ATRX tolerates activity-dependent histone H3 methyl/phos switching to maintain repetitive element silencing in neurons

Abstract

ATRX (the alpha thalassemia/mental retardation syndrome X-linked protein) is a member of the switch2/sucrose nonfermentable2 (SWI2/SNF2) family of chromatin-remodeling proteins and primarily functions at heterochromatic loci via its recognition of “repressive” histone modifications [e.g., histone H3 lysine 9 tri-methylation (H3K9me3)]. Despite significant roles for ATRX during normal neural development, as well as its relationship to human disease, ATRX function in the central nervous system is not well understood. Here, we describe ATRX’s ability to recognize an activity-dependent combinatorial histone modification, histone H3 lysine 9 tri-methylation/serine 10 phosphorylation (H3K9me3S10ph), in postmitotic neurons. In neurons, this “methyl/phos” switch occurs exclusively after periods of stimulation and is highly enriched at heterochromatic repeats associated with centromeres. Using a multifaceted approach, we reveal that H3K9me3S10ph-bound Atrx represses noncoding transcription of centromeric minor satellite sequences during instances of heightened activity. Our results indicate an essential interaction between ATRX and a previously uncharacterized histone modification in the central nervous system and suggest a potential role for abnormal repetitive element transcription in pathological states manifested by ATRX dysfunction.

ATRX (the alpha thalassemia/mental retardation syndrome X-linked protein) was initially identified through genetic linkage studies indicating that numerous mutations located within the ATRX coding sequence result in a rare mental retardation syndrome characterized by a wide array of developmental disabilities, alpha thalassemia, and severe cognitive deficits (1). The ATRX protein consists of numerous conserved protein binding domains that are critical to its function, including an N-terminal globular domain (2), referred to as the ATRX-DNMT3-DNMT3L (ADD) domain. The ATRX ADD (ADD_ATRX) domain further comprises multiple subdomains including a GATA-1–like domain and plant homeodomain (PHD) zinc finger (3, 4). The C terminus of ATRX similarly contains numerous conserved collinear domains, which confer its ATPase activity, as well as other structures that are important for mediating protein–protein interactions and its subnuclear localization patterns (5).

In most cell types, ATRX resides exclusively within the nucleus and predominantly localizes to highly repetitive heterochromatic sequences throughout the genome (e.g., pericentromeric satellite sequences, telomeres, ribosomal DNA, etc.) (6, 7). Several mechanisms controlling ATRX tethering to heterochromatic loci have recently been identified. Specifically, it has been demonstrated that ATRX can target chromatin directly through DNA template interactions with its GATA-1–like domain (7, 8). Furthermore, we previously showed that ATRX assists the histone chaperone Daxx in the deposition of the histone variant, H3.3, at specific heterochromatic loci (e.g., telomeres) (9). ATRX is also known to target chromatin through direct binding interactions with specific histone posttranslational modifications (PTMs), either directly or in complex with secondary interacting proteins, such as HP1 (10) or MeCP2 (11). Recently, the ADD_ATRX domain was demonstrated to efficiently bind N-terminal histone H3 lysine 9 tri-methylation (H3K9me3)-containing peptides (a known heterochromatic histone PTM) in the absence of H3K4me3/2 due to an inability of the ATRX PHD finger to recognize H3K4 methylation (12–14). However, it is unclear whether chemical modifications adjacent to H3K9me3, such as H3S10 phosphorylation (H3S10ph), a mark that is dramatically induced in the central nervous system (CNS) in response to various modes of environmental stimulation, might impact ATRX localization to chromatin in vivo. Given the combinatorial nature of histone modifications, it is likely that alternative mechanisms of ATRX recruitment/displacement are possible. Furthermore, the presence of numerous mutations within the ADD_ATRX domain in individuals with X-linked alpha thalassemia mental retardation syndrome, many of which would be predicted to disrupt binding interactions between ADD_ATRX and heterochromatic targets (i.e., those harboring PTMs, such as H3K9me3) (15), suggests a crucial role for ATRX in the maintenance of heterochromatic stability in the CNS. Therefore, we hypothesized that the ADD_ATRX domain might confer additional flexibility to allow for adjacent histone PTMs, such as H3S10ph, to maintain heterochromatic silencing during periods of increased cellular activity in postreplicative neurons.

Here, using a combination of biophysical, genome-wide, and functional approaches, we demonstrate a direct role for ATRX in maintaining heterochromatic transcription/stability during periods of heightened neuronal activity via “protective” recognition of the activity-dependent combinatorial histone PTM histone H3 lysine 9 tri-methylation/serine 10 phosphorylation (H3K9me3S10ph). Our findings shed light on a potential causative mechanism underlying cognitive deficits associated with X-linked alpha thalassemia mental retardation syndrome.

