Human Endometrial DNA Methylome Is Cycle-Dependent and Is Associated With Gene Expression Regulation

Abstract

Human endometrium undergoes major gene expression changes, resulting in altered cellular functions in response to cyclic variations in circulating estradiol and progesterone, largely mediated by transcription factors and nuclear receptors. In addition to classic modulators, epigenetic mechanisms regulate gene expression during development in response to environmental factors and in some diseases and have roles in steroid hormone action. Herein, we tested the hypothesis that DNA methylation plays a role in gene expression regulation in human endometrium in different hormonal milieux. High throughput, genome-wide DNA methylation profiling of endometrial samples in proliferative, early secretory, and midsecretory phases revealed dynamic DNA methylation patterns with segregation of proliferative from secretory phase samples by unsupervised cluster analysis of differentially methylated genes. Changes involved different frequencies of gain and loss of methylation within or outside CpG islands. Comparison of changes in transcriptomes and corresponding DNA methylomes from the same samples revealed association of DNA methylation and gene expression in a number of loci, some important in endometrial biology. Human endometrial stromal fibroblasts treated in vitro with estradiol and progesterone exhibited DNA methylation changes in several genes observed in proliferative and secretory phase tissues, respectively. Taken together, the data support the observation that epigenetic mechanisms are involved in gene expression regulation in human endometrium in different hormonal milieux, adding endometrium to a small number of normal adult tissues exhibiting dynamic DNA methylation. The data also raise the possibility that the interplay between steroid hormone and methylome dynamics regulates normal endometrial functions and, if abnormal, may result in endometrial dysfunction and associated disorders.

Human endometrium, the obligate tissue for blastocyst implantation and embryonic development (1), undergoes complex molecular, cellular, and functional changes on a cyclic basis in response to the ovarian steroid hormones, estradiol (E₂) and progesterone (P₄) (2–4). In the proliferative phase, E₂ induces cellular proliferation and angiogenesis (2), and after ovulation in the secretory phase, P₄ induces secretory transformation of the epithelium and differentiation (decidualization) of the stromal fibroblasts (2–4). Extensive changes in the transcriptome and proteome are required for these morphologic and functional changes, which involve well-orchestrated gene expression regulation by well-characterized transcription factors, nuclear receptors, and coregulators (5–10). In addition to these well-characterized regulatory mechanisms, there is increasing evidence in steroid hormone–responsive tissues (11, 12), including endometrium (13), of changes in gene activity due to epigenetic modifications, although the involvement of epigenetic mechanisms such as DNA methylation is less well understood in cycling endometrium.

Methylation of the carbon-5 position of cytosine, mostly in the context of CpG dinucleotides, is the main epigenetic modification of DNA (14–17) and is essential for a properly functioning genome, including maintenance of chromosome stability and transcriptional repression, and is a rate-limiting step for many cellular transitions (eg, embryonic and germ cell development and nuclear reprogramming) (14, 18–20). DNA methylation at the promoter-associated CpG islands of genes is usually associated with gene silencing (21), whereas promoter CpG island hypomethylation (22, 23) and gene body CpG dinucleotide hypermethylation are associated with gene activation (24, 25). About 10% to 20% of genes display DNA methylation patterns in a tissue-specific manner, usually associated with tissue-specific patterns of gene expression (26). In human endometrium, DNA methyltransferase (DNMT1, DNMT3A, and DNMT3B) expression is cycle dependent (8, 27) and in endometrial explants DNMTs are regulated by P₄ (27). Furthermore, the DNA demethylating agent 5-aza-2′-deoxycytidine increases ESR2 (estrogen receptor-β) expression in human endometrial stromal fibroblasts (eSFs) (28), underscoring the importance of epigenetic modifications in the endometrial cellular steroid hormone response. Whether the human endometrial DNA methylome is cycle dependent has not been established and is the subject of this article.

Herein, using a bisulfite-based quantitative assay that measures DNA methylation at 27 578 CpG sites, we investigated the DNA methylome of human endometrium to better understand the involvement of epigenetic mechanisms in hormonally regulated cyclic endometrial activity. Our results show dynamic, cycle phase–dependent DNA methylation correlating with gene expression changes in the endometrium at specific loci, underscoring an important role of this epigenetic modification in the steroid hormone response of this tissue.

Materials and Methods

Sample collection and processing

This study was approved by the Committee on Human Research of the University of California, San Francisco (UCSF). Eighteen eutopic endometrial tissue samples, n = 6 in each of the proliferative (PE), early secretory (ESE), and midsecretory (MSE) phases, were obtained after informed consent from subjects undergoing endometrial biopsy, hysterectomy, or gynecologic surgery for a benign condition (Table 1). All subjects were documented as not having endometriosis and not being pregnant and had no hormonal treatments within 3 months before sample acquisition. Tissue acquisition, processing, and storage were conducted by the UCSF National Institutes of Health Specialized Cooperative Centers Program in Reproductive and Infertility Research Human Endometrial Tissue and DNA Bank, as described previously (29). All subjects were nonsmokers with one exception (Table 1). Menstrual cycle phase was determined by histologic evaluation by 3 independent readers according to the Noyes criteria (30), by blood levels of estradiol and progesterone, and/or by unsupervised principal component analysis and independent hierarchical clustering analysis of microarray gene expression data (31). The mean ages of participants were 42.5 ± 1.44 years for the PE phase, 40.5 ± 2.72 years for the ESE phase, and 46.3 ± 1.44 years for the MSE phase (P > .05).

Table 1.

Sample Characteristics.

Sample No.	Age, y	Weight, kg	BMI, kg/m2	Ethnicity	Histology Reading	Microarray Phasing
Z19	39	82.6	25.5	White	PE	PE
Z20	42	NA	NA	White	PE	PE
Z21	44	59.0	23.9	White	PE	PE
Z22	43	88.0	32.5	Black	PE	PE
Z23	40	67.1	25.3	White	PE	PE
Z24	47	111.6	NA	White	PE	PE
Z25	44	65.0	24.5	White	ESE	ESE
Z26	46	82.3	30.2	White	ESE	ESE
Z27	40	83.4	31.4	White	ESE	ESE
Z28	34	95.7	35.2	White	ESE	ESE
Z29	45	65.8	NA	Black	ESE	ESE
Z30	34	96.6	36.4	White	ESE	ESE
Z31	49	64.5	25.2	White	MSE	MSE
Z32	50	60.8	21.5	White	MSE	MSE
Z33	42	61.3	21.2	White	MSE	MSE
Z34	45	56.3	20.7	White	MSE	MSE
Z35	46	79.0	28.0	White	MSE	MSE
Z36	46	61.9	23.3	White	MSE	MSE

All subjects were nonsmokers (except Z21, PE, a past smoker), were not exposed to hormonal medications within 3 months before tissue sampling, and were confirmed as being not pregnant. Cycle phase was determined by histologic evaluation according to Noyes criteria and also by sample clustering in unsupervised principal component analysis of microarray data. Dating by both methods was consistent for all subjects.

DNA extraction

Genomic DNA was extracted from endometrial tissue samples using a NucleoSpin Tissue Kit (Macherey-Nagel Inc), according to the manufacturer's protocol, and was stored at −20°C until use.

