Evf2 (Dlx6as) lncRNA regulates ultraconserved enhancer methylation and the differential transcriptional control of adjacent genes

DNA methylation is an epigenetic modification associated with transcriptional changes in plants and animals (Feng et al., 2010; Smith and Meissner, 2013). Although factors involved in DNA methylation have been studied in depth, it is still not known how specific sites are selected. Multiple lines of evidence demonstrate the involvement of long non-coding RNAs (lncRNAs), specifically those known to regulate dosage compensation and imprinting through control of DNA methylation in cis (Barlow, 2011; Lee and Bartolomei, 2013). In addition to their cis regulation, lncRNAs that control imprinting also act in trans, as shown by studies with the H19 lncRNA (Forné et al., 1997). One possible mechanism for lncRNA control of DNA methylation has been proposed for Kcnq1ot1 (Mohammad et al., 2010), an lncRNA that forms a complex with DNA methyltransferase 1 (DNMT1), recruiting DNMT1 to chromatin and increasing paternal-specific CpG island methylation. The RNA-binding properties of DNMTs and methyl-CpG binding protein 2 (MECP2) (Nan et al., 1997) raised the possibility that large subgroups of non-coding RNAs might be directly involved in DNA methylation (Jeffery and Nakielny, 2004).

Besides imprinting and gene dosage compensation, lncRNA control of DNA methylation is less well understood. Genome-wide roles for antisense lncRNAs and other lncRNAs in gene regulation have been proposed (Lee, 2012), increasing their potential for controlling DNA methylation. In embryonic stem cells, the majority of lncRNAs are close to, or overlapping with, protein-coding genes (Sigova et al., 2013), adding further to this potential.

Evidence that antisense lncRNAs can differentially control DNA methylation in trans was provided by work on two different lncRNAs: Khps1a (Imamura et al., 2004) and p15AS (Yu et al., 2008). Whereas Khps1a facilitates CpG island demethylation in the promoter of its sense partner Sphk1, p15AS increases methylation in the p15 (Cdkn2b) promoter. In a screen for functional relationships between sense/antisense pairs in the genome, both positive and negative regulatory effects were found (Katayama et al., 2005), a result supported by the differential effects of Khps1a and p15AS on methylation. Therefore, lncRNAs can mediate changes in methylation, possibly depending on DNMT and methyl-binding protein recruitment.

Work from our laboratory showed that Evf2 (Dlx6as; also known as Dlx6os1) (Feng et al., 2006) is an ultraconserved, developmentally regulated Dlx6 antisense lncRNA, that regulates activity of Dlx5/6 DNA regulatory sequences ei and eii (Zerucha et al., 2000). Evf2 recruits both known activator (DLX) (Panganiban and Rubenstein, 2002) and repressor (MECP2) proteins (Bond et al., 2009). It is unknown whether this recruitment of a methylation binding protein (MECP2) has an effect on methylation in this region. In addition, the precise mechanism of action of the DLX activator and MECP2 repressor proteins on the enhancer is unknown.

The biological significance of Evf2-Dlx5/6ei interactions is supported by altered adult hippocampal GABA circuitry in mice lacking Evf2 (Bond et al., 2009), transcriptional effects of a single nucleotide polymorphism (SNP) in Dlx5/6ei linked to autism (Poitras et al., 2010), and the established role of MECP2 in autism (Guy et al., 2007; Guy et al., 2001). Loss of Evf2 results in increased Dlx5 and Dlx6 expression in E13.5 MGE, a major site of sonic hedgehog-activated Dlx and Evf2 gene regulatory events crucial for GABAergic interneuron development (Anderson et al., 1997a; Feng et al., 2006; Kohtz et al., 1998).

The ability of Evf2 to recruit MECP2 to Dlx5/6ei and eii (Bond et al., 2009), and the role of MECP2 in chromatin organization in the Dlx5/6 region (Horike et al., 2005) raised several questions regarding the relationship between transcriptional effects of MECP2 recruitment and Dlx5/6ei DNA methylation. In this article, we show that loss of Mecp2 in E13.5 MGE results in increased expression of Evf2 and Dlx5, with no change in Dlx6 expression. Evf2 prevents CpG methylation in Dlx5/6ei, indicating that methylated CpG sites are not responsible for MECP2 recruitment. Using a transgenic mouse model with reduced Evf2 expression (∼0.38× wild-type expression levels), we show that Evf2 trans activity inhibits methylation of the Dlx5/6 ultraconserved enhancer in mice lacking Evf2 (Evf2^TS/TS), without altering Dlx5/6 gene expression. Based on both genetic epistasis and rescue experiments, we propose a model in which Evf2 lncRNA inhibits Dlx5/6 ultraconserved DNA methylation, facilitating antagonistic interactions between repressive and activating transcription factors MECP2 and DLXs. These interactions allow differential control of adjacent genes by shared DNA regulatory elements.

