Participants
Our participants were from the Ma’anshan Birth Cohort (MABC) study, a population-based prospective study that recruited pregnant women during early pregnancy in the city of Ma’anshan, China, from May 2013 until September 2014. Inclusion criteria for this cohort were as follows: ≤14 gestational weeks; ≥18 years old; living in Ma’anshan city for more than 6 months; planning on delivering at the Maternal and Child Health Care Centre of Ma’anshan; and good communication and interpersonal skills. A total of 3474 pregnant women who met the inclusion criteria were included in this cohort and followed-up for their physical and mental health information at the second and third trimesters of pregnancy. Postpartum, 3273 single live-birth children were invited for follow-up to assess their development and growing environment postnatally at: 0 and 42 days; 3, 6, 9, 12 and 18 months; and 3, 4 and 5 years. A total of 2405 (69.2%) mother–child pairs who completed both the assessment of pregnancy-related anxiety at the third trimester and emotional symptoms and hyperactivity at 4 years old were included in this study. Comparison of the basic maternal characteristics between those who were recruited in the final data analysis and those who dropped out is shown in Supplementary Table 1. This study was approved by the ethics committee of Anhui Medical University (Reference number: 20180084) and written informed consent was obtained from each pregnant woman.
Pregnancy-related anxiety
Pregnancy-related anxiety in the third trimester of pregnancy was measured using the Pregnancy-Related Anxiety Questionnaire, which comprised 13 items across three subscales: “fear of woman’s own health” (six items); “fears related to the health of the fetus” (five items); and “fear of childbirth” (two items). Participants were required to rate their answers on a four-point Likert scale from 1 (never worried) to 4 (always worried). Scores of all the items were summed, with a total score of 13–52; a higher score indicates a higher level of anxiety. The pregnant woman whose score reached or exceeded the 75th percentile of the total score will be evaluated as having pregnancy-related anxiety. The scale was given in Chinese, which was developed by our team based on the Chinese population and has been verified in 7017 pregnant women in Anhui Province, China, with Cronbach’s alpha of 0.81 and a test–retest reliability of 0.79 [30].
Children’s emotional symptoms and hyperactivity
Children’s emotional symptoms and hyperactivity were evaluated by the Strengths and Difficulties Questionnaire (SDQ), which is a brief behavioral screening instrument used to measure emotional and behavioral difficulties and prosocial behavior of 4–16-year-olds [31]. The SDQ contained 25 items and covered five subscales relating to the child’s emotional symptoms, conduct problems, hyperactivity, peer relationship problems and prosocial behavior. Each subscale consists of five questions rated on a three-point Likert scale (not true = 0; somewhat true = 1; certainly true = 2) and scores in the range 0–10. Higher scores represent greater symptom severity. In our study, we used the Emotional Symptoms and Hyperactivity subscales of the SDQ, which was filled out by parents (91.6%), grandparents (7.6%) or others (0.8%) of the child. Children were assessed at a mean of 48.96 months (standard deviation was 2.70 months). The preschooler whose subscale score is above the 80th percentile (indicating borderline and abnormal) will be identified as having problems on this subscale [32]. Specifically, the cut-off scores in our study were 3 and 6, respectively. SDQ used in our study was given in Chinese, translated from English. As described in Du et al. 2008 [33], the Cronbach's α coefficients for the Emotional Symptoms and Hyperactivity subscales of the Chinese translation of the SDQ were 0.60 and 0.76, respectively. Besides, the convergent and discriminant validity proved to be acceptable [33].
CpG islands selection, sample collection and DNA methylation detection
We selected CpG islands located in the promoter of the FKBP5, NR3C1 and HSD11B2 genes from 2 kb upstream of the transcriptional start site (TSS) to 1 kb downstream of the first exon according to the following criteria [34]: (1) 200 bp minimum length; (2) 50% or higher cytosine–guanine content; (3) 0.60 or higher ratio of observed/expected CpG dinucleotides. Finally, four regions from CpG islands of the NR3C1 gene (111 CpG sites), two regions from CpG islands of the HSD11B2 gene (48 CpG sites) and five regions from CpG islands of the FKBP5 gene (104 CpG sites) were selected and sequenced (Figure 1).
Placental lobules from the full-thickness placenta 5 cm around the umbilicus were collected by trained personnel within minutes after delivery. The placental lobules were then immediately snap-frozen in liquid nitrogen, transported to the laboratory within 24 hours and preserved at −80°C until further analysis. The 2405 subjects included were ranked in descending order according to the Pregnancy-Related Anxiety Questionnaire scores. The placentas of the top 300 subjects (pregnancy-related anxiety group) and the bottom 300 subjects (control group with no pregnancy-related anxiety) were then analyzed for DNA methylation.
