Prenatal antiepileptic exposure associates with neonatal DNA methylation differences

Abstract

Antiepileptic drugs (AEDs) are used to treat a variety of neuropsychiatric illnesses commonly encountered in women during their reproductive years, including epilepsy and bipolar disorder. Despite their widespread use, the impact of prenatal exposure on fetal development remains obscure. To evaluate whether AEDs taken by pregnant mothers influence DNA methylation patterns in their neonates, DNA was extracted from the umbilical cord blood of 201 neonates whose mothers were treated for neuropsychiatric illness during pregnancy and interrogated across 27,578 CpG sites using the Illumina HumanMethylation27 BeadChip. The association of each methylation value with the cumulative duration of prenatal AED exposure was examined using a linear mixed model. The average methylation level across all CpG sites was calculated for each subject, and this global methylation measure was evaluated similarly. Neonates with a longer duration of AED exposure in pregnancy showed a decrease in average global methylation (p = 0.0045). Further, DNA methylation of CpG sites in 14 genes significantly decreased with the duration of prenatal AED exposure even after adjusting for multiple comparisons (FDR < 0.05). For a small subset (n = 19) of these neonates, a second tissue, placenta, was available in addition to cord blood. Methylation of 3 of these 14 CpG sites was also significantly decreased in placental tissue. These novel data suggest decreased DNA methylation in neonates of mothers who took AEDs during pregnancy. The long-term stability and potential impact of these changes warrant further attention, and caution may be warranted before prescribing AEDs to pregnant women.

Introduction

Antiepileptic drugs (AEDs) are used to treat a variety of medical illnesses commonly encountered in women during reproductive years, including epilepsy, bipolar disorder, migraine headaches, fibromyalgia, and other pain syndromes.^1-4 Some of these illnesses (most notably epilepsy and bipolar disorder) almost always require continued pharmacological treatment during pregnancy.^5,6 AEDs cross the human placenta, producing considerable fetal medication exposure.^7-9 The implications of fetal AED exposure on child development warrant scrutiny.

As a class, there have been long-standing concerns about the impact of prenatal AED exposure on the risk for congenital malformations (also known as birth defects) and offspring neurodevelopment.¹⁰ Recent findings suggest that prenatal AED exposure may be associated with dose-dependent verbal and nonverbal cognitive deficits in children.¹¹ In particular, valproate exposure predicts significant decreases in IQ¹² and congenital malformations.¹³ Preclinical studies of several AEDs (i.e., phenobarbital, phenytoin, vigabatrin, and valproate) suggest they can induce widespread neuronal apoptosis and impair neurogenesis in immature animal brains.^14-16 Further, lamotrigine, valproate and vigabatrin may impair neuronal migration.^17,18 These observations raise serious concern that AEDs, which are commonly used in women of childbearing potential, could produce similar adverse effects in children with prenatal or early postnatal exposure.

Valproate alters the epigenome of exposed cells by inhibiting histone deacetylases (HDACs) and inducing DNA demethylation.^19-22 These epigenetic mechanisms participate in fundamental developmental and regulatory processes. Thus, increasing attention has been focused on the epigenetic changes resulting from other commonly prescribed AEDs.^20,23 However, no previous study has examined associations between prenatal exposure to AEDs and DNA methylation in human neonates. Therefore, we examined DNA methylation patterns throughout the genome in umbilical cord blood and placentas from neonates with varying durations of AED exposure during gestation.

Results

A total of 201 mother-neonate pairs were included (Table 1). Of these, 53 neonates were exposed to AEDs during pregnancy (range of exposure: 4–40 weeks).

Table 1. Clinical and Demographic Characteristics of the Mother/Neonate Pairs (n = 201).