Total RNA was extracted from pairs of mouse embryonic day (E) 13.5 medial ganglionic eminence (MGE) tissue using the EZNA MicroElute Total RNA Kit (Omega). E13.5 MGE dissections were as previously described (Feng et al., 2006). cDNA was synthesized using the qScript cDNA Synthesis Kit (Quanta Biosystems). The resulting cDNA served as a template for the amplification of the genes of interest normalized to the constitutive gene (Actb, encoding β-actin) by quantitative real-time PCR, using TaqMan or SYBR Green reagents. For TaqMan PCR, we used the following: Dlx5 (Mm00438430_m1), Dlx6 (Mm01166201_m1) and Actb (Mm00607939_s1). Rat-specific Evf2 primers were previously described (Feng et al., 2006). For SYBR Green PCR, we used the following primers: mEvf2-F (0.2 μM, 5′-CTCCCTCCGCTCAGTATAGATTTC-3′); mEvf2-R (0.2 μM, 5′-CCTCCCCGGTGAATATCTCTT-3′); β-actin-F (0.3 μM, 5′-GCGAGCACAGCTTCTTTGC-3′); and β-actin-R (0.3 μM, 5′-TCGTCATCCATGGCGAACT-3′).

For Dlx5 imprinting analysis, primers spanning a SNP in Dlx5 were used to amplify transcripts from E13.5 MGE cDNA as previously described (Horike et al., 2005), cut with HindIII and fragments analyzed by agarose gel electrophoresis. Ratios of resulting products were quantified by ImageJ software. Values were averaged from three embryos for each genotype, as indicated.

Evf2 rescue (Evf2R) mice were generated using the Dlx1/2 enhancer (Ghanem et al., 2007), floxed TS (transcription stop) sequence (Soriano, 1999), and full-length rat Evf2 cDNA (Feng et al., 2006). The TS sequence was removed by crossing Evf2R^floxedTS to EIIAcre mice [Jackson, FVB/N-TgN (EIIa-Cre) C5379Lmgd] to allow expression of rat Evf2 in Dlx1/2-expressing cells. Evf2R^EIIAcre was crossed to Evf2^TS/TS, to generate Evf2^TS/TS;R^+/- and Evf2^TS/TS;R^-/- embryos for the gene expression and methylation analysis shown in Fig. 4. Evf2^TS/TS mice were generated as previously described (Bond et al., 2009). B6.129P2(C)-Mecp2^tml.lBird/J mice (Guy et al., 2001) were obtained from the Jackson laboratory. As Mecp2 is X-linked, Mecp2 null mice were generated by crossing Mecp2^+/- females to Bl6 males, and analyzing male E13.5 embryos (Mecp2^-/y). Dlx1/2^+/- mice (Anderson et al., 1997a) were generously given by Dr Kenneth Campbell (Children’s Hospital Research Foundation, Cincinnati, OH, USA).

Genomic DNA from mouse E13.5 MGE tissue was modified with the methylSEQr Bisulfite Conversion Kit (Applied Biosystems). The genomic region spanning the Dlx5/6 ultraconserved enhancer (ei) was amplified by using nested PCR, using the following primers: External primers: F: 5′-GATTTGGGTATTTTTTAAATTATG-3′ and R: 5′-AAAATAAATACAAAAACATCAACC-3′; Internal primers: F: 5′-GTTATTTATTAGAAGTTAATAGAG-3′ and R: 5′-TAAACATTTTCTAATTTCAAAATTC-3′. The PCR products were cloned into pGEM T-easy vector (Promega) and individual clones were sequenced. A minimum of 45 clones was sequenced from each genotype.

