Maternal atopy and offspring epigenome-wide methylation signature

ABSTRACT

The increase in the prevalence of allergic diseases is believed to partially depend on environmental changes. DNA methylation is a major epigenetic mechanism, which is known to respond to environmental factors. A number of studies have revealed that patterns of DNA methylation may potentially predict allergic diseases.

Here, we examined how maternal atopy is associated with methylation patterns in the cord blood of neonates.

We conducted an epigenome-wide association study in a cohort of 96 mother-child pairs. Pregnant women aged not more than 35 years old, not currently smoking or exposed to environmental tobacco smoke, who did not report obesity before conception were considered eligible. They were further tested for atopy. Converted DNA from cord blood was analysed using Infinium MethylationEPIC; for statistical analysis, RnBeads software was applied. Gestational age and sex were included as covariates in the final analysis.

83 DM sites were associated with maternal atopy. Within the top DM sites, there were CpG sites which mapped to genes SCD, ITM2C, NT5C3A and NPEPL1. Regional analysis revealed 25 tiling regions, 4 genes, 3 CpG islands and 5 gene promoters, (including PIGCP1, ADAM3A, ZSCAN12P1) associated with maternal atopy. Gene content analysis revealed pointwise enrichments in pathways related to purine-containing compound metabolism, the G1/S transition of the mitotic cell cycle, stem cell division and cellular glucose homoeostasis.

These findings suggest that maternal atopy provides a unique intrauterine environment that may constitute the first environment in which exposure is associated with methylation patterns in newborn.

KEYWORDS:

• Atopy
• maternal effect
• EWAS
• epigenetics
• programming

Introduction

In recent decades, we have observed substantial increases in the prevalence of allergic diseases. Environmental changes and rapid development of industrialization and urbanization, together with modern lifestyle factors, are considered to be the principal factors in this process[1].

The first environment to which human beings are exposed occurs in utero, where individuals are surrounded by maternal metabolic and immune milieu before birth. It has been shown that the in utero environment in asthmatic mothers is very specifically different from non-asthmatic mothers, and direct effects of the altered cytokine environment, induced by asthma, exist in pregnancy[2]. Maternal effects have been described in relation to atopy and other allergy outcomes [3,4]. Also, epidemiological studies concerning early allergy phenotypes indicate that the prenatal period is the part of life that seems to be most crucial for promoting immune system maturation [5,6]. Both atopy and asthma begin in utero [2,7]. This observation is similar to those made in other studies, where prenatal programming conditioned disease phenotypes, such as diabetes or metabolic syndrome, later in life [8,9].

DNA methylation (DNAm) is a major epigenetic mechanism that influences gene expression. However, this relationship is complicated and not fully understood. DNAm is known to respond to environmental factors and connects fundamental mechanisms of genetics and the environment to allergy development, with both genes and the environment mediating interactions on the molecular level. A number of studies have revealed that the DNAm patterns of allergy genes are associated with allergic diseases [10–13].

Many factors contribute to DNAm changes in the foetus. Exposure to maternal smoking[14], BMI[15], DHA levels[16], vitamin D levels [17] and folate supplementation [18] have been shown to be associated with DNAm in cord blood. Thus, we hypothesized that maternal atopy influences prenatal programming.

Here we present genome-wide DNAm patterns in cord blood in a cohort of newborns in relation to the occurrence of maternal atopy to identify both general methylation patterns and the state of specific genes that are differentially methylated.

Results

All recruited pregnant women were tested for atopy at enrolment. Apart from having or not having pets, there were no significant differences between atopic and non-atopic mothers. In the atopic group, asthma was reported in 12%, allergic rhinitis in 30% and atopic dermatitis in 20%. The mean age was 29 years in both the atopic and non-atopic groups. The mean BMI was 20.5 in the atopic group and 21.5 in the non-atopic group, and the mean GWG (gestational weight gain) was 14.81 and 12.9, respectively, in the atopic and non-atopic groups. The proportion of caesarean sections was relatively high: 42% and 47%, respectively. In comparison, in the general population, the national proportion of caesarean sections in the year 2017 was 44% (https://ec.europa.eu/eurostat/). The majority of women were highly educated – they graduated from University with a degree. In both groups, more than 80% of women were taking vitamin supplements – the majority being multivitamins which contain vitamin D and folate. Characteristics of the groups studied are presented in Table 1.