Maternal	N	%
Psychiatric Diagnosis*	196	97.5
No Axis I Diagnosis	12	6.1
Bipolar Disorder	40	20.4
Major Depressive Disorder	102	52.0
Panic Disorder	8	4.1
Generalized Anxiety Disorder	16	8.2
Anxiety Disorder NOS	4	2.0
Other Disorders	14	7.1
Epilepsy	26	12.9
Antiepileptics During Pregnancy	53	26.4
Lamotrigine	36	67.9
Valproate	3	5.7
Levetiracetam	3	5.7
Carbamazapine	2	3.8
Topirimate	1	1.9
Phenytoin	1	1.9
Gabapentin	1	1.9
Multiple AEDs	6	11.3
Maternal Age at Delivery, Mean (SD), yrs	33.5	5.00
Parity, Mean (SD)	0.85	0.99

Neonatal	N	%
Race, %
Caucasian	183	91.0
African American	18	9.0
Sex, %
Male	98	48.8
Female	103	51.2
Neonatal Birth Weight, Mean (SD), kg	3.37	0.45
Gestational Age (EGA), Mean (SD), weeks	38.80	1.10
APGAR at 5 min, Mean (SD)	8.74	1.19

^* 9 of the 12 subjects with no axis I diagnosis have epilepsy. Psychiatric diagnosis was not available for 5 mothers.

Prenatal exposure to AEDs

Investigations of medication exposure during pregnancy typically are conducted in clinical populations; therefore, individual use of multiple medications from different classes is common. The potential interactions between medication classes were examined using hierarchical clustering to determine if methylation patterns were influenced by use of concurrent medications taken daily (i.e., antidepressants) or as needed (i.e., benzodiazepines, hypnotics or antiemetics). Other environmental risk factors (tobacco or alcohol) were similarly evaluated, but no significant patterns were evident. Therefore, we examined the AUC for the number of weeks of exposure to AEDs as a continuous variable for each of the 27,578 CpG sites.

Global methylation levels were decreased in the umbilical cord blood of neonates exposed to AEDs (t = -2.88; p = 0.0045; Figure 1). Additionally, methylation of 14 individual CpG sites was decreased in subjects with greater duration of AED exposure (FDR < 0.05; Table 2). Increased methylation was not observed at any CpG site (FDR < 0.05). We observed no evidence that changes in DNA methylation patterns were specific to a single AED nor did the observed changes appear more extreme in those exposed to more than one type of AED (data not shown).

Figure 1. Decrease in global methylation in umbilical cord blood (vertical axis) with increased duration of exposure to AED (horizontal axis).

Table 2. CpG sites differentially methylated in umbilical cord blood with increased duration of exposure to AEDs.

CpG ID	Gene	Umbilical Cord Blood			Placenta
		Δβ *	t-statistic	p-value	Δβ *	t-statistic	p-value
cg05157725	COL21A1	-0.018	-4.95	1.85 × 10−6	-0.085	-0.69	0.25
cg15711744	ANP32D	-0.029	-4.74	4.67 × 10−6	-0.081	-0.84	0.21
cg01206970	PPFIA3	-0.009	-4.62	7.72 × 10−6	-0.037	-1.62	0.066
cg17391877	PGC	-0.029	-4.56	9.82 × 10−6	-0.076	-3.79	0.0013
cg19523692	ATP1B4	-0.028	-4.49	1.33 × 10−5	0.130	1.01	0.83
cg27394486	C15orf2	-0.064	-4.48	1.42 × 10−5	-0.296	-1.79	0.050
cg00353953	ZNF384	-0.036	-4.42	1.81 × 10−5	-0.107	-2.40	0.017
cg11042320	PDGFRB	-0.042	-4.41	1.88 × 10−5	-0.019	-0.34	0.37
cg07388493	NDUFS5	-0.030	-4.48	2.14 × 10−5	0.017	0.45	0.67
cg00536175	GATA1	-0.031	-4.34	2.14 × 10−5	-0.013	-0.25	0.40
cg23124451	CBX7	-0.031	-4.32	2.68 × 10−5	0.009	0.12	0.55
cg23307338	MS4A12	-0.034	-4.31	2.75 × 10−5	0.071	1.06	0.16
cg21690892	GUCA1B	-0.013	-4.30	2.90 × 10−5	0.031	1.99	0.97
cg15277108	SDS	-0.020	-4.29	3.08 × 10−5	0.056	1.84	0.95

