Genome-wide DNA methylation of the liver reveals delayed effects of early-life exposure to 17-α-ethinylestradiol in the self-fertilizing mangrove rivulus

ABSTRACT

Organisms exposed to endocrine disruptors in early life can show altered phenotype later in adulthood. Although the mechanisms underlying these long-term effects remain poorly understood, an increasing body of evidence points towards the potential role of epigenetic processes. In the present study, we exposed hatchlings of an isogenic lineage of the self-fertilizing fish mangrove rivulus for 28 days to 4 and 120 ng/L of 17-α-ethinylestradiol. After a recovery period of 140 days, reduced representation bisulphite sequencing (RRBS) was performed on the liver in order to assess the hepatic genome-wide methylation landscape. Across all treatment comparisons, a total of 146 differentially methylated fragments (DMFs) were reported, mostly for the group exposed to 4 ng/L, suggesting a non-monotonic effect of EE2 exposure. Gene ontology analysis revealed networks involved in lipid metabolism, cellular processes, connective tissue function, molecular transport and inflammation. The highest effect was reported for nipped-B-like protein B (NIPBL) promoter region after exposure to 4 ng/L EE2 (+ 21.9%), suggesting that NIPBL could be an important regulator for long-term effects of EE2. Our results also suggest a significant role of DNA methylation in intergenic regions and potentially in transposable elements. These results support the ability of early exposure to endocrine disruptors of inducing epigenetic alterations during adulthood, providing plausible mechanistic explanations for long-term phenotypic alteration. Additionally, this work demonstrates the usefulness of isogenic lineages of the self-fertilizing mangrove rivulus to better understand the biological significance of long-term alterations of DNA methylation by diminishing the confounding factor of genetic variability.

Introduction

Early-life exposure to environmental stressors, encountered during the sensitive period of embryogenesis or in juveniles, can be critical in shaping the long-term control of tissue physiology and homoeostasis. Although this paradigm, referred to as the Developmental Origins of Health and Disease (DOHaD) [1], is now widely recognized, the molecular mechanisms by which early exposures influence the propensity of disease and phenotype later in life remain elusive. Identifying these mechanisms has become extremely important to understand the long-term effects of toxicants in human and wildlife and ensure the proper risk assessment of xenobiotic exposure [2].

Nowadays, endocrine disrupting chemicals (EDCs) – xenobiotics able to interfere with the proper functioning of the endocrine system – are ubiquitous in the environment and our everyday life. The timely release and tightly regulated concentrations of hormones in early life are crucial for the proper development of the organism, including the reproductive, nervous, and immune systems. Therefore, early-life exposure to EDCs can have dramatic consequences on homoeostasis and physiology. In fact, mounting epidemiological evidence link exposures to EDCs to the increased incidence of metabolic diseases, immune diseases, neurological disorders, cancer and alteration of fertility in humans [3–6]. Of the many EDCs present in the aquatic environment, 17-α-ethinylestradiol (EE2) – a synthetic derivative of oestradiol used in oral contraceptives – is of particular concern due to its high potency and resistance to degradation [7]. In treated seawage, EE2 is often the major compound with oestrogenic activity and its impact on aquatic wildlife has been deeply investigated (reviewed in [8]). In aquatic species, field studies and laboratory experiments show that exposure to EE2 affects fecundity, fertility, reproductive behaviour and induces intersex, endangering natural populations [9].

Potential long-term and persistent effects of early-life EDC exposure involve the stable alteration of gene expression. This is possible through epigenetic regulation, involving histone modifications, non-coding RNAs, and DNA methylation. The latter refers to the transfer of a methyl group from a methyl-donor, S-adenosyl-methionine (SAM), to a cytosine in the context of CpG dinucleotides, forming a methylcytosine [10,11]. This process is catalysed by two families of DNA methyltransferases (DNMTs): whereas DNMT3 enzymes are responsible for de novo methylation during development and differentiation, the DNMT1 family is maintaining methylation through each cell division by copying hemimethylated DNA, making DNA methylation a stable and heritable modification. On the other hand, the addition of a hydroxyl group to methylcytosine by ten-eleven translocation (Tet) enzymes is one of the most common mechanisms of active demethylation [12,13]. In mammals, most DNA methylation occurs in the symmetric CpG context, while only a small amount of non-CpG methylation is observed [10,11]. DNA methylation regulates gene expression by altering chromatin state and accessibility to CpG sites. For instance, methylation occurring at promoter regions can prevent the binding of transcription factors, resulting in gene silencing. Yet, the relationship between methylation and gene expression is more complex, as methylation also occurs at intragenic sites, enhancers or suppressor elements [13,14]. DNA methylation is dynamically regulated throughout life: during gametogenesis and embryogenesis, two waves of demethylation/remethylation deeply reprogram the methylome to produce a totipotent zygote [14–16]. Then, DNA methylation plays an important role throughout development as it directs cellular differentiation after reprogramming. During early development and juvenile stages, DNA methylation is thus thought to be particularly sensitive to environmental factors [17].

An increasing number of studies highlight the potential roles of DNA methylation in mediating long-term effects of sub-toxic developmental exposure to xenobiotics and roles in the aetiology of diseases including cancer, obesity, cardiovascular disease, diabetes, hypertension, and neurodegenerative disorders [18–22]. Indeed, DNA methylation could act as a long-term memory of past exposures mediating stable changes in gene expression [23,24]. In ecotoxicology, the role of epigenomics is receiving more and more attention to explain the long-term delayed and potential transgenerational effects of xenobiotics [25–27]. In particular, accumulating evidence indicates that hormones and endocrine disrupting compounds can alter the epigenome [21,28,29]. For instance, the regulation of DNMT transcription by ESR1 (an oestrogen receptor that acts as a transcription factor) may represent one of the possible mechanisms by which hormones influence methylation [30]. Other potential mechanisms may involve the reduction of SAM availability [31], the alteration of histone activity as a result of membrane receptor oestrogenic signalling [32,33], the expression of miRNAs, or direct interactions between oestrogen receptors and enzymes involved in the methylation machinery, such as thymine-DNA glycosylase (TDG) [34].

So far, relatively few studies have examined the effects of endocrine disruptors on DNA methylation in aquatic species. Using bisulphite conversion and pyrosequencing, Strömqvist et al. [35] have shown decreased methylation levels of three CpG sites located in the 5' flanking region of the vitellogenin I gene, a known biomarker of oestrogenic exposure, in the liver of male and female zebrafish, following a 14-day exposure to 100 ng/L EE2. Using Methylated DNA Immunoprecipitation (MeDIP) and high-throughput sequencing, followed by validation using bisulphite sequencing PCR (BSP) and RT-PCR, Mirbahai et al. [36] investigated the whole-genome methylation and gene expression in tumours in the liver of the Common dab (Limanda limanda), sampled in various sites in English rivers. Genes involved in pathways related to cancer, including apoptosis, wnt/β-catenin signalling and genomic and non-genomic oestrogen responses, were altered both in methylation and transcription. In that case the exact mixture of environmental contaminants causing the liver tumours was not identified but the molecular responses point towards oestrogenic disruption. On zebrafish early life stages, Falisse et al. [37] showed that exposure to the antibacterial agent and EDC triclosan modified the methylome after 7 days exposure, as well as the expression of related genes, using Reduced Representation Bisulphite Sequencing (RRBS).

The identification of environmentally-induced alterations in the epigenome ideally requires the use of individuals with no existing or very low genetic variability to rule out the confounding factors of the genotype on the observed phenotype [38]. Such a model species is provided by the mangrove rivulus, Kryptolebias marmoratus, one of the only known self-fertilizing vertebrates (the other one being its sister species, K. hermaphroditus). Although males and hermaphrodites are present in natural populations, most reproduction occurs by self-fertilization of hermaphrodites. Exclusive selfing of hermaphrodites during several generations results in highly homozygous and virtually ‘clonal’ lineages that can occur naturally or be bred in a laboratory setting [39,40]. A naturally clonal and sexually reproducing vertebrate species allows the study of epigenetic mechanisms ruling out the confounding factor of genetic variability, which is especially useful in ecotoxicology experiments in the sublethal range and/or dealing with delayed effects where non-dramatic molecular changes are expected. Little is known about the epigenetic mechanisms in K. marmoratus but it has been reported that the sex-ratio can be modulated by temperature-sensitive DNA methylation [41]. Recent studies described the DNA methylation reprogramming event during embryogenesis [42], as well as the time course expression of enzymes involved in epigenetics during its development [43,44]. Another study has shown that parasite load can modify DNA methylation [45], while these authors also found a higher differentiation of the methylation landscape between different genotypes than between environments [46].

In previous studies, we successfully used the mangrove rivulus to assess the delayed effects of BMAA (beta-N-Methylamino-L-alanine) [47] and 17-α-ethinylestradiol (EE2) [48,49] in adults after an early-life stage exposure. Hatchlings were exposed to 4 and 120 ng/L EE2 for 28 days, and allowed to recover for 140 additional days in clean water. The 28 days exposure has been chosen to include a 4 weeks period of exposure during early life stages that would be long enough to potentially induce endocrine effects and to be environmentally relevant. At 168 dph, all the fish are sexually mature. This design represents a ratio 1:5 between the exposure and the recovery periods. We reported delayed and persistent effects of EE2 on growth, reproduction and steroid levels as well as liver, ovotestis and brain molecular phenotypes assessed by label-free quantitative proteomics. Effects of EE2 were tissue and dose dependent, with most effects occurring at the environmentally relevant concentration (4 ng/L) in brain and liver, and at the higher concentration in the gonads. Among the three studied organs, the liver showed the most sensitive response to EE2. In this organ, EE2 affected known oestrogen-responsive pathways such as lipid, fatty acid and steroid metabolism, apolipoproteins, innate immune system and inflammation. Interestingly, several proteins involved in SAM metabolism were affected in the liver, providing further indications of the potential effect of EE2 on DNA methylation. The liver serves an essential and conserved role for metabolic homoeostasis in all vertebrates, and reproduction in the case of oviparous species. Several studies have shown that the liver is an important target of endocrine disruption in mammals [50] and fish [36,51–53]. Using an RRBS approach, the present study aimed at characterizing the hepatic methylation landscape in adults and investigating the genome-wide changes in DNA methylation in the liver of 168 dph adults that were exposed to EE2 as hatchlings to test the hypothesis of the potential involvement of DNA methylation in the delayed effects of EDCs.