Results

The ADD Domain of ATRX Is Insensitive to K9me3/S10phos Switching.

To identify ADD_ATRX domain-mediated interactions with chromatin in vitro, we first performed modified histone peptide immunoprecipitation (IP) assays using recombinant ADD_ATRX to monitor its binding interactions with the N-terminal tail of histone H3 (Fig. 1A). In agreement with previous reports (12–14), ADD_ATRX domain binding to the unmodified H3 N terminus was disrupted by the “active” histone PTM H3K4me3 whereas H3K9me3 markedly enhanced ADD_ATRX histone H3 interactions under high salt conditions. Methylation of either H3R2 or H3R8 had minimal effects on ADD_ATRX binding to histone H3 peptides. Further analysis of N-terminal phosphorylation events on ADD_ATRX domain binding to the histone H3 tail indicated that H3T6ph, and to a lesser extent H3T3ph, are capable of disrupting ADD_ATRX domain binding to histone H3 whereas H3S10ph is unexpectedly compatible with such associations (Fig. 1B). Moreover, previously reported enhancements in binding of the ADD_ATRX domain to histone H3 by H3K9me3 were found to be unaffected by the addition of adjacent S10 phosphorylation (Fig. 1B). These data indicate that such site-specific H3 phosphorylation affords differential effects on ADD_ATRX domain interactions with histone H3.

Fig. 1.

ADD_ATRX binding to the H3 N terminus is maintained in the presence or absence of H3S10ph. (A) Primary amino acid sequence of the histone H3 N-terminal tail (1–14). The H3K9methyl-S10phos switch is shown above the sequence. (B) IP assays examining binding interactions between the ADD_ATRX domain and modified histone H3 peptides, as indicated. High salt: IP assays were performed in buffer containing 500 mM NaCl. (C and D) Cocrystal structures of domain–peptide complexes for the CD of HP1ɑ (C, prepared from PDB ID code 1KNE) and the ADD_ATRX domain (D) bound to H3K9me3 (1–15) and H3K9me3S10ph (1–15) peptides, respectively. In both panels, H3 peptides are indicated in yellow, and amino acids of the polar residues comprising the S10ph recognition state are displayed in pink. (E and F) ITC curves for the HP1ɑ CD (E) and ADD_ATRX domain (F) with H3 (1–15) peptides containing various modifications. (G) Peptide IPs examining binding between the HP1ɑ CD/ADD_ATRX domain and modified histone H3 peptides in the presence or absence of phosphatase treatment.

Because the ADD_ATRX domain is exempt from H3K9methyl/S10phosphorylation “switching,” a phenomenon that differs from the chromodomain (CD) of heterochromatic protein 1 (HP1α) (16), we next sought to compare the molecular structure of the ADD_ATRX domain to a previously reported HP1α CD crystal structure in complex with an H3K9me3 peptide (Fig. 1C) [PDB ID code 1KNE) (17, 18)]. To this end, we solved the cocrystal structure of the ADD_ATRX domain in complex with an H3K9me3S10ph (1–15) peptide at 2.6 Å (Table S1) and further traced the amino acid residues (1–11) of the bound peptide (Fig. S1A). The intermolecular features of the ADD_ATRX domain bound to H3K9me3 are identical to those in recent reports (12–14), demonstrating that H3K9me3 resides within a composite pocket formed between the GATA-like domain and the PHD zinc finger (Fig. S1B). Phosphorylated H3 serine 10 is positioned away from the core peptide-binding surface and forms minimal contacts with ADD_ATRX (Fig. 1D). Strikingly, unmodified arginine 8 on histone H3 (H3R8) was found to form relayed salt bridges or multiple hydrogen bonds with a surface residue, E225, in the ADD domain-binding pocket, as well as with phosphorylated H3S10 (Fig. 1D). Such neutralization effects thus allow for ADD_ATRX domain binding to the combinatorial H3 signature of “K4me0-K9me3-S10ph” (Fig. S1B). Subsequent isothermal titration calorimetry (ITC) data support our structure-based observations indicating that, whereas HP1α CD binding is radically diminished by H3S10ph, likely due to its charge repulsion with E56 of the HP1α CD (105× lower affinity; K_D, 170 μM for H3K9me3S10ph vs. 1.61 μM for H3K9me3) (Fig. 1E and Fig. S2), ADD_ATRX domain binding is insensitive to H3S10 phosphorylation when bound to the H3K9me3S10ph peptide displaying a similar affinity to binding interactions observed with the H3K9me3 peptide alone (K_D, 0.19 μM for H3K9me3S10ph vs. 011 μM for H3K9me3); moreover, substitution of H3R8 with an alanine (H3R8A) significantly reduced binding by ∼30-fold (Fig. 1F and Fig. S2). Peptide IPs with the recombinant HP1α CD vs. the ADD_ATRX domain similarly confirmed contrasting abilities to recognize and bind to H3K9me3S10ph because phosphatase-treated H3K9meS10ph peptides regained HP1α CD binding that was previously not observed with the H3K9me3S10ph peptide (Fig. 1G). In summary, these data indicate that the ADD_ATRX domain is a distinct “reading module” that can recognize and bind to H3K9me3S10ph in vitro.