Global methylation analysis of mouse B1 line elements (1-4) from Evf2^+/+ and Evf2^TS/TS E13.5 MGE was performed by EpigenDx by pyrosequencing PCR products from bisulfite-treated DNA (Uhlmann et al., 2002). E13.5 MGE tissue was digested with proteinase K (40 μg) at 50°C for 30 minutes in a volume of 40 μl, and cell debris pelleted by centrifugation at 14,000 g for 10 minutes. Digested tissue (20 μl) was used in the bisulfite conversion reaction; E13.5 MGE DNA (500 ng) was bisulfite treated by EpigenDx using a proprietary bisulfite salt solution. DNA was diluted to 45 μl, and 5 μl of 3 N NaOH for 30 minutes at 42°C to denature the DNA. Bisulfite salt solution (100 μl) was added to the DNA and incubated for 14 hours at 50°C. Bisulfite-treated DNA was purified using Zymogen DNA columns and eluted in 20 μl TE (10 mM Tris pH 7.5, 0.1 mM EDTA); 1 μl was used for each PCR.

The PCR was performed with 0.2 μM of each mouse B element primer (1-4). One of the PCR primers was biotinylated to purify the final PCR product using Sepharose beads. The PCR product was bound to Streptavidin Sepharose HP (Amersham Biosciences), and the Sepharose beads containing the immobilized PCR product were purified, washed and denatured in 0.2 M NaOH solution, and rewashed using the Pyrosequencing Vacuum Prep Tool (Pyrosequencing, Qiagen). Pyrosequencing primer (0.2 μM) was annealed to the purified single-stranded PCR product. PCR products (10 μl) were sequenced by Pyrosequencing PSQ96 HS System (Biotage), following the manufacturer’s instructions (Pyrosequencing, Qiagen). The methylation status of mouse B1 line elements was analyzed individually using QCpG software (Pyrosequencing, Qiagen).

Although it has been shown that MECP2 is a transcriptional repressor of Dlx5 and Dlx6 in postnatal brain (Horike et al., 2005), the role of Mecp2 in embryonic brain development has not been defined. In addition, MECP2 exhibits both activator and repressor roles in vivo, raising the question of whether MECP2 represses or activates Dlx5/6 expression in E13.5 MGE (Chahrour et al., 2008). Previous results showed that loss of Evf2 in E13.5 MGE decreases binding of MECP2 to Dlx5/6ei and eii, with decreased histone deacetylase (HDAC1) binding to Dlx5/6eii (Bond et al., 2009). Thus, decreased HDAC1 binding to Dlx5/6eii is one of the possible explanations for increased expression of Dlx5 and Dlx6.

In order to determine the effect of MECP2 on Dlx5/6 enhancer activity in vivo, we compared Evf2, Dlx5 and Dlx6 expression in E13.5 MGE of Mecp2 null mice [Fig. 1, mutant (gray bars) and wild-type littermates (black bars)]. Loss of Mecp2 [Mecp2 null (Mecp2^-/y)] increases Evf2 (∼2-fold) and Dlx5 (∼2.5-fold) transcripts, with no detectable changes in Dlx6. Horike et al. (Horike et al., 2005) reported ∼2-fold increase in Dlx5 and ∼1.8-fold increase in Dlx6 in Mecp2 null postnatal cortex. Therefore, E13.5 MGE and postnatal cortex are similar in that MECP2 moderately inhibits Dlx5 expression (∼2- to 2.5-fold), but these regions differ in control of Dlx6.

In E13.5 MGE, removal of one copy of Dlx1/2 from Mecp2 null mice (Mecp2null; Dlx1/2^+/-) prevents activation of Evf2 and Dlx5. This result suggests that increased Evf2 and Dlx5, in the absence of MECP2, is mediated by DLX1/2. Removal of one copy of Dlx1/2 (Dlx1/2^+/-) inhibits Evf2 and Dlx5 expression, indicating dose-dependent Dlx1/2 regulation of Evf2 and Dlx5. Complete loss of Dlx1/2 (Dlx1/2^-/-) results in loss of Evf2, Dlx5 and Dlx6, in agreement with previous demonstrations that Dlx1/2 is crucial for Dlx5/6 gene activation (Anderson et al., 1997b; Zerucha et al., 2000). Loss of Mecp2 does not change levels of Dlx1 or Dlx2 transcripts (supplementary material Fig. S1). Therefore, increased expression of Evf2 and Dlx5 in Mecp2 null mice does not result from increased Dlx1/2 expression.

These genetic epistasis experiments suggest that DLX1/2 and MECP2 interactions are antagonistic rather than cooperative. Antagonistic interactions suggest that MECP2 and DLX1/2 compete for binding on the same allele, arguing against parent of origin-specific Dlx5 regulation. Although Dlx5 imprinting has been demonstrated in humans (Okita et al., 2003), there has been controversy in the literature regarding Dlx5 imprinting in mice (Horike et al., 2005; Kimura et al., 2004; Schüle et al., 2007).