Table 1. Characteristic of study group. C p < 0.05 were considered significant; A chi2 test was performed to assess categorical data, a Man-Whitney U test was used for quantitative variables. Data are presented as mean ± SD or as numbers and frequencies (%). GWG gestational weight gain

	Atopic n = 50	Non-atopic n = 46	p
Maternal age (years)	29.38 (± 1.96)	29.84 (± 2.38)	0.33
Asthma	6 (12%)	1 (2%)
Allergic rhinitis	15 (30%)	7 (15,21%)
Atopic dermatitis	10 (20%)	2 (4,35%)
Maternal pre-pregnancy BMI (kg/m^2)	20.9 (± 2.24)	21.5 (± 3.03)	0.75
GWG (kg)	14.81 (± 4.34)	12.90 (± 4.081)	0.06
Healthy diet index	5.75 (± 1.97)	5.85 (± 1.75)	0.84
History of smoking	17 (34%)	21(45%)	0.2
Pets at home	11 (22%)	21 (45%)	0.01
Birth weight (g)	3416 (± 592)	3476 (± 434)	0.97
Apgar-median value	10	10	0.97
Child sex-female	18 (36%)	19 (41%)	0.59
Gestational age (week)	39,6 (± 1.27)	39.5 (± 1.19)	0.61
Parity-first child	44 (88%)	43 (94%)	0.36
Delivery Caesarean section	21 (42%)	22 (47%)	0.57
Tertiary education	49 (98%)	43 (93%)	0.27
Smoking ever before pregnancy	17 (34%)	21 (45,65%)	0.21
Vitamin supplementation	42 (84%)	43 (93,48%)	0.15
Multivitamin	34 (68%)	34 (73,91%)	0.29
Folic acid only	4 (8%)	7 (15,21%)
Vitamin D only	4 (8%)	1 (2,17%)

Global methylation profile differentiation

Probe filtering based on quality resulted in a final set of 827,101 interrogated methylation sites across all 96 samples. An initial test that detected an association between traits (covariates) and principal components of methylation variability within samples showed that gestational age and sex were associated with methylation profile characteristics. In the final analysis, gestational age and sex were used to correct obtained methylation β-values. Additionally, a correction for cell-type heterogeneity was implemented, as proposed by Houseman et al. (2014)[19]. Differentiation of global methylation profiles of the study groups (as shown using principal component analysis) was low, with a visible subgroup of samples showing distinct methylation profiles that included subjects predominantly from non-atopic mothers (Supplementary File 1). This sub-cluster was also distinct when methylation profiles were analysed on regional levels, which corresponded to promoters, genes and Chg. islands. This sub-cluster remained separated when unsupervised hierarchical clustering was performed based on Euclidean distance (Supplementary File 2). The distribution of methylation values across different genomic regions (CpG contexts) was also similar between groups (Supplementary File 3).

Site-level differential methylation analysis

Analysis of differential methylation revealed very few significantly differentially methylated sites after FDR (p < 0.05, n = 43). Nevertheless, following the guidelines of Müller et al. [20], we have used the RnBeads software score to identify DM sites with high levels of confidence. As differentially methylated, we selected the top 0.01% of sites based on the empirical distribution of their scores. This resulted in the selection of 83 sites which differed between offspring born to atopic mothers and controls with a high confidence (Figure 1; Table 2; Supplementary File 4). The analysis of site annotations showed that they were predominantly localized within gene bodies (48.2%) and intergenic regions (33.7%). Only 19.3% of the DM sites considered were located within known CpG islands. Most of the DM sites identified were hypomethylated in the atopic group (n = 49; 59%). The average absolute delta beta value for hypermethylated sites in the atopic group was low (0.054), while for hypomethylated sites it was 0.047. The distribution of hyper- and hypomethylated DM sites in different functional elements differed highly significantly (Chi-Square test p < 0.01). The hypermethylated sites were more commonly located in intergenic regions (42.4% in hyper- vs. 28.6% in hypomethylated sites), while hypomethylated sites were more common in gene bodies (55.1% in hypo- vs. 39.4% in hypermethylated sites) (Supplementary Table 1). Within the 83 significant individual sites identified, there were 25 mQTL associated CpG sites[21].