Materials and methods

Experimental fish, EE2 exposure and sample collection

Individuals used in this study were from the same experiment reported in Voisin et al. [48], which also described the generation of breeding stock population and collection of eggs. Upon hatching, mangrove rivulus were transferred to 300 mL glass jars filled with 100 mL of 25 ppt reconstituted salt water (Instant Ocean™ sea salt), and the corresponding treatment: vehicle control (0.000012% ethanol), 4 ng/L and 120 ng/L 17-α-ethinylestradiol (EE2) (Sigma-Aldrich E4876-1 G). Fish were raised at a temperature of 26°C and a 12:12 h photoperiod. Detailed preparation of EE2 solutions and the measurement of actual exposure concentrations by ELISA are reported in Voisin et al. [48]. Average measured concentration of the nominal 120 ng/L exposure solution was 132.3 ± 14.7 ng/L. Since 4 ng/L is situated below the detection limit (20 ng/L), we refer to the measure of the stock solution, which was 10.15 ± 0.15 µg/L for the nominal 10 µg/L stock solution. Throughout the paper, we will refer to the nominal concentrations of 4 and 120 ng/L. Fish were exposed for a period of 28 days with a 100% water renewal 3 times a week. At 28 dph, fish were transferred to individual 1.2 L plastic containers filled with 400 mL of 25 ppt clean salt water and reared until 168 dph. Rivulus were fed 1 mL of concentrated newly hatched brine shrimp (Artemia nauplii) until 56 dph, 2 mL starting at 56 dph and 4 mL starting at 91 dph. At 168 dph, fish were sacrificed by a rapid transfer to 4°C 25 ppt salt water and sectioning the spinal cord. For each experimental condition, livers from five individuals were dissected out at 168 dph, snap-frozen in liquid nitrogen and stored at −80°C until DNA extraction. All rivulus husbandry and experimental procedures were performed in accordance with the Belgian animal protection standards and were approved by the University of Namur Local Research Ethics Committee (UN KE14/230). The agreement number of the laboratory for fish experiments is the LA1900048.

DNA extraction, RRBS library preparation and post-processing

Genomic DNA from individual livers was extracted using the NucleoSpin Tissue XS kit (Macherey-Nagel, Germany) following the manufacturer’s protocol. DNA concentration and quality were assessed using the NanoDrop 8000 spectrophotometer (Thermo Scientific, Waltham, MA) and 1% agarose gel electrophoresis. A total of 15 RRBS libraries (5 for each experimental condition) were prepared following established protocols [54,55]. In brief, genomic DNA was digested with MSPI (New England Biolabs, Ipswich, MA) followed by end repair, addition of 3′ A overhangs and addition of methylated adapters (Illumina, San Diego, CA) to the digested fragments. Following adapter ligation, size selection was performed by cutting 40–220 bp DNA fragments (pre-ligation size) from a 3% (w/v) NuSieve GTG agarose gel (Lonza, Basel, Switzerland). Subsequently, libraries were bisulphite converted using the EZ DNA methylation kit (Zymo Research, Irvine, CA) with an extended incubation time of 18–20 h. Bisulphite converted libraries were amplified by PCR and sequenced (100-bp single ended reads) by New Zealand Genomics Limited (University of Otago, New Zealand) on an Illumina HiSeq2000 sequencer. Sequences were obtained in FASTQ format.

The quality of sequenced reads was performed using the FastQC application (Babraham Institute, Cambridge, UK). Adaptor sequences were removed from the reads using the cleanadaptors program of the DMAP package [56]. Single-ended bisulphite reads were aligned against the Kryptolebias marmoratus assembly (GCF_001649575.1_ASM164957v1) using the Bismark software v0.17.0 (default mode: alignments to complementary strands were ignored – i.e., not performed) [57]. Proximal genes and genomic positions of the fragments were identified using the identgeneloc programme from the DMAP package [58]. Fragments were linked to the nearest protein-coding gene and the genomic position was attributed as follows: internal (within the gene body), promoter (0–5 kb upstream of the transcription start site), upstream (5–10 kb upstream of the TSS) and intergenic (>10 kb). The annotation of intergenic fragments or fragments located upstream of genes (5–10 kb) and in intergenic regions (>10 kb) is provided as an indication and should be considered with caution as methylation of these regions may not have any influence on the corresponding gene expression. Fragments with a methylation level ≥ 80% are qualified as highly methylated, while those with a methylation level ≤ 20% are qualified as lowly methylated.

Analysis of differential DNA methylation

Base resolution differences in DNA methylation following developmental exposure to EE2 were assessed based on MSPI fragments (40–220 bp range) as a unit of analysis. To filter the fragments suitable for differential methylation analysis, we selected the fragments with ≥ 2 CpG sites, ≥10 reads per fragment (CpG10) and present in at least three of five individuals for each experimental condition. These fragments are referred to as ‘analysed fragments’. The 102,018 analysed fragments form our reduced representative genome (RR genome) and corresponds to a total of 14,679 unique RefSeq identifiers, as more than one fragment often maps to a same gene. Differential methylation analysis was performed as previously described using the diffmeth program from DMAP [56,58] between each experimental group, resulting in three group comparisons: Ctl versus 4 ng/L EE2 exposed individuals, Ctl versus 120 ng/L and 4 versus 120 ng/L. Determination of significantly differentially methylated fragments (DMFs) was based on ≥ 10% differential methylation between the groups and P-value <0.01, and are qualified as hyper- or hypomethylated fragments.

Network analysis

RefSeq identifiers were retrieved using BLAST for all analysed DMFs. To establish a reference gene list for the network analysis, a list including all DMFs was filtered to include only gene body and promoter genomic locations and contained 68,762 fragments mapped to rivulus RefSeq and human RefSeq identifiers. As more than one fragment often mapped to a same gene, this list corresponded to 10,440 unique human RefSeq identifiers. The list, containing P-values and methylation difference for each of the three treatment comparisons was uploaded in the Ingenuity Pathway Analysis software (IPA, QIAGEN, www.qiagen.com/ingenuity). To gain an overall view of the pathways and functions associated with promoter and gene body DMFs, as well as the connections between differentially abundant proteins, we performed a network analysis for each group comparison.

Results

Characterization of the hepatic DNA methylation landscape in K. marmoratus

Using RRBS, we generated 15 K. marmoratus liver methylomes with an average of 29.5 million sequenced reads (between 19.7 and 40 million reads) (Table 1, Supplementary Table 1). Unique alignment efficiency of these libraries ranged from 40.3% to 46.8% (except the library 23–4 with 28.5%, which was discarded in further analyses), with an average of 44.4%. This corresponds to a total of 102,018 fragments suited for the analysis after MSPI digestion of genomic DNA and 40–220 bp size-selection (Figure 1a). The total size of the RR genome was 9.37 Mb, which accounted for 1.4% of the published full genome (GCF_001649575.1_ASM164957v1) [59]. Within this RR genome, we counted 701,862 CpG dinucleotides, representing 6.1% of the 11.5 million CpG in the reference genome. Consequently, the enrichment factor in CpGs of the RR genome was 4.3 fold. Of all the analysed fragments, the majority was located within gene bodies (47%), followed by intergenic regions (30%), promoter (16%) and upstream regions (7%) (Figure 1b). The distribution of CpGs followed exactly the same pattern as the fragments (data not shown). We consider this dataset of about 6% of all genomic CpGs, albeit not uniformly distributed across the genome, as a representative sample that permits the investigation of changes in DNA methylation in liver due to exposure to toxicants. Global CpG methylation levels in liver tissue ranged from 63.5% to 66.4% in all samples. On the contrary, methylation level of non-CpG cytosines remained low between 1.7 and 2.4%. It should be noted that non-CpG methylation calculated here included the true genomic non-CpG methylation and the CpGs that failed to convert during bisulphite treatment.

Table 1.

Characteristics of the reduced representative genome and comparisons with the genome

Genome
Size in scaffolds (Mb)	680
GC content	37%
Number of CpGs (millions)	11.5
Reduced genome
Size (Mb)	9.37
Number of reads (min-max; millions)	19.7–40
Uniquely mapped reads (min-max; millions)	8.9–17.8
Mapping efficiency (min-max)	40.3–47.1%
Number of analysed fragments	102 018
% of analysed genome	1.4
Number of CpGs	701,862
% of genomic CpGs	6.1
CpG enrichment factor	4.3
% CpG methylation (min-max)	63.5–66.4%
% non-CpG methylation	1.7–2.4%
% of hypomethylated fragments	20
% hypermethylated fragments	67

Figure 1.

Hepatic DNA methylation landscape in K. marmoratus . (a) Frequency distribution of MSPI fragment size. Mean fragment size was 95 bp and median 88 bp; (b) Overall distribution of analysed fragments in genomic elements; (c:)) Frequency distribution of DNA methylation % for MSPI fragments. Distribution of MSPI fragments follows a bimodal distribution, which is expected given the binary character of DNA methylation in cells; (d) Boxplot of DNA methylation (%) of fragments according to their location in genomic elements (including medians and quantile 25–75%).

For the following description, we took only the control group into consideration, in order to avoid any possible effect of the EE2 treatment. The average DNA methylation level of most MSPI fragments in mangrove rivulus were above 80% (67% of all fragments – considered as highly methylated) or below 20% of methylation (20% of all fragments – considered as lowly methylated) (Figure 1c). The level of DNA methylation was the lowest in fragments located in promoter regions (mean = 42.3%; median = 12.8%) compared to upstream (mean = 66.2%; median = 88.4%), intergenic (mean = 72.6%; median = 89.1%) and gene body (mean = 78.0%; median = 91.5%). Figure 1d depicts the distribution and spread (25–75 quantiles) of DNA methylation level for fragments within each region. We can observe that the methylation range was the highest for fragments located in the promoter region, showing that even if the majority of these fragments were lowly methylated (52%), a considerable part was highly methylated (38%). Despite the fact that fragments in promoter regions represented only 16% of the total analysed fragments, they accounted for 42% of all lowly methylated fragments (Figure 2a), compared to fragments located in gene bodies (27%), intergenic (23%) and upstream (8%) regions. In contrast, only 9% of all highly methylated fragments were located in the promoter regions, compared to 53% in gene bodies, 31% in intergenic regions and 7% in upstream regions (Figure 2b). Supplementary Figure 1 depicts the coefficient of variability of the methylation level within each genomic region, for lowly methylated fragments on the one hand, and for highly methylated fragments on the other hand. Highly methylated fragments presented a low level of variability (< 3%), in opposition to lowly methylated fragments for which the coefficient of variability was higher than 33%. This pattern was the same for all the fragments, independently of the genomic region.

Figure 2.

Distribution of lowly methylated (<20% methylation) (a) and highly methylated (> 80% methylation) (b) fragments among the genomic regions (gene body, promoter, upstream, intergenic) of the liver reduced representative genome of the mangrove rivulus.

