DNA Methylation and Demethylation Underlie the Sex Difference in Estrogen Receptor Alpha in the Arcuate Nucleus

A transient exposure to testosterone during perinatal life underlies the emergence of many sex differences in the mammalian brain [1, 2]. For example, male mice have more neurons expressing calbindin and vasopressin in several forebrain regions [3, 4], while females have more neurons expressing estrogen receptor (ER) α [5, 6]. In some cases, sex differences in cell number are due to differential cell death in males and females [7, 8], but other differences persist even when developmental cell death is prevented, for example [9, 10]. In the latter case, sexual differentiation is often a process of establishing stable, sex-specific patterns of gene expression in neurons, that is, the differentiation of neurochemical phenotype. Although sex differences in neurochemical phenotype are common, the underlying molecular processes are largely unknown. Several recent findings suggest that epigenetic modifications, which play prominent roles in the differentiation of cell type throughout the body, may underlie sexual differentiation of cell phenotype in the brain [11-15].

The mediobasal hypothalamus contains prominent populations of ERα-expressing neurons in the arcuate (ARC) nucleus and ventrolateral area of the ventromedial nucleus (VMHvl) [5, 6, 16]. ERα neurons in these regions play important, and often sex-specific, roles in the control of feeding behavior, energy homeostasis, sexual behavior, and the regulation of bone density [17-19]. We recently found that inhibiting DNA methylation in newborn mice reduces or eliminates the usual female bias in the number of neurons expressing ERα in the VMHvl [13, 15].

DNA methylation is catalyzed by DNA methyltransferases (DNMTs) and usually occurs at the fifth carbon position of cytosine residues followed by a guanine (CpG sites) [20]. DNA methyl marks (5-methylcytosine, 5mC) can be quite long-lasting [21], but can also be removed through a series of oxidative steps catalyzed by ten-eleven translocation enzymes (TETs). In the first step, 5mC is converted to 5-hydroxymethylcytosine (5hmC), which is both an intermediary in demethylation and an independent epigenetic mark in its own right [22]. Although a growing number of exceptions are reported [23-28], 5mC is typically associated with gene repression, whereas 5hmC is most often associated with gene expression [25, 29-31]. Thus, the emergence of a particular cell phenotype in part depends on the balance between methylation (controlled by DNMTs) and demethylation (controlled by TETs) at specific genes.

The methylation landscape of the brain changes dynamically during the first few weeks (rodents) or years (humans) of life [32]. We recently found peak expression and enzyme activity of DNMTs and TETs in the mouse hypothalamus during the first postnatal week [33], which coincides with the critical period for sexual differentiation of the brain and behavior. Furthermore, we found sex differences in Tet gene expression in the neonatal hypothalamus, with greater expression of Tet2 and Tet3 in males [33]. Here, we manipulated DNMTs and TETs neonatally to test the hypothesis that DNA methylation and/or demethylation underlie the development of ERα cell number in the ARC. We also examined DNA methylation of Esr1 promoter regions in the ARC and neighboring VMHvl to determine if sex differences in ERα expression correlate with differences in CpG methylation.

C57BL6/J mice were bred in our vivarium and checked daily for pups. Mice were housed in cages with corn cob bedding in a room maintained at 22°C with food and water available ad libitum. Animal procedures were performed in accordance with the National Institutes of Health animal welfare guidelines and were approved by the Georgia State University Institutional Animal Care and Use Committee.