^* Δβ = average change in β-value associated with 40 vs. 0 weeks exposed to AEDs

Because both the umbilical cord and placenta are integral in the transport of nutrients and drugs to the neonate, epigenetic changes in the umbilical cord may also be observed in the placenta. Therefore, in a follow-up analysis using 19 placental samples from this population, we evaluated the methylation pattern for each of the CpG sites that were differentially methylated in the umbilical cord blood. Methylation of cg17391877 (PGC; p = 0.0013), cg00353953 (ZNF384; p = 0.017), and cg27394486 (C15orf2; p = 0.050) were decreased in the placental tissue of subjects with increased numbers of weeks of exposure to AEDs. Methylation of cg01206970 (PPFIA3) also appeared to decrease with duration of exposure, but fell short of statistical significance (p < 0.07). We did not observe an association between global placental DNA methylation and duration of AED exposure.

Maternal illness

Of the 53 mother-neonate pairs exposed to AEDs, 26 of the mothers were being treated for epilepsy and 27 were being treated for psychiatric disorders (20 with bipolar disorder, 5 with major depression and 2 for an anxiety disorder). There was a moderate rate of comorbid psychiatric illnesses in the 26 women with epilepsy, including major depressive disorder (n = 8), an anxiety disorder (n = 4), and other psychiatric diagnoses (n = 2). Therefore, we sought to evaluate whether maternal illness could be responsible for the changes in neonatal DNA methylation. No CpG site was differentially methylated in subjects whose mothers had epilepsy and/or any current psychiatric diagnosis. We next evaluated the potential effects of prenatal seizure activity by comparing the neonatal DNA methylation patterns of women who reported having seizures during pregnancy (n = 14) to those who did not, but no CpG site was associated. Initially the number of seizures during pregnancy associated with 138 CpG sites independent of AED exposure (FDR < 0.05), but after correction for influential outliers, no CpG site remained significant (data not shown).

Discussion

These novel data suggest that prenatal AED exposure is associated with differences in the methylation patterns of DNA obtained from both umbilical cord blood and placental tissue. Commonly prescribed AEDs including valproate^21,22 and topiramate²³ can act as histone deacetylase (HDAC) inhibitors, which result in widespread changes in DNA methylation.^21-23 However, only three subjects in this study were exposed to valproate and only one to topiramate, while 77.4% of subjects taking AEDs were exposed to lamotrigine (67.9% on monotherapy; Table 1). Removal of 37 individuals exposed to AEDs other than lamotrigine did not substantially change the effect sizes observed in Table 1. Lamotrigine does not appear to act as an HDAC inhibitor in rat astrocytes,²⁰ suggesting that changes in DNA methylation resulting from prenatal AED exposure may be due to biological mechanism other than HDAC inhibition.

Several individual CpG sites had decreased methylation levels with increased duration of AED exposure. Methylation of a CpG site in progastricsin (PCG; also known as pepsinogen C) was significantly lower in both umbilical cord blood and placental DNA. PCG is developmentally regulated²⁴ and is expressed in the gastrointestinal tract, lung, pancreas and genitals as well as secreted into serum.²⁵ It has a central role in processing and activation of surfactant protein B, a critical component of pulmonary surfactant.²⁶