In order to address whether Dlx5 is imprinted in E13.5 MGE, and whether Evf2 regulation of Dlx5 is parentally controlled, we used a previously identified Dlx5 SNP in JF1 mice (Horike et al., 2005) to distinguish between parental alleles. We find that Dlx5 is not imprinted in E13.5 MGE, as the ratio between maternal Dlx5 and paternal Dlx5 transcripts is either 1.18 or 1.10 (Fig. 2A, lanes 1 and 2, corresponding to genotypes in schematics 1 and 2). When Evf2 is truncated on the paternal (Fig. 2A, lane 3) or maternal (Fig. 2A, lane 4) alleles, Dlx5 increases (Fig. 2A, green arrows). These data suggest that Evf2 repression of Dlx5 is equal on maternal and paternal alleles (Fig. 2A, ratio between maternal Dlx5 and paternal Dlx5 transcripts is ∼2.3 for genotypes in schematics 3 and 4). Fig. 2B shows E13.5 MGE qRT-PCR analysis of Dlx5 expression in Evf2TS^pat/+^mat (pink bars), Evf2TS^mat/+^pat (blue bars) and Evf2^+/+ littermates (black bars). Dlx5 increases to 1.5× wild-type levels upon maternal or paternal Evf2 loss. Together, these data show that Dlx5 is not imprinted at this time in development, and that Evf2 represses Dlx5 on both maternal and paternal alleles, arguing against MECP2 allele-specific or parentally controlled Dlx5 expression in E13.5 MGE.

Given that MECP2 is a methyl-CpG binding protein, we next asked whether Evf2 recruits MECP2 by increasing CpG methylation within Dlx5/6ei. We used previously described mice lacking Evf2 (Evf2^TS/TS) (Bond et al., 2009), in which a triple poly(A) transcription stop site is inserted into the 5′ end of Evf2. Bisulfite sequencing analysis of 890 bp spanning Dlx5/6ei (13 potential CpG methylation sites) on E13.5 MGE DNA was performed on Evf2 wild-type littermates (Evf2^+/+) and compared with Evf2^TS/TS mutants (Fig. 3A,B). Increased methylation is observed in Evf2^TS/TS E13.5 MGE DNA at two specific CpG sites, ⁵⁷⁶CpG and ⁷⁵⁷CpG (Fig. 3B), showing that Evf2 inhibits site-specific CpG DNA methylation in the Dlx5/6 ultraconserved enhancer ei. Global methylation analysis of mouse B1 line elements at four sites (1-4) shows that Evf2 loss does not increase global methylation (Fig. 3C), indicating that methylation increase at Dlx5/6ei does not result from global effects.

We next asked whether Evf2 controls Dlx5/6ei CpG methylation through trans or cis mechanisms. In order to answer this question, we developed an Evf2 rescue transgenic model, in which rat Evf2 cDNA (Feng et al., 2006) is driven by a Dlx1/2 enhancer (Ghanem et al., 2007). Evf2 rescue transgenic mice (Evf2R) were crossed to Evf2^TS/TS mice to generate mice that only expressed Evf2 from the transgene (Evf2^TS/TS;R). Fig. 4A,B shows a schematic of the Evf2 transgene and genotyping results of cre-mediated removal of the triple poly(A) stop site (TS) placed at the 5′ end of the rat Evf2 transgene (Evf2^TS/TS;R). Although we have not ruled out the possibility that transgene integration is in chromosome 6, transgene inheritance does not appear to be linked to Dlx5/6, as Evf2^TS/+;R offspring from crosses of Evf2^TS/+ × Evf2^+/+;R are obtained at expected frequencies.

Quantitative RT-PCR analysis of E13.5 MGE from Evf2^TS/TS;R mice shows that rat-specific Evf2 is expressed from the transgene (Fig. 4C). Comparison of Evf2 transcript levels expressed by the transgene and wild-type Evf2 levels is possible using primers that recognize both rat and mouse transcripts (Fig. 4D, Evf2); the transgene is expressed at 0.38× wild-type levels (Fig. 4D). Dlx5 and Dlx6 expression does not differ between Evf2^TS/TS;R and Evf2^TS/TS E13.5 MGE (Fig. 4E). However, Dlx5/6ei ⁵⁷⁶CpG and ⁷⁵⁷CpG methylation is significantly decreased in E13.5 MGE DNA of Evf2^TS/TS;R transgenic rescue mice compared with Evf2^TS/TS (Fig. 4F). There is a slight (<15%), but statistically significant, increase in ⁶²⁶CpG methylation, raising the possibility of opposing methylation effects of Evf2 on sites other than ⁵⁷⁶CpG and ⁷⁵⁷CpG. The ability of the Evf2 transgene, even at 0.38× wild-type levels, to significantly decrease ⁵⁷⁶CpG and ⁷⁵⁷CpG methylation of Dlx5/6ei, suggests that Evf2 inhibits methylation in trans.