H3K9me3S10ph Is an Activity-Dependent Heterochromatic Histone PTM in Neurons.

Given existing links between ATRX dysfunction and associated mental retardation syndromes, as well as previous evidence for activity-dependent expression of H3S10ph in the CNS, we next sought to investigate potential links between H3K9me3S10ph and the recruitment of full-length ATRX to neuronal chromatin. Although H3K9me3S10ph has extensively been studied in mitotic cells, its existence and function have not yet been reported in postmitotic neurons. Because ATRX is critical to numerous neurodevelopmental processes (e.g., dendritic spinogenesis) and its expression is enriched in neurons (19, 20), we aimed to better understand whether ATRX cellular and genomic localization is affected by the presence of this dual PTM. Similar to many other histone phosphorylation events, H3K9me3S10ph was not expressed in quiescent (i.e., nonstimulated) neurons. Therefore, we sought to examine whether its expression could be induced in response to pharmacological manipulations that increase neuronal activity, similar to that of other histone phosphorylation marks (e.g., H3S10ph). To this end, Western blot analysis using an antibody that specifically recognizes H3K9me3S10ph revealed that this combinatorial modification occurs exclusively during periods of heightened neuronal activity. Forskolin, an activator of adenylyl cyclase and protein kinase A (PKA), but not TPA, an activator of protein kinase C (PKC), induced a marked up-regulation of H3K9me3S10ph expression (Fig. 2A). Similarly, neuronal depolarization via potassium chloride (KCl) also increased the expression of H3K9me3S10ph in primary neurons. Moreover, such inductions of H3K9me3S10ph after KCl treatment were found to be mediated by PKA-dependent signaling because PKA inhibition (H89), but not inhibition of other kinases, was found to attenuate KCl-mediated increases in the mark’s expression (Fig. 2A). In addition, stimulation-induced H3K9me3S10ph (and H3S10ph alone, as previously reported) was found to be specific because H3T3ph (Fig. 2A) was not similarly increased after PKA activation or KCl depolarization. H3T3ph was observed only after treatments with Calyculin A, a phosphatase inhibitor, which similarly induces the expression of H3K9me3S10ph in neurons. Neither HP1α nor Atrx expression was affected by the treatment conditions examined. Given the fundamental importance of PKA signaling to synaptic plasticity (e.g., long-term potentiation), both during neurodevelopment and in the adult brain, it is likely that altered recognition of this combinatorial mark may be associated with ATR-X syndrome-associated pathologies.

Fig. 2.

Activity-dependent H3K9me3S10ph colocalizes with Atrx, but not HP1ɑ, in primary neurons. (A) Western blot analysis of acid-extracted histones from primary cortical neurons in the presence or absence of PKA activation (Forskolin) or depolarization (KCl) in combination with numerous pharmacological inhibitors (Bis, Bisindolylmaleimide I, a PKC inhibitor; H89, H-89 dihydrochloride, a PKA inhibitor; PD98059, a MEK kinase inhibitor; Calyculin A, a potent PP1 and PP2A inhibitor; and EGTA, Ethylene glycol tetraacetic acid, a chelator for calcium ions). HP1α and Atrx protein expression were similarly examined. (B and C) Double immunofluorescence of neuronal nuclei ± stimulation (Forskolin) using antibodies reactive against H3K9me3S10ph (red) and Atrx (B) or HP1ɑ (C) (green). (Scale bar: 5 μm.) Neuronal nuclei were identified with DAPI. Percentages of colocalization between either Atrx or HP1α and H3K9me3S10ph are included to the right (n = 8 cells per group). Data are presented as mean ± SEM. N.D., not detectable.