Changes in DNA methylation following developmental exposure to EE2

Global DNA methylation levels of rivulus liver averaged 65.1 ± 1.0% in Ctl samples, 66.0 ± 0.2% in 4 ng/L and 65.5 ± 0.6% in 120 ng/L exposed individuals and were not significantly different (Anova on arcsin squareroot transformed values). Compared to the Ctl group, 58 DMFs were reported for the low EE2 concentration of 4 ng/L, 33 for the higher concentration of 120 ng/L, while 62 DMFs were significant between 120 and 4 ng/L groups. The majority of DMFs had a higher methylation level at 4 ng/L compared to Ctl and 120 ng/L groups, and at 120 ng/L compared to Ctl (Table 2, Figure 3a). Figure 3b depicts the DMF location within the RR genome and showed a similar pattern as the general distribution of the analysed fragments (Figure 1b) with the DMFs located in the order gene body > intergenic > promoter > upstream regions. None of the significant DMF was common to all three group comparisons (Figure 3c). However, 2 DMFs were significant at both EE2 treatments compared to Ctl: one located in an intergenic region, in proximity of the liprin-alpha-4 (PPFIA4) gene, and one located in the gene body of the ATP-binding cassette sub-family A member 1 gene (ABCA1) (Figure 3d). In addition, there were three common DMFs to 4 ng/L vs Ctl and 4 ng/L vs 120 ng/L comparisons, of which two were found within gene bodies: calcium/calmodulin-dependent protein kinase type 1D (CAMK1D), testis-specific serine/threonine-protein kinase (TSSK1B) and one in the intergenic region in proximity of the lymphoid-restricted membrane protein-like gene (LRMP). Finally, one fragment was significantly differentially methylated in both 120 vs Ctl and 120 vs 4 ng/L, located in the intergenic region in proximity to the PTK2B (protein-tyrosine kinase 2) gene. Tables 3–5 report the DMFs for the three comparisons: 4 ng/L vs Ctl, 120 ng/L vs Ctl, and 4 ng/L vs 120 ng/L, respectively. Among all group comparisons, the maximum effect size (difference in mean methylation) observed for hypermethylation was 21.9% (promoter region of nipped-B-like protein B, NIPBL) and −19.0% for hypomethylation (intergenic region, eukaryotic translation initiation factor 4 gamma 3 EIF4G3). Both DMFs were significant between 4 ng/L and Ctl groups. It is also important to report the fact that very few DMFs were classified as lowly methylated fragments. When comparing exposed groups to Ctl, only NIPBL was lowly methylated in Ctl at 19.6% and increased to 41.5% after exposure to 4 ng/L EE2.

Table 2.

Number of analysed fragments, analysed CpGs and number of DMFs in each group comparison

	4 ng/L vs Ctl	120 ng/L vs Ctl	120 vs 4 ng/L
Fragments with P < 0.01	1,674 (1.6%)	787 (0.75%)	1,471 (1.4%)
Fragments with ≥ 10%	383 (0.36%)	311 (0.30%)	471 (0.46%)
Differentially methylated fragments (DMFs; P < 0.01 and methylation difference ≥ 10%)**	58 (0.06%) (46↑, 12↓)	33 (0.03%) (20↑, 13↓)	62 (0.06%) (13↑,49↓)
Gene body DMFs	26 (22↑, 4↓)	12 (9↑, 3↓)	31 (6↑, 25↓)
Promoter (>5kb) DMFs	6 (4↑, 2↓)	4 (3↑, 1↓)	11(3↑, 8↓)
Upstream (5–10 kb) DMFs	3↑	4 (2↑, 2↓)	3↓
Intergenic (>10 kb) DMFs	23 (17↑, 6↓)	13 (6↑, 7↓)	17 (4↑, 13↓)

Figure 3.

(a) Volcano plot representations of significance and differential methylation in analysed fragments. Dotted lines represent thresholds of significance: −10LogP20, corresponding to P0.01 and mean methylation difference 10% (a) 4 ng/L vs control; B:120 ng/L vs control; C: 4 ng/L vs 120 ng/L. (b) Overall distribution of DMFs in genomic elements; (c) Venn Diagram showing specific and common DMFs between group comparisons; (d) Methylation levels (%) of common fragments between 4 vs Ctl and 120 vs Ctl comparisons (Da,Db), between 4 vs Ctl and 4 vs 120 comparisons (c–e) and between 120 vs Ctl and 4 vs 120 ng/L comparisons (f). For clearer representation of methylation differences, y axes start at 40%.

Table 3.

Differentially methylated fragments (DMFs) between 4 ng/L EE2 and control individuals. Methylation difference represents the mean methylation at 4 ng/L minus the mean methylation in controls. Fragments discussed in the text are marked with *

GeneID	Len	CpGs	Description	Human gene symbol	Ctl (% Meth.)	4 ng/L (% Meth.)	Meth. Difference (%)	P-value	Test
Gene body
XP_017282219.1	126	4	Protein kinase C-binding protein NELL1	NELL1	62.6	79.4	16.8	3.20E-03	F(1.8) = 17.24
XP_017293079.1	186	11	Sphingosine 1-phosphate receptor 1	S1PR1	59.4	74.9	15.5	9.64E-03	F(1.6) = 13.98
XP_017282327.1	144	8	Drebrin	DBN1	72.2	86.9	14.7	5.12E-03	F(1.8) = 14.56
XP_017297110.1	139	7	Protein phosphatase 1 regulatory subunit 12 C-like	PPP1R12C	51.8	66.0	14.3	1.79E-03	F(1.7) = 23.81
XP_017263166.1	57	2	Astrotactin-2	ASTN2	23.2	36.7	13.5	1.38E-04	F(1.8) = 46.24
XP_017269942.1	112	6	ATP-binding cassette sub-family A member 1-like	ABCA1	70.5	83.0	12.4	5.99E-03	F(1.8) = 13.73	*
XP_017273856.1	72	2	Rhomboid-related protein 1	RHBDL1	71.5	83.5	12.0	5.64E-03	F(1.5) = 21.51
XP_017287079.1	95	2	AT-rich interactive domain-containing protein 2	ARID2	53.3	65.3	12.0	3.68E-03	F(1.4) = 37.07
XP_017290801.1	47	2	Nuclear receptor corepressor 2	NCOR2	53.9	65.4	11.5	3.41E-03	F(1.8) = 16.86
XP_017259611.1	127	8	Protein CASC3	CASC3	79.0	90.5	11.5	4.87E-03	F(1.6) = 18.84
XP_017284950.1	143	6	Zinc finger protein 385 C	ZNF385C	81.1	92.5	11.4	5.43E-04	F(1.8) = 30.77
XP_017290159.1	154	5	Testis-specific serine/threonine-protein kinase 1-like	TSSK1B	79.0	90.4	11.4	3.12E-03	F(1.7) = 19.44	*
XP_017272906.1	101	5	Calcipressin-3	RCAN3	71.5	82.8	11.3	9.95E-06	F(1.8) = 95.83
XP_017269734.1	152	12	Neogenin	NEO1	79.8	90.8	11.1	6.99E-03	F(1.8) = 12.95
XP_017263758.1	72	6	Disks large homolog 3	DLG3	71.4	82.2	10.8	1.39E-03	F(1.8) = 22.84
XP_017270265.1	98	7	Round spermatid basic protein 1-like protein	RSBN1	65.6	76.3	10.7	1.37E-03	F(1.8) = 22.96
XP_017280982.1	184	13	Whirlin	WHRN	74.4	85.1	10.7	8.98E-03	F(1.4) = 22.56
XP_017294972.1	126	5	Friend leukaemia integration 1 transcription factor-like	FLI1	63.3	73.8	10.5	8.18E-03	F(1.8) = 12.19
XP_017274567.1	173	6	Matrix metalloproteinase-15-like	MMP15	82.8	93.1	10.3	3.53E-03	F(1.6) = 21.56
XP_017269638.1	153	8	Late secretory pathway protein AVL9 homolog	AVL9	61.6	71.9	10.3	9.28E-03	F(1.6) = 14.22
XP_017261650.1	148	13	Calcium/calmodulin-dependent protein kinase type 1D-like	CAMK1D	79.5	89.8	10.3	6.41E-04	F(1.8) = 29.22	*
XP_017288959.1	106	4	Neurotensin receptor type 1	NTSR1	80.6	90.7	10.2	1.07E-04	F(1.8) = 49.78
XP_017273249.1	136	9	Lethal [3]malignant brain tumour-like protein 1	L3MBTL1	87.8	77.6	−10.2	3.30E-03	F(1.6) = 22.17
XP_017275235.1	161	14	PH and SEC7 domain-containing protein 2	PSD2	37.6	26.7	−10.9	8.71E-03	F(1.8) = 11.89
XP_017288462.1	144	5	Adhesion G-protein coupled receptor D2	ADGRL2	92.5	81.3	−11.2	4.30E-03	F(1.7) = 17.22
XP_017293232.1	169	8	Amyloid beta A4 protein	APP	38.5	27.2	−11.3	4.05E-03	F(1.8) = 15.85
Promoter					0.0	0.0
XP_017263260.1	53	3	Nipped-B-like protein B	NIPBL	19.6	41.6	21.9	8.51E-03	F(1.8) = 12.00	*
XP_017266946.1	172	10	G1/S-specific cyclin-D2-like	CCND2	53.5	72.5	19.0	3.66E-03	F(1.6) = 21.24	*
XP_017267503.1	176	13	Translation initiation factor IF-3, mitochondrial	MTIF3	73.3	85.2	11.9	4.16E-03	F(1.7) = 17.43
XP_017259959.1	165	9	Protein SCAI	SCAI	83.6	94.2	10.6	9.07E-03	F(1.6) = 14.36
XP_017283065.1	126	4	Tyrosine-protein kinase STYK1-like	STYK1	57.7	46.4	−11.3	4.51E-03	F(1.8) = 15.25
XP_017283302.1	138	4	Natterin-3-like		83.0	66.3	−16.7	1.22E-03	F(1.6) = 32.87
Upstream					0.0	0.0
XP_017279598.1	153	5	Homeobox protein CDX-1	CDX1	51.5	68.9	17.4	4.36E-03	F(1.7) = 17.12
XP_017280347.1	97	7	Peroxisome proliferator-activated receptor alpha	PPARA	52.5	64.7	12.1	2.58E-03	F(1.8) = 18.58
XP_017295153.1	154	5	Cortactin-binding protein 2-like	CTTNBP2	83.3	94.7	11.5	3.35E-03	F(1.6) = 22.04
Intergenic					0.0	0.0
XP_017271697.1	116	3	Sorting nexin-29	SNX29	66.2	83.1	16.9	2.67E-03	F(1.6) = 24.16
XP_017286782.1	132	4	Dedicator of cytokinesis protein 8	DOCK8	57.3	73.4	16.2	9.74E-03	F(1.8) = 11.38
XP_017275594.1	95	4	Ephrin-A1-like	EFNA1	54.0	69.8	15.8	3.34E-03	F(1.8) = 16.98
XP_017270106.1	162	7	Armadillo repeat protein deleted in velo-cardio-facial syndrome	ARVCF	73.4	88.7	15.3	6.87E-03	F(1.6) = 16.26
XP_017275116.1	105	4	Teashirt homolog 3	TSHZ3	35.0	49.8	14.8	7.20E-03	F(1.7) = 14.03
XP_017275238.1	182	8	TGF-beta receptor type-2-like	TGFBR2	46.1	60.7	14.7	7.26E-06	F(1.6) = 205.01
XP_017272930.1	98	2	Uncharacterized protein C4orf45 homolog		60.7	75.0	14.4	7.28E-03	F(1.7) = 13.97
XP_017268339.1	172	6	Zinc finger protein 511	ZNF511	68.3	80.9	12.6	6.16E-03	F(1.8) = 13.59
XP_017272056.1	139	11	Protein Wiz-like	WIZ	81.7	94.0	12.4	2.63E-03	F(1.7) = 20.72
XP_017271447.1	135	4	Liprin-alpha-4	PPFIA4	78.6	90.8	12.2	3.28E-03	F(1.8) = 17.10
XP_017285408.1	52	4	Lymphoid-restricted membrane protein-like	LRMP	75.7	87.7	11.9	1.83E-04	F(1.8) = 42.61
XP_017272912.1	179	9	Protocadherin Fat 4	FAT4	84.6	95.7	11.1	3.98E-03	F(1.5) = 25.36
XP_017263052.1	157	6	Protein C9orf69 homolog	TMEM250	70.9	81.9	10.9	8.97E-04	F(1.8) = 26.31
XP_017287805.1	122	4	Collagen alpha-1(XI) chain	COL11A1	67.4	78.1	10.8	8.22E-03	F(1.8) = 12.17
XP_017279615.1	103	4	Transmembrane protein 126A-like	TMEM126A	60.1	70.5	10.4	7.63E-03	F(1.8) = 12.52
XP_017286707.1	100	3	Homeobox protein SIX3	SIX3	34.8	45.1	10.3	6.88E-03	F(1.7) = 14.30
XP_017272585.1	118	13	Transmembrane protein 132 C	TMEM132C	43.4	53.5	10.2	7.24E-03	F(1.8) = 12.78
XP_017293370.1	155	10	Lethal(2) giant larvae protein homolog 2	LLGL1	90.4	80.3	−10.2	6.85E-03	F(1.8) = 13.06
XP_017288632.1	176	7	Large neutral amino acids transporter small subunit 2-like	SLC7A8	89.8	78.4	−11.4	1.80E-03	F(1.7) = 23.80
XP_017268028.1	127	4	Protein FAM222B	FAM222B	41.6	29.1	−12.5	1.47E-04	F(1.8) = 45.41
XP_017269623.1	166	17	Cadherin-20-like	CDH20	42.7	29.6	−13.0	9.02E-03	F(1.7) = 12.79
XP_017270388.1	147	8	Protein Wnt-5b	WNT5B	56.0	39.9	−16.1	5.14E-03	F(1.7) = 16.05
XP_017264233.1	167	3	Eukaryotic translation initiation factor 4 gamma 3	EIF4G3	61.4	42.4	−19.0	5.78E-03	F(1.8) = 13.92