DNA methylation analysis

Endometrial sample genomic DNA was bisulfite-converted using the EZ-96 DNA Methylation Kit (Zymo Research), according to the manufacturer's protocol, at the University of Southern California (USC) Epigenome Center. Quality controls for the amount and completeness of bisulfite conversion were conducted by a panel of MethyLight reactions, as described previously (32). All samples passed the initial quality control test and were further assayed by the Illumina Infinium HumanMethylation27 DNA methylation platform at the USC Epigenome Center, based on Illumina's specifications, as described previously (33). This Illumina platform interrogates DNA methylation levels of 110 microRNA-coding and 14 475 protein-coding genes at 27 578 CpG sites. Illumina platform probes are selected based on their differentially methylated status in cancers and during development.

DNA methylation values were scored as β values, calculated for each probe as the ratio of the methylated signal over the total fluorescent signal. β values range from 0 to 1, with lower levels of DNA methylation close to 0 and higher levels of DNA methylation close to 1 (33). DNA methylation measurement quality was assessed by the detection P value of each probe in each sample. Detection P values were calculated based on the difference in signal intensity of each probe compared with that of a set of 16 negative control probes. Probes with detection values of P < .05 were considered to have statistically significant differences from background and were retained for further data analysis. Probes with a detection value of P > .05 were marked with NA (not acceptable) and were excluded from analysis as nonsignificant DNA methylation detection values. By stringent technical quality criteria, 1 MSE sample with the highest probe dropout rate (1.8%) and an overall deviant DNA methylation distribution pattern on density plot compared with the rest of the samples was eliminated from subsequent analysis (Supplemental Figure 1).

The Illumina Infinium platform has been used in several studies (34, 35), and data obtained with this platform have been validated by other assays such as bisulfite genomic sequencing (36). The reproducibility of these data has been shown previously (33, 37), and we observed similar reproducibility with technical replicates of a subset of our samples as well (2 PE, 2 ESE, and 2 MSE; data not shown).

DNA methylation clustering and visualization

To investigate DNA methylation profiling across all samples, we selected the probes with the highest variation across all samples. The SD of each probe across all samples was calculated to identify the most variable probes, and 10% of all probes (2758) with the most variable DNA methylation levels across all samples were selected for cluster analysis. Probes with more than 1 NA dropout data point were eliminated. Two-way average-linkage unsupervised hierarchical clustering analysis was performed on the selected probes for both CpG sites and samples, using JMP software (38).

Global DNA Methylation Profile and Patterns of Endometrium

To better understand the global pattern of endometrial methylome, we identified loci that were highly methylated (β >0.8), were highly unmethylated (β <0.2), or had intermediate methylation levels (0.2 ≤ β ≤ 0.8) in each phase. We removed 25 probes with missing data in >1 sample from further analysis. We identified 2881 CpG sites to be highly methylated and 16 730 CpG sites to be highly unmethylated in each phase. We next identified 828 CpG sites that were highly methylated and 569 CpG sites that were highly unmethylated in 1 or 2 but not all 3 phases. The remaining 6746 CpG sites had intermediate levels of DNA methylation, with either similar or variable β values across the phases. Next, we determined what fractions of CpG sites within each group were located within or outside CpG islands.

Phase-specific changes in DNA methylation

CpG sites showing differential DNA methylation between cycle phases were identified using median β value differences as described previously (39). In brief, the median β value was calculated for each probe in each phase of the cycle excluding from the analysis probes with >2 missing values across all samples. Median β value differences between cycle phases were derived by subtracting the median β value of each probe in 1 phase from the corresponding value in another phase for the following comparisons: ESE vs PE, MSE vs ESE, and MSE vs PE. In all comparisons analysis included only differentially methylated probes with a detectable difference of 95% confidence (Δβ of ≥0.136), as described previously (33).

DNA methylation of human eSFs treated with hormones in vitro

Human endometrial tissue was obtained from normal controls through the UCSF Human Endometrial Tissue and DNA Bank, after written informed consent (as mentioned above). eSFs were treated with 10 nM E₂, 10 nM E₂ + 1 μM P₄, or vehicle for 15 days in culture, as described previously (40). IGFBP1, a decidualization marker, was determined by ELISA (Alpha Diagnostic International). Cells from n = 4 subjects exhibiting robust decidualization after 15 days and parallel treatment groups and controls were selected for further analysis. DNA was extracted, bisulfite converted, and assayed on an Illumina Infinium 450K methylation platform at USC as described above. Genes with differential DNA methylation in E₂ + P₄ (corresponding to secretory phase) vs E₂ (corresponding to proliferative phase) were selected and compared with genes observed in PE vs ESE and PE vs MSE from eutopic endometrium.

Gene expression microarray

RNA isolation, cDNA conversion, and microarray

Portions of the same tissue samples used for DNA methylation analysis were processed using TRIzol reagent (Invitrogen), and total RNA was isolated with DNase treatment using an RNeasy Plus Kit (QIAGEN), according to the manufacturer's protocol. RNA quality was assessed using the Bioanalyzer 2100 (Agilent Technologies). RNA samples were prepared for microarray analysis according to Affymetrix specifications. For each sample, 5 μg of total RNA was reversed transcribed to cDNA, followed by second-stand DNA generation using DNA polymerase and cRNA generation by in vitro transcription. Sample quality was assessed using the Agilent Bioanalyzer, and hybridization to an Affymetrix HU133 Plus 2.0 gene array interrogating >38 500 genes (Affymetrix, Inc) was conducted at the UCSF Genome Core facility as described previously (41).

Comparison of DNA methylation and gene expression data

The raw .cel expression data were normalized by the GCRMA method using GeneSpring GX 12.0 software (Agilent Technologies). Normalized expression and DNA methylation data were imported into R, and the corresponding probes from the 2 platforms were matched using the transcription unit identifier. The Spearman correlation was used to investigate the relationship between DNA methylation and gene expression values for each locus. In this analysis every DNA methylation probe for a given locus was compared individually with all transcripts of the corresponding locus to discern the effect of the localization of methylation changes on the corresponding gene expression.

Pathway analysis

Differentially methylated loci that were associated with gene expression changes (either positively or negatively) in each phase of the endometrium were examined using the DAVID (Database for Annotation, Visualization and Integrated Discovery) database (42) to elucidate functionally important genes and/or pathways.

Results

DNA methylation profiles of endometrium during the menstrual cycle

Cluster analysis

Genome-scale DNA methylation profiles of 27 578 CpG sites, associated with 14 475 genes, were assessed in 17 samples from human endometrium. Two-dimensional unsupervised hierarchical cluster analysis of all samples and the top 10% (2758) most variable probes revealed 2 main branches (Figure 1). The first branch included only secretory phase samples (4 of the 6 ESE and 4 of the 5 MSE); whereas the second branch included all of the PE samples and 3 secretory phase samples. It should be noted that for gene transcription changes, the greatest differences are seen in PE vs secretory endometrium and that ESE clusters with PE and not with MSE (31). The observation that cluster 1 included only secretory endometrium but not any PE samples indicates that the secretory phase has a methylation profile different from that of the proliferative phase, suggesting that DNA methylation profiles change during the menstrual cycle.

Figure 1.