The brains used in this portion of the study were the same as those from a prior study reporting on the effects of neonatal inhibition of DNA methylation on ERα in the POA and VMHvl [15]. Male (n = 23) and female (n = 19) controls received a subcutaneous injection of 25 µL of peanut oil on the day of birth and following day, while androgenized females (n = 21) received testosterone propionate (100 µg in 25 µL peanut oil; Millipore Sigma, St Louis, MO, USA). Half of the animals in each of the 3 hormonal groups also received intracerebroventricular (ICV) injections of either vehicle or a DNMT inhibitor, zebularine, on the day of birth (postnatal day [P] 0) and P1. Vehicle (500 nL per hemisphere; 10% dimethyl sulfoxide in 90% saline) and zebularine (300 ng in 500 nL vehicle per hemisphere; Millipore Sigma) were injected into the lateral ventricles of cryoanesthetized pups, as described previously [15]. Zebularine is a cytosine analog, which incorporates into DNA and traps DNMTs [34]. In postmitotic cells such as neurons, cytosine analogs are incorporated via the base excision repair pathway, a mechanism that replaces entire cytosine bases during active methylation/demethylation cycles [35-37]. All animals were tattooed for identification and euthanized prior to puberty at P25. This age was chosen because it allowed us to examine relatively long-lasting effects of the neonatal inhibition of DNA methylation in the absence of confounding effects of postpubertal gonadal hormones.

Tet mRNA expression in the hypothalamus is elevated during the first postnatal week, and sex differences in Tet2 and Tet3 expression (male > female) are observed during this time [33]. To test the role of demethylation in the development of the sex difference in ERα expression in the ARC and VMHvl, we used small interfering RNAs (siRNAs) to reduce Tet expression in males to female-like levels. ICV injections were performed on cryoanesthetized pups as described above. On P5, females (n = 7) received a control injection of nontargeting RNAs in delivery media, while males received either the control (n = 15) or a cocktail (n = 14) of SMARTPool siRNAs targeted against Tet2 and Tet3 (400–500 pmol; Accell; Horizon Discovery, Cambridge, UK). Each SMARTPool contained a mixture of 4 siRNAs targeting the same mRNA. Previous studies using Accell siRNAs report a neuron-selective knockdown that is maximal 2–4 days after a single ICV injection [38, 39]. The brains of half of the control- and siRNA-treated males were collected 2 days after injection (P7), and punches were taken of the anterior and posterior hypothalamus to confirm gene knockdown by RT-qPCR, as described in Ref. [33]. The remaining brains were harvested at P25.

Brains were fixed in 5% acrolein in 0.1 M phosphate buffer for 24 h and then transferred to 30% sucrose for several days. Brains were frozen-sectioned into 4 series (zebularine-treated animals) or 2 series (siRNA-treated animals) of 30 μm thickness and stored in cryoprotectant (30% sucrose, 30% ethylene glycol, and 1% polyvinylpyrrolidone in 0.1 M phosphate buffer) until immunohistochemical staining. One series of zebularine-treated tissue and one series from the siRNA experiment were stained for ERα (rabbit anti-ERα, 1:20,000; EMD Millipore, Billerica, MA, USA), as described previously [15]. For the zebularine experiment, images through the ARC were captured using Stereo Investigator software (MBF Bioscience, Williston, VT, USA), and ERα labeling was quantified using ImageJ (version 1.52/1.53; National Institutes of Health, Bethesda, MD, USA). The ARC was identified using well-defined landmarks including the shape of the third ventricle and presence of the median eminence. We drew a contour around the ARC in each section, and pixels above background per square micrometer were determined using the Shanbhag algorithm. The 4 sections through the ARC with the greatest labeling were summed for each animal. To validate this method, we also manually counted ERα cells in a subset of male and female animals using Stereo Investigator and confirmed a very similar pattern of results. For the siRNA experiment, images through the ARC and VMHvl were captured, contours were drawn around each brain region, and pixels above background per square micrometer was determined from the 7 (ARC) or 5 (VMHvl) sections with the greatest labeling.

To test whether our treatments had a generalized effect on cell types in the ARC, additional series of sections were labeled for the detection of kisspeptin (rabbit anti-kisspeptin, 1:2,000; EMD Millipore) or calbindin (mouse anti-calbindin-D28k, 1:20,000; Millipore Sigma), as previously [15]. The calbindin-labeled tissue included only males and females in the zebularine experiment (i.e., testosterone-treated females were not included) and was counterstained with thionin to aid in visualization. Brains from a separate cohort of P1 and P25 mice (P1: n = 7 males, 6 females; P25: n = 8 males, 10 females) were also collected and immunostained for ERα to determine if sex differences were present in untreated animals.