Table 4.

Differentially methylated fragments (DMFs) between 120 ng/L EE2 and control individuals. Methylation difference represents the mean methylation at 120 ng/L minus the mean methylation in controls. Fragments discussed in the text are marked with *

GeneID	Len	CpGs	Description	Human gene symbol	Ctl (% Meth.)	120 ng/L (% Meth.)	Meth. Difference (%)	P-value	Test
Gene body
XP_017286106.1	136	4	Mitogen-activated protein kinase kinase kinase kinase 3 0	MAP3K3	74.2	87.5	13.3	2.34E-03	F(1.8) = 19.20
XP_017285971.1	74	2	Nuclear factor 1 X-type-like	NFIX	41.8	55.2	13.3	3.76E-03	F(1.8) = 16.29
XP_017259709.1	123	6	LMBR1 domain-containing protein 2-B-like	LMBRD2	79.7	92.0	12.3	9.18E-03	F(1.7) = 12.70
XP_017291042.1	172	15	Neuronal PAS domain-containing protein 2-like	NPAS2	72.9	85.0	12.1	3.67E-04	F(1.8) = 34.66
XP_017286691.1	65	2	A disintegrin and metalloproteinase with thrombospondin motifs 6	ADAMTS6	79.0	91.0	12.0	6.47E-03	F(1.8) = 13.34
XP_017287636.1	158	7	Leucine-rich PPR motif-containing protein, mitochondrial	LRPPRC	81.8	93.7	11.9	6.48E-03	F(1.6) = 16.68
XP_017259612.1	145	6	Potassium Voltage-gated channel subfamily H member 5-like	KCNH5	79.2	90.7	11.5	5.41E-03	F(1.7) = 15.75
XP_017284196.1	159	8	AT-rich interactive domain-containing protein 3B-like	ARID3B	66.9	77.7	10.8	2.05E-03	F(1.8) = 20.08
XP_017269942.1	112	6	ATP-binding cassette sub-family A member 1-like	ABCA1	70.5	81.1	10.6	5.20E-03	F(1.8) = 14.48	*
XP_017282238.1	72	2	Neuron navigator 2	NAV2	86.7	76.3	−10.4	1.43E-03	F(1.7) = 25.80
XP_017275214.1	174	12	Teneurin-1-like	TENM1	86.5	75.3	−11.2	5.48E-03	F(1.7) = 15.66	*
XP_017280416.1	108	5	Ephrin-A3-like	EFNA3	73.9	62.4	−11.4	2.53E-03	F(1.8) = 18.70
Promoter
XP_017296617.1	106	5	LIM domain Transcription factor LMO4.1-like	LMO4	72.2	84.2	12.0	4.29E-03	F(1.8) = 15.53
XP_017280337.1	48	2	Protein NDRG1-like	NDRG1	29.2	41.1	11.9	3.72E-03	F(1.7) = 18.19	*
XP_017273723.1	137	3	Solute carrier family 22 member 6-like	SLC22A6	85.0	96.5	11.5	9.45E-03	F(1.6) = 14.10
XP_017278883.1	158	9	Collectrin	CLTRN	73.2	59.0	−14.2	3.86E-03	F(1.5) = 25.71
Upstream
XP_017276084.1	156	4	Protein FAM102B	FAM102B	84.5	97.0	12.6	1.33E-03	F(1.6) = 31.80
XP_017266436.1	65	2	Supervillin-like	SVIL	79.8	89.9	10.1	4.29E-03	F(1.7) = 17.23
XP_017277864.1	46	2	Receptor-interacting serine/threonine-protein kinase 1	RIPK1	80.5	67.0	−13.6	6.50E-04	F(1.8) = 29.11
XP_017261155.1	151	7	Golgin subfamily A member 6-like protein 2	GOLGA6L2	73.8	58.8	−15.0	2.85E-03	F(1.7) = 20.11
Intergenic
XP_017279686.1	122	5	Ras-GEF domain-containing family member 1 C	RASGEF1B	72.8	85.5	12.7	9.78E-03	F(1.7) = 12.36
XP_017277156.1	163	11	Coiled-coil domain-containing protein 85A	CCDC85A	81.3	93.9	12.7	1.20E-03	F(1.5) = 43.51
XP_017271447.1	135	4	Liprin-alpha-4	PPFIA4	78.6	89.7	11.2	5.86E-03	F(1.7) = 15.25
XP_017288708.1	170	5	Protein turtle homolog B-like	IGSF9B	78.0	88.3	10.2	8.82E-03	F(1.7) = 12.91
XP_017297201.1	96	3	Williams-Beuren syndrome chromosomal region 27 protein-like		69.3	79.5	10.2	1.17E-03	F(1.8) = 24.14
XP_017263439.1	143	4	Rho GTPase-activating protein 24	ARHGAP24	79.6	89.8	10.2	3.37E-04	F(1.7) = 42.13
XP_017282790.1	124	10	Protein-tyrosine kinase 2-beta	PTK2B	60.3	49.6	−10.7	7.58E-03	F(1.8) = 12.56
XP_017292488.1	165	11	Fibrinogen alpha chain	FGA	56.9	45.8	−11.1	8.99E-03	F(1.6) = 14.42	*
XP_017281773.1	112	2	Alpha-crystallin A chain-like	CRYAA	75.2	62.5	−12.7	6.19E-03	F(1.7) = 14.92
XP_017270131.1	151	10	Semaphorin-4 C	SEMA4C	62.6	48.3	−14.3	1.33E-03	F(1.6) = 31.80
XP_017283010.1	154	6	General transcription factor II–I repeat domain-containing protein 2-like	GTF2IRD2	53.2	37.7	−15.5	4.35E-03	F(1.8) = 15.45
XP_017276780.1	127	4	THAP domain-containing protein 10	THAP10	93.8	77.4	−16.4	6.03E-04	F(1.6) = 42.98
XP_017292195.1	140	6	Cullin-5	CUL5	88.4	70.1	−18.3	9.85E-03	F(1.6) = 13.84

Table 5.

Differentially methylated fragments (DMFs) between 120 ng/L EE2 and 4 ng/L EE2 exposed individuals. Methylation difference represents the mean methylation at 120 ng/L minus the mean methylation at 4 ng/L. Fragments discussed in the text are marked with *