Two-dimensional unsupervised hierarchical cluster analysis of 17 samples and the top 10% (2578) of probes with the most variable DNA methylation levels. Cluster 1 includes only secretory samples, whereas cluster 2 includes all PE samples and 1 ESE sample in 2a and 1 PE, 1 ESE, and 1 MSE sample in 2b. Diamonds, PE; triangles, ESE; circles, MSE. The heat map represents less methylated (dark blue) to more methylated (dark red).

DNA methylation profile of the endometrium

The following analyses examined the changes in profiles and in patterns of DNA methylation in the endometrium and across the cycle by assessing the frequency of differentially methylated loci across the phases, whether changes involved gain or loss of methylation, chromosomal distribution, and genetic location of these changes, and whether differentially methylated loci were located within or outside a CpG island.

We compared the methylation level of every CpG site across all 3 phases to assess the overall methylation pattern and profile of endometrium (see Materials and Methods). We observed (Figure 2A) that more than 60% of CpG sites are highly unmethylated (β value <0.2), whereas only 10% of the CpG sites are highly methylated (β value >0.8) in all 3 phases. Only 4% of the CpG sites were highly methylated (2%) or highly unmethylated (2%) in 1 or 2 but not all 3 phases. The remaining 25% of the CpG sites showed moderate levels of methylation (0.2 ≤ β value ≤ 0.8), with either similar or variable levels across the 3 phases (Figure 2A).

Figure 2.

Global DNA methylation pattern of human endometrium. A, percentage of CpG sites that are highly methylated or highly unmethylated in all 3 phases, CpG sites that are either highly methylated or highly unmethylated in some but not all phases, and CpG sites with intermediate levels of methylation. B and C, distribution of highly unmethylated or highly methylated sites in all phases of the cycle based on their location within or outside CpG islands, respectively. D, distribution within or outside CpG islands of the highly methylated and highly unmethylated CpG sites in some (but not all) phases plus the CpG sites with intermediate DNA methylation levels. β, β value; CGI, CpG island; In CGI, within CpG island; non-CGI, outside CpG island.

We further investigated whether the methylation state of these CpG sites is associated with their location within or outside CpG islands. In the highly unmethylated group, 94% of CpG sites are located within a CpG island (Figure 2B). The highly unmethylated CpG islands also represent 79% of total CpG islands in the platform (Figure 3). Furthermore, of the highly methylated group, only 35% are within a CpG island (Figure 2C), representing a mere 5% of total CpG islands in the platform (Figure 3). On the other hand, the majority of highly methylated CpG sites are not associated with a CpG island (Figure 2C). This observation indicates that only a small fraction of the CpG islands is highly methylated in the endometrium, whereas most CpG islands are unmethylated. The data also indicate that hypermethylated CpG sites are mostly located outside CpG islands, in line with previous reports in somatic tissues that a small portion of CpG islands are methylated, whereas the majority remain unmethylated (19, 26) and most hypermethylated CpG sites are located outside CpG islands (19, 26) (see Supplemental Figure 2 for the overall distribution of DNA methylation levels in relation to location within or outside CpG islands).

Figure 3.

Distribution of CpG sites based on their location within or outside CpG islands. The middle pie chart represents the distribution of probes within or outside CpG islands on the 27K platform. Smaller pie charts depict the frequency of hypermethylation, hypomethylation, or intermediate methylation within each category. Light green and light red represent hypomethylated sites, dark green and dark red represent hypermethylated sites, and checkered green and checkered red represent intermediate methylated sites. β, β value; CGI, CpG island; In CGI, within CpG island; non-CGI, outside CpG island.

The remaining 29% of the CpG sites with variable and/or intermediate DNA methylation levels (Figure 2D) were subsequently analyzed to find differentially methylated CpG sites between the 3 cycle phases. In this group, 41% of the CpG sites are located within and 59% located outside CpG islands (Figure 2D), which corresponds to 16% of total CpG islands and 62% of all non-CpG islands, respectively (Figure 3).

Differentially methylated loci in different phases

We compared DNA methylation profiles and patterns in ESE vs PE, MSE vs ESE, and MSE vs PE to better understand epigenetic differences corresponding to distinct hormonal environments across the cycle, ie, the E₂-dominant proliferative phase and the P₄-dominant secretory phase.

We observed the most differences between MSE vs PE (66 CpG sites), followed by ESE vs PE (27 CpG sites), and the fewest changes between MSE vs ESE (22 CpG sites) (Figure 4). This is consistent with the unsupervised cluster analysis (Figure 1). Changes involved both gain and loss of methylation in a manner that roughly half of differentially methylated loci displayed gain of methylation and the other half loss of methylation in each comparison (Figure 4 and Supplemental Table 1). These changes were distributed across the genome, and we did not observe concentration of DNA methylation changes on specific chromosomes or on specific chromosomal hot spots in any of the comparisons (Supplemental Table 1). Of note are genes with known importance in endometrial biology in different phases of the cycle (Figure 4).

Figure 4.

Heat map of differentially methylated CpG sites between cycle phases. A, Differentially methylated CpG sites between ESE and PE. B, Differentially methylated CpG sites between MSE and ESE. C, Differentially methylated CpG sites between MSE and PE. Data are mean-centered and arranged top to bottom from low to high median β value in PE in panel A, in ESE in panel B, and in PE in panel C. Blue represents less methylation, and yellow represents more methylation compared with the mean. The location of CpG sites within or outside CpG islands is shown on the right. Black represents within and white represents outside CpG islands; gray represents NA. CGI, CpG island.

Assessment of the location of differentially methylated loci within or outside a CpG island revealed that methylation changes involved both CpG islands and non-CpG islands (Figures 4 and 5). Nearly 60% of the probes differentially methylated in ESE vs PE and in MSE vs ESE and 47% of the probes differentially methylated in MSE vs PE were located within a CpG island (Figure 5 and Supplemental Table 1).

Figure 5.

Distribution of differentially methylated loci between phases based on location within or outside a CpG island. The top panel shows the frequency of location within or outside a CGI for all differentially methylated loci between each comparison regardless of gain or loss of methylation, which is further divided into 2 pie charts based on gain or loss of methylation shown in the bottom panel. CGI, CpG island; Non-CGI, outside CpG island; In CGI, within CpG island.

In MSE vs PE, although the differentially methylated CpG sites are located almost equally within or outside CpG islands, we observe a different pattern when assessing the direction of the methylation changes. CpG sites that show gain of methylation in MSE compared with PE are mostly located within a CpG island (Figure 5), whereas CpG sites that show loss of methylation in MSE compared with MSE compared with PE are mostly located outside CpG islands (Figure 5). This is different from what is observed in MSE vs ESE or ESE vs PE (Figure 5); however, the number of differentially methylated CpG sites is smaller in these comparisons making conclusions less robust.

Differentially methylated genes between secretory and proliferative endometrium

Differentially methylated loci were surveyed to identify genes that differed in methylation status in the secretory phases (ESE and MSE) compared with those in the proliferative phase. As shown in Table 2, some loci including COL11A2, ID2, NFAM1, RUNX3, and ZNF57 were more methylated in MSE and ESE vs PE (Figure 6, panel I), whereas other loci including ADORA1, C21orf128, SMCR7, and TRPV3 were less methylated in MSE and ESE vs PE (Figure 6, panel II). Methylation changes of most of these loci were associated with gene expression changes (see below).