Atrx, but Not HP1α, Costains with H3K9me3S10ph After Neuronal Stimulation.

Next, to examine the relationship between cellular distribution patterns of H3K9me3S10ph and Atrx/HP1α in neurons, we performed coimmunofluorescence analyses in the presence or absence of neuronal stimulation. Consistent with previous Western blot analyses, H3K9me3S10ph expression was absent in nonstimulated neurons. Atrx, however, was observed to be highly expressed and constitutively localized to the nucleus in primary cortical neurons, with occasional speckles seen within heterochromatin foci (Fig. 2B). In stimulated neurons, H3K9me3S10ph was markedly increased in expression and appeared at centromeric and pericentromeric regions within the nucleus (e.g., heterochromatic structures that surround the nucleolus and nuclear membrane). Notably, in response to neuronal stimulation, Atrx was found to reorganize within the nucleus whereby speckled Atrx signal was observed to now overlap significantly with H3K9me3S10ph. Partial colocalization patterns between Atrx and H3K9me3S10ph were also evident at the centromeric repeat region encircling the nucleolus (Fig. 2B). Conversely, HP1α existed primarily as nuclear puncta under control conditions whereas stimulation led to more diffusive patterns of HP1α staining, resulting in minimal overlap with H3K9me3S10ph (Fig. 2C). Together, these data demonstrate that Atrx, but not HP1α, is a heterochromatin-associated remodeling protein that interacts with H3K9me3S10ph during periods of enhanced neuronal activity in vivo.

Atrx and H3K9me3S10ph Are Enriched at Heterochromatic Repeat Sequences of Stimulated Neurons.

To assess the genome-wide impact of Atrx enrichment at sites of enhanced H3K9me3S10ph during periods of increased neuronal activity, we next applied chromatin immunoprecipitation coupled to massively paralleled DNA sequencing (ChIP-seq). Under nonstimulated conditions, Atrx was found to enrich most heavily throughout intergenic regions of the genome (Fig. 3A and Dataset S1) (62% of total occupancy), consistent with previous ChIP-seq data in primary human erythroid cells and mouse embryonic stem cells (ESCs) (7), suggesting that Atrx is primarily heterochromatic. Additionally, Atrx was observed to localize at a small subset of genic promoters and gene bodies. After further classification of transcripts based upon their levels of expression (i.e., high, medium, low), Atrx was found to most significantly enrich at weakly expressed genes (Fig. 3B and Dataset S1), consistent with previous reports indicating a repressive transcriptional role for Atrx (21). After neuronal stimulation, Atrx was observed to display significant patterns of differential enrichment throughout percentromeric heterochromatin, with ∼14% of Atrx differential events occurring within this genomic region (Fig. 3C and Dataset S2). Because Atrx expression is unaffected by neuronal stimulation, such differential enrichment likely reflects pericentromeric reorganization of Atrx binding at common repetitive sequences. Centromeric and pericentromeric regions are composed of highly repetitive minor satellite and major satellite DNA sequences, respectively, and produce noncoding RNA required for maintaining the heterochromatic state necessary for centromeric integrity (22).

Fig. 3.

Heterochromatic patterns of Atrx enrichment are maintained in the presence of H3K9me3S10ph after neuronal stimulation. (A) Genomic distribution of Atrx peaks in nonstimulated primary neurons. (B) Odds ratio analysis of correlations between Atrx peak enrichment and gene expression (binned as high, medium, and low) in nonstimulated neurons. P values were determined using the Fisher’s exact test and are labeled in red. (C) Genomic distribution of Atrx differential sites after neuronal stimulation (KCl depolarization). (D) Relative enrichment of Atrx and H3K9me3S10ph at different classes of repetitive sequences in response to neuronal stimulation (control enrichment values are normalized “1” represented by a dashed line). To do so, ChIP reads were directly aligned to a library of mouse consensus repetitive sequences (i.e., the repeatome). Data are represented as stimulation-induced fold enrichments over input in comparison with control neurons. (E) ChIP-qPCR validation (biological triplicates) of H3K9me3S10ph enrichment in response to neuronal stimulation at specific repetitive sequences, as well as at a highly active gene (Rps19), which served as a negative control. N.D., not detectable. Where appropriate, data are represented as mean ± SEM.