GeneID	Len	CpGs	Description	Human gene symbol	4 ng/L Meth (%)	120 ng/L Meth (%)	Diffmeth(%)	Pr	Test
Gene body
XP_017288867.1	154	13	Scavenger receptor class F member 1	SCARF1	21.9	35.2	13.3	8.77E-03	F(1.8) = 11.86
XP_017272657.1	145	13	Dihydropyrimidinase-related protein 2-like	DPYSL2	22.9	34.9	12.0	6.20E-03	F(1.8) = 13.55
XP_017280995.1	45	2	Glutamate receptor ionotropic, NMDA 1-like	GRIN2D	58.2	70.0	11.8	7.75E-03	F(1.8) = 12.45
XP_017271462.1	61	3	Selenocysteine-specific elongation factor	EEFSEC	29.4	41.2	11.8	3.18E-03	F(1.8) = 17.28
XP_017286963.1	148	8	G patch domain and ankyrin repeat-containing protein 1	ANKRD1	84.4	95.8	11.4	9.38E-03	F(1.6) = 14.15
XP_017289196.1	101	7	Pre-B-cell leukaemia transcription factor 1	PBX1	40.2	50.8	10.6	8.05E-03	F(1.8) = 12.27
XP_017290510.1	51	4	CREB-regulated transcription coactivator 3	CRTC3	45.2	35.1	−10.0	3.68E-03	F(1.8) = 16.40
XP_017285590.1	138	4	Fibroblast growth factor 14-like	FGF14	85.6	75.5	−10.1	6.02E-04	F(1.7) = 34.77
XP_017261320.1	140	9	Tight junction protein ZO-1-like	TJP1	95.8	85.6	−10.2	1.65E-03	F(1.4) = 57.07
XP_017295247.1	155	11	Protein AF-17-like	MLLT6	92.7	82.4	−10.3	9.40E-03	F(1.5) = 16.77
XP_017288897.1	108	5	Methyltransferase-like protein 16	METTL16	69.2	58.8	−10.4	6.98E-03	F(1.8) = 12.96
XP_017283711.1	156	15	Large neutral amino acids transporter small subunit 3	SLC43A1	92.5	81.9	−10.6	1.29E-03	F(1.7) = 26.73
XP_017261650.1	148	13	Calcium/calmodulin-dependent protein kinase type 1D-like	CAMK1D	89.8	79.1	−10.7	6.59E-03	F(1.8) = 13.25	*
XP_017261484.1	88	4	Zinc finger protein AEBP2	AEBP2	95.2	84.2	−11.1	3.41E-03	F(1.8) = 16.86
XP_017290127.1	142	7	Phospholysine phosphohistidine inorganic pyrophosphate phosphatase	LHPP	95.1	83.9	−11.2	2.65E-03	F(1.5) = 30.59
XP_017286486.1	60	4	Rho guanine nucleotide exchange factor 17	ARHGEF17	41.4	29.6	−11.8	4.21E-03	F(1.8) = 15.64
XP_017269775.1	137	5	C-Jun-amino-terminal kinase-interacting protein 1	MAPK8IP1	83.3	71.5	−11.9	7.69E-03	F(1.7) = 13.66
XP_017283446.1	101	3	Multiple C2 and transmembrane domain-containing protein 1	MCTP1	87.0	75.0	−12.0	2.07E-03	F(1.8) = 20.03
XP_017291190.1	148	9	Nuclear factor 1 X-type-like	NFIX	89.8	77.5	−12.3	8.63E-03	F(1.6) = 14.69
XP_017296764.1	142	6	Serine-rich coiled-coil domain-containing protein 2-like	CCSER2	91.8	79.4	−12.4	1.78E-03	F(1.6) = 28.40
XP_017284506.1	137	5	Neutral amino acid transporter B(0)	SLC1A5	19.2	6.8	−12.4	5.05E-03	F(1.8) = 14.64
XP_017272705.1	90	3	G-protein coupled receptor 98	GPR98	90.8	78.3	−12.5	2.84E-03	F(1.6) = 23.57
XP_017263193.1	131	3	Regulator of G-protein signalling 3	RGS3	88.3	75.8	−12.6	8.92E-03	F(1.6) = 14.48
XP_017295850.1	52	6	Cathepsin E-like	CTSE	75.2	62.2	−13.0	2.30E-03	F(1.8) = 19.32
XP_017279844.1	102	7	Rab11 family-interacting protein 2	RAB11FIP2	91.7	77.4	−14.3	1.18E-05	F(1.8) = 91.50
XP_017289883.1	140	5	Partitioning defective 3 homolog B	PARD3B	80.4	65.9	−14.5	8.65E-03	F(1.6) = 14.68
XP_017290159.1	154	5	Testis-specific serine/threonine-protein kinase 1-like	TSSK1B	90.4	75.9	−14.6	5.69E-03	F(1.7) = 15.43	*
XP_017279357.1	128	3	Ephrin type-A receptor 6	EPHA6	77.8	62.7	−15.1	5.71E-03	F(1.8) = 13.98
XP_017262478.1	135	4	Unconventional myosin-IXb	MYO9B	66.6	51.3	−15.3	6.94E-03	F(1.8) = 12.99
XP_017265799.1	136	5	Serine/threonine-protein phosphatase 2A regulatory subunit B’’ subunit beta-like	PPP2R2B	38.1	22.6	−15.4	5.97E-03	F(1.7) = 15.14
XP_017268748.1	131	7	Zinc finger protein ZFPM2	ZFPM2	68.2	52.7	−15.5	1.51E-03	F(1.8) = 22.22
XP_017285362.1	44	2	Death-associated protein 1	DAP	80.3	62.0	−18.4	1.59E-03	F(1.6) = 29.70	*
Promoter
XP_017274091.1	81	3	Inactive rhomboid protein 1-like	RHBDF1	64.2	78.7	14.5	7.55E-03	F(1.6) = 15.59
XP_017294893.1	50	2	Ribosomal RNA processing protein 36 homolog	RRP36	50.7	62.8	12.1	8.34E-03	F(1.8) = 12.10
XP_017292323.1	84	6	E3 ubiquitin/ISG15 ligase TRIM25-like	ZFHX3	19.0	29.0	10.0	3.02E-03	F(1.8) = 17.59
XP_017296523.1	125	6	Synaptotagmin-1-like	SYTL1	86.0	75.7	−10.3	3.01E-03	F(1.8) = 17.62
XP_017290352.1	136	9	Synaptotagmin-14	SYT14	81.5	70.7	−10.8	7.25E-03	F(1.8) = 12.78
XP_017272703.1	99	3	Transcriptional activator MN1	MN1	82.5	71.6	−10.9	8.16E-03	F(1.8) = 12.20
XP_017294308.1	158	4	Uncharacterized protein LOC108249428	TCR alpha	89.1	77.3	−11.9	4.12E-03	F(1.5) = 24.94
XP_017280009.1	157	8	Collagen alpha-2(VI) chain-like	COL6A2	85.8	70.4	−15.4	2.67E-04	F(1.7) = 45.45
XP_017261271.1	125	4	Repulsive guidance molecule A-like	RGMA	88.9	73.0	−15.9	4.05E-03	F(1.7) = 17.62
XP_017285485.1	168	12	Synaptotagmin-12	SYT12	75.0	57.4	−17.6	1.68E-03	F(1.7) = 24.38
XP_017294454.1	168	15	Zinc finger and BTB domain-containing protein 4-like	ZBTB4	65.3	45.2	−20.2	8.65E-03	F(1.6) = 14.68
Upstream
XP_017284506.1	126	8	Neutral amino acid transporter B(0)	SLC1A5	54.0	43.9	−10.1	8.77E-05	F(1.8) = 52.63
XP_017265262.1	179	9	Nuclear receptor subfamily 2 group C member 2	NR2C2	93.6	83.1	−10.5	1.63E-03	F(1.6) = 29.41
XP_017279091.1	165	10	T-lymphoma invasion and metastasis-inducing protein 2-like	TIAM2	74.7	60.7	−14.1	8.95E-03	F(1.8) = 11.77
Intergenic
XP_017264766.1	68	3	Leucine-rich repeat transmembrane protein FLRT2-like	FLRT2	29.3	42.8	13.5	7.29E-03	F(1.7) = 13.96
XP_017295097.1	105	3	Uncharacterized protein LOC108249937		71.5	85.1	13.5	7.43E-03	F(1.6) = 15.70
XP_017280372.1	78	3	Sugar phosphate exchanger 3	SLC37A3	36.3	46.9	10.6	5.03E-03	F(1.8) = 14.66
XP_017282790.1	85	4	Protein-tyrosine kinase 2-beta	PTK2B	53.1	63.4	10.3	4.62E-03	F(1.8) = 15.12
XP_017263868.1	112	4	Serine protease 23-like	PRSS23	89.1	79.0	−10.1	1.30E-03	F(1.8) = 23.38
XP_017268118.1	115	6	RNA-binding protein Musashi homolog 2	MSI2	72.8	62.6	−10.2	2.68E-03	F(1.8) = 18.33
XP_017285408.1	52	4	Lymphoid-restricted membrane protein-like	LRMP	87.7	77.2	−10.4	9.80E-04	F(1.8) = 25.58
XP_017274772.1	103	5	Hepatic leukaemia factor	HLF	78.1	67.6	−10.5	2.36E-03	F(1.7) = 21.55
XP_017295868.1	162	8	Protein diaphanous homolog 3	DIAPH3	85.8	74.9	−10.8	8.12E-03	F(1.5) = 18.03
XP_017261032.1	108	3	NACHT, LRR and PYD domains-containing protein 12-like	NLRP12	94.8	83.9	−10.9	5.96E-03	F(1.6) = 17.28
XP_017282790.1	124	10	Protein-tyrosine kinase 2-beta	PTK2B	61.3	49.6	−11.7	1.92E-03	F(1.8) = 20.53
XP_017293105.1	121	4	Uncharacterized protein LOC108248683		81.2	69.3	−11.9	2.76E-03	F(1.8) = 18.15
XP_017284405.1	137	8	Aryl hydrocarbon receptor nuclear translocator-like protein 1	ARNTL	95.6	82.8	−12.9	1.03E-03	F(1.7) = 28.95
XP_017277282.1	134	7	Stromal membrane-associated protein 2-like	SMAP2	68.9	55.6	−13.3	7.38E-03	F(1.8) = 12.69
XP_017296945.1	92	3	DNA repair protein RAD51 homolog 3	RAD51 C	61.1	46.3	−14.8	1.54E-03	F(1.8) = 22.08
XP_017264564.1	145	11	GDNF-inducible zinc finger protein 1-like	GZF1	80.1	64.7	−15.4	3.02E-03	F(1.8) = 17.58
XP_017260964.1	158	6	Myotubularin-related protein 10	MTMR10	66.8	46.4	−20.4	2.43E-04	F(1.7) = 46.87

Networks associated with gene body and promoter DMFs

A network analysis was performed with the IPA software using genes corresponding to DMFs located in gene body and promoter regions. Only networks with the highest number of associated DMFs are reported (Table 6). The highest network scores were found in the 4 ng/L vs Ctl comparison: the first network involved 16 molecules, with functions related to organismal abnormalities, cell morphology and nervous system function and is represented in Figure 4. The second network was composed of 15 molecules and was related to gene expression, drug metabolism and lipid metabolism. In 120 ng/L exposed individuals compared to Ctl, only one network with more than nine molecules was established in IPA, related to lipid metabolism, molecular transport and small molecule biochemistry. Finally, four networks were created when comparing 120 ng/L to 4 ng/L exposed individuals, endorsing the differences in the response at these two concentrations. The first network was related to antigen presenting, cell-to-cell signalling and the inflammatory response. The 2^nd and 3^rd networks were mostly associated with cellular processes (e.g., cellular organization and morphology, cell-to-cell signalling) and the 4^th network was linked to cellular death and survival and connective tissue function.

Table 6.