The proximity of ⁵⁷⁶CpG to previously defined DLX1/2 binding sites (Fig. 3A, D1 and D2, green) in Dlx5/6ei (Zerucha et al., 2000) raises the possibility that Evf2 recruitment of DLX1/2 plays a role in ⁵⁷⁶CpG and ⁷⁵⁷CpG methylation. However, loss of DLX1/2 loss also results in Evf2 loss (Fig. 1), making it difficult to distinguish between the role of DLX1/2 binding to Dlx5/6ei and the role of Evf2 in methylation. Analysis of Dlx1/2^-/- E13.5 MGE (Fig. 5A) shows that ⁵⁷⁶CpG and ⁷⁵⁷CpG Dlx5/6ei methylation increases in the absence of Dlx1/2. This result is consistent with increased ⁵⁷⁶CpG and ⁷⁵⁷CpG Dlx5/6ei methylation observed in Evf2^TS/TS mice, and further supports the idea that Evf2 prevents site-specific ⁵⁷⁶CpG and ⁷⁵⁷CpG Dlx5/6ei methylation.

Given that Evf2 also recruits MECP2 to Dlx5/6ei (Bond et al., 2009), we next asked whether MECP2 recruitment plays a role in Dlx5/6ei methylation. Analysis of Dlx5/6ei methylation in Mecp2null;Dlx1/2^+/- mice (Fig. 5B) shows no significant difference in CpG methylation profile across Dlx5/6ei, compared with wild-type littermates. Although there appears to be slightly increased methylation at sites ⁶²⁶CpG and ⁷⁵⁷CpG, these differences are not statistically significant. It is interesting to note that Evf2 expression in Mecp2null;Dlx1/2^+/- and Evf2^TS/TS;R mice (Fig. 1) are similar, and that Dlx5/6ei methylation profiles are also similar [Mecp2null;Dlx1/2^+/- (Fig. 5B) and Evf2^TS/TS;R (Fig. 4F)]. Although we did not determine Dlx5/6ei methylation profiles in Dlx1/2^+/- mice, it would be expected that Mecp2null;Dlx1/2^+/- and Dlx1/2^+/- methylation profiles would also be similar, as Evf2 levels are similar between Mecp2null;Dlx1/2^+/- and Dlx1/2^+/- mice. Together, these data support the idea that Evf2, but not MECP2, controls Dlx5/6ei methylation.

Six key results presented in this paper are listed below, followed by their implications.

1. MECP2 represses Dlx5 and Evf2, but not Dlx6. Evf2 recruitment of MECP2 to Dlx5/6 intergenic enhancers inhibits Dlx5 expression, whereas Evf2 antisense transcriptional regulation inhibits Dlx6 expression. Therefore, the Evf2 lncRNA enables differential regulation of genes with shared regulatory elements.

2. Loss of one copy of Dlx1/2 in Mecp2 null E13.5 MGE rescues increased expression of Dlx5 and Evf2. MECP2 inhibits Dlx5 and Evf2 expression by antagonizing DLX1/2 transcriptional activation.

3. Dlx5 is not imprinted in E13.5 MGE tissue. Evf2 represses Dlx5 equally on maternal and paternal alleles. Evf2 recruitment of MECP2 is unlikely to be involved in Dlx5 imprinting, and therefore unlikely to be allele specific, at this time in development.

4. Evf2 prevents methylation at two specific sites in the Dlx5/6ei ultraconserved enhancer. Evf2 transgene expression at 0.38× wild-type levels reduces site-specific methylation of the Dlx5/6ei ultraconserved enhancer. An lncRNA can determine site-specific methylation pattern across an enhancer, at least in part, through trans-acting mechanisms.

5. Despite reduced site-specific enhancer methylation, Evf2 transgene expression at 0.38× wild-type levels does not change Dlx5 and Dlx6 expression. The lncRNA-dependent methylation pattern across an enhancer might not reflect the activity state of the enhancer, but rather the presence of the lncRNA, which can have activator or repressor activity.