To determine whether zebularine treatment altered developmental cell death, brains from a separate cohort of vehicle- and zebularine-treated male (n = 11) and female (n = 9) pups were collected 6 h after the last injection on P1 and immunolabeled for activated caspase-3 (AC3), as previously described [40]. We drew contours around the ARC and counted the number of AC3-positive cells bilaterally. Cell death density was obtained by dividing the number of AC3-positive cells by the area sampled and is expressed as AC3 cells per square millimeter.

One series of brain sections from vehicle- and zebularine-treated females was stained with thionin to determine whether zebularine altered the total number of cells in the ARC at weaning. Because zebularine specifically affected ERα in the ARC of females, only females were included in this analysis. Unbiased stereology using the optical fractionator probe was performed using Stereo Investigator software (MBF Biosciences). Contours were drawn around the ARC at low power, and counts of cells exhibiting a neuronal morphology were performed with a ×100 oil objective using a 324 μm² counting frame and 4,900 μm² sampling grid. The Gundersen coefficient of error for counts ranged from 5 to 7% [41].

A separate cohort of male (n = 84) and female (n = 84) pups received ICV injections of vehicle or zebularine on P0 and P1. Forty-eight vehicle-treated pups (24 of each sex) were euthanized ∼4 h after ICV injection on P1. The 120 remaining animals were collected on P25, and body weight was recorded for about half of these animals (n = 33 males and 35 females, randomly selected). Brains were removed, flash-frozen in 2-methylbutane cooled to −20°C, and cut on a cryostat to the level of the ARC and VMHvl. Punches were taken from the ARC and VMHvl and kept at −80°C until processing. Because a punch of the ARC damages the adjacent VMHvl, and vice versa, separate animals were used for each brain region. Punches from 3 animals of the same sex were pooled per sample. Bisulfite conversion followed by pyrosequencing was performed by EpigenDx (Hopkinton, MA, USA) to obtain single-nucleotide resolution of modified cytosines in Esr1. Specifically, we assessed the modification of cytosines in 16 CpG sites in the untranslated regions of exons A and C. Because bisulfite sequencing cannot distinguish between 5mC and 5hmC, we refer to the results of this analysis as “total methylation” (as in Refs. [25, 42]). Methylation of these regions was previously shown to associate with changes in Esr1 gene expression in the mouse brain during development or after bisphenol A exposure [43, 44].

Independent, 2-tailed t tests were used to compare ERα labeling in the ARC of untreated males and females on P1. Two-way ANOVAs (group-by-treatment) were used to compare body weight, and ERα, kisspeptin, or calbindin labeling at P25 in animals that received either vehicle or zebularine neonatally. Results of pyrosequencing were analyzed using a 2-way repeated measures ANOVA for P1 samples, with sex as the between-subjects factor and CpG site as the repeated measure. Pyrosequencing at P25 was analyzed with a 3-way repeated measures ANOVA with sex and zebularine treatment as between-subjects factors and CpG site as the repeated measure. A significant main effect of sex was followed by post hoc tests of individual CpG sites. Developmental changes in mean methylation across CpG sites were analyzed with 2-way ANOVAs (age-by-sex). Effect of Tet siRNA treatment on ERα and kisspeptin expression was analyzed by 1-way ANOVA. Fisher’s LSD and Tukey’s post hoc tests were used where appropriate, and all analyses were performed using Prism version 9 (GraphPad software, San Diego, CA, USA).

A female bias in estrogen binding, ERα mRNA expression, or ERα immunolabeling in the ARC of rodents has previously been reported, but is not seen in all studies or at all ages [5, 6, 16, 45-47]. We observed prominent ERα labeling in the ARC of untreated animals that appeared confined to cell nuclei (Fig. 1). Among untreated mice, we found no sex difference in ERα immunolabeling at P1, but females had more labeling than males at weaning (P25; t₁₆ = 3.64, p = 0.002; Fig. 1). In a preliminary study, we also observed a greater expression of Esr1 mRNA in the ARC of control females (n = 4) than of males (n = 4) at P25 using single-molecule in situ hybridization (online suppl. Fig. 1; see www.karger.com/doi/10.1159/000519671 for all online suppl. material).