Table 2.

Selection of Differentially Methylated Probes Between Cycle Phases

Illumina Probe ID	Gene Symbol	Product	CpG Island	Median Difference (Absolute Value)
PE < ESE
cg03958979	NR2E1	Nuclear receptor subfamily 2; group E; member 1	Y	0.14
cg16823701	IFNA1	Interferon; α 1	N	0.16
cg12626411	PRG4	Proteoglycan 4	N	0.16
cg17901463	GSTM1	Glutathione S-transferase M1 isoform 1	Y	0.18
cg17568996	NFAM1	NFAT activation molecule 1 precursor	N	0.18
cg03171924	RUNX3	Runt-related transcription factor 3 isoform 2	Y	0.14
cg10143146	COL11A2	Collagen; type XI; α 2 isoform 3 preproprotein	Y	0.18
cg13055278	ID2	Inhibitor of DNA binding 2	Y	0.17
cg22074666	C21orf7	Chromosome 21 open reading frame 7	N	0.16
PE > ESE
cg11719784	ADORA1	Adenosine A1 receptor	N	0.16
cg18085517	TRPM1	Transient receptor potential cation channel; subfamily M; member 1	N	0.16
cg12728032	ELF5	E74-like factor 5 ESE-2b	Y	0.14
cg16706631	HIST1H4E	H4 histone family; member J	Y	0.14
cg19766460	C21orf128	Hypothetical protein LOC150147	N	0.27
cg26771272	SMCR7	Smith-Magenis syndrome chromosome region; candidate 7	Y	0.15
cg05790038	TRPV3	Transient receptor potential cation channel; subfamily V; member 3	N	0.15
cg08158289	KIAA0141	Hypothetical protein LOC9812	Y	0.16
cg00852964	VNN1	Vanin 1 precursor	N	0.18
cg00204262	ECH1	Peroxisomal enoyl-coenzyme A hydratase-like protein	Y	0.21
ESE < MSE
cg20022541	FAM181A	Hypothetical protein LOC90050	N	0.26
cg08085267	C17orf57	Hypothetical protein LOC124989	Y	0.19
cg20098118	UXT	Ubiquitously-expressed transcript isoform 2	Y	0.19
cg06074920	KRT34	Type I hair keratin 4	N	0.17
cg16242770	KRTAP17-1	Keratin associated protein 17-1	N	0.15
cg15928398	ST6GAL1	Sialyltransferase 1 isoform a	Y	0.15
cg00662775	TCEAL4	Transcription elongation factor A (SII)-like 4	Y	0.14
ESE > MSE
cg27202708	C1orf65	Hypothetical protein LOC164127	Y	0.16
cg05333568	C1orf65	Hypothetical protein LOC164127	Y	0.16
cg27371741	TDGF1	Teratocarcinoma-derived growth factor 1	Y	0.17
cg26511075	RANBP3 liter	Hypothetical protein LOC202151	N	0.17
cg19195724	SNORD109A		N	0.18
cg03882305	TRIM74	Hypothetical protein LOC378108	Y	0.22
cg13181019	MPP7	Palmitoylated membrane protein 7	N	0.23
cg01120761	CLEC4C	C-type lectin domain family 4; member C isoform 1	N	0.29
PE < MSE
cg16242770	KRTAP17-1	Keratin associated protein 17-1	N	0.21
cg04797323	SOCS2	Suppressor of cytokine signaling-2	Y	0.18
cg15928398	ST6GAL1	Sialyltransferase 1 isoform a	Y	0.21
cg10143146	COL11A2	Collagen; type XI; α 2 isoform 3 preproprotein	Y	0.14
cg03171924	RUNX3	Runt-related transcription factor 3 isoform 2	Y	0.14
cg00662775	TCEAL4	Transcription elongation factor A (SII)-like 4	Y	0.14
cg08209724	ZNF207	Zinc finger protein 207 isoform b	Y	0.19
cg15308737	ARSG	Arylsulfatase G	N	0.16
cg12111714	ATP8A2	ATPase; aminophospholipid transporter-like; class I; type 8A; member 2	Y	0.14
cg25569462	TRIML2	Hypothetical protein LOC205860	N	0.18
cg20022541	FAM181A	Hypothetical protein LOC90050	N	0.26
cg08634464	ZNF57	Hypothetical protein LOC126295	Y	0.14
cg20098118	UXT	Ubiquitously-expressed transcript isoform 2	Y	0.18
cg06074920	KRT34	Type I hair keratin 4	N	0.15
cg17568996	NFAM1	NFAT activation molecule 1 precursor	N	0.15
cg13055278	ID2	Inhibitor of DNA binding 2	Y	0.14
PE > MSE
cg11719784	ADORA1	Adenosine A1 receptor	N	0.14
cg21296602	TAF1D	Hypothetical protein MGC5306	N	0.16
cg25141490	IL17B	Interleukin 17B precursor	N	0.19
cg24512973	MUC1	MUC1 mucin isoform 1 precursor	Y	0.14
cg12743398	SULT1A2	Sulfotransferase family; cytosolic; 1A; phenol-preferring; member 2	N	0.17
cg12493906	MMP26	Matrix metalloproteinase 26 preproprotein	N	0.14
cg26771272	SMCR7	Smith-Magenis syndrome chromosome region; candidate 7	Y	0.17
cg19766460	C21orf128	Hypothetical protein LOC150147	N	0.24
cg15928132	CCKAR	Cholecystokinin A receptor	N	0.14
cg03870862	ZC3H3	Zinc finger CCCH-type domain containing 3	Y	0/14
cg03270204	DDR1	Discoidin domain receptor family; member 1 isoform b	N	0.15
cg04576021	HLA-DOB	Major histocompatibility complex; class II; DO β precursor	N	0.15
cg20993403	EPB41L1	Erythrocyte membrane protein band 4.1-like 1 isoform a	N	0.14
cg00035347	NT5C2	5′-Nucleotidase; cytosolic II	N	0.14
cg05790038	TRPV3	Transient receptor potential cation channel; subfamily V; member 3	N	0.15
cg03665457	HSD17B7P2	17-β-Hydroxysteroid dehydrogenase type VII-like isoform 2	Y	0.17
cg25527547	PLOD3	Procollagen-lysine; 2-oxoglutarate 5-dioxygenase 3 precursor	N	0.18
cg14862827	SUSD1	Sushi domain containing 1	Y	0.19
cg16173109	FLJ38379		Y	0.20
cg13181019	MPP7	Palmitoylated membrane protein 7	N	0.30
cg01120761	CLEC4C	C-type lectin domain family 4; member C isoform 1	N	0.24
cg27202708	C1orf65	Hypothetical protein LOC164127	Y	0.14

Figure 6.

Heat map of a selected number of loci differentially methylated between phases. I, Genes more methylated in MSE and ESE compared to PE. II, Genes less methylated in MSE and ESE than in PE. III, Genes more methylated in MSE than in PE and ESE. IV, Genes less methylated in MSE than in PE and ESE. Blue represents less methylation, and yellow represents more methylation compared with the mean.

