw.edu).

Code availability

Representative code for conducting the analyses are available at the Zenodo repository under the GNU General Public License v3.0 (https://doi.org/10.5281/zenodo.14183453).⁹⁹