6. Loss of Dlx1/2, but not Mecp2, increases site-specific methylation of the Dlx5/6ei ultraconserved enhancer. Evf2 recruitment of MECP2 is not involved in regulating site-specific methylation of the Dlx5/6ei ultraconserved enhancer.

These results are based on analysis of gene expression and methylation patterns in mouse mutants either lacking or expressing different doses of Evf2, Dlx1/2 or Mecp2. This information, combined with previously published chromatin immunoprecipitation, leads to models (Fig. 6A,B) that describe how Evf2 methylation control and transcription factor recruitment contribute to Dlx5/6ei enhancer-dependent gene regulation. These models are described in detail below.

Based on genetic manipulation of Mecp2 and Dlx1/2 in vivo (Fig. 1), the schematics in Fig. 6A describe possibilities for MECP2 and DLX1/2 occupancy of Dlx5/6ei/eii. Fig. 1 shows that loss of MECP2 activates Evf2 (∼2-fold) and Dlx5 (∼2.5-fold) expression. When one copy of Dlx1/2 is also lost (double mutant, Mecp2null;Dlx1/2^+/-), increased Evf2 and Dlx5 expression is lost. This supports the idea that MECP2 inhibits Evf2 and Dlx5 expression by preventing DLX1/2 activation of Dlx5/6ei/eii, suggesting antagonism between MECP2 and DLX1/2. There is an additional effect resulting from removal of one copy of Dlx1/2, whereby both Evf2 and Dlx5 decrease further compared with wild type (∼0.5-fold). One possible explanation of why Evf2 and Dlx5 do not return to wild-type levels, but decrease even further is their dose-dependent regulation by DLX1/2. Dose-dependent regulation of Evf2 and Dlx5 by DLX1/2 is supported by removing one copy of Dlx1/2 (Dlx1/2^+/-), which also results in Evf2 and Dlx5 expression at ∼0.5× wild-type levels. The fact that levels of Evf2 and Dlx5 are equal in Dlx1/2^+/- and Mecp2null;Dlx1/2^+/- shows that loss of MECP2 has no effect when DLX1/2 levels are limiting.

The schematics in Fig. 6A illustrate a possible explanation of why Evf2 and Dlx5 levels are the same in Dlx1/2^+/- and Mecp2null;Dlx1/2^+/- mice. In the wild-type situation, there are three possible states of Dlx5/6ei/eii occupancy: inactive (both copies are bound by MECP2; Fig. 6Aa); low activity (one copy is bound by MECP2 and the other by DLX1/2; Fig. 6Ab); high activity (both copies are bound by DLX1/2; Fig. 6Ac).

In Mecp2null;Dlx1/2^+/+ mice (Fig. 6Ae), MECP2 is absent, but DLX1/2 is at wild-type levels; DLX1/2 is expected to bind both copies, generating a high activity state (Fig. 6Ac,e). In Mecp2null;Dlx1/2^+/- mice (Fig. 6Ad), MECP2 is still absent, but DLX1/2 levels are limiting; DLX1/2 only binds one copy of Dlx5/6ei/eii, generating a low activity state (Fig. 6Ab,d*). In Dlx1/2^-/- mice (Fig. 6Af), DLX1/2 is absent, generating an inactive state (Fig. 6Aa,f). MECP2 is shown bound to Dlx5/6ei/eii in (Fig. 6Af); however, given previous chromatin immunoprecipitation (ChiP) data showing that Evf2 recruits MECP2 to Dlx5/6ei and eii (Bond et al. 2009), it is also possible that MECP2 does not bind Dlx5/6ei and eii in the absence of Evf2 and DLX1/2. In Dlx1/2^+/- mice, DLX1/2 levels are limiting (Fig. 6Ag); DLX1/2 binds one copy of Dlx5/6ei/eii, whereas MECP2 binds the other copy, generating a low activity state (Fig. 6Ab,d*,g*). The fact that genotypes d^* and g^* generate similar states of low Dlx5/6ei/eii activity supports the idea that the major role of MECP2 is to antagonize DLX1/2. Although this model is an over simplification of complex interactions that influence Dlx5/6ei/eii activity, they are consistent with both present genetic data (Fig. 1) and previous ChIP data showing binding of MECP2 and DLX1/2 to Dlx5/6ei and eii (Bond et al., 2009).