An additional search identified genes whose methylation status remained unchanged throughout the proliferative (peak E₂) and early secretory (low P₄) phases but differed in the MSE (peak P₄) phase (ie, MSE vs ESE and in MSE vs PE). Loci with more methylation in MSE were FAM181A, KRT34, KRTAP17–1, ST6GAL1, TCEAL4, and UXT; loci with less methylation in MSE compared with that in the earlier phases of the cycle included C1orf65, CLEC4C, and MPP7 (Table 2 and Figure 6, panels III and IV; see Supplemental Table 1 for the full list of phase-dependent differentially methylated genes).

Relationship between DNA methylation status and gene expression

To investigate the relationship between DNA methylation status and gene expression in normal cycling endometrium, DNA methylation data from the Infinium platform were compared with the corresponding gene transcript levels on the Affymetrix U133 Human Plus 2.0 array (Tables 3 and 4), using the same (paired) endometrial samples. Although it is customary to merge the data of the different probes for individual loci, this practice precludes discriminating localization of methylation changes within loci and the potential effect on gene regulation. Therefore, in this analysis, data from each probe for every locus on the Illumina platform were matched to all the transcripts of the corresponding locus in the Affymetrix platform (see Materials and Methods).

Table 3.

Gene Expression and DNA Methylation Association Based on Location Within or Outside CpG Island

	ESE vs PE			MSE vs PE			MSE vs ESE
	Mean ρ	Total Probe	%	Mean ρ	Total Probe	%	Mean ρ	Total Probe	%
All negative association	−0.34	34	63	−0.32	82	58	−0.26	12	30
All positive association	0.25	20	37	0.29	60	42	0.26	28	70
Within CpG island (negative association)	−0.35	14	54	−0.24	33	51	−0.26	6	30
Within CpG island (positive association)	0.28	12	46	0.29	32	49	0.27	14	70
Outside CpG island (negative association)	−0.34	20	71	−0.37	49	64	−0.26	6	30
Outside CpG island (positive association)	0.22	8	29	0.30	28	36	0.25	14	70

Table 4.

Selected Loci With Strong and Moderate Correlation of Gene Expression and DNA Methylation

Illumina Probe ID	Affymetrix Identification	Gene Symbol	Gene Name	CpG Island	Spearman ρ
ESE vs PE
cg11719784	216220_s_at	ADORA1	Adenosine A1 receptor	No	−0.433
cg12728032	220625_s_at	ELF5	E74-like factor 5	Yes	0.300
cg17901463	215333_x_at	GSTM1	Glutathione S-transferase μ 1	Yes	−0.367
cg09879869	236004_at	MUDENG	Adaptor-related protein complex 5, μ 1 subunit	No	−0.700
cg17568996	230322_at	NFAM1	NFAT activating protein with ITAM motif 1	No	−0.533
cg03958979	207443_at	NR2E1	Nuclear receptor subfamily 2, group E, member 1	Yes	−0.317
cg12626411	206007_at	PRG4	Proteoglycan 4	No	−0.483
cg03171924	234928_x_at	RUNX3	Runt-related transcription factor 3	Yes	−0.417
cg01009664	206622_at	TRH	Thyrotropin-releasing hormone	Yes	−0.400
cg05790038	1555291_at	TRPV3	Transient receptor potential cation channel, subfamily V, member 3	No	−0.500
cg00186701	213122_at	TSPYL5	TSPY-like 5	Yes	−0.667
MSE vs PE
cg11719784	205481_at	ADORA1	Adenosine A1 receptor	No	0.627
cg00629217	223611_s_at	LNX1	Ligand of numb-protein × 1, E3 ubiquitin protein ligase	No	0.500
cg08477744	203417_at	MFAP2	Microfibrillar-associated protein 2	No	−0.700
cg12493906	220541_at	MMP26	Matrix metallopeptidase 26	No	−0.745
cg24512973	207847_s_at	MUC1	Mucin 1, cell surface associated	Yes	−0.655
cg17568996	243099_at	NFAM1	NFAT activating protein with ITAM motif 1	No	−0.527
cg00035347	236703_at	NT5C2	5′-nucleotidase, cytosolic II	No	−0.791
cg03171924	234928_x_at	RUNX3	Runt-related transcription factor 3	Yes	−0.455
cg21296602	222728_s_at	TAF1D	TATA box binding protein (TBP)-associated factor, RNA polymerase I, D	No	−0.545
cg06194808	235817_at	TMEM184A	Transmembrane protein 184A	No	−0.818
cg25569462	1552580_at	TRIML2	Tripartite motif family-like 2	No	−0.609
cg05790038	1555291_at	TRPV3	Transient receptor potential cation channel, subfamily V, member 3	No	−0.327
MSE vs ESE
cg25598083	202982_s_at	ACOT2	Acyl-CoA thioesterase 2	No	−0.714
cg05130485	207686_s_at	CASP8	Caspase 8, apoptosis-related cysteine peptidase	No	0.500
cg13181019	238778_at	MPP7	Membrane protein, palmitoylated7 (MAGUK p55 subfamily member 7)	No	−0.333
cg27371741	206286_s_at	TDGF1	Teratocarcinoma-derived growth factor 1	Yes	−0.452

Changes in DNA methylation show positive or negative correlation with changes in gene expression

Because the relationship between DNA methylation and gene expression is not linear (16), the Spearman correlation was used to examine this relationship. Hypermethylation of CpG dinucleotides located in a CpG island at the 5′ region of a gene has been shown to be associated with gene silencing, whereas hypermethylation in the body of the gene is suggested to be associated with gene activation (19) (see Discussion). Therefore, the relationship between changes in DNA methylation and changes in gene expression was assessed for both positive and negative associations as well as the location of the interrogated CpG in the context of CpG islands (Tables 3 and 4).

We compared the frequency of loci that showed a negative association vs those with a positive association. In ESE vs PE, the percentage of loci with negative association (63%) was higher than that with positive association (37%), and negative association was also stronger (mean ρ = −0.341 vs mean ρ = 0.252, respectively) (Table 3). Similar results were observed in the MSE vs PE comparison (Table 3). However, in MSE vs ESE, the percentage of loci showing a negative association was less than that of those with a positive association (30% vs 70%, respectively); however, associations in each direction were of similar strength (ρ = −0.262 vs ρ = 0.255) (Table 3).

Association of DNA methylation with gene expression based on location within or outside a CpG island

The association of DNA methylation with gene expression was further analyzed by dividing differentially methylated CpG sites into 2 groups: within a CpG island or outside a CpG island (Table 3).

For ESE vs PE and MSE vs PE, equal numbers of differentially methylated loci showed positive or negative association with gene expression, when located within a CpG island. However, when outside a CpG island, most differentially methylated loci showed a negative association with gene expression (Table 3). In MSE vs ESE, most differentially methylated loci showed a positive association with gene expression, either within or outside a CpG island (Table 3).

In summary, overall, in the proliferative vs each of the secretory subphases, equal numbers of differentially methylated loci located within a CpG island had positive or negative correlation with gene expression, whereas a majority of the differentially methylated sites located outside a CpG island had a negative association. This result is in line with current understandings that changes in CpG island methylation, sometimes, but not always, result in negative association with gene expression, whereas changes in DNA methylation in non-CpG island CpG sites, mostly at the 5′ positions of the genes are usually negatively associated with gene expression (19, 26).