We next compared ERα in the ARC at P25 in the males, females, and females treated neonatally with testosterone from our previous study to examine the hormonal control of this sex difference [15]. Control females again had greater ERα labeling than did control males (p < 0.0001 for vehicle female vs. vehicle male; Fig. 2). Neonatal treatment of females with testosterone markedly reduced ERα labeling at P25 (p < 0.0001 for vehicle female vs. vehicle female + T), indicating that the sex difference is due to programming effects of testosterone. Our testosterone treatment, in fact, hypermasculinized females relative to vehicle-treated males (p < 0.01).

To test the role of DNMTs in the development of ERα labeling in the ARC, we compared animals that were treated neonatally with zebularine or vehicle. There was no main effect of neonatal zebularine treatment, but a highly significant interaction between group (male, female, and female + T) and treatment (vehicle vs. zebularine; F_2,54 = 15.12, p < 0.0001; Fig. 2). Neonatal zebularine treatment decreased ERα labeling in control females (p = 0.0005) to male-like levels, while having no effect on males. Zebularine also slightly increased labeling in testosterone-treated females (p = 0.035). As a result, group differences in ERα labeling in the ARC were eliminated among the zebularine-treated animals (Fig. 2). The same pattern of results was obtained using direct cell counts of ERα, rather than automated analyses of labeling (not shown).

Our neonatal treatments overlapped with the peak of developmental cell death in the mouse brain [48], but we previously found that effects of zebularine on ERα cell number in the POA are independent of effects on developmental cell death or total cell number at weaning [13]. Here, we also found no effect of zebularine on developmental cell death at P1 (Fig. 3a) or total cell number at P25 in the ARC (Fig. 3b). Thus, neonatal inhibition of DNA methylation reduced the proportion of neurons that express ERα in the ARC of females, rather than the number of surviving neurons.

To test whether effects of neonatal zebularine generalized to other cell types in the ARC, we immunohistochemically labeled alternate sections from the animals in this study for kisspeptin and calbindin. We found main effects of sex (F_2,49 = 115.2, p < 0.0001) and treatment (F_1,49 = 10.88, p < 0.0018), as well as a sex-by-treatment interaction (F_2,49 = 4.895, p = 0.012) on kisspeptin labeling in the ARC. Females had more kisspeptin than males (p < 0.0001), as expected [49], and labeling was mainly present within fibers (online suppl. Fig. 2a). Neonatal zebularine treatment reduced kisspeptin labeling of females (p = 0.002), but a robust sex difference in kisspeptin remained in zebularine-treated animals (p < 0.0001). Females also had more calbindin cells than did males (main effect of sex: F_1,26 = 14.31, p < 0.001), but we found no effect of zebularine and no zebularine-by-sex interaction on calbindin cell number in the ARC (online suppl. Fig. 2b). Thus, effects of zebularine treatment were cell-type specific, as we, and others, have seen previously [15, 50]. We also found the expected effect of sex, but no effect of zebularine and no zebularine-by-sex interaction on body weight (online suppl. Fig. 2c), suggesting that gross development is normal in zebularine-treated animals, as has also been seen previously by us and others [13, 51].

DNA methylation is dynamically regulated during brain development [32, 52, 53], indicating an active methylation/demethylation cycle. TET enzymes control the turnover of DNA methyl marks, and we previously found higher expression of Tet2 and Tet3 mRNA in the POA and mediobasal hypothalamus of neonatal male mice [33]. To test for a role for DNA demethylation in the development of ERα labeling, we administered siRNAs targeting Tet2 and Tet3 to newborn males, while control males and females received injections of nontargeting RNAs. Tet2/Tet3 expression was significantly reduced in punches of the anterior and posterior hypothalamus 48 h after siRNA injection (main effect of siRNA, Tet2: F_1,10 = 6.72, p = 0.027; Tet3: F_1,10 = 5.21, p < 0.05; online suppl. Fig. 3). The reduction was fairly subtle (∼15–30%), but was comparable to the magnitude of sex differences in Tet expression seen previously [33]. To determine whether this partial knockdown was functionally significant, ERα labeling was examined in the ARC and VMHvl at P25.

