Methylation analysis for postpartum depression: a case control study

Women were recruited from Nagoya University Hospital and its associated institutes. Inclusion criteria were participation in perinatal classes for pregnant Japanese women before week 25 of pregnancy, age ≥ 20 years, and the ability to understand and speak Japanese. The study was explained both orally and in writing to all participants during the perinatal classes, and all participants provided written informed consent.

Participants were provided with a consent form and a self-reporting questionnaire during a perinatal class. The questionnaire included the EPDS and demographic questions, such as age, parity and current and previous mental illness (Additional file 1). Women were asked to complete the questionnaire in early pregnancy (before week 25) and return their responses, along with the agreement document, via mail using the provided self-addressed envelope. We provided self-addressed envelopes and postage on these envelopes. Questionnaires were mailed to women again around week 36 of pregnancy and 1 month after delivery, and returned in the same manner.

The EPDS is a 10-item self-reporting questionnaire that assesses PPD. Each item is scored on a 4-point Likert scale, with higher scores indicating worse depression [21]. The EPDS was translated into Japanese and then reverse translated back into English, and was determined to be the same as the English version [22]. The Japanese version of the EPDS showed good internal consistency (Cronbach’s alpha = 0.78) and test-retest reliability (Spearman’s correlation coefficient = 0.92) [22]. In the original version, the cut-off point is 12/13 [21], and the EPDS cut-off point 12/13 is recommended for English-speaking women [23]. In the non-English version, a cut-off point lower than 12/13 has been reported to be appropriate [24‐28], as cut-off point requires validation for specific populations [29]. Although one report regarding the EPDS recommended the cut-off point 12/13 in the Japanese version [30], the cut-off point in that study was calculated by subjects who excluded pregnant women with an EPDS ≦8 during pregnancy. Participant in this study is different from that study [30]. For example, we included women with an EPDS ≦ 8 during pregnancy. As a result, we used a cut-off point 8/9 in Japanese version of the EPDS by Okano et al. [22]. A score ≥ 9 on the Japanese version of the EPDS has been used as an indication of major depressive episodes. The test has a sensitivity of 75% and a specificity of 93% [22]. We used the 8/9 cut-off point and defined a score ≥ 9 as depressive episodes.

We evaluated depressive symptoms using the EPDS during early pregnancy (before week 25), late pregnancy (around week 36), and 1 month after birth. We measured EPDS scores twice in pregnancy, to confirm that participants were not depressed in pregnancy both cases and controls. The incidence of peak depression in the postnatal period has been debated [31],and one report found a peak 6 weeks after birth [32]. On the other hand, “maternity blues” frequently occur during the period immediately after childbirth to 10 days after childbirth [33]. Prevalence rates of maternity blues vary from 40 to 80% [34]. We considered that the time period immediately after childbirth, when many pregnant women experience maternity blues, was not a suitable time to search for PPD biomarkers. Therefore, we chose 1 month after birth as time point for postpartum EPDS and blood collection. We defined cases and controls based on the results of the EPDS scores. The controls scored under the cut-off point before week 25, around week 36, and 1 month after birth. The cases scored under the cut-off point before week 25 and around week 36, but scored over the cut-off point at 1 month postpartum. The control group excluded participants with current or previous mental illness. In addition, we matched cases and controls in terms of age, parity, gestational weeks, and the day of blood collection after birth. A total of 18 cases and 18 controls were identified.

We collected peripheral blood from all participants about 1 month after birth for extraction of genomic DNA. Venous blood was collected into tubes containing EDTA, and genomic DNA was extracted from peripheral blood using the Qiagen QIAamp DNA bloods Kit (Qiagen, Hilden, Germany). The extracted DNA was stored at − 30 °C until the start of the experiment. We performed DNA extraction from October 2006 to March 2013, and conducted experiments from March 2013 to November 2015.

Genomic DNA was extracted from peripheral blood. Bisulfite conversion of genomic DNA from peripheral blood was done using the EpiTect Plus Bisulfite Kits (Qiagen Ltd., Hilden, Germany). Genome-wide DNA methylation profiles were obtained using the Infinium Human Methylation 450BeadChip according to the manufacturer’s instructions (Illumina, San Diego, CA, USA). To minimize potential batch effects, cases and controls were processed together and run on the same BeadChip. Samples were randomly positioned on each BeadChip to reduce any possible position effects within chips. We used a correction method applied in a previous study to reduce technical bias in the DNA methylation array data [35]. Briefly, DNA methylation values were corrected for background and normalized with the Subset-quantile Within Array Normalization function [36] of minfi in R to adjust for color bias [37]. The minfi is a flexible and comprehensive R package for the analysis of Infinium DNA methylation microarrays, and R is an open-source, free software program for statistical analysis.

DNA methylation level was quantified as an M value, which can be converted to a β value according to the equation: β = 2^M/(2^M + 1) [38]. The M value for all DNA methylation sites was corrected for batch effects with the ComBat function of sva in R. sva is an R package that contains functions for removing batch effects [39] and the ComBat function adjusts for known batches using an empirical Bayesian framework [40].