Loci with high correlation between DNA methylation and gene expression

In response to E₂ and P₄, human endometrium undergoes global transcriptomic changes involving genes in various signaling pathways, accounting for many of the dynamic histologic and functional events observed across the cycle (31, 43). A goal herein was to investigate the biological importance of genes whose DNA methylation changes show a correlation with changes in gene expression. Table 4 shows a partial list of the genes with strong and moderate correlation of the changes in DNA methylation and gene expression in the different phase comparisons (see Supplemental Table 2 for the full gene list).

To better understand the biological relevance of these genes, we interrogated their involvement in molecular pathways and their potential functions within endometrium. Pathway analysis using the DAVID database (42) revealed principal pathways represented by genes with correlation between DNA methylation and gene expression changes in ESE vs PE and included regulation of transcription (ID2, ELF5, NFAM1, IFI16, NR2E1, and RUNX3), regulation of apoptosis (IFI16, ADORA1, NR2E1, and RUNX3), and cell proliferation and regulation of RNA metabolic process (ID2, ELF5, IFI16, NR2E1, and RUNX3). Main pathways associated with changes in MSE vs PE are transmembrane (MUC1, DDR1, ST6GAL1, DRD5, TRPV3, CLIC6, TMEM184A, NFAM1, ADORA1, HLA-DOB, and ST6GALNAC1) and extracellular matrix and secreted proteins (MUC1, DDR1, ST6GAL1, MMP26, MFAP2, and COL11A2). Main pathways related to changes in MSE vs ESE include blood vessel morphogenesis (TDGF1, CASP8, and WT1). Several of these genes are also known to be important either in endometrial cyclicity, preparing for receptivity to embryonic implantation, and apoptosis or are endometrial cell specific (see Discussion), including RUNX3, important in TGF-β signaling; GSTM1, important in the response to oxidative stress; members of the matrix metalloproteinase family (MMP26); nuclear receptor subfamily 2 (NR2E1); and others (see Table 4 for a selected list by cycle phase and Supplemental Table 2 for the full gene list).

DNA methylation changes in eSFs in different hormonal milieux

To gain insight beyond global endometrial tissue changes in DNA methylation in different hormonal milieux, we investigated the DNA methylation status of human eSFs treated with E₂ + P₄ vs E₂ in vitro (mimicking secretory vs proliferative phases, respectively, in vivo). Our preliminary analysis revealed DNA methylation changes for a number of genes, corresponding to either gain or loss of methylation similar to the observed changes in the endometrial tissue across the phases (Supplemental Table 3). Several of these genes also showed association of DNA methylation with changes in gene expression in the endometrium (Supplemental Table 2). These data support our whole-tissue observations of changes in DNA methylation across the cycle in different hormonal milieux.

Discussion

Dynamic DNA methylation and association with cyclic ovarian hormones

Endometrial DNA methylation changes during the menstrual cycle

The finding of dynamic changes in DNA methylation in human endometrium in different hormonal milieux is unique. Although extensive epigenetic reprogramming occurs in the preimplantation embryo (18, 20, 44) and during gametogenesis in primordial germ cells (45, 46), dynamic genome-wide changes in DNA methylation in normal adult tissues have been reported only in brain (47–49) and hematopoietic stem cell differentiation (50) and now in cycling human endometrium. This is in addition to DNA methylation changes that occur in cancer (17, 51, 52), infertility (39), aging (53, 54) and in response to environmental factors (55). How the DNA methylation changes reported here occur within the relatively short time span of a human menstrual cycle is not clear and warrants further investigation in longitudinally collected samples.

Epigenetic mechanisms and modifications in human endometrium

Various enzymes important in epigenetic processes have been reported in human endometrium. These include constitutive expression of histone deacetylase inhibitor 3, modest elevation of histone deacetylase 2 in the secretory phase (56) and varying expression of DNA methyltransferases (DNMT1, DNMT3A, and DNMT3B) during the cycle (8, 27) and with E₂ and P₄ treatment of endometrial explants (57). In addition, levels of immunoreactive 5-methylcytosine vary in glandular epithelial and stromal cells across the cycle (58). Herein, we found that the DNA methylation profiles of human endometrium differ in different phases of the cycle, resulting in the segregation of proliferative from secretory phase samples by unsupervised cluster analysis of the differentially methylated loci. Furthermore, the greatest DNA methylation differences are between the proliferative (peak E₂) and midsecretory (peak P₄) phases, consistent with periods of greatest changes in gene expression and suggesting a relationship between DNA methylation and endometrial gene expression across the cycle. In addition, our preliminary in vitro experiments corresponding to hormonal milieux of peak E₂ and peak P₄ show DNA methylation changes of several of the same genes. The fact that these epigenetic modifications occur in human endometrium during its hormonally regulated cyclic activity further suggests a relationship between steroid hormone action and endometrial DNA methylation dynamics (see below).

Steroid hormones and epigenetic changes in human endometrium

A major characteristic of human endometrium is that it undergoes cellular proliferation, differentiation, and regeneration on a cyclic basis. The question arises as to whether changes observed in the endometrial methylome occur in response to ovarian hormones or whether the DNA methylation changes facilitate endometrial responsiveness to these hormones, or perhaps it is a combination of both.

A growing body of literature supports that the epigenome can be affected by hormones or their disruption, such as by estrogen agonists. For example, bisphenol A (BPA), a weak estrogen agonist (59) induces hypomethylation and increased expression of the Agouti gene (60) in mice exposed prenatally. Mouse models of prenatal diethylstilbestrol exposure and genital tract neoplasia show hypomethylation and overexpression of EGF and SRF, estrogen-sensitive LTF (61), and HOXA10 hypermethylation, resulting in altered expression in the developing mouse uterus (62). Diethylstilbestrol, an estrogen agonist, causes vaginal clear-cell adenocarcinoma in adult women exposed in utero (63) and has been associated with epigenetic modifications in uterine tissues. Methoxychlor, another estrogenic endocrine disruptor, affects DNA methylation and alters reproductive senescence and estrous cyclicity in mice exposed prenatally (64).

In contrast to steroid hormones affecting the epigenome as part of their mechanism of action, it is possible that epigenetic changes could modulate steroid hormone action per se. For example, during early mouse postnatal development, Esr1 mRNA expression decreases within the cortex of the brain, coinciding with progressive gain of methylation in several Esr1 gene promoters (65). In addition, the methyl-DNA binding protein, MeCP2, is associated with the Esr1 promoter as it becomes methylated, suggesting a role for DNA methylation in Esr1 mRNA suppression in the postnatal brain (65). The preoptic area and hypothalamus are organized by early estradiol exposure (48) and methylation of Esr1 is altered by estradiol (49). Moreover, the brain can undergo epigenetic changes during development (47) and across the lifespan in response to varying hormonal milieu (47, 48). In addition, hypomethylation of an ESR2 promoter CpG island was shown to result in higher mRNA and protein levels in endometriosis (28).