We found a significant difference in ERα labeling across the 3 groups in the ARC (F_2,20 = 4.66, p < 0.025; Fig. 4). Females again had greater ERα labeling than control males (p < 0.04). In males with Tet2/Tet3 knockdown, ERα labeling was significantly increased relative to control males (p < 0.015), to a level very similar to that of females (p = 0.97; Fig. 4). In contrast, siRNAs against Tet2 and Tet3 had no effect on ERα labeling in the VMHvl, although we did find the expected sex difference, with greater ERα labeling in females (F_2,21 = 14.94, p < 0.0001). To test whether effects of Tet downregulation generalized to another cell type in the ARC, we immunohistochemically labeled the remaining brain sections of these animals for kisspeptin. We found a significant difference in labeling across groups, with females having greater kisspeptin immunoreactivity relative to control males (p < 0.0001; online suppl. Fig. 4). In contrast to effects on ERα however, kisspeptin labeling did not differ between control and siRNA-treated males. Thus, neonatal downregulation of Tets appears to disrupt sexual differentiation of specific cell types.

A neonatal inhibition of DNA methylation leads to high, female-like ERα expression across groups in the VMHvl [15], but uniformly low, male-like ERα expression across groups in the neighboring ARC of the same animals (current report). Moreover, neonatal inhibition of TET enzymes, which is expected to increase 5mC, increased ERα in the ARC of males. This suggested that 5mC regulates ER expression differently in the ARC and VMHvl, and may paradoxically increase ERα expression in the ARC. The mouse ERα gene (Esr1) can be transcribed from 6 promoters (untranslated exons A, B, C, F1, F2, and H), resulting in 5 mRNA variants but a single protein product [54]. Methylation levels at exons A and C have been associated with changes in gene expression in the brain [43, 55]. We therefore used bisulfite conversion of DNA followed by pyrosequencing to compare effects of age, sex, and zebularine treatment on total methylation (5mC + 5hmC) of exons A and C in the ARC and VMHvl, focusing on 16 CpG sites previously shown to regulate expression of Esr1 in the mouse brain: CpGs 1–11 in exon A and CpGs 40–44 in exon C (see online suppl. Table 1) [43, 44].

The levels of total methylation of Esr1 in the ARC and VMHvl observed at P1 were similar to those reported by Westberry et al. [43] in the neonatal mouse cortex using the same technique (Fig. 5). In exon A, repeated measures ANOVA indicated no overall sex difference in total methylation across the 11 CpG sites in either the ARC or VMHvl on P1 (Fig. 5a). However, small sex differences in both brain regions emerged at P25, in opposite directions: total methylation levels across the 11 CpG sites were significantly higher in females than in males in the ARC (main effect of sex, F_1,16 = 8.63, p < 0.01), and higher in males than in females in the VMHvl (F_1,16 = 7.69, p < 0.015; Fig. 5a). Analyses of individual CpG sites showed that the pattern (F > M in the ARC, and M > F in the VMHvl) was consistent across most CpG sites (Fig. 5b). However, the differences were subtle and reached significance only for CpG 10 in the ARC, and for CpGs 6, 9, 10, and 11 in the VMHvl (see online suppl. Fig. 5 for absolute percent methylation values at each CpG site). The effects of sex were also specific to exon A as we found no sex differences in exon C at P1 or at P25 in either the ARC or VMH (all p values >0.05; online suppl. Fig. 6a, b).

