2.1. Sample characteristics
The characteristics of the 60 mother-infant pairs examined in this study are presented in Table 1. There were no pre-pregnancy overweight or obese women (BMI > 25) enrolled in our study. The mothers experienced no complications during pregnancy and newborns were all full-term births. Morisaki et al. estimated optimal and acceptable range of GWG for lowest probability of adverse outcomes such as complicated delivery, preeclampsia, and preterm delivery with each pre-pregnancy BMI category in 104,700 Japanese women [13]. According to the study of Morisaki et al., we considered upper limit of GWG recommendations of the Japanese Ministry of Health, Labour and Welfare (JMHLW) guidelines are theoretical: women with a pre-pregnancy BMI between 18.5 and 25 should gain approximately 7– 12 kg and those less than 18.5 should gain 9–12 kg body weight during the gestational period. To see insufficient GWG effects compared to optimal GWG, we extracted samples whose GWG were equal to or less than the upper limit of recommended GWG; 12 kg (n = 51). The distribution of GWG of 51 subjects is shown in Figure S1 (6.8 ± 3.4 kg). Figure S2 shows that the results of the estimation of cell type composition differed between individuals. Therefore, we considered the composition of each cell type as confounders in the linear regression model described in Methods.
2.2. Continuous GWG in the equal and less than upper limit of recommendation was associated with infant cord blood methylation
We hypothesized that GWG effects on infant epigenome might plot U-shape or reach plateau at a certain weight. Therefore, to see insufficient GWG effects, we extracted subjects whose GWG were insufficient or in optimal range as referring upper limit of recommendation (ULO; under upper limit of optimal range). In these subjects, GWG was significantly associated with cord blood methylation at five CpG sites with an
FDR-adjusted P-value < 0.05. No CpG site was significant after Bonferroni correction adjusted the P-value < 0.05 (Figure 1A and Figure S3, Table 2). Meanwhile, the regression analysis of methylation values against continuous GWG among all 60 175 subjects, which includes 9 subjects whose GWG exceeded optimal range, resulted in no significant association after multiple test correction (Figure 1B).
The methylation status were negatively associated with continuous GWG at all these significant five CpGs in 51 ULO subjects (Figure 1C and Table 2). Assessment of the methylation values of these five CpG sites in all 60 subjects revealed that the methylation plot reached a plateau or traced a U-curve near the border of the upper limit of the GWG recommendations (Figure 2). Meanwhile, non-linear categorical analysis showed that both insufficient and excessive GWG (n = 22 and 9, respectively) did not exhibit any significant differences when compared to the adequate (n = 29) group (Figure S4). These results indicated that insufficient GWG may gradually affect the fetal development as detectable range by DNA methylation levels and those effects may be reversed in excessive GWG. Within the five significant CpG sites, cg00599163 exists at the promoter site of LINC01806 and the methylation levels at this site in cord blood has been reported to be negatively associated with gestational age (GA) [19] (Table 2). The four other sites were in intergenic regions or gene body sites; out of these four, cg10370704 is in an intragenic enhancer chromatin state in several types of blood cells according to NIH Roadmap Epigenomics Mapping Consortium [20] (Figure S5).
2.3. Neither pre-pregnancy BMI nor birth weight associated with infant cord blood DNA methylation in our subjects
Regression analysis against pre-pregnancy BMI was also analyzed. An association between continuous pre-pregnancy BMI and DNA methylation in cord blood was not found at any CpG site (Figure S6A). Non-linear pre-pregnancy BMI distinguishing samples between lean (BMI < 18.5; n = 17) and normal (BMI ≥ 18.5 to ≤ 25; n = 43) also showed no association with DNA methylation (Figure S6B). None of our samples were collected from obese women (BMI > 25). The effects of pre-pregnancy BMI were not associated with methylation status in our samples. Simultaneously, there was no association between cord blood DNA methylation sites and birth weight in our samples (Figure S6C).
2.4. The CpG sites associated with GWG were also GA-related methylation sites
GA at birth is known to be associated with methylation levels at several CpG sites in cord blood [19, 21, 22]. Therefore, we conducted additional regression analysis on GA at birth. Continuous GA was significantly associated with methylation levels at two CpG sites (Figure S6D) located very close to one another (30 bp distance) on the promoter of the PLCH1 gene (Table 3). One of these two, cg11932158, was reported to be associated with GA at birth in two different cohorts [19, 22] on which a DNA methylation assay using a 450K array (i.e., a previous version of the array) was performed. Another cg21262198 was assayed with a newly designed probe in the EPIC array employed in this study; therefore, information regarding the association between the DNA methylation of cord blood cells and GA may not exist yet. However, it has been suggested that the promoter region of the PLCH1 gene, including the cg21262198 locus, might be associated with GA even after 37 weeks. GWG was not correlated with GA (P = 0.339) in our samples. Thus, we added GA as a covariate to an association study with continuous GWG and methylation, in which the minimum FDR-adjusted p value slightly elevated to 0.055 that was confirmed at 16 CpG sites. The above five CpG sites were also included in these 16 as top minimum nominal p values (Table 4). Meanwhile, an additional 11 CpG sites showed the same FDR-adjusted p value after considering GA as a covariate. Methylation at these 11 CpG sites also showed a negative association with continuous GWG, except for one of the CpG sites:
cg00445959 was in the enhancer chromatin modifications and was reported to be associated with last menstrual period-estimated GA [19]. These results indicated that lower GWG may slow fetal epigenetic development separately from GA.
2.5. Validation of methylation values at 5 CpG sites that associated with GWG in ULO
We validated GWG-associated methylation at the 5 CpG sites using an alternative methylation-measurement method to BeadChip namely deep targeted bisulfite sequencing. For the validation, we again converted unmethylated cytosines into uracils by treating subject’s genome DNA with bisulfite. Then targeted PCR was performed as described in method. Deep sequencing of bisulfite PCR amplicons using next generation sequencer enabled us to detect 1% methylation difference with minimum 6 reads to maximum 153 reads (supplementary table 2). As a result, within 20 CpGs that was included in PCR amplicons for 5 targeted CpGs, only 2 CpG sites which is adjacent to cg00599163 were significantly correlated with GWG in ULO negatively, although correlation of methylation at cg00599163 with GWG was not significant by validation analysis (Figure 3). Significant negative correlation between methylation at each two CpG site and GWG were also confirmed within whole subjects when added 9 subjects who gained weight more than upper limit of recommendation during pregnancy in validation experiments (supplementary Figure 7). Regarding to the region around cg00599163, we acquired 4998 ± 2391 reads (min. 2384 to Max. 15108) and the half of them on the middle 3 CpGs including cg00599163 and the most 5’ and 3’ CpGs, respectively. The distribution of methylation percentage of the cg00599163 in whole subjects was 81.4 to 96.5 % in deep targeted bisulfite sequencing which was wider than b value distribution which was 0.88 to 0.99 with BeadChips. The other 4 CpG sites other than cg00599163 were not significantly correlated in negative direction with GWG. It was the same as for neighborhood CpGs of each 4 CpG site (Supplementary table 2).