Table 2. Top individual CpG sites differentially methylated between the analysed groups and their annotations in mQTL and EWAS Atlas databases

cg id	CHR	Mean difference	Regulation due to maternal atopy	p value	Combined Rank	UCSC Ref Gene Name	UCSC Ref Gene Group	mQTL	ATLAS db
cg19373347	1	−0,054530219	Hyper	1,8E-05	987		Intergenic
cg04990378	20	0,060359753	Hypo	0,000377	1914	C20orf166	Body
cg07997860	3	0,052077231	Hypo	0,00043	2116	STAC	Body		Asthma
cg13655169	14	0,056791611	Hypo	0,000534	2552		Intergenic
cg15668967	5	−0,062960699	Hyper	0,000399	3341		Intergenic	YES
cg04311686	5	−0,065457165	Hyper	0,000804	3690		Intergenic
cg10528424	11	0,142,814,783	Hypo	0,001013	4343	SYT8	Body		Asthma
cg07601484	20	−0,055840455	Hyper	0,001019	4370		Intergenic
cg12746557	21	0,030175735	Hypo	0,000876	4405	KCNJ15	Body	YES
cg10507965	10	0,036684685	Hypo	0,001035	4428	SCD	Body
cg13092397	18	−0,078980897	Hyper	0,001429	5786	LINC00669	Body
cg15717617	8	0,053380751	Hypo	0,001435	5805	PLEKHA2	Body
cg18548103	2	−0,028037937	Hyper	0,000779	5817	ITM2C	?		Asthma
cg04036196	7	−0,027301153	Hyper	0,000817	5934	NT5C3A	Body
cg27367142	10	0,05675362	Hypo	0,001609	6390		Intergenic	YES
cg01584086	11	−0,046425801	Hyper	0,001676	6612		Intergenic
cg01400671	9	0,046056889	Hypo	0,001687	6654		Intergenic	YES
cg02956194	6	−0,228,330,076	Hyper	0,001747	6840		Intergenic
cg20198384	20	0,030402417	Hypo	3,58E-05	6881	NPEPL1	Body
cg25830305	11	−0,060954357	Hyper	0,001798	7018	TNNI2	?	YES	Atopy

Figure 1. Plot of mean methylation difference between groups against – log₁₀p-value for differential methylation of (a) CpG sites, (b) tailing regions, (c) genes, (d) promoters and (e) CpG islands. Colour gradient corresponds to combined rank based on RnBead software score.

Analysis of differentially methylated regions

Region-based tests included tiling regions, genes, promoters and GpG islands. From each category, the top 0.01% of regions with the best identified RnBead scores were selected as DMRs. This resulted in the selection of 25 tiling regions, 4 genes, 3 CpG islands and 5 gene promoters, which contained at least two probes inside each region (Supplementary File 5). Most DM tiles (n = 20) were hypomethylated in the atopic group, which had an average delta beta value of 0.028. The number of sites within tiled regions ranged from 2 to 12, and the regions were distributed throughout 18 different autosomes. All gene regions that were differentially methylated were hypomethylated in the atopic group with an average delta beta value of 0.025. The genes included PIGCP1, ADAM3A, AL390719.2 and ZSCAN12P1. DM promoters with at least two analysed probes corresponded to CMYA5 and A2M-AS1 genes and AC005481.1, AC114291.1 and AC005280.1 novel transcripts. Promoters were both hypo- and hypermethylated with an average absolute delta beta value of 0.023. The three identified CpG islands had a mean absolute delta beta value of 0.032 and were localized on chromosomes 7 (158,854,447–158,854,967 bp), 9 (90,621,589–90,622,101) and 12 (9,217,329–9,217,715).