Although it would be interesting to perform an analysis of double mutants lacking both Mecp2 and Dlx1/2 (Mecp2null; Dlx1/2^-/-), Mecp2null; Dlx1/2^+/- females proved to be very poor breeders. Therefore, analysis of the double mutants proved unfeasible. However, future experiments to determine the state of MECP2 binding to Dlx5/6ei/eii in Dlx1/2^-/- E13.5 MGE might be informative. As Evf2 is lost in Dlx1/2^-/- E13.5 MGE, lack of MECP2 binding to Dlx5/6ei/eii would further support Evf2-dependent recruitment of MECP2. However, if MECP2 remains associated to Dlx5/6ei/eii in the absence of both DLX1/2 and Evf2, this would support an alternate state in which Evf2 is not required for MECP2 recruitment in the absence of DLX1/2.

In previous experiments, ectopic assays using cultured neural cell lines (Feng et al., 2006) or E13.5 MGE explants (Bond et al., 2009), suggested that Evf2 transcriptional activity occurs in trans. Co-transfection of Evf2 and Dlx2 cDNAs into C17 neural cells demonstrated increased activity of Dlx5/6ei and eii in luciferase reporter assays (Feng et al., 2006). Electroporation of Evf2 cDNA into E13.5 MGE increased expression of Dlx5 and Dlx6, also supporting the observation that Evf2 increases Dlx5/6ei and eii enhancer activity in trans (Bond et al., 2009). However, the levels of Evf2 expressed by electroporation could not be determined. Evidence in this paper supports the involvement of trans-acting mechanisms during Evf2 interactions with the Dlx5/6ei enhancer. Using Evf2^TS/TS;R transgenic mice, we show that Evf2 transgene expression at these reduced levels does not change Dlx5 or Dlx6 expression, but does prevent methylation of the Dlx5/6ei enhancer. Multiple factors, including culture-induced conditions, could contribute to the differences found between the two assays. The finding that Evf2 expressed by the transgene in Evf2^TS/TS;R mice is reduced (∼0.38×) compared with wild type raises the possibility that low levels of Evf2 can inhibit Dlx5/6ei methylation, but higher levels may be necessary to elicit transcriptional changes. Therefore, one possibility is that inhibition of Dlx5/6ei methylation is necessary, but not sufficient during Evf2 transcriptional regulation. By lowering the level of Evf2 in vivo, transgenic expression might distinguish between methylation and transcriptional effects of Evf2 on Dlx5/6ei. Another possibility is that complete methylation inhibition is not achieved by the transgene, as Evf2^TS/TS;R methylation (Fig. 4F) profiles are slightly different compared with those of Evf2^+/+ (Fig. 3B).

DNA methylation analyses have focused on CpG islands and the role of site-specific CpG methylation across enhancers is not established. Our results on the relationship between Evf2 lncRNA, Dlx1/2, Mecp2, Dlx5/6ei enhancer CpG DNA methylation and transcriptional activity suggest a complex relationship in which Evf2 is a key regulator of enhancer methylation profile.

Based on the results in this article combined with previous ChIP experiments showing Evf2-dependent recruitment of DLX and MECP2 proteins to Dlx5/6ei (Bond et al., 2009), a model for interactions at Dlx5/6ei is proposed (Fig. 6B). Evidence from this article and others demonstrates that Dlx1/2 is required for Evf2, Dlx5 and Dlx6 expression (Fig. 6B, green arrows). Support that Evf2 represses Dlx6 expression through antisense inhibition (Fig. 6B, red arrow), is supported by Bond et al. (Bond et al., 2009), who showed that Evf2 inhibition of Dlx6 is greater than Dlx5 inhibition, and that electroporated Evf2 is unable to rescue Dlx6 expression. The transgenic model in this article further supports the involvement of Evf2 cis regulatory activity (Evf2^TS/TS;R), i.e. Dlx6 remains unaltered by Evf2 transgene expression (Fig. 4), and Dlx5 adjacent to Evf2TS increases (Fig. 2A, lanes 3 and 4). Evf2ncRNA-trans inhibits Dlx5/6ei methylation at two specific sites: ⁵⁷⁶CpG and ⁷⁵⁷CpG (Fig. 3A, red). This is supported by the analysis provided in Fig. 3B (⁵⁷⁶CpG and ⁷⁵⁷CpG methylation increases in Evf2^TS/TS mutants compared with wild types), and Fig. 4F (⁵⁷⁶CpG and ⁷⁵⁷CpG methylation decreases in Evf2^TS/TS;R compared with Evf2^TS/TS).