The sex differences that emerged at P25 were superimposed upon large increases in average total methylation of CpG sites of the exon A promoter region in both sexes and regions between P1 and P25 (main effect of age; F_1,24 > 200, p < 0.0001 in each region; Fig. 5), consistent with our previous observation of an overall increase in global 5mC between birth and P25 in the mouse hypothalamus [33]. In addition, we found an interaction between age and sex in both regions (ARC: F_1,24 = 5.01, p = 0.035; VMH: F_1,24 = 7.59, p = 0.011): females gained more methylation with age than males in the ARC, and males gained more methylation with age than females in the VMHvl. There was a smaller increase in total methylation with age in exon C when collapsing across sex (main effect of age; ARC: F_1,24 = 5.41, p = 0.029; VMHvl: F_1,24 = 6.52, p = 0.017; online suppl. Fig. 6a, b). Surprisingly, we also found that neonatal zebularine treatment increased total methylation in exon A at P25 in the ARC (F_1,16 = 10.43, p < 0.01; online suppl. Fig. 6c), with no sex-by-treatment interaction. Post hoc tests indicate that the effect was significant for CpG sites 6 and 9 (not shown). There was no effect of neonatal zebularine treatment on total methylation in the VMHvl.

ERα neurons in the ARC and VMHvl play important roles in estradiol-mediated effects on energy homeostasis, physiology, and behavior. A knockdown of Esr1 in the ARC increases food intake and bone density in female mice [18, 19], whereas blocking ERα signaling in the VMHvl reduces energy expenditure and thermogenesis [19, 56, 57]. In both regions, effects are sex-specific, with little or no effect of Esr1 knockdown in males [18, 19, 56]. Females have more ERα neurons in the VMHvl [5], and some, but not all, previous studies find more ERα protein or mRNA in the ARC of rodents [5, 16, 45]. Here, we found greater ERα labeling in weaning age females in 3 independent cohorts. A similar female bias in ERα cell number has been reported in a variety of brain regions and vertebrate species [58-62]. In several cases, the sex difference requires differential exposure to testosterone during perinatal life [47, 63, 64], but the mechanism by which testosterone programs sex differences in ERα expression, or other sex differences in neurochemistry for that matter, remains largely unknown.

We previously reported that sex differences in ERα in the POA and VMHvl of mice develop postnatally and are epigenetically regulated [13, 15]. Both sexes express high levels of ERα at birth, and ERα cell number decreases in males over the next few weeks. This decrease can be prevented by treating neonatal males with an inhibitor of DNMTs, with no effect on females and no change in developmental cell death or total cell number [13]. We showed here that a sex difference in ERα, favoring females, also develops postnatally in the ARC and is abolished by inhibiting DNMTs during the first 2 days of life. However, DNMT inhibition eliminated the sex difference in the ARC by decreasing (i.e., masculinizing) ERα expression in females. As previously, zebularine influenced ERα in the ARC without altering neonatal cell death or total cell number, and long-term effects were both sex- and cell-type specific. It is unknown what makes a given gene susceptible to a transient epigenetic disruption. Neonatal zebularine treatment in rats acutely alters the expression of <2% of all genes in the POA [11], and a similar selectivity has been reported in other studies following treatment with zebularine or other epigenetic inhibitors [50, 65-67]. One hypothesis is that genes actively undergoing regulation at the time of intervention are particularly affected [65, 68], and this may include genes undergoing sexual differentiation at birth.

In addition to the sex differences in ERα protein and mRNA at weaning, sex differences in total methylation of a promoter region of Esr1 emerged at weaning. Across 11 CpG sites of exon A, males had a slightly higher percent of total methylation than females in the VMHvl at P25. This is consistent with the hypothesis that 5mC marks may contribute to lower ERα labeling in the VMHvl of males. In the ARC however, females had a slightly higher percent of total methylation across exon A, despite their higher levels of ERα protein and mRNA. In addition, neonatal inhibition of DNMTs, which is expected to decrease 5mC, reduced ERα expression in the ARC of females, and neonatal knockdown of Tet2/Tet3 expression in males, which is expected to increase 5mC, increased ERα cell number at weaning. Taken together, these observations suggest that 5mC is associated with Esr1 gene activation in the ARC. Although contrary to the canonical association of DNA methylation with gene repression, there are a growing number of cases where DNA methylation of specific gene regions promotes transcription [24-28, 69].