Taken together, these findings suggest a dynamic interplay between DNA methylation and a changing hormonal milieu. We postulate that, in human endometrium, hormonal changes affect DNA methylation and changes in DNA methylation patterns play a crucial role in endometrial responsiveness to hormonal changes. This hypothesis is of great interest in our laboratory and awaits further experimental testing and validation.

DNA methylation and gene expression in human endometrium

Global methylation profile and pattern in the endometrium

Besides the small fraction of CpG sites with dynamic DNA methylation levels, we observed that across all phases, most CpG islands are unmethylated and only a very small percentage are hypermethylated. In addition, most of the hypermethylated loci are in non-CpG islands. This observation is in line with somatic tissue global DNA methylation patterns of hypermethylation of non-CpG island CpG sites and a small fraction of CpG islands, with most CpG islands remaining unmethylated (19, 66). Furthermore, there are tissue-specific CpG island methylation and tissue-specific unmethylated regions (26). Understanding which fractions of these highly methylated CpG islands or the unmethylated regions in our data are shared with other somatic tissues and which are endometrium specific requires further investigation and comparison with other somatic tissue methylomes and remains to be elucidated.

In addition to tissue-specific CpG islands or unmethylated regions, partially methylated domains (PMDs) (67) usually covering long genomic regions that could mark transcriptionally repressed regions, including genes with methylation levels of 0.2 ≤ β ≤0.8, are also potentially tissue specific (68). Although there is evidence suggesting they may be found in fibroblasts (69, 70), they may not be present in all adult tissues (71). Whether some of the repressed genes observed in endometrium with intermediate methylation are in fact associated with PMDs specific to endometrium is yet to be determined, using in-depth sequencing methods that investigate large regions of the genome. It is possible that the localization of the genes within the PMDs may contribute to their down-regulation during the menstrual cycle.

Mechanisms of de novo methylation and demethylation in endometrium

De novo methylation.

We observed specific de novo methylation in different cycle phases. In other somatic cells, 2 main mechanisms are involved in de novo methylation of targeted CpG islands: (1) the CpG islands must contain signals to direct the methylation, eg, mediated by histone methyltransferases that recruit DNA methylases (72–74); or (2) removal of transcription factors and/or nucleosomes containing histone 3 lysine 4 trimethylation (H3K4me3) in response to hormonal changes that would otherwise protect against de novo methylation. The latter is based on data suggesting that many unmethylated CpG islands near transcription start sites contain transcription factor binding sites (66) and nucleosomes containing H3K4me3 (75, 76) that may be inhibiting DNA methylation. Methylation of non-CpG island regions can have a direct impact on binding to their recognition sequences of transcription factors that play crucial roles during the menstrual cycle.

Demethylation.

Besides de novo methylation of CpG islands, we also observed demethylation. In the case of tissue-specific demethylation, it has been suggested that tissue-specific transacting factors (77, 78) and transcription factor binding may play a role (79, 80). Demethylation of genes is usually associated with gene activation, as was observed in our analysis, although its causal role is not clear. In the endometrium, phase-specific transcription factors may be involved in this process. Demethylation with regard to hormonal changes has been reported in enhancers, resulting in their subsequent activation (79, 81). Enhancers, located at variable distances to transcription start sites (TSSs), are generally CpG-poor with variable levels of methylation (82, 83) and play crucial roles during development. The relationship between methylation, transcription factor binding, and enhancer activity is complex and not well understood. Further investigation is anticipated to understand better how ovarian hormones affect transcription factor binding and enhancers/insulators related to changing levels of DNA methylation at various locations of a transcriptional unit in human endometrium.

Relationship of DNA methylation with gene expression

To better understand the role of DNA methylation in cycling endometrium, we investigated its relationship to gene expression changes in paired samples and found that for some genes, increased DNA methylation correlated with decreased levels of transcription, but as observed before (84), for a number of other loci there was a positive correlation between methylation and gene expression. DNA methylation of CpG islands at the promoter is generally, but not always, assumed to be negatively associated with gene expression. We also observed that methylation at CpG islands showed negative, positive, or no association with gene expression. Besides changes in CpG islands, we observed DNA methylation changes at non-CpG island regions. Unlike CpG islands at the 5′ end of genes, substantial fluctuations occur in these regions; however, usually an inverse relationship between methylation of the non-CpG island loci and gene expression exists (19). Again, we observed that most of these changes are negatively associated with gene expression, particularly when changes in the proliferative vs secretory phases are compared.

The precise role of DNA methylation in gene expression control is not fully understood. Although strong evidence suggests that there is no transcription initiation when CpG islands at the TSS are methylated (85), it is not clear which comes first, methylation or silencing. It has been suggested that DNA methylation serves to lock the silent state (19), but there is evidence suggesting it may also be involved in initiation of silencing (50). This is important in the interpretation of our results. Given the relatively short time period between the phases of the cycle and the higher frequency of gain of methylation at the CpG islands than of non-CpG islands in our study, it is possible that these changes may play a more instructive role in gene silencing than just serving as a locking tool for already repressed transcription induced by other mechanisms.

Biologically important genes in cycling endometrium

Differentially methylated genes with functional importance

We observed herein that DNA methylation changes affecting gene expression involve genes in signaling pathways including regulation of transcription and proliferation and transmembrane proteins. Several of these genes, such as RUNX3, DDR1 (NEP), MMP26, and MUC1, have been shown by many groups to play significant roles in endometrial biology, function, and embryonic implantation. Others such as NR2E1 and GSTM1 are also known to be important in biological processes with potential significance in endometrium, which remain to be elucidated. Several of the genes also show changes in DNA methylation when eSF cells are treated with E₂ + P₄ vs E₂ (Supplemental Table 3). These genes are discussed below.

RUNX3.

RUNX3 (runt-related transcription factor 3) is involved in TGF-β signaling (86) important in cycling endometrium (87). Knockout of Runx3 in mice resulted in atrophic uteri compared with those in the wild type with less developed endometrial layer, uterine glands, thinner stromal layer, and no E₂-induced epithelial proliferation (88). The lower mRNA levels of TGF β1 and TGFβ3 and lack of E₂-induced EGF expression in stromal cells may be associated with the suppressed E₂-dependent epithelial proliferation and suggests a regulatory role of Runx3 in the E₂-induced uterine growth (87). In wild-type mice, Runx3 showed differential expression during early pregnancy with low expression on days 1 to 4 and increased expression in the stromal cells surrounding the implanting blastocyst on day 5, followed by high expression in the decidual cells from days 6 to 8. Similar results were observed in artificial decidualization and activated implantation in delayed uterus, suggesting an important role for Runx3 during mouse implantation and decidualization (89). Multiple transcripts of this gene exist, for some of which the CpG island is located before the TSS, whereas for others it is located within the gene body (analysis on the UCSC genome database). Herein, we observed that RUNX3 is hypermethylated in the secretory phase in both ESE and MSE. Of the 3 transcripts represented on the Affymetrix platform, the CpG island hypermethylation was associated with decreased gene expression for 1 transcript and showed no association with the others. We also observed increased methylation of RUNX3 in eSFs treated with E₂ + P₄ (decidualized) vs E₂, confirming our initial observation. Hypermethylation and the subsequent decrease in transcription of RUNX3 may be involved in the suppression of endometrial epithelial/stromal cell proliferation in secretory phase endometrium.