We estimated the cell type composition for each sample with the estimate Cell Counts function [41] of minfi in R. For each sample, probes with a detection p value ≥0.05 were assigned a status of “not detected”; this led to the exclusion of 1316 probes (5% of samples). We also excluded 30,819 probes previously found to be cross-reactive (≥47 bases) [42] and probes on the Y chromosome. Probes containing single nucleotide polymorphisms have been found to influence the assessment of DNA methylation status when using the Infinium Human Methylation 450BeadChip [43]. We therefore filtered out probes that contained single nucleotide polymorphisms with a minor allele frequency > 0.01 based on 1000 Genomes project samples of ASN (ASN: Han Chinese individuals in Beijing, Southern Han Chinese, and Japanese in Tokyo) [42] to reduce the frequency of false positives. Finally, a total of 379,277 DNA methylation sites remained.

The association of PPD with DNA methylation status at each DNA methylation site was assessed using a linear regression analysis, with DNA methylation status at each site as a dependent variable, and PPD label (case = 1, control = 0) and covariates as independent variables. The covariates comprised age and the cell type composition of samples. All p values were corrected for genomic control. The significance level α for the epigenome-wide association study analysis was determined by dividing 0.05 by the number of DNA methylation sites for Bonferroni correction (α = 0.05/379,277 = 1.32 × 10^− 7). A p value < 0.05 was considered nominally significant.

Gene ontology (GO) terms are used to describe 3 categories of gene products: biological process, molecular function, and cellular component [44]. These terms are widely used in gene function studies. Identifying GO terms that are overpresented within a given list of genes can increase the understanding of the functional relevance of these genes [45].

For genes that showed differential DNA methylation between cases and controls, gene set enrichment analysis was used to identify enriched GO terms. Analyses were done using the Database for Annotation, Visualization and Integrated Discovery (DAVID) 6.8 [46, 47]. DAVID is a free, online bioinformatics resource that lists a comprehensive set of functional annotations that can be used to identify the biological significance of a list of genes. We selected 1000 sites with the lowest p values. A total of 711 genes that were annotated within these 1000 sites were analyzed using DAVID. GO terms with a false discovery rate < 0.05, as calculated by the Benjamini-Hochberg adjustment method, were considered significant.

The datasets used and analyzed during the current study are available from the corresponding author on reasonable request.

Table 1 shows characteristics of the study subjects. The mean ages of cases and controls were 33.3 ± 4.4 and 34.6 ± 4.6 years, respectively. There were no significant differences in age, parity, gestational weeks,baby’s weight (g), and blood collection day after birth (days) between cases and controls. EPDS score at 1 month after birth was significantly higher among cases compared with controls. The range of EPDS scores at early pregnancy and late pregnancy was 0–8 for cases and controls. The range of EPDS scores at 1 month after birth was 9–25 for cases and 0–8 for controls.

After initial processing, 18 cases and 18 controls as well as 379,277 DNA methylation sites remained for subsequent analysis. We performed an association analysis for DNA methylation level and PPD and found no genome-wide significant association (Fig. 1). In addition, no outliers were detected in the quantile-quantile plot of −log₁₀(p) for the 379,277 tests of association between DNA methylation status and cases-controls (Fig. 2).

As a result of DAVID 6.8 analysis of the 711 genes associated with the 1000 sites with the lowest p values (Additional file 2: Table S1), we found four GOTERM DIRECT terms and 16 GOTERM FAT terms with a false discovery rate < 0.05 (Table 2). GOTERM DIRECT provides GO mappings directly annotated by the source database and GOTERM FAT filters out very broad GO terms based on a measured specificity of each term. We showed 22 genes of the axon guidance in which the Fold Enrichment was largest among the four GOTERM DIRECT terms (Table 3).

In this study we started recruiting participants before week 25 of pregnancy, so we could collect detailed information from participants and match cases and controls in terms of age, parity, gestational weeks, and the day of blood collection after birth. The association between PPD and age [58, 59], parity [60, 61], and past mental illness [62, 63] has been reported. However, in this study, there were no significant differences in age and parity between the two groups. We excluded participants with current or previous mental illness and who had an EPDS score≧9 during pregnancy. Therefore, it was possible to exclude the influence of age, parity and history with mental illness to depression. Furthermore, there was no significant difference in the blood collection date of peripheral blood used for methylation analysis, and it was possible to compare the methylation status of cases and controls around 30 days postpartum. Additionally, we placed cases and controls on the same chip but in a random position to mitigate a batch effect and minimize technical bias.

This study has several limitations. First, the sample size was small. In this study, we evaluated depressive symptoms between pregnancy and the postpartum period using the EPDS, and performed blood collection after childbirth. When the number of cases is small, a large number of controls is needed to avoid bias. However, it was difficult for us to obtain a large sample. We attempted to overcome the limitation of sample size by appropriate analysis. However, it is possible that the negative findings in this study could be due to the small sample size. It is difficult to make a conclusion based on the analysis of small objects, so results of this study can only be considered preliminary data. Future studies with larger samples are needed. Second, to clarify the difference in postpartum depression, the conditions during pregnancy cases and controls were matched. That is, we excluded participants with current or previous mental illness and who had EPDS score ≥ 9 during pregnancy. Therefore, our study is subject to selection bias and the results do not reflect a general pregnant population. Depression during pregnancy and medical history are considered to be related to PPD, and it is necessary to consider these subjects in the future. Third, we defined cases and controls based on the results of the EPDS scores in this study. However, EPDS is a screening tool as opposed to diagnostic tool, and cases of PPD cannot be determined from EPDS score alone. It was a limitation of this study that diagnostic interviews could not be conducted for all the subjects. Finally, we used peripheral blood as opposed to brain tissue in this study. However, the methylation status of peripheral blood and brain may be different, and although using peripheral blood as a biomarker has a clinical advantage, it is also a study limitation.