Because ChIP-seq enables more focused analyses of protein localization at repetitive genomic sequences, we next analyzed Atrx and H3K9me3S10ph enrichment throughout different classes of repetitive sequences after construction of the mammalian “repeatome.” Consistent with predictions from Western blot and immunofluorescence analyses, H3K9me3S10ph was found to increase in its enrichment at various repetitive sequences after neuronal stimulation, primarily those representing endogenous retrovirus (ERV) elements, satellite sequences, and simple repeats (Fig. 3D and Dataset S3). Such increases in H3K9me3S10ph at specific repetitive sequences were further validated via quantitative ChIP in independent samples (Fig. 3E). Although Atrx did not display considerable alterations in its enrichment at repetitive sequences after stimulation, likely due to its reorganization throughout regions of common repeat sequences, its patterns of occupancy were appropriately maintained in the presence of increased H3K9me3S10ph. These data are consistent with previous biochemical and structural analyses (Fig. 3D). Taken together, these data indicate that, during periods of enhanced neuronal activity, Atrx enrichment at, and recruitment to, specific heterochromatic loci are effectively maintained and can tolerate phosphorylation-induced changes in the local chromatin environment, thereby distinguishing its function from other heterochromatic proteins, such as HP1α.

miR-Mediated Depletion of Atrx Increases Minor Satellite Transcription in Stimulated Neurons.

In yeast and mammalian cells, it has previously been shown that low levels of noncoding RNA from centromeric and pericentromeric regions are transcribed and are necessary for maintaining repressive chromatin states during cell division. To further test whether maintenance of Atrx at H3K9me3S10ph marked loci represents a physiologically relevant blockade to repetitive element transcription in neurons, we next assessed the impact of Atrx depletion on such events in primary neurons. We first generated four microRNA (miR) constructs with a GFP reporter directed against distinct Atrx coding sequences to examine their efficacy in depleting Atrx expression in mESCs in comparison with a negative control (miR_Neg). In doing so, we identified three constructs that efficiently depleted Atrx expression in GFP⁺ FACS-sorted mESCs 24 h after transfection (miR_Atrx_2, miR_Atrx_3, and miR_Atrx_4) (Fig. 4A). miR_Atrx_3 was further subcloned into a lenti-GFP viral construct, packaged at high titer, and transduced into primary neurons for 7 d in vitro (DIV). High levels of Atrx knockdown were confirmed in primary neurons at DIV 11, with no effects observed on the expression of numerous loading controls, such as tubulin, H3K9me3, and histone H3 (Fig. 4B).

Fig. 4.

Atrx depletion increases minor satellite transcription in stimulated neurons. (A) Western blot analysis of GFP⁺ FACS-sorted mouse ESCs transfected with a negative control (miR_Neg) or Atrx-specific miR constructs coexpressing with GFP (sets 1–4 from Invitrogen). (B) Western blot analysis of primary cortical neurons transduced with Lenti-GFP-miR_Atrx_3 vs. miR_Neg. (C) RT-qPCR analysis of repeat element transcription in the presence or absence of KCl stimulation after infection with Lenti-GFP-miR_Neg vs. miR_Atrx_3. n = 3 biological replicates per group. Ct values were normalized to Gapdh. **P < 0.01; n.s., nonsignificant. (D) Model of H3K9me3S10ph associations with Atrx, but not HP1ɑ, at minor satellite sequences after periods of heightened neuronal activity. HP1ɑ is comprised of an N-terminal chromodomain (CD) and a C-terminal chromoshadow domain (CSD), separated by a hinge region. ATRX contains a conserved N-terminal ATRX-DNMT3-DNMT3L (ADD) domain and C-terminal collinear domains with ATPase activity found in the switch2/sucrose nonfermentable2 (SWI2/SNF2) family of proteins (SNF2h). Where appropriate, data are represented as mean ± SEM.

If indeed the role of Atrx is to maintain chromatin associations at repetitive sequences marked by H3K9me3S10ph after periods of enhanced activity, then one might predict that depletion of Atrx would abrogate heterochromatic stability by allowing for increased repeat transcription at Atrx-associated H3K9me3S10ph loci. To address this possibility, we next assessed the impact of Atrx depletion on repeat transcription via reverse transcription-quantitative PCR (RT-qPCR) after 5 h of KCl-mediated depolarization. Using previously reported primers for analyzing transcription of several repeat elements in mammalian cells (23), we observed that stimulation alters the expression of numerous repeat transcripts, including ERV elements [e.g., intracisternal A-particles (IAPs)] and major satellites, whereas other repeat transcripts are decreased in expression (e.g., 18S ribosomal RNA) or remain unchanged (e.g., minor satellites) (Fig. 4C). Although Atrx knockdown had little effect on repeat transcription in nonstimulated neurons, presumably due to homeostatic maintenance of stable heterochromatic states under control conditions, Atrx depletion was found to significantly increase the expression of minor satellite sequences after neuronal stimulation. Thus, it seems that Atrx inhibition directly potentiates the heterochromatic transcription of minor satellite sequences, indicating a causal role for Atrx in the silencing of such elements during periods of heightened neuronal activity and enhanced H3K9me3S10ph enrichment. These observations indicate a critical role for Atrx in the maintenance of heterochromatic stability in developing neurons and suggest a potentially causative mechanism underlying the severe pathologies associated with X-linked alpha thalassemia mental retardation syndrome.