One limitation of our findings is that pyrosequencing of sodium bisulfite-treated DNA does not distinguish 5mC marks from 5hmC. The concern in this case is lessened by the fact that 5hmC in neurons primarily accumulates in gene bodies and is depleted in promoter regions [32, 42]. In addition, 5mC is 4–5 times more abundant than 5hmC in CpG dinucleotides [70]. Because we examined promoter regions and modifications in the CpG context, most of the total methylation we detected was presumably 5mC. Nonetheless, one would ideally like to sequence using techniques that distinguish 5mC from 5hmC [71], although the amount of starting material required is a serious impediment for this type of analysis.

A second limitation is that the sex differences in total methylation we observed were quite small (on the order of 2% in absolute terms and 20% relative differences between sexes across the CpG sites of exon A). Although these differences are similar in magnitude to what was previously reported for sex differences in methylation of steroid receptors in the rat POA and mediobasal hypothalamus [53], it is reasonable to question whether such differences could meaningfully affect gene expression. In our study and related previous studies [43, 53, 72], pyrosequencing of Esr1 was examined from brain punches, which contain many cells types. ERα neurons comprise only a minority of all cells, even in regions such as the VHM and ARC, where they are relatively abundant. Any “signal” (i.e., a sex difference in methylation of Esr1) must therefore be detected over quite a bit of noise. Thus, we do not know the methylation status of Esr1 in neurons specifically expressing ERα or – equally interesting for this study – neurons that expressed high levels of ERα at birth in males but no longer do at P25. Sex differences in those cells could be much larger, or smaller, than sex differences in the aggregate of all cell types that we examined. In clinical studies, very small mean differences in DNA methylation (often, smaller than those reported here) have consistently been associated with disease susceptibility or environmental exposures, for example, Refs. [73-75]. Subtle differences in methylation have also been associated with shifts in the transcription start sites and resulting changes in mRNA stability and, hence, protein expression of steroid receptors [75]. In addition, while small differences in the methylation of a single CpG site may not be meaningful, when hyper- or hypomethylation is spread over several adjacent CpG sites, as was seen here, effects are compounded [75, 76]. Ultimately however, any measure of DNA methylation provides evidence that is correlative in nature. To test whether any given methyl mark causes a sex difference in ERα cell number would require specifically manipulating that mark in the cells of interest, and while this is now theoretically possible [77, 78], it may be some time before this can be achieved site- and cell-specifically, in vivo, in a newborn mouse brain.

In addition to the sex differences in total methylation that emerged at P25, we found that both sexes accumulate modified cytosines in exon A, and to a lesser degree in exon C, of Esr1 from birth to weaning. This is consistent with previous findings of marked increases in 5mC and 5hmC throughout the genome in the mouse hypothalamus and neocortex from birth to adolescence [32, 33]. The fact that females have persistently high levels of ERα in both the ARC and VMHvl despite the accumulation of methylation in a brain-relevant Esr1 promoter suggests that there are other regulatory mechanisms that allow for gene expression across development, and underscores the difficulty in extrapolating from DNA methylation to mRNA or protein expression.

We expected zebularine, which inhibits DNMTs, to decrease 5mC marks and therefore total DNA methylation of Esr1 but did not observe that in either the VMHvl or ARC. In the VMHvl, there was no effect of neonatal zebularine treatment on total methylation of exon A or exon C, even though this treatment previously led to a lasting increase in ERα in males [13, 15]. Possible explanations are that zebularine may act indirectly (e.g., by inhibiting an inhibitor of Esr1 in males), or that the Esr1 promoter regions we examined do not capture the direct effects of zebularine. In addition, recent studies suggest that tissue-specific DNA methylation patterns are often observed outside of CpG islands, for example, in CpG shores or “shelves” [79, 80]. In blood cells of postmenopausal women, decreasing estradiol levels influence CpG shore methylation, while sparing promoter regions of Esr1 [81]. Thus, while methylation of promoters A and C has previously been implicated in developmental regulation of Esr1 in the brain, methylation in nonpromoter regions may also play important roles in regulating gene expression [27, 82, 83].