DDR1.

DDR1 (discoidin domain receptor discoidin domain receptor tyrosine kinase 1; NEP) is suggested to play an important role in regulating endometrial cellular proliferation. It cleaves and inactivates ET-1 (endothelin-1), which stimulates phosphorylation of Akt and DNA synthesis in endometrial stromal cells via the ET(A) receptor and phosphatidylinositol-3 kinase signaling pathways (90). It is expressed in decidualized eSFs but not in endometrial epithelium or nondecidualized stromal fibroblasts (90). Our data show that it is more methylated in whole endometrial tissue in PE and becomes less methylated in MSE, resulting in increased expression in MSE. Furthermore, our in vitro results show higher methylation when endometrial stromal fibroblasts are treated with E₂ and lower methylation with E₂ + P₄ treatment, consistent with the in vivo data.

MMP26.

MMP26 (matrix metalloproteinase-26) activates MMP9, which exhibits proteolytic activity on a large number of extracellular and basement membrane proteins (91, 92). Its expression pattern begins in the proliferative phase, peaks at midcycle, and then decreases to nondetectable levels in the late secretory and menstrual phases (93). In vitro expression of MMP26 in endometrial explants requires both E₂ and P₄. In endometrium of pregnant rhesus monkeys (94) the average endometrial levels of MMP26 mRNA (and protein) are high on day 12 of pregnancy but decrease significantly on days 18 and 26 with intense localization in the glandular epithelium on day 12 and in the walls of spiral arterioles adjacent to the implantation site on day 26. The spatiotemporal expression in early pregnancy suggests an important role of MMP26 in tissue remodeling regulation of glandular epithelium and spiral arteries. The MMP26 CpG site is located at a non-CpG island promoter region and is demethylated in MSE vs PE, with a strong negative correlation with its gene expression. Our in vitro eSF data confirm these in vivo results. This is an example of a biologically important gene that shows that demethylation of a non-CpG island promoter is highly associated with increased gene expression, as discussed above.

MPP7.

MPP7 (membrane protein, palmitoylated 7) is a member of the membrane-associated guanylate kinases, important in cell-cell contact. MPP7 plays a role in epithelial cell polarity and tight junction formation in endometrium and other tissues (95). The MPP7 interrogated CpG site is not located in a CpG island and becomes less methylated in MSE vs PE, and this change is negatively associated with its gene expression.

MUC1.

MUC1 (mucin 1) is a glycoprotein in human endometrium that interacts with ESR1, stabilizing it and stimulating ESR1-mediated transcription (96). Suppression of MUC1 inhibits expression of EGFR (97). In vivo and in vitro studies have shown that P₄ increases MUC1 expression in humans (98). The importance of MUC1 expressed in endometrial epithelium has been extensively studied and shown to be important in endometrial receptivity and embryo attachment (99) with abnormalities in infertile women with polycystic ovary syndrome or endometriosis (100). Herein, the MUC1 CpG island is less methylated in MSE vs PE, which is strongly correlated with an increase in gene expression. We did not detect changes in DNA methylation in our in vitro eSF study, suggesting that the changes observed in whole endometrial tissue reflect changes in the epithelial compartment, consistent with the known expression of MUC1 in endometrial epithelium.

NR2E1 and GSTM1.

NR2E1 (nuclear receptor subfamily 2, group E, member 1) and GSTM1 (glutathione S-transferase mu 1) are examples of genes with known biological importance, but their exact function in the endometrium remains to be elucidated. NR2E1 affects many downstream genes such as zinc finger transcription factor BCL11A (CTIP1/Evi9) (101), and it recruits lysine-specific demethylase 1 (LSD1), an important transcriptional coregulator that modulates histone methylation (102, 103). The CpG island located before the TSS of NR2E1 is hypermethylated in ESE vs PE and is negatively associated with its gene expression. We also observed higher methylation in eSF treated with E₂ + P₄ vs E₂ confirming the in vivo results.

GSTM1 functions in detoxifying carcinogens and products of oxidative stress. Interestingly, glutathione S-transferases are cycle dependent in human endometrium (31). Alterations in GSTM1 methylation result in changes in gene expression and have been associated with various cancers (104, 105). Our data showed that the GSTM1 CpG island located at the TSS is more methylated in the secretory vs the proliferative phase, and this was correlated with decreased gene expression in the secretory phase. Because there are multiple GSTMs in human endometrium, the physiologic significance of GSTM1 in the context of its detoxification function in proliferative vs secretory phase tissue remains to be determined.

Important endometrial genes reported in other transcriptomic studies

Many highly differentially expressed genes across the cycle in our previous transcriptome studies (31, 40, 106) were not detected as differentially methylated in the current study. There are several potential explanations for this observation. First, many genes with differential gene expression are the direct or downstream targets of steroid hormones, which may not involve DNA methylation. Second, it is reported that the higher the level of expression, the less likely is a gene's promoter CpG island to become de novo methylated and vice versa (107, 108). Third, cellular heterogeneity may have a greater impact on DNA methylation data than it has on gene expression data. This occurs because a given CpG dinucleotide may either be completely methylated or unmethylated in any given cell, varying widely among its own population, as well as across different cell types, which leads to detection of less robust methylation changes in the whole tissue. Last, in addition to DNA methylation, various other mechanisms are known to regulate gene expression through different pathways and other epigenetic mechanisms. Therefore, we would not necessarily expect to find DNA methylation changes in the same genes reported in previous transcriptomic studies.

Limitations of this study

There are limitations and caveats to our study, including the small sample size, cross-sectional study design, cellular heterogeneity of tissue samples, and Illumina platform probe bias with higher representation of (1) loci reported to be differentially methylated in cancer and during development and (2) CpG islands but not loci with variable partial methylation. Despite these shortcomings, the observed DNA methylation changes pave the way to expand our analysis using an unbiased comprehensive genome-wide DNA methylation platform in a larger number of samples collected longitudinally and of isolated cell types to better understand the extent and role of epigenetic changes in cycling human endometrium and the role of steroid hormones in these changes in normal endometrial physiology and in endometrium-related disorders.

In summary, the data presented herein provide, to our knowledge, the first report of dynamic DNA methylation changes in human endometrium in different hormonal milieux. These include both gain and loss of methylation either within or outside CpG islands, with proportions of DNA methylation within vs outside a CpG island varying across cycle phases. Several genes, such as RUNX3, DDR1, MUC1, MMP26, and MPP7, with predicted biological relevance in endometrial biology, show DNA methylation changes associated with gene expression changes, underscoring the importance of epigenetic regulation of gene expression in different hormonal milieux in human endometrium. In addition to advancing our understanding of molecular changes in cycling endometrium, the current data are interesting from a broader biological perspective, as the cyclic changes in DNA methylation in normal adult endometrium described herein add to the limited number of methylome changes documented in normal adult human tissues. The extent and level of these changes in the different cell types within human endometrium remain to be elucidated and are under investigation in our laboratory. The data raise the possibility that DNA methylome modifications by exogenous hormonal exposures may contribute to altered endometrial cellular function and associated endometrial disorders.

Additional material

Supplementary data supplied by authors.