Genomic DNA was extracted from the placenta tissue of the above two groups using a QIAGEN kit (QIAGEN, Hilden, Germany). DNA methylation detection was performed using MethylTargetTM (Genesky Biotechnologies Inc., Shanghai, China). In brief, the procedure was as follows: DNA was subjected to sodium bisulfite treatment using an EZ DNA Methylation™-GOLD kit (Zymo Research) according to the manufacturer’s instructions; multiplex polymerase chain reaction (PCR) was performed with an optimized primer set combination; PCR amplicons were diluted and amplified using indexed primers; index PCR amplicons were separated by agarose gel electrophoresis and purified using a QIAquick gel extraction kit (QIAGEN); and libraries from different samples were quantified and pooled together, followed by sequencing on the Illumina NextSeq platform according to the manufacturer’s instructions. Sequencing was performed with a 2 ´ 150 bp paired-end mode. The methylation levels of each CpG were equal to the ratio of methylated cytosine to total cytosine.
Confounding factors
Based on existing literature and the results of our univariate analysis, we considered the following variables as confounders: maternal age, pre-pregnancy body mass index (BMI), gestational weight gain, education, family monthly income, smoking, drinking, gestational diabetes, pregnancy-induced hypertension, delivery mode and exclusive breastfeeding in the first 6 months. Information on maternal age, pre-pregnancy BMI, education, family monthly income, smoking and drinking was obtained through self-assessment questions in the first trimester questionnaire. Pregnancy complications, including gestational diabetes, pregnancy-induced hypertension and child gender, birthweight and gestational age at delivery were extracted from medical records. Information on gestational weight gain and feeding patterns at 6 months was derived from the postnatal questionnaire filled out by parents or other caregivers of preschoolers. The distribution of the covariates is shown in Table 1.
Statistical analysis
The participants’ characteristics are presented as mean ± standard deviation (mean ± SD) or number (frequency). Differences in the distribution of demographic characteristics between the pregnancy-related anxiety group and the control group were assessed using the t-test for continuous variables and chi-square tests for proportions. Logistic regression models were used to estimate the odds ratio (OR) with 95% confidence interval (95% CI) in the relationship between maternal pregnancy-related anxiety in the third trimester and children’s emotional symptoms and hyperactivity. The OR and 95% CI values were adjusted for several confounding factors, including maternal age, pre-pregnancy BMI, gestational weight gain, education, family monthly income, smoking, drinking, gestational diabetes, pregnancy-induced hypertension, delivery mode and exclusive breastfeeding in the first 6 months.
In this study we used factor analysis, which described variability between 263 CpG sites as a lower number of latent factors, to reduce the number of comparisons made. Factor analysis uses factor rotation of maximize orthogonal rotation (maximum variance method). Factor significance is defined by eigenvalues of > 2. Factor load, representing the correlation between individual CpG methylation and factor scores, was used to determine the contribution of individual CpGs to each factor; CpGs with an absolute factor loading of ≥ 0.3 were retained. Logistic regression was used to investigate the relationship between potential methylation variables generated by factor analysis and pregnancy-related anxiety in the third trimester and children’s emotional and hyperactivity stratified by infant gender. Then, the PROCESS program of mediation was used to perform a mediation analysis [35]. To test the mediating roles of potential methylation variables in the relationship between pregnancy-related anxiety in the third trimester and children’s emotional and hyperactivity. This approach uses bootstrapping to estimate all of the parameters. The mediating effect was tested using a bootstrap estimation approach with 5000 repetitions. When the 95% CI did not contain 0, the indirect effect was considered significant.
We also performed sensitivity analyses to check the robustness of our results. First, we excluded preterm birth (gestation < 37 weeks). Maternal anxiety in the third trimester is associated with premature delivery [36], and premature infants have increased risk of long-term neurodevelopmental problems [37]. To the extent that the unmeasured pathology that triggers preterm birth also harms the fetus directly, preterm birth can be confused with neonatal outcome. Direct adjustment of gestational age as a mediating variable will lead to bias when analyzing the relationship between risk factors and neonatal outcome. Second, the interaction between the severity of birthweight and maternal anxiety had significant impact on infant development[38]. Direct adjustment of birthweight can cause bias, therefore we did not adjust for birthweight in the main analysis but carried out sensitivity analysis instead.
All statistical analyses were performed with SPSS 23.0 software. The level of significance was P < 0.05.