Gene content analysis of DM sites and regions

The sites that were DM between atopic and control groups were associated with 50 different genes. Together, the genes were not significantly enriched (after correction for multiple testing) in any biological processes, KEGG pathways or disease phenotypes. However, some pointwise enrichments have been observed for purine-containing compound metabolic processes (ADCY8, ESRRB, HK1, MRI1, NT5C3A, and SCD), the G1/S transition of the mitotic cell cycle (CUL3, ESRRB, NACC2, and PRIM2), stem cell division (CUL3, ESRRB) and cellular glucose homoeostasis (ADCY8, HK1, and PTPRN2). The subset of genes with hypermethylated sites (n = 16) was mainly associated with the cell cycle G1/S phase transition and DNA replication, while hypomethylated sites (n = 34) were associated with purine ribonucleotide biosynthetic processes (Supplementary File 6).

The tiling regions between atopic and control groups that were identified as DM spanned 19 different genes. These genes did not enrich any biological processes after FDR; however, pointwise significance was found for cell death regulation (NCK adaptor protein 2, EYA4, MEAK7, GRIN2A, DAPK1, GRK5), the glutamate receptor signalling pathway (GRIN2A, DAPK1) or phenotypes associated with sudden cardiac death (EYA4, DTNA) (Supplementary File 7). DM regions covering known CpG islands spanned two different genes, namely VIPR2 and LINC00612, which, like genes and promoters identified via regional tests, were either not classified in GO databases or were not enriched in any annotation categories.

Discussion

This study demonstrated that maternal atopy is associated with specific epigenetic signatures in offspring. Individual site analysis identified 83 CpG sites which map to 50 genes associated with maternal atopy. The most significantly differentially methylated sites annotated to specific genes included C20orf166, STAC, SYT8, KCNJ15, SCD, LINCOO669, PLEKHA2, ITM2C, NT5C3A and NPEPL1. The range of methylation differences were 5–22%. Within this group, DMs mapped to four genes known to impact the immune system, allergy and asthma. These identified genes were SCD, ITM2C, NT5C3A and NPEPL1.

The SCD gene encodes a stearoyl-coenzyme A desaturase, which has been reported to associate with two major allergic phenotypes: atopic dermatitis and asthma. The gene plays an important role in lipid synthesis. Disrupted expression of SCD is associated with patients presenting atopic dermatitis and causes a malfunction in the skin barrier of these patients[22]. In experimental models of asthma, inhibition of the enzyme encoded by SCD leads to airway hyper-responsiveness and reduced immune defence against viruses[23]. ITM2C is a target gene of GATA-3, a T-cell specific transcription factor. ITM2C deficiency has little impact on the function of polyclonal T cells, but attenuates T helper cell-dependent immune responses[24], which seems to be specifically important for the future development of Th2 phenotypes such as allergy. Additionally, altered methylation of ITM2C, STAC, PRIM2 and CCD81 genes has been revealed in the lung cells of patients with asthma[25]. NT5C3A is a gene encoding a protein that affects erythrocyte function. Recently, studies have shown a different regulatory function for the gene. NT5C3A expression is induced by IFN gamma and acts as an anti-inflammatory regulator via an epigenetic mechanism[26]. IFN gamma is one of the most important cytokines affecting allergic outcomes by regulating the maturation of T-cells. Another gene, NPEPL1, is associated with FEV1/FVC ratio in smokers, implying a possible biological function for the gene in the occurrence of asthma. FEV1/FVC is the most sensitive indicator of bronchial obstruction in clinical practice. It is also widely used in research as a proxy of asthma severity and lung function[27].