Top molecular networks and functions associated with our dataset. Only differentially methylated fragments situated in promoter or gene body and their corresponding annotated genes were taken into account

	Top Diseases and Functions	Score	#	Molecules in Network
4 ng/L vs CTL	Organismal Injury and Abnormalities, Cell Morphology, Nervous System Development and Function	40	16	26s Proteasome, ↑ABCA1, ↓ADGRL2, Akt, ↓APP, Calmodulin, ↑CAMK1D, ↑CASC3, ↑CCND2, ↑DBN1, ↑DLG3, ERK, ERK1/2, oestrogen receptor, ↑FLI1, GPR61, GPR78, HISTONE, Hsp90, IgG, Jnk, LDL, ↑MMP15, ↑NCOR2, ↑NIPBL, Nr1h, P38 MAPK, PI3K (complex), ↑RHBDL1, RNA polymerase II, ↑S1PR1, stigmasterol, ↓STYK1, Tgf beta, ↑TSSK1B
	Gene Expression, Drug Metabolism, Lipid Metabolism	37	15	AGT, ↑ARID2, ↑ASTN2, ↑AVL9, CREB1, EGFR, ESR1, FOXP3, GPR12, GPR85, HNRNPK, IFNG, IGSF10, ↓L3MBTL1, MAPK1, MATN2, ↑MTIF3, ↑NELL1, ↑NEO1, ↑NTSR1, ↑PPP1R12C, ↓PSD2, Pvr, ↑RCAN3, RELA, ↑RSBN1, S1PR, SAMD11, ↑SCAI, SMAD3, TNFRSF1A, TP53, UBL3, ↑WHRN, ↑ZNF385C
120 ng/L vs CTL	Lipid Metabolism, Molecular Transport, Small Molecule Biochemistry	23	9	↑ABCA1, Act1, ↑ARID3B, benzoic acid, CHL1, cholesterol sulphate, Cxcl15, ERK, FPR2, GPNMB, GPR119, hydrocortisone, Jnk, LMBRD2, LXR ligand-LXR-Retinoic acid-RXRŒ±, ↑MAP3K3, mir-33, MTX3, ↓NAV2, ↑NDRG1, ↑NPAS2, palmitoleic acid, PAPPA, SCARB1, SFXN5, SLC1A4, ↑SLC22A6, SPARCL1, STEAP4, stigmasterol, ↓TENM1, TGFB1, TNF, tretinoin, UTS2R
120 ng/L vs 4 ng/L	Antigen Presentation, Cell-To-Cell Signalling and Interaction, Inflammatory Response	29	13	ACOX1, ↓ADGRV1, ↓AEBP2, APP, Aspartyl Protease, Camk, ↓CAMK1D, CAMK1G, ↓CTSE, DOCK3, GPR50, ↑GRIN2D, HSP90AA1, ↓LHPP, ↓MCTP1, ↓MLLT6, NMDA Receptor, NTSR2, PPARG, ↓PPP2R2B, QPCTL, QRFPR, RASL11B, ↑RRP36, ↑SCARF1, SET, SLC5A7, SNAP25, TAS1R3, TMOD4, TSSK2, ↓TSSK1B, ↑ZFHX3, ZKSCAN8, ZNF445
	Cellular Assembly and Organization, Cellular Movement, Cell-To-Cell Signalling and Interaction	21	10	3 beta HSD, Akt, ↑ANKRD1, ↓ARHGEF17, ARHGEF18, ↓DAP, ↑DPYSL2, EFNB, ↓EPHA6, ERK, gelatinase, GJC1, GJD2, Gsk3, indole, Insulin, KLF11, MAP3K13, ↓MAPK8IP1, miR-130a-3p (and other miRNAs w/seed AGUGCAA), MYZAP, NFkB (complex), NTN4, PDGF (family), PDK2, ↓RGS3, SETD6, SH3RF1, TACSTD2, TDO2, ↓TJP1, Vegf, VMP1, ↓ZBTB4, ↓ZFPM2
	Cell Morphology, Cellular Compromise, Cellular Function and Maintenance	21	10	ANO3, APP, ARHGEF17, BCL2L11, CCSER2, CLK4, DIXDC1, DOCK3, ↑EEFSEC, EFNB3, ENTPD7, FBXL7, GRN, HIPK2, MAPK14, MASTL, METTL16, MN1, NBEAL1, NTN4, NXF1, NXPE3, PARD3B, PARD6B, PDK2, PDK3, QRICH1, RAC1, RGMA, ↑RHBDF1, RTN4R, SAMD4B, SYT12, SYTL1, ZNF445
	Cell Death and Survival, Cellular Compromise, Connective Tissue Development and Function	18	9	AR, ARFGEF1, COL5A2, ↓COL6A2, COL9A1, Collagen type VI, ↓CRTC3, CTNNB1, DHH, DQX1, EHD3, ↓FGF14, FOS, Glucocorticoid-GCR, HSPA5, KSR1, LYPD1, MYO5B, ↓MYO9B, MYZAP, ↓NFIX, Nuclear factor 1, ↑PBX1, PKNOX2, PLAU, ↓RAB11FIP2, RBM33, SCGB1D4, ↓SLC1A5, SLC25A46, ↓SLC43A1, SPDEF, TDO2, TP53, WNT10B

Figure 4.

Network of genes corresponding to significant DMFs (located in promoter region or in gene body) at 4 ng/L compared to control. Blue: hypermethylation at 4 ng/L compared to control, Yellow: hypomethylation at 4 ng/L compared to control.

DNA methylation of genes involved in growth, steroidogenesis and reproduction

As mentioned above, a total of 14,679 unique RefSeq identifiers were attributed among the 102,018 analysed fragments. It is therefore possible to search for genes of interest regardless of their significance. Based on the effects observed on growth, steroid hormone levels and reproduction reported in Voisin et al. [48] following EE2 exposure, we extracted DNA methylation levels of selected genes or biomarkers involved in the response to EE2 (oestrogen receptor), reproduction (vitellogenin), steroidogenesis (aromatase) and the somatotropic axis (insulin-like growth factor I and II, Igf1 receptor, Igf binding protein and Igf2 mRNA binding proteins) (Table 7). This list is non-exhaustive and contains 24 fragments located in gene bodies [21] or promoter regions [3]. The methylation difference for all the fragments is very low (below 5%). However, a fragment of 153 bp located in vitellogenin-1-like gene body showed a hypomethylation of 9.4% after exposure to EE2 at 120 ng/L compared to control.

Table 7.

DNA methylation levels of fragments related to genes involved in growth, steroidogenesis or reproduction

					% Methylation				Methylation difference (%)
ID	Len	CpGs	Description	DMF	Ctl	4 ng/L	120 ng/L		4 – Ctl	120 – Ctl	120–4	P-value	Test
XP_017294595.1	85	9	Oestrogen receptor beta-like	Gene body	94.71	96.28	95.30		1.57	0.59	−0.98	0.07493	F(2.12) = 3.24
XP_017267390.1	54	8	Oestrogen receptor beta-like	Promoter	97.06	97.82	96.54		0.76	−0.52	−1.28	0.13056	F(2.12) = 2.42
XP_017267390.1	72	6	Oestrogen receptor beta-like	Gene body	94.50	95.72	95.28		1.22	0.78	−0.44	0.14477	F(2.12) = 2.28
XP_017267390.1	101	6	Oestrogen receptor beta-like	Gene body	29.99	33.30	29.14		3.31	−0.85	−4.16	0.39458	F(2.12) = 1.01
XP_017260454.1	104	4	Oestrogen-related receptor gamma-like	Gene body	85.19	89.50	86.40		4.31	1.21	−3.10	0.0425	F(2.12) = 4.16
XP_017291458.1	96	4	Oestrogen-related receptor gamma-like	Gene body	81.14	85.59	81.55		4.45	0.41	−4.04	0.0446	F(2.11) = 4.18
XP_017262455.1	153	5	Vitellogenin-1-like	Gene body	91.30	89.95	81.87		−1.35	−9.43	−8.08	0.0050	F(2.10) = 9.45
XP_017262455.1	89	6	Vitellogenin-1-like	Gene body	90.45	93.45	89.99		3.00	−0.46	−3.46	0.0105	F(2.12) = 6.82
XP_017262455.1	75	6	Vitellogenin-1-like	Gene body	92.58	92.50	92.81		−0.08	0.23	0.31	0.6645	F(2.12) = 0.42
XP_017295559.1	72	6	Aromatase	Gene body	87.84	87.62	84.35		−0.22	−3.49	−3.27	0.0093	F(2.12) = 7.08
XP_017295559.1	137	12	Aromatase	Gene body	92.69	93.60	94.71		0.91	2.02	1.11	0.1607	F(2.12) = 2.14
XP_017267185.1	131	5	Insulin-like growth factor I	Gene body	4.51	3.68	2.51		−0.83	−2.00	−1.17	0.2681	F(2.11) = 1.49
XP_017267185.1	150	9	insulin-like growth factor I	Promoter	91.94	93.76	94.54		1.82	2.60	0.78	0.3066	F(2.12) = 1.31
XP_017259346.1	44	3	Insulin-like growth factor II	Gene body	95.35	96.63	93.88		1.28	−1.47	−2.75	0.1261	F(2.12) = 2.47
XP_017274518.1	64	4	Insulin-like growth factor II	Gene body	5.53	4.72	5.07		−0.81	−0.46	0.35	0.7446	F(2.12) = 0.30
XP_017274518.1	44	3	Insulin-like growth factor II	Gene body	4.86	5.91	4.55		1.05	−0.31	−1.36	0.1565	F(2.12) = 2.17
XP_017284163.1	86	7	Insulin-like growth factor 1 receptor	Gene body	92.18	94.49	91.72		2.31	−0.46	−2.77	0.0390	F(2.12) = 4.30
XP_017284163.1	153	13	Insulin-like growth factor 1 receptor	Gene body	94.08	94.82	91.76		0.74	−2.32	−3.06	0.0701	F(2.12) = 3.34
XP_017261561.1	164	13	Insulin-like growth factor 1 receptor	Gene body	91.60	95.83	92.94		4.23	1.34	−2.89	0.0823	F(2.10) = 3.24
XP_017284163.1	138	10	Insulin-like growth factor 1 receptor	Gene body	88.79	92.11	90.51		3.32	1.72	−1.60	0.1541	F(2.12) = 2.19
XP_017279935.1	55	7	Insulin-like growth factor 2 mRNA-binding protein 2	Gene body	95.81	97.37	95.80		1.56	−0.01	−1.57	0.0029	F(2.12) = 9.85
XP_017279935.1	98	10	insulin-like growth factor 2 mRNA-binding protein 2	Gene body	90.32	93.25	88.80		2.93	−1.52	−4.45	0.0129	F(2.12) = 6.40
XP_017279586.1	97	6	Insulin-like growth factor 2 mRNA-binding protein 3	Gene body	91.79	93.45	90.36		1.66	−1.43	−3.09	0.0669	F(2.12) = 3.42
XP_017292570.1	75	10	Insulin-like growth factor-binding protein 1	Promoter	1.26	1.66	1.19		0.40	−0.07	−0.47	0.1667	F(2.12) = 2.09