Even more surprising, neonatal zebularine treatment slightly increased the percent total methylation in exon A of Esr1 in the ARC at P25. If 5mC indeed activates Esr1 in the ARC, the increased total methylation after zebularine would be consistent with the increased ERα seen in testosterone-treated females that received zebularine. However, the effect of zebularine was seen across all groups, which is harder to explain. We note that in post hoc analyses, the individual CpG sites affected by zebularine did not overlap with those that showed a significant sex difference in the ARC. In the current study, for example, CpG-10 was significantly different by sex, but not by zebularine treatment, in both the ARC and VMHvl. Differences in Esr1 expression between peripheral tissues have been attributed to the methylation status of a single CpG site [84], although this is uncommon.

DNA demethylation and 5hmC marks are just beginning to be explored in the context of sexual differentiation. Given its unusually high abundance in gene bodies of neurons [85], 5hmC is likely to have an important role in shaping neural gene expression. We recently found more consistent sex differences in the expression of Tets than of Dnmts in the neonatal mouse brain [33], with higher Tet2 and Tet3 expression in males. The fact that suppressing Tet2 and Tet3 expression in males increased ERα in the ARC to female-like levels indicates that this sex difference is functionally meaningful, and gives further credence to the idea that the sex difference in ERα expression in the ARC is related to levels of DNA methylation early in development. More generally, this finding suggests that demethylation, or possibly stable 5hmC marks, contributes to sexual differentiation of the brain.

Overall, our data confirm a sex difference in the ERα cell number in the ARC at weaning (female > male) that depends on neonatal testosterone. We further show that neonatal inhibition of DNA methylation decreases ERα labeling in females and neonatal inhibition of demethylation increases ERα labeling in males. Females also had slightly higher methylation in a promoter region of Esr1 at weaning than did males, which was opposite to the pattern seen in the VMHvl. It is possible that neurons with the potential to express ERα are normally inhibited from doing so by the presence of DNA methyl marks in the VMH and the absence of such marks in the ARC, although additional studies would be required to demonstrate this. Since sex differences in both regions are due to perinatal gonadal steroids, testosterone may program ERα cell number through epigenetic mechanisms. In cancer cells, total methylation levels are reduced in response to estradiol [86], and one mechanism for this hormone-dependent hypomethylation is the upregulation of Tet2 expression and co-binding at ER-target genes [87, 88]. Neonatal estradiol also dampens the catalytic efficiency of DNMTs in the POA of the rat brain [11]. Finally, gonadal steroids alter histone modifications [89, 90], which may, reciprocally, affect DNA methylation levels [91, 92]. Thus, there are likely to be multiple parallel and interacting epigenetic mechanisms by which sex steroid hormones orchestrate stable changes in neurochemistry in the developing brain.

We thank Taylor Hite for technical assistance and members of the Forger Lab for feedback on the manuscript.

All procedures were performed in accordance with the National Institutes of Health animal welfare guidelines and were approved by the Georgia State University Institutional Animal Care and Use Committee (protocol number: A18062).

There is no conflict of interest regarding this work.

This work was supported by a National Science Foundation Graduate Research Fellowship and a National Institute of Neurological Disorders and Stroke Grant F99/K00 (F99NS120531) awarded to L.R.C., and a National Institutes of Health Grant R01 (MH068482) and Brains & Behavior Seed Grant from Georgia State University awarded to N.G.F.

L.R.C., C.D.C., and N.G.F. conceptualized the experiments. L.R.C., I.N.K.V.C., and N.G.F. wrote the manuscript. L.R.C., C.D.C., I.N.K.V.C., D.M., E.K.R., and A.C.-R. assisted in data collection and/or analysis. All authors edited and approved the manuscript.

All data generated or analyzed during this study are included in this article and its online supplementary files. Further inquiries can be directed to the corresponding author.