Furthermore, beyond the top most significantly differentially methylated sites, were others mapped to genes previously reported for their association with allergy including CCDC81[28], LGR6 [25,29,30], MIR166[31], ESRRB [13,25,32], HK1[33], TNFRSF17[34], PTPRN2 [13,25,35–37], GRK5 [38,39], CAMTA1 [40,41], NCK2[42], COLEC11[43], VIPR2 [25,44–46], and PRIM2[41]. Changed methylation patterns corresponding with allergic and respiratory phenotypes were also identified for SYT8 [25,30], TNN12, NXN, INP5A and ARHGAP30 [13,25], ADYC8 [25,47], NACC2 [13,25,30,48], SRPRB[49], ERICH1 [13,25,47], CUL3 and AHRR [25]. AHRR was revealed in many studies as the most significant marker of smoking exposure in both neonate and adult populations [50,51]. As our group was not exposed to tobacco smoke during pregnancy, this is an unexpected result. Also, we did not observe a significant difference between groups in smoking habits before pregnancy. Possibly, tobacco exposure prior to pregnancy could play a role or some air pollutant has a similar effect.

While comparing our results with other EWAS studies, we found three CpG sites which had been previously revealed as associated with allergy. The first, cg04453550, mapped to TNFRSF17 (a 12% methylation difference was observed). Hypermethylation within this site has previously been shown to be associated with food allergies[34]. TNFRSF17 protein plays a role in B-cell maturation, signal transduction, cell survival and proliferation[52]. The second, cg01400671, is an intragenic region [13] and the third, Cg13575925, mapped to LOC144571[53]. For some other differentially methylated CpG sites identified, associations have been shown between methylation patterns and phenotypes such as ageing, maternal smoking, maternal diabetes and air pollution.

In our initial analysis, we excluded methylation markers which overlapped with SNPs. Further, we used an mQTL database for the assessment of genetic influence on methylation signatures in newborns. In the 10 most highly ranked CpG sites identified, cg12746557 mapped to KCNJ15 was mQTL related. CpG sites with lower rank within genes – CCDC81, MIR166, ESRRB and PTPRN2 – were also mQTL related. This type of epigenetic marker has been described as highly heritable and constant through and are indicative of the impact of genetics on DNAm[21].

Via regional analysis, we identified 37 differentially methylated regions. These regions map to genes and gene promoters including PIGCP1, ADAM3A, ZSCAN12P1, CMYA5, A2M-AS1, as well as new transcripts including AC005481.1, AC114291.1, AC005280.1, PIGCP1, ADAM3A, ZSCAN12P1 and pseudogenes. Pseudogenes act as traps for transcripts and may affect gene regulation. ADAM33 and ZSCAN1, other members from the ADAMs and ZSCAN superfamilies, have been reported in GWAS and EWAS studies to being associated with allergic diseases [54,55]. However, according to current knowledge, ADAM3 protein function is related to fertility and cell-to-cell adhesion. Also, this particular member of the ADAM superfamily is lacking the metalloprotease activity [56] which seems to be important according to the functional role of ADAM33 in asthma[57]. There are studies referring to ADAM3A as being associated with specific neoplastic diseases, pointing to its regulatory role[58]. ZSCAN proteins are transcription factors that may be involved in angiogenesis, cell apoptosis and differentiation, cell migration and cell invasion, cell proliferation and stem cell properties[59], but the function of ZCAN12P1 is unknown. A2M-AS1, an RNA gene, has been reported to associate with chronic lung disease[60].

In the enrichment analysis of 50 genes associated with DM sites, we found that genes did not significantly enrich any biological processes, KEGG pathways or disease phenotypes collectively. However, pointwise enrichment has been observed in purine-containing compound metabolic processes (ESRRB, HK1, NT5C3A, SCD genes), the G1/S transition of the mitotic cell cycle (ESRRB, PRIM2), stem cell division (ESRRB) and cellular glucose homoeostasis (HK1, PTPRN2). This is indicative of the complex character of variation observed.