Methylation of a CpG site in collagen, type XXI, α 1 (COL21A1) was also decreased in the umbilical cord blood of neonates with more exposure to AEDs. COL21A1 is a developmentally regulated gene that is highly expressed during fetal development and is hypothesized to contribute to the extracellular matrix assembly of the vascular network during blood vessel formation.²⁷ COL21A1 is localized to tissues containing type I collagen, and a CpG site in zinc finger protein 384 (ZNF384), which binds and regulates the promoter of type 1 collagen, was also differentially methylated in both umbilical cord blood and placental DNA. This transcription factor is expressed in osteocytes, osteoblasts, and chondrocytes²⁸ and is believed to facilitate their formation through extracellular matrix remodeling. ZNF384 may also bind to numerous genes involved in osteogenesis, hematopoiesis, and gonadal development.²⁹

We also observed decreased methylation of a CpG site in (C15orf2), which is part of the Prader-Willi region of chromosome 15q11.2, in both umbilical cord blood and placental DNA. Interestingly, microdeletions in this area result in idiopathic generalized epilepsy syndromes, which are believed to be highly heritable.³⁰ C15orf2 is imprinted such that only paternal alleles are expressed in the fetal brain, but both maternal and paternal alleles are expressed in adult tissues including several brain regions and testes.³¹ It is feasible that deletions or epigenetic silencing may contribute to vulnerability to heritable forms of epilepsy. While the function of this gene remains unknown, prenatal exposure to AEDs may result in increased expression thereby conferring some protection to the neonate.

Because AEDs cross the placenta and can be detected in the developing fetus, we hypothesize the differences in DNA methylation result from direct exposure.^7-9 However, it is also possible that maternal AED exposure prior to conception influenced the epigenetic patterns of her oocytes. Animal studies will be helpful in delineating these hypotheses and in exploring the underlying mechanisms. Our study examined DNA extracted from peripheral tissues, and we identified alterations in DNA methylation in genes related to peripheral functions that may be tissue-specific and have long-term developmental consequences. While the DNA in this study was from heterogeneous tissues and should be interpreted with caution, we hypothesize that prenatal exposure to AEDs may correlate with clinically significant consequences to the offspring. Epigenetic differences associated with prenatal AED exposure should be evaluated in additional tissues, including those of the brain and other organ systems through animal or post-mortem studies. Differential methylation of the genes identified in this study may be a precursor to clinically significant differences in the neonate, and further studies will be needed to evaluate the stability of this pattern over time. This study had a limited control group because the samples were obtained from a clinical population. However, the study benefited from laboratory confirmation of AED exposure and prospective documentation of other exposures and maternal illnesses. It is also important to note that we observed no impact of the maternal diagnoses for which these AEDs were prescribed, including epilepsy or bipolar disorder that could account for the observed patterns in offspring DNA methylation.

It remains unclear whether there is a dose effect of these medications on neonatal DNA methylation or the extent to which the medication and illness may interact, and further studies in this area are warranted. However, these data may have broad implications given the wide range of conditions treated by AEDs.

Methods

Subject ascertainment and clinical assessment

Subjects were recruited as part of the Specialized Center of Research for Sex and Gender Effects (SCOR) and the Translational Research Center for Behavioral Sciences (TRCBS) through the Emory Women’s Mental Health Program. Women with a history of psychiatric illness and/or epilepsy were enrolled prior to 14 weeks gestation. At study entry, mothers completed an intake questionnaire for demographic, socioeconomic, medical and psychiatric history, and were administered the Structured Clinical Interview for DSM-IV (SCID).³² They were evaluated at 4–6 week intervals throughout pregnancy to assess psychiatric symptoms, seizure history and compliance with medication regimen. Maternal medication use during pregnancy was documented at each visit and recorded by gestational week. Maternal inclusion criteria included: 1) > 17 y of age, 2) written and verbal fluency in English, 3) folate supplementation, 4) term delivery of a live, singleton infant, and 5) availability of DNA from umbilical cord blood. Women were excluded for: 1) unstable non-neuropsychiatric illnesses requiring pharmacological treatment during pregnancy (e.g., asthma, autoimmune disorders); 2) abnormal thyroid stimulating hormone; and 3) positive urine drug screen or use of illicit drugs or narcotics during pregnancy.