Previous experiments (Bond et al., 2009) suggested that one possible explanation for Evf2 repression of Dlx5/6eii is that Evf2 recruits MECP2, recruiting HDAC1 to Dlx5/6eii, and inhibiting Evf2 and Dlx5 expression (red). However, in this article, we show that Dlx6 levels do not change upon loss of MECP2 (Fig. 1). Therefore, it is possible that either MECP2 does not inhibit Dlx6 expression, or, by inhibiting Evf2 expression, decreased Dlx6 antisense inhibition balances MECP2 repressive effects. Whereas Evf2, Dlx5 and Dlx6 expression requires Dlx1/2 (Fig. 1), loss of Evf2 results in decreased DLX1/2 binding to Dlx5/6ei and eii (Bond et al., 2009). This leads us to propose that Evf2 stabilizes, but is not required for DLX1/2 binding, counteracting MECP2 repressive effects. In this model, Evf2 lncRNA facilitates differential dosage of adjacent genes that are under the control of the same DNA regulatory enhancers through balanced recruitment of both activator and repressor, and antisense inhibition of one of the genes in a bi-gene cluster. A major question for future studies will be to determine how the levels of Evf2 activator and repressor recruitment are determined.

Methylation studies of Dlx5/6ei in this article suggest that the methylation profile across the enhancer is determined by the Evf2 lncRNA. However, the functional state of the Dlx5/6ei enhancer cannot be assessed by the CpG methylation profile. Dlx5/6ei methylation increases when the enhancer is inactive (Dlx1/2^-/-) and also when enhancer activity increases (Evf2^TS/TS). This is consistent with the idea that methylation reflects the presence of an RNA regulator that recruits both positive and negatively acting transcription factors. The model is complicated by the fact that when the Evf2 transgene is expressed at 0.38× wild-type levels, Dlx5/6ei methylation significantly decreases, but Dlx5 or Dlx6 expression levels are not altered (Fig. 4D,E). There are many possible explanations. For instance, it could be argued that, although statistically reduced, the transgene does not completely inhibit methylation to the levels seen in wild type (compare Dlx5/6ei methylation profiles obtained in Fig. 3B and Fig. 4F). Therefore, the Evf2 dose might need to be higher. Alternatively, Evf2 cis-regulation might play a role in methylation inhibition.

An alternative possibility is that Dlx5/6ei enhancer methylation does not play a role in transcriptional regulation, reflecting only the presence of Evf2 RNA. However, an argument against this is that Evf2 prevents site-specific ⁵⁷⁶CpG and ⁷⁵⁷CpG Dlx5/6ei methylation, rather than random methylation over the 13 possible CpG sites tested. If methylation status is just a consequence of the presence of the RNA and does not play a functional role, site-specific methylation effects would not be expected. At this point, nothing is known about additional factors necessary for methylation of these two specific sites.

Taken together, the model described in Fig. 6B raises an intriguing possibility that the relationship between Evf2 and enhancer methylation is dynamic, reflecting the dynamic nature of positive and negative transcription factor occupancy of Dlx5/6ei. If so, such dynamic changes might make it difficult to correlate on/off enhancer activity states with DNA methylation profiles in vivo.

Given the identification of diverse classes of lncRNAs at the genome-wide level, this work raises many questions regarding lncRNA-dependent site-specific DNA methylation. (1) Do different classes of lncRNAs, in addition to ultraconserved lncRNAs, regulate site-specific methylation patterns in enhancer and non-enhancer regions? Or, is this mechanism unique to Evf2 and Dlx5/6 ei ultraconserved enhancer? (2) How do site-specific enhancer methylation, lncRNA presence, enhancer activity and histone modifications correlate? Is there a histone modification signature that is associated with RNA-dependent enhancer methylation? (3) Are DNMTs and additional factors involved in RNA-mediated site-specific enhancer methylation? Future experiments will be necessary to define the significance of these relationships to gene regulation.

We thank Dr Marc Ekker (University of Ottawa) for Dlx1/2 enhancer plasmid; Dr Kenneth Campbell (University of Cincinnati, Children’s Hospital) for sending Dlx1/2^+/- mice; the Northwestern transgenic facility for pronuclear injections generating conditional Evf2;R^{floxed TS} mice; Fabricio Costa and Bento Soares (Lurie Children’s Research Center, Northwestern University) for advice on bisulfite sequencing; and Liying Yang (EpigenDx) for global methylation analysis of mouse B1-4 elements.