To our knowledge, there are two studies concerning the maternal effect on DNAm at birth in the current literature. Both investigated the effects of the occurrence of maternal asthma. In the first study, 12 CpG regions within the 10 genes were described, that had methylation differences greater than 10%. The study was conducted in 12-month-old infants born to 25 asthmatic mothers and 12 mothers that were not asthmatic. There were atopic mothers in both groups; however, a higher proportion occurred in the asthmatic group. The authors determined the relationship between the methylation status of PM20D1 and atopy[4]. A second study included a cohort of 36 children born to asthmatic mothers, some of whom were atopic. The authors compared methylomes at birth and prospectively evaluated the development of asthma later in life. Within this study, 589 DM regions were found[55]. Both studies identified two differentially methylated genes that were significantly associated with the presence of maternal asthma, PIWIL1 and SMAD3. However, neither gene was identified in our study. Population differences could be the reason why we did not replicate these findings. In both mentioned papers, the population of mothers were mixed according to atopy status, so atopic mothers were included in asthma and non-asthma groups. Also, other parameters such as metabolic status, smoking and maternal age were not included in those analyses. Both PIWIL1 and SMAD3 are specifically related to asthma. PIWIL1 is overexpressed in asthmatic airway epithelia, SMAD3 functions as a regulator of TGF beta, thus impacting Treg and Th17 differentiation, and also correlates with the level of IL1 beta, a pro-inflammatory mediator in asthma.

This study had some advantages over previous reports. We selected a specific group of pregnant women, whose children could be estimated to be within a high-risk population for the development of allergy. These children were born in an industrialized city and to an atopic mother, but they also lacked risk factors such as pre-pregnancy obesity, metabolic complication of pregnancy or tobacco smoke exposure. These criteria made the group more homogeneous but also very unique, with no adequate currently available replication cohort. In VIVA, ALSPAC and R generation cohorts, the data for maternal atopy were collected as self-reported asthma, allergic rhinitis or eczema, which are slightly different phenotypes[48]. We performed our analysis while considering factors previously reported to impact methylation in cord blood, such as maternal age, BMI[61], GWG [62] and healthy diet index[63]. Also, we have applied cell heterogeneity adjustments in cord blood[19], which increased the reliability of our results in comparison to unadjusted analyses. We used the RefFreeEWAS method proposed by Houseman et al. i. This method was shown to work equally well as other methods using reference datasets for cord blood[64]. Additionally, according to Houseman et al., RefFreeEWAS, may adjust for detailed cell type differences that may be unavailable even in existing reference datasets [65]. Umbilical cord blood contains mixed leukocyte populations with nucleated red blood cells. The latter is only a minor fraction of the cord blood cell population (3.4–9.3%) [66] but could increase due to prenatal stress or hypoxia in situations such as maternal smoking, obesity or preterm birth. As our mothers were not smoking, not obese and in general had children born on term, these complications were not the case in our group. Presumably, variations in nucleated red blood cells should not significantly affect global methylation profiles.

In this study, the FDR method was used for multiple testing correction. However, we decided to use RnBeads ‘combined rank’ as it combines statistical significance as well as size and quotient of methylation difference, which presumably strengthens differential methylation analysis. In our results, 43 sites had significant differences after FDR between groups; however, they poorly covered sites identified by rank analysis. The sites which were significant after FDR had a mean methylation difference among groups of 2.5E-05. The calculation of combined rank ‘addresses the problem that minimal but consistent differences tend to achieve low p-value that doesn’t reflect biological significance.’[67]

We have used allergic sensitization as a proxy for atopy. It is an objective measure and the very first allergic outcome reflecting an alternation in the immune system. Atopy, in consequence, leads to allergic symptoms, such as asthma, allergic rhinitis or atopic dermatitis. However, not all atopic individuals have clinical manifestations of inappropriate IgE immune response. In some epidemiological studies, up to 43% of subjects sensitized to common inhalant allergens did not present any respiratory symptoms[68]. But, on the other hand, IgE from asymptomatic subjects have been revealed as functional, meaning that they initiate the cascade of cells, cytokine and inflammation even in the absence of symptoms [69–71].