Discussion

Characterization of the mangrove rivulus RR genome

Cytosine methylation is a predominant epigenetic regulation of gene expression. It occurs mainly on CpG dinucleotides and is classically associated with gene silencing when it occurs at CpG rich promoter regions [60]. Whole genome bisulphite sequencing (WGBS) provides global genome coverage of DNA methylation at single-base resolution, and is consequently referred as the ‘gold standard’ method. Despite its benefits, including a high coverage of CpGs (>90% in human) and an unbiased representation, it remains expensive and difficult to use with large number of samples, mostly because the detection of changes in methylation between samples demands high sequencing depth [61]. In the present study, we opted for reduced representation bisulphite sequencing (RRBS), an approach that also resolves single-base DNA methylation, combining the use of restriction enzymes and bisulphite sequencing. Unlike WGBS, RRBS can only sequence rich-CpG regions, and has limited coverage of the genome [62]. However, the fact that it enriches the analysed fragments with a high density of CpGs makes this technique cost-effective and allows to work on a higher number of samples with a high sequencing depth [63]. Here we reported the DNA methylation pattern of mangrove rivulus liver using RRBS. This species shows a very peculiar reproduction strategy as it is the only vertebrate which alternates between cross-fertilization and selfing, the latter being the most common. It results in several lineages of isogenic individuals, and epigenetic mechanisms have been proposed to explain the occurrence of phenotypic variability despite a low level of genetic diversity [42]. We reported a RR genome representing 1.4% of the whole genome, which was comparable to what was obtained in mouse (1.4%) or in zebrafish (2.2%) brain [54,64]. The percentage of CpG covered in the RR genome reached 6.1% of the total number of genomic CpGs, reaching a 4.3-fold enrichment in CpG sites. This CpG coverage is similar to what was found in mouse and zebrafish (7.0% and 5.3%, respectively [54],), but higher than what other studies reported for diverse fish species such as stickleback [1% [65]], Atlantic salmon [2.75% [66]], rainbow trout [<1% [67]] and guppies [1.5–2% [68]]. In mangrove rivulus brain, Berbel-Filho et al. [46] also reported a lower CpG coverage at 1.2%. This might be astonishing considering that different tissues of the same species usually show the same methylation profile [69] and methodological aspects can not be rejected to explain this discrepancy between those studies. Most noticeable is the fact that we obtained significantly more reads in the present study (explaining that fragments aligned to a higher proportion of the genome) but also a lower alignment efficiency (44.4% in average vs 62.9%) as many fragments could not be mapped. This relatively low level of alignment efficiency was also observed in zebrafish brain following the exact same technical workflow [54]. RRBS reads are relatively short (40–220 bp with majority being < 100 bp) and its plausible that this contributes to large proportion of unmapped reads. Another possible explanation is the proportion of RRBS reads that are derived from repeat regions in the genome as the RRBS reads from repeat region aligns with less efficiency [70,71].

The distribution of the analysed fragments across the RR genome showed 16% located in promoter region. The analysed fragments in this region are also low in zebrafish liver at 12%, and even lower in Atlantic salmon (4%) [66]. On the contrary, this percentage is higher in mammal species such as mouse liver (52%) [69]. This proportion of analysed fragments located in promoter region can explain the observed difference of the mean global methylation level reported in fish species and in mammals, the latter having a lower value (28% in mouse vs 65% in the mangrove rivulus, 70% in zebrafish, 75% in Atlantic salmon). The promoter regions being less methylated on average than other regions in both fish and mammals, a higher proportion of fragments from this region decreases the average global methylation as observed in mouse RR genome. These differences between mammals and fish can also be reflected in the proportion of highly and lowly methylated fragments. If the majority of fragments are either highly or lowly methylated in all reported vertebrates, a larger proportion of fragments present a methylation level higher than 80% in fish than in mammals (67% of fragments in our study) [69].

Regarding the mangrove rivulus, even if the majority of promoter fragments were lowly methylated, a significant proportion were highly methylated (38%), which indicates an elevated range of the methylation level in promoter regions, and consequently a regulatory activity of gene expression. In contrast, other genomic regions presented a narrower range in methylation levels, with the majority of fragments being highly methylated. When compared to zebrafish, an interesting difference appeared regarding the proportion of analysed fragments located in gene bodies. While it reached 47% in mangrove rivulus, zebrafish has only 25% of CpGs in introns and exons. Conversely, the proportion of analysed intergenic fragments is higher in zebrafish (63%) than in mangrove rivulus (37% including intergenic and upstream regions). A possible explanation could reside in the higher proportion of transposable elements (TEs) in zebrafish genome. There is increasing evidence that TEs are important targets of DNA methylation to silence their expression and consequently show a high level of CpG methylation. The zebrafish genome is rich in TEs as 52% of its genomic sequence is composed of the different classes of TEs [72,73], compared to 27.3% in the mangrove rivulus [59]. This difference could explain the relatively lower proportion of analysed fragments located in intergenic regions in the mangrove rivulus compared to zebrafish. When considering other fish species, the part of TEs in genomes is very variable and values as low as 5% have been reported in Tetraodon nigroviridis [73]. This variability of TE amount is explained by the large range of genome size among fish, and Chalopin et al. [72] reported a positive correlation between TE content and genome size in teleosts species, which also applies to the mangrove rivulus. Contrary to what was advanced by Zhang et al. [69], it is therefore unlikely that the pattern of high global methylation observed in fish RR genomes is the result of an enrichment in TEs. Nevertheless, the role of DNA methylation in TEs should be further investigated, including a precise look at the different types of TEs. For example, the rolling-circle transposons, known as Helitrons, are particularly abundant in the mangrove rivulus genome (0.65%) compared to other fish species including its phylogenetic relatives [Japanese medaka (0.03%) and Turquoise killifish (0.06%)]. From all studied fish species, only zebrafish showed a higher level of Helitrons at 1.5% [73]. It is known that TEs in general, and the Helitron subfamily in particular, are mediator of host genes regulation and contribute to genome evolution and adaptation [74]. It has been suggested that the mobilization of TEs and the changes in epigenetic pattern could be important in the process of organism’s rapid adaptation to environmental changes [75]. The exact role of these elements and their methylation level in the mangrove rivulus is a promising perspective to investigate how this species adapt and evolve using a mixed-mating reproduction system and consequently a low genetic diversity within lineages.

Effects of EE2 on the liver methylome

Growing evidence suggests that EDCs can modify DNA methylation and consequently gene expression pattern. Recently, experiments on zebrafish embryos clearly established a link between exposure to bisphenol A, DNA methylation and behavioural impairments [76]. Data also showed that EE2 can modify the methylation status of target genes [35]. In a previous study, we showed that the liver proteome of adult mangrove rivulus was impaired by exposure to EE2 during early life stages (ELS) [49]. It has been hypothesized that exposure to EE2 during ELS can impair DNA methylation and induce long-term consequences later in life as observed on the proteome and on the phenotype [48]. To test this hypothesis, we aimed at detecting potential long-term changes in the hepatic DNA methylation of 168 dph adults following a 28-day early-life exposure of mangrove rivulus hatchlings. We obtained a total of 146 DMFs among all group comparisons, from which a majority were observed between the two EE2 treatments and between 4 ng/L exposed individuals and Ctl. The larger observed impacts of the lowest EE2 concentration on DNA methylation is in accordance with a non-monotonic dose–response relationship often reported for exposure to endocrine disruptors. Hormones are effective at very low concentration, having a high affinity for their receptors, and so are the endocrine disrupting chemicals [77]. The present results are reinforced by a similar response detected in brain and liver protein expression profiles of the mangrove rivulus exposed in the same conditions to EE2 [49]. To our knowledge, only one other study investigated the long-term effects of early-life exposure to chemicals on fish methylome [78]. This study investigated the effects of two compounds, mono(2-ethylhexyl) phthalate (MEHP, 30 µM) and 5-azacytidine (5AC, 10 µM, an inhibitor of DNMT1) in zebrafish. They found that most methylation changes occurred at conserved non-genic regions with cis-regulatory functions (i.e., enhancers, silencers, transcription factor binding sites).

Among the differentially methylated fragments (DMFs) found in adult mangrove rivulus liver following early-life exposure to EE2, none was significantly affected in all three group comparisons (Ctl vs 4 ng/L; Ctl vs 120 ng/L; 4 vs 120 ng/L). Six DMFs were significant in two out of three comparisons, of which three were located in gene bodies. A DMF corresponding to ATP-binding cassette sub-family A member 1 (ABCA1) was hypermethylated at both 4 ng/L and 120 ng/L compared to Ctl. ABCA1 is a known oestrogen-responsive gene [79]. It was included in networks in both comparisons and appeared to play a central role in the response to EE2, showing interactions with the oestrogen receptor and several other molecules (Figure 4). It is a major regulator of cholesterol and phospholipid homoeostasis by regulating efflux of cholesterol and phospholipids outside of the cell, that are then taken up by apolipoproteins A1 and E to form high density lipoprotein (HDL) molecules. Namely, oestradiol exposure of smooth muscle cells has been shown to enhance cholesterol efflux to APOA1 and HDL, which was associated to ABCA1 overexpression via ESR2 and liver X receptor (LXR) activation [80]. We suggest that the alteration of methylation level of ABCA1 gene might be related to the altered abundances of liver apolipoproteins and other downstream proteins involved in lipid metabolism previously reported [49]. The fact that both concentrations of EE2 impacted this gene methylation level indicates that this mechanism might be a general mode of action of EE2, and could induce effects at environmental concentrations as low as 4 ng/L.

Two DMFs were significantly hypermethylated at 4 ng/L in comparison to both Ctl and 120 ng/L treatments: the calcium/calmodulin-dependent protein kinase type 1D (CAMK1D), and the testis-specific serine/threonine-protein kinase (TSSK1B). Both DMFs were included in the first network of 4 ng/L vs Ctl and 120 vs 4 ng/L comparisons. CAMK1D is a protein kinase that participates in the calcium-regulated calmodulin-dependent kinase cascade, regulating calcium-mediated granulocyte function and respiratory burst and activating the transcription factor CREB1. According to Figure 4, CAMK1D interacts not only with calmodulin, but also with the amyloid precursor APP, which can be induced by oestrogens through extracellular-regulated kinase 1 and 2 (ERK1/2) phosphorylation [81]. Finally, TSSK1B is essential for spermatid development and spermatogenesis. While TSSK1B is likely not expressed in the liver, it can be possible that an epigenetic alteration of this gene could be found in the ovotestis following EE2 exposure, with potential consequences on reproduction. Although no study to date has linked EE2 and TSSK1B, a study in rats has shown that bisphenol A exposure decreases the expression of TSSK1B mRNA [82].

We must nevertheless be careful when interpreting the changes of methylation in these 3 DMFs. First, the observed differences in methylation, if significant, ranged between 10.3 and 14.6%. Further studies should determine whether these differences have a biological impact on gene expression. Moreover, these three DMFs occurred in gene bodies, as did the majority of DMFs in this study (and MSPI fragments). While methylation of DNA at promoters is well known to suppress transcription, the role of DNA methylation in gene bodies is more difficult to interpret [83]. Gene body methylation is conserved among plants and animals, with a general preference for exons [84]. Potential roles include the stimulation or inhibition of transcription elongation [85] and the regulation of splicing [86]. In mammals, most genes possess alternative transcription start sites that can be located within the gene bodies, complicating the link between methylation and expression [86]. Nevertheless, it is now recognized that gene body methylation is positively correlated with transcriptional activity in most species [12] and we should not disregard the possible implication of these observed changes in methylation after exposure to EE2.