The following AEDs were included in this study: carbamazepine, valproate, gabapentin, lamotrigine, levetiracetam, oxcarbazepine, phenytoin, pregabalin and topiramate (Table 1). Medication use was confirmed by objective laboratory assay of maternal serum AED concentrations during pregnancy. Subjects were excluded if serum analysis from maternal and/or umbilical cord blood was consistently below the limits of detection of laboratory assays suggesting noncompliance with the medication regimen. All participants provided written informed consent prior to study enrollment, and the Emory University Institutional Review Board approved this study.

Biological sample collection and DNA extraction

Umbilical cord blood samples were collected at birth, stored on ice, and processed within 2 h of delivery. Plasma was separated by centrifugation at 4°C, and the cellular fraction was frozen at -80°C until processing. DNA was extracted from the cellular fraction at the Emory Biomarker Service Center using a Qiagen Biorobot M48. Placental cotyledons from planned cesarean section deliveries were dissected immediately following delivery after ablation of the decidua basalis (maternal tissue), and aliquots were snap frozen at -80°C. Placental samples were homogenized using a TissueLyser II, and DNA was isolated using an AllPrep Kit (Qiagen).

DNA methylation

DNA methylation assays have been previously described in detail.³³ Briefly, 1 µg DNA was bisulfite-treated, whole-genome amplified, fragmented, and hybridized to the HumanMethylation27 BeadChip (Illumina). One sample of pooled female DNA was included on each BeadChip as a control throughout the experiment and assessed for reproducibility using a Pearson R² coefficient. Three samples with probe detection call rates < 90% or with an average intensity value of either < 50% of the experiment-wide sample mean or < 2000 arbitrary units (AU) were excluded from the analysis. For individual sample i and CpG site j, the signals from methylated (M) and unmethylated (U) bead types are used to calculate a β value where β = M/(U+M) and is the proportion of CpG dinucleotides methylated at a particular site. Prior to computing β values, we normalized the signal data to adjust for any technical variability between samples by utilizing the information from 16 negative control probes that are included on the Illumina BeadChip and are designed to detect a true methylated and unmethylated signal of zero.

Statistical analysis

We worked with the log ratio of β-values log[β /(1-β)], which is equivalent to the log signal ratio log(M/U) for normalized values.³⁴ To identify CpG sites for which methylation varied significantly with AED exposure, we fit a separate linear mixed effects model for each CpG site, where log[β /(1-β)] was treated as the outcome. Our independent variable was area under the curve (AUC) for weeks of AED exposure during pregnancy, normalized to a standard 40-week pregnancy, because maternal daily dose (mg/day) varied across pregnancy. Maternal age, preconception and delivery body mass index (BMI), current diagnosis of epilepsy or a neuropsychiatric condition, method of delivery, neonatal sex and gestational age (GA) at delivery were examined as potential independent predictors of umbilical cord and placental DNA methylation using the model described above. Similarly, we assessed maternal use of caffeine, alcohol and tobacco during pregnancy. Any variable that yielded differentially methylated genes that met experiment-wide criteria was included as a covariate in subsequent analyses. For each CpG site, we regressed log[β /(1-β)] on our measure of AUC, as well as sex, race and GA.^33,35 Chip-specific random effects were included in the model to allow for chip-to-chip differences in measurement of the proportion of DNA methylated. The above model was fit for each of 27,578 CpG sites. To define an initial set of CpG sites for which methylation varied significantly with each outcome, we applied a false discovery rate (FDR) cutoff of 0.05 using the method described by Storey.³⁶ To follow up significant results in placental samples, we computed 1-sided p-values using the model described above. To examine global methylation, we also fit the above model with a measure of average methylation across all 27,578 CpG sites as the dependent variable: log[ β̅ / (1− β̅)], where β̅ represents the average β value over all 27,578 sites for an individual.