The majority of our group of pregnant women had higher education. Possibly, this group of people are more prone to participate in scientific research but, also, they are attending birth schools more eagerly than others. The proportion is high in the general population anyway; in the year 2018, 44% of young people graduated from tertiary educational institutions in our country[72].

We have shown that maternal atopy is associated with DNAm in cord blood and that methylation patterns are clearly distinguishable between neonates born to atopic and non-atopic mothers. Although the observed methylation differences in the majority of sites are not very high (2–22%), we considered that a cumulative effect of methylation was likely to occur.

Methods

Study population

Study participants were recruited as a part of the ongoing ELMA – Epigenetic Hallmark of Maternal Atopy and Diet study, a prospective study of women and their children. ELMA was established to identify exposures that may contribute to allergic disease susceptibility in children. Women living in the metropolitan area, not currently smoking or exposed to ETS (environmental tobacco smoke), who did not report obesity before conception, gestational diabetes or hypertension were considered eligible and were recruited during their third trimester (≥28 week of gestational age) of pregnancy by attendants of a childbirth schools situated in obstetrics clinics and hospitals in Wroclaw. From November 2016 to January 2019, 158 women were enrolled and the majority (72.15%) were successfully followed to delivery. Umbilical cord blood samples were collected by midwifes or obstetricians at birth. Children from these pregnancies are being followed at 3, 6 and 12–18 months of age.

Allergic sensitization (atopy) in mothers was defined as the presence of sensitization to allergens via measurement of serum allergen-specific immunoglobulin E (IgE) for 20 allergens including cow milk, egg, wheat flour, peanuts, dust mites, trees, grasses, cat, dog, Alternaria and Cladosporium allergens (POLYCHECK, Biocheck, Germany). Women were designated atopic if they displayed at least one positive reaction (sIGE ≥0.35kUL⁻¹) to any allergen.

All women completed food frequency questionnaires FFQ containing queries about diet that was based on the FFQ KOMPAN questionnaire. The questionnaire included 38 questions regarding the estimated frequency of intake of food groups and beverages. Finally, analyses were performed using a derivative variable (healthy diet index), which was based on the frequency of whole meal bread, buckwheat, oatmeal, whole grain pasta or other coarse cereal, milk, fermented milk drink (yoghurt, kefir, cottage cheese), white meat (chicken, turkey, rabbit, fish), legume seed (bean, pea, soybean, lentil), fruit and vegetable consumption[73]. The index is ranged from 0 to 20, with higher values reflecting a healthier diet.

The cord blood samples from 96 mother-child pairs were selected for DNAm analysis, 50 from atopic mothers and 46 from mothers that were non-atopic. The selection of probes was based on DNA availability and quality. Mothers aged more than 35 years were excluded from this analysis.

The study was approved by the Ethical Committee of the Wroclaw Medical University and all participants signed informed consent forms.

Methylome analysis

Cord whole blood was secured in EDTA probes during delivery, according to standard procedures. DNA was extracted from whole blood using the QIAamp DNA Blood Mini Kit (Qiagen), according to the manufacturer’s protocol. DNA was further quantified using a Qubit 2.0 fluorimeter. DNA probes with unsatisfying quality were excluded. DNA was converted for bisulphite sequencing using the EZ DNA Methylation™ Kit (Zymo Research). Converted DNA from 96 cord blood samples from newborns with atopic (n = 50) and non-atopic control (n = 46) mothers were finally analysed using the Infinium MethylationEPIC Kit (Illumina, San Diego, CA). The hybridized and stained EPIC arrays were ultimately scanned using a HiScanSQ system (Illumina).

Infinium MethylationEPIC arrays covered over 850,000 methylation sites across the human genome. Apart from sites within known CpG islands, they included probes identifying CpG sites outside CpG islands, non-CpG methylated sites identified in human stem cells (CHH sites), DMs identified in tumour versus normal cells, FANTOM5 enhancers, ENCODE open chromatin and enhancers, DNase hypersensitive sites and sites located in miRNA promoter regions.