Beside significant changes of DNA methylation within gene bodies, we also reported several DMFs belonging to intergenic regions. Interpreting alterations in DNA methylation in these regions is challenging as the gene annotation may not directly apply. Intergenic regions are typically highly methylated in all vertebrates and bear a conserved role of maintaining genome stability, by preventing the transposition of repetitive TEs and silencing cryptic promoters and cryptic splice sites [87]. Moreover, TEs possess strong promoters that may be able to transcribe neighbouring genes (transcriptional interference). Finally, a compact chromatin state in these regions can prevent recombination. Most (>90%) methylcytosines occur in transposons, and low methylation of these regions could reduce genomic stability [88]. As above mentioned, the diversity and composition of TEs in the mangrove rivulus could contribute to the genetic recombination and evolutionary adaptation in this self-fertilizing species [59]. To gain further insights into the role of intergenic methylation, future research should investigate the overlapping of DMFs with potential TEs.

In addition, we pointed out the very low proportion of DMFs classified as lowly methylated fragment (< 20% methylation). The nipped-B-like protein B (NIPBL) is the exception as it showed 19.6% methylation in Ctl but increased to 41.5% in fish exposed to 4 ng/L EE2. It is remarkable that this fragment is also the one showing the highest effect size of our study (+ 21.9%) and is located in the promoter region. Few more lowly methylated fragments were significant when comparing the 2 treatments, but all others were either highly methylated or showed an intermediary methylation level. Two explanations can be advanced. First, the lowly methylated fragments were about 1/3 less numerous than the highly methylated ones in the RR genome as reported in Figure 2. This is related to the proportionally low occurrence of fragments located in the promoter region, which have the lowest methylation level. Second, we also showed that lowly methylated fragments have a 10-fold higher coefficient of variability (33% vs 3% in highly methylated fragments). It results that the statistical power to detect changes of lowly methylated fragments is lower and that we can consequently report only DMFs with a high effect size, such as NIPBL. To overstep this limitation, one should include a higher number of replicates in future RRBS analyses. It is even more important for this kind of studies for which effects are searched long after the end of the exposure, which decrease the probability to observe high effect sizes. Nevertheless, our study is the first to report the effect of an endocrine disruptor exposure on the methylation level of NIPBL promoter region. This effect, which was significant while the exposure had stopped 140 days earlier, could have serious long-term consequences on adult fish. NIPBL is associated to cohesin in order to mediate chromatin cohesion, important for mitosis. This complex plays an important role in histone acetylation, chromosome architecture and therefore in regulation of gene expression [89,90]. In human, a mutation of NIPBL is associated to the ‘Cornelia de Lange’ syndrome, a developmental disorder [91]. In zebrafish, the zygotic genome activation is facilitated by cohesin and was a key for the embryogenesis [92]. Intriguingly, the cohesin complex has been proposed to modulate oestrogen-dependent gene transcription and a link has been established with breast cancer [93]. Our results suggest that NIPBL hypermethylation could be a mode of action of EE2 exposure and could potentially modify gene regulation. Further studies should investigate the long-term links between the change in NIPBL methylation level and the overall changes in gene expression in adults, long after the exposure to EE2.

Additionally, we reported the methylation level of fragments corresponding to oestrogen receptors, the aromatase, vitellogenin and several genes involved in the somatotropic axis, which is the main regulator of growth in vertebrates. Most differences in mean methylation of these fragments between experimental conditions were comprised between 0 and 5% and were not significant. Consequently, we cannot link the previously observed effects of EE2 exposure on growth, steroid hormone levels and reproduction [48] with the DNA methylation of these genes. However, one fragment located in gene body of vitellogenin 1 was 9.4% less methylated at 120 ng/L compared to control, while there was no significant effect at 4 ng/L. Vitellogenin is the precursor of egg yolk proteins, produced in the liver of sexually mature female fish in response to endogenous oestrogens. It is well known that it can also be induced by exogenous oestrogens such as EE2 [94]. Stromqvist et al. [35] reported a significant hypomethylation of three CpGs sites located between two potential predicted oestrogen responsive elements (EREs) upstream of the first vitellogenin exon, in the liver of adult zebrafish after exposure to 100 ng/L EE2 during 14 days. In mangrove rivulus, further studies should investigate whether the methylation changes of the vitellogenin 1 fragment could influence vitellogenin expression and further the reproduction of hermaphrodites. The different effect between the two tested concentrations can be associated with the reported impact on reproduction by Voisin et al. [48] who showed fewer laid eggs by hermaphrodites after exposure to 4 ng/L, but not to 120 ng/L. The implication of some compensatory mechanisms at the highest tested concentration involving DNA methylation of vitellogenin 1 CpGs might be hypothesized.

As previously mentioned, DMFs reported as being significant in the present study mostly showed different methylation level between 10 and 20%. These effect sizes could be considered as low, mostly if we compare with experiments comparing normal and pathological tissue such as tumour [95] which can be over 20%. Berbel-Filho et al. [46] also reported higher effect sizes on the rivulus reared during 10 months in differentially enriched environments (up to 48.9% of hypermethylation). If it is necessary to question the biological relevance of changes in DNA methylation below 20%, it is not surprising to have low effect sizes using an experimental design where fish were reared in the same environment for most of their life after an initial exposure during 28 days. This is a similar approach as in studies searching for the role of pregnancy and early-life exposures on later-in-life health outcomes in human. These studies show small-magnitude epigenetic effect sizes below 5%, while it can still persist and be replicated across populations and time and be correlated with gene expression [96]. The individuals used in the present study being issued from the same isogenic lineage, the confounding factor of genetic variability is reduced, which may help to reveal more subtle difference in DNA methylation level, and stress the usefulness of this species in epigenetic studies as model to investigate the developmental origin of health and diseases.

Little correspondence was found between the proteome previously published [49] and the epigenome (Supplementary Figure 2). Although the number of identified genes was about 10 x greater in the methylome analysis, only about half of the total identified proteins in the liver was found in the methylome analysis. Similarly, we could retreive the methylation level of fragments belonging to approximately half (52%) of the differentially abundant proteins. None of these differentially abundant proteins could be related to a change in DNA methylation level of the promotor region. Conversely, only five of the significantly DMFs were also identified, but not significant, in the proteomic analysis in liver: G1/S-specific cyclin-D2-like (CCND2, promoter region, significant between 4 ng/L and Ctl), teneurin-1-like (TENM1, gene body, significant between 120 ng/L and Ctl), protein NDRG1-like (NDRG1, promoter, significant between 120 ng/L and Ctl), fibrinogen alpha chain (FGA, intergenic, significant between 120 ng/L and Ctl), and death-associated protein 1 (DAP, gene body, significant between 4 and 120 ng/L). Comparison between DNA methylation and protein expression pattern is rare but is fully justified. The proteome is the main functional product of gene expression and is closely related to the molecular phenotype [97]. As such, the proteomic level can generate new hypothesis on the modes of action of xenobiotics and on the adaptive responses of organisms [98]. Epigenetics being at the interface between the genotype and the environment, a more systematic link with the proteome would improve the understanding of the adverse outcome pathways (AOPs) and the outcomes in terms of adaptation to the environment. However, it is challenging to correlate these two levels, because of technical variability, and because differentially abundant proteins are not necessarily regulated by differential methylation, but rather the end product of upstream regulators, themselves differentially methylated. Also, protein abundance results not only from transcription, which can be influenced by DNA methylation, but also from mRNA stability, alternative splicing, translation, protein turnover and post-translational modifications. Nevertheless, comparison of proteome and methylation data can be eased by the analysis of gene ontology. Some biological processes were identified in both datasets, such as lipid metabolism, inflammation, connective tissue development and molecular transport, which makes their long-term impairment plausible mechanisms of EE2 toxicity.

Conclusions

In conclusion, our study provides further evidence on the capacity of EDCs to alter the methylome and shows that these changes can be apparent several months after the exposure, supporting the hypothesis of possible long-term modulation of gene expression through epigenetics. It also confirms the non-monotonic response to EDCs as most significant effects were observed at the environmental concentration of 4 ng/L. The DMFs belonging to promoter and gene bodies revealed networks involved in several biological functions potentially regulated by oestrogens, including lipid metabolism, cellular processes (death and survival), connective tissue function, molecular transport and inflammation, many of which were also identified in a previous proteomic analysis. Importantly, we reported that methylation of nipped-B-like protein B (NIPBL) might be an important mode of action of EE2 and, as involved in chromosome organization and gene expression, could have long-term impact on gene regulation and on organism’s general functions. To gain a more comprehensive view of the significance of DMFs in promoter, gene body, upstream and intergenic regions, future studies should investigate the overlapping of the DMFs with potential enhancers, insulators, transcription start sites, TEs, oestrogen-responsive elements and other transcription factor binding sites that may interact with oestrogen receptors. Among them, TEs, and more particularly helitrons, are potentially important targets of environmental stressors that could be involved in the response of organisms to environmental changes. Despite its limitations, RRBS has proved to be an efficient method to investigate effects of EE2 on DNA methylation in the mangrove rivulus. This self-fertilizing fish should be further considered as a model species to investigate the interplay between environmental stressors, epigenetics and adaptation. The possibility to naturally discard the confounding factor of genetic variability makes this species a top model to better understand the roles of epigenetics in the long-term response to environmental xenobiotics and in the developmental origin of health and diseases.

Funding Statement

This work was supported by the Fonds de la Recherche Scientifique - FNRS.

Data accessibility statement

All the data generated by the RRBS workflow will be submitted to Dryad repository and will be publicly accessible. The BAM files after Bismark alignment and the excel sheet containing the values for all analysed fragments have already been submitted and are available at this link: https://datadryad.org/stash/share/k7Lq6mHbxBMF5BHdk8cOp1EFGSGVK3tfJoBx4tPYnzc.

The DOI is: https://doi.org/10.5061/dryad.v41ns1rsv

The Fastq files will be submitted when a decision will have been given for the acceptance of the manuscript.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Author contributions

Anne-Sophie Voisin: investigation, manuscript writing and editing, experimental design, data acquisition and analysis

Victoria Suarez Ulloa: data analysis, manuscript editing

Peter Stockwell: bioinformatic data analyses

Aniruddha Chatterjee: supervision, methodology, RRBS analyses, manuscript editing

Frédéric Silvestre: investigation, supervision, funding acquisition, manuscript writing and editing, experimental design, submission and revision

The authors report no conflict of interest

Supplementary material

Supplemental data for this article can be accessed here.