Discussion

In this study, we first sought to identify whether other N-terminal histone modifications might influence binding interactions between the ADD_ATRX domain and histone H3. In agreement with previous studies, H3K4me3 was found to disrupt ADD_ATRX binding to histone H3 whereas H3K9me3 promoted enhanced ADD_ATRX binding in comparison with that of the unmodified H3 N terminus. Additionally, we found that ADD_ATRX binding was markedly disrupted by H3T3ph and H3T6ph but not by H3S10ph, either in the presence or absence of H3K9me3. These findings are in contrast to HP1α, whose chromodomain interactions with H3K9me3 were dramatically impaired by the addition of H3S10ph in vitro. Structural comparisons between the HP1ɑ CD bound to an H3K9me3 peptide and the cocrystal structure of ADD_ATRX in complex with H3K9me3S10ph further elucidated the specific amino acid residues that allow for H3S10ph/H3K9me3S10ph state recognition by the ADD_ATRX domain.

In assessing such interactions in vivo, as well as the physiological relevance of these binding properties in the CNS, we further showed that H3K9me3S10ph, like H3S10ph alone, represents an activity-dependent histone PTM in primary neurons, and that ATRX, but not HP1, colocalizes within heterochromatic compartments of the nucleus with H3K9me3S10ph during periods of heightened cellular activity. Subsequent genome-wide studies further supported the ability of the ADD_ATRX domain to tolerate enhancements of H3K9me3S10ph at heterochromatic loci, indicating that ATRX is capable of maintaining its associations with silenced repetitive sequences in the presence of increased neuronal activity. In neurons, repetitive element transcription is generally limited by such binding interactions, a homeostatic process that is required to maintain chromatin structural stability and neurophysiological outcomes. However, after virally mediated ATRX knockdown in neurons, which mimicked many of the loss-of-function mutations observed in human patients, specific subsets of repeat transcripts were found to increase in their expression in response to neuronal stimulation, representing an unsilencing event typically uncharacteristic of neurons. Combined, our structural and functional studies indicate that ATRX is essential to the regulation of heterochromatic stability in neurons and that its insensitivity to H3K9me3S10ph upon neuronal stimulation acts as an additional protective blockade to aberrant patterns of heterochromatic unsilencing and cellular dysfunction.

Atrx Mutations, Heterochromatic Transcription, and Disease Pathology.

The precise molecular mechanisms by which Atrx is targeted to heterochromatin have not yet fully been resolved. Previous studies have demonstrated that the ADD domains of ATRX and the de novo DNA methyltransferases recognize methylated vs. unmodified lysine residues in the histone H3 N terminus (13, 14, 24); however, the impact of neighboring serine/threonine phosphorylation events is not known. Our current studies demonstrate that threonine phosphorylations (T3 and T6) proximal to lysine 4 are capable of disrupting binding interactions between ADD_ATRX and the H3 N terminus whereas serine phosphorylation (S10), which lies adjacent to lysine 9, has little effect on its binding. Because these site-specific phosphorylation interactions are similarly preserved in ADD domains of numerous DNA methyltransferases (e.g., Dnmt3a, Dnmt3b, and Dnmt3l), it is likely that histone phosphorylation may play an additional role in the regulation of DNA methylation.

Mutations in ADD domains of numerous proteins have been linked to a variety of pathological states, including immunodeficiency syndromes, solid and blood cancers, and neurological disorders (25–27). Nearly half of all of the mutations associated with ATR-X syndrome occur within the ADD domain (15), suggesting a functional significance for this histone modification-reading module in disease pathology. Disease-causing mutations in the ADD_ATRX domain reduce affinities for chromatin recognition, thereby resulting in reduced ATRX protein expression levels and/or weakened activity at target loci (25). Therefore, our findings that Atrx knockdown derepresses minor satellite repeat transcription during periods of increased neuronal activity may provide clues to the molecular mechanisms promoting pathological states associated with ATR-X syndrome. However, further examinations of the impact of patient-associated ATRX mutations on repetitive element stability in the CNS are needed.

In addition to direct chromatin recognition properties, the ADD_ATRX domain interacts with MeCP2 in neurons (11, 28) and is recruited to methylated DNA. The functional consequence of this interaction, however, remains largely unknown. Because loss of MeCP2 expression in mature neurons disrupts Atrx enrichment at heterochromatic foci, but not vice versa (11), it is possible that Atrx function in the CNS might also depend on MeCP2 associations with chromatin. To this end, the ADD_ATRX domain is notably absent from Atrx orthologs in Caenorhabditis elegans and Drosophila, organisms that are known to lack DNA methylation (29, 30). Future studies examining the interplay between histone phosphorylation, DNA methylation, MeCP2 recruitment, and Atrx function at heterochromatic loci (e.g., minor satellites, major satellites, IAP repeats, etc.) will be required to gain a complete picture of how altered epigenetic landscapes resulting from Atrx dysfunction associate with human disease.

Heterochromatic Stability and Activity-Dependent Neuronal Responses.

Many noncoding RNAs are now known to exist as integral components of chromatin–protein binding complexes, many of which are critically important to transcriptional maintenance. For example, noncoding RNAs associate with sex chromosome dosage compensation complexes (31), the telomere complex (32), pericentromeric heterochromatin (32), and nuclear stress bodies (33). Although it is well known that tightly regulated noncoding transcription of heterochromatin plays an important role during mitosis (34), the regulation of the repetitive element transcription has not yet been studied in neurons.

Previous studies have demonstrated that heterochromatic transcription is intimately linked to cellular proliferation and differentiation, as well as responses to cellular stress (23, 35, 36). More recently, altered repetitive element transcription in the adult CNS of rodents exposed to chronic stress or drugs of abuse has also been observed; however, the functional consequences of these events have not been fully delineated (37, 38). Minor satellite transcripts form an integral portion of centromeric protein complexes, and, in the absence of these noncoding sequences, appropriate assembly of heterochromatin at chromosomal centromeres is severely impaired in mammalian cells (39, 40). Therefore, minor satellite RNAs may serve as a molecular scaffold for chromatin-remodeling complexes at the centromere and may be essential to chromosomal stability. Therefore, in neurons, such repression of these heterochromatic sequences by ATRX in the face of activity-dependent methyl/phos switching (e.g., H3K9me3S10ph) may play a prominent role in gene regulation, centromeric integrity, and neuronal maturation processes necessary to preserve normal patterns of synaptic development and organization.

Conclusion

In conclusion, our findings demonstrate a previously unidentified molecular mechanism linking a combinatorial histone modification to the ADD_ATRX domain, an interaction that is important for maintaining repeat stability in postmitotic neurons. Because many nuclear proteins contain histone-reading modules similar to that of ATRX, we anticipate that further characterizations of these regulatory domains will greatly enhance our understanding of epigenetic mechanisms controlling integral processes associated with neurodevelopment and neural plasticity.

Materials and Methods

Recombinant Protein Preparation.

The ADD domain encompassing residues 167–289 of human ATRX was cloned into a pGood6p vector (an in house-modified vector based on pGEX-6p-1) containing an N-terminal GST tag and was verified by sequencing. The GST construct was then overexpressed in Escherichia coli BL21 (DE3) cells (Novagen). After overnight induction with 0.4 mM isopropyl β-d-thiogalactoside at 20 °C in LB media supplemented with 0.1 mM ZnCl₂, cells were harvested in buffer containing 0.4 M KCl, 20 mM Tris⋅HCl, pH 7.5, and disrupted by the EmulsiFlex-C3 homogenizer (Avestin). After centrifugation, supernatants were loaded onto a GST affinity column. Bound protein was eluted with 20 mM glutathione and cleaved using PreScission proteases (GE Healthcare). The ADD_ATRX domain was separated from the cleaved GST tag and purified via size-exclusion chromatography on a Hiload 16/60 Superdex 75 column (GE Healthcare) in elution buffer: 20 mM Tris⋅HCl, pH 7.5, 200 mM KCl, and 2.5 mM DTT. Peak fractions corresponding to ADD_ATRX were then concentrated to ∼10–15 mg⋅ml⁻¹ for crystallization.

The Drosophila HP1α CD corresponding to amino acids 17–76 of the full-length protein was cloned into pGood6p with an N-terminal GST tag. Expression and purification of the CD_HP1α domain were performed in an essentially identical procedure to that described for ADD_ATRX, with the exception that the ITC titration buffer (see below: 0.1 M KCl, 20 mM Hepes-Na, pH 7.0, 5 mM β-mercaptoethanol) was used as the elution buffer during gel filtration.

Crystallization, Data Collection, and Structural Determination.

Crystallization was performed via the sitting drop vapor diffusion method at 4 °C by mixing equal volumes (1-2 μL) of protein and reservoir solution. WT ADD_ATRX was premixed with H3_(1-15) peptides (1:2) at a concentration of 15 mg⋅ml⁻¹ in a buffer containing 200 mM KCl, 20 mM Tris⋅HCl, pH 7.5, and 2.5 mM DTT. Complex crystals were obtained using the following reservoir conditions: 13–20% PEG 4000, 0.1 M Mes, pH 6.0, and 3 mM MgCl₂. Diffraction data of the complexed crystals were collected under the cryo-protectant Santovac Cryo Oil (Hamptom Research). ADD_ATRX-H3_(1-15) complex data were collected at beamline BL17U at the Shanghai Synchrotron Radiation Facility at 0.9789 Å. All data were indexed, integrated, and merged using the HKL2000 software package (www.hkl-xray.com). Detailed data-collection statistics are summarized in Table S1.

Cocrystal structures were solved by molecular replacement using the MOLREP program from the CCP4 software suite (www.ccp4.ac.uk) with free-form ADD_ATRX as the search model. Structural refinement was performed using the PHENIX software suite (www.phenix-online.org), with iterative manual model building using COOT (www.biop.ox.ac.uk/coot). Final structural refinement statistics are summarized in Table S1. All structural figures were created using PYMOL (www.pymol.org/).

ITC Measurements.

All calorimetric experiments with ADD_Atrx and CD_HP1ɑ proteins were conducted at 25 °C using a MicroCal iTC200 instrument (GE Healthcare). ADD_ATRX samples were dialyzed in the following buffer: 20 mM Hepes-Na, pH 7.0, 100 mM KCl, and 5 mM β-mercaptoethanol. Protein concentration was determined by absorbance spectroscopy at 280 nm. Peptides were quantified by weighing on a large scale and then aliquotted and freeze-dried for individual use. Acquired calorimetric titration curves were analyzed using Origin 7.0 (OriginLab) using the “One Set of Binding Sites” fitting model. Detailed thermodynamic parameters of each titration are listed in Fig. S2.

Peptide IP Assays.

Briefly, biotin-conjugated histone peptides were incubated overnight at 4 °C with High Capacity Streptavidin Agarose (Pierce) and washed five times to remove unbound peptides. For IPs of the ADD_ATRX and CD_HP1ɑ proteins, 10 μL of packed peptide-streptavidin resins were incubated with 50 µg of protein for 4 h at 4 °C in a buffer containing 20 mM Hepes, pH 7.9, 150 mM KCl, 0.01% Nonidet P-40, 10% (vol/vol) glycerol, 5 mM BME, 0.4 mM PMSF, and protease inhibitors (Roche). After extensive washing, proteins bound to the beads were extracted with elution buffer at pH 2.5, separated by SDS/PAGE, and examined by Coomassie blue staining. Peptide gels were similarly run and examined by Coomassie blue staining to control for equal IP conditions.

Animals.

Pregnant female C57BL/6 mice were singly housed in a colony with a 12-h light/dark cycle (lights on from 0700 hours to 1900 hours) at constant temperature (23 °C) with ad libitum access to water and food. All animal protocols were approved by the Institutional Animal Care and Use Committee at The Rockefeller University.

Primary Neuronal Culture and Pharmacological Manipulations.

Pregnant female mice were euthanized with carbon dioxide on embryonic day 16.5. Embryonic cortices were harvested, dissociated, and plated at a density of approximately one brain (∼6 × 10⁶ cells) per poly-d-lysine (0.1 mg/mL; Sigma) coated six-well or 10-cm cell-culture plate (Corning). For KCl stimulations, neurons were cultured through DIV 8–11 (depending on the experiment) before depolarization. Control and KCl plates were pretreated with 100 µM D-AP5 (Tocris) and 1 µM Tetrodotoxin (Tocris) for 1 h at 37 °C. For plates to be stimulated, 1/3 of the media was removed and saved. To collected media, 170 mM KCl, 2 mM CaCl₂, 1 mM MgCl₂, and 10 mM Hepes, pH 7.9, were added. The isotonic KCl-containing media was then restored to each plate at 1×, and neurons were incubated in a cell culture incubator for 5 h at