Data analysis

Intensities obtained after probe scanning were added to strictly control quality, filtering and normalization. Initial quality control was performed via the analysis of array controls using the BeadArray Reporter Software (Illumina). These analyses did not reveal any significant deviations from expected values. Further filtering was accomplished using RnBeads software[74]. Initially, Infinium probes overlapped with SNPs, and sites with missing β values were removed. Both sites and samples were filtered using a Greedy approach, which iteratively removes probes and samples of highest impurity from the dataset. These correspond to the rows in the detection p-value table that contain the largest fraction of unreliable measurements. Additionally, probes located on sex chromosomes were excluded as a result of expected variation resulting from sex differences. The obtained data were normalized using the BMIQ procedure [75] and batch effects, such as array and array position as well as other hidden confounders, were identified and removed using the surrogate variable analysis (SVA) method[76]. The significance of covariates including healthy diet index, age, BMI, gestational age, birth weight, GWG, parity, type of delivery, having pets and sex were evaluated using a procedure that was also implemented in RnBeads software. While only gestational age and sex traits were significantly associated with methylation of samples, they were included in a final analysis in which the refFreeEWAS method [65] was used for beta-value (methylation values) corrections regarding cell-type heterogeneity and the inclusion of the two mentioned covariates. The refFreeEWAS method allows reference-free deconvolution that provides proportions of putative cell types defined by their underlying methylomes and allows an explicit quantitation of the mediation of phenotypic associations with DNAm by cell composition effects. Finally, RnBead software scores were used to detect sites that were differentially methylated between the groups considered. RnBeads combines statistical testing with a priority ranking scheme that is based on the absolute and relative effect size of the differences between groups, and it assigns a combined rank score for differential DNAm to each analysed CpG site and genomic region. This combined rank is defined as the maximum (i.e., worst) of three individual rankings: (i) by absolute difference in mean DNAm levels, (ii) by the relative difference in mean DNAm levels, which is calculated as the absolute value of the logarithm of the quotient of mean DNAm levels, and (iii) by the CpG-based or region-based p-value, calculated as described above[67].

The smaller the combined rank for a site, the greater the evidence for differential methylation exhibited. The top 0.01% of differentially methylated sites with best ranks were analysed as well as the top 0.01% of differentially methylated regions (DMRs). DMR types were defined as tiling regions (with a window size of 5 kb across the genome), genes, promoters (Ensembl gene definitions were used that define promoter regions as the 1,500 bases upstream and 500 bases downstream of the transcription start sites of corresponding genes) and GpG islands (obtained from tracks in UCSC Genome Browser). The genes associated with DM sites and DM regions were analysed using the WEB-based GEne SeT AnaLysis Toolkit [77] to identify enriched biological processes, pathways and disease phenotypes. Overrepresentation tests were performed with respect to all known genes (genome) using an FDR [78] correction for multiple testing. Additional gene annotations were performed using the Panther Classification System[79].

Additionally, we have performed a search using the EWAS ATLAS database [80] to identify previous epigenetic studies investigating atopy- and respiratory-related phenotypes.

Conclusion

In this study, we have identified several DM CpG sites and regions in cord blood of infants born to atopic mothers. These DM sites map to genes previously revealed to be associated with allergic phenotypes and also to new ones, which can act as a focus of new studies. These findings suggest that maternal atopy constitutes a unique intrauterine environment that is associated with methylation patterns.

Disclosure statement

Danielewicz H, Gurgul A, Dębińska A, Myszczyszyn G, Szamtoła T, Myszkal A, Jasielczuk G, Drabik-Chmaerska A report grant and personal fees from the National Science Center, Poland, during the conduct of the study; Danielewicz H reports lecturer fees from Mead Johnson Nutrition outside the submitted work; Hirnle L, Boznanski A report a grant from the National Science Center, Poland, during the conduct of the study.

Supplementary material

Supplemental data for this article can be accessed here.

Correction Statement

This article has been republished with minor changes. These changes do not impact the academic content of the article.

Additional information

Funding

This study was funded by the National Science Centre, Poland [DEC-2015/19/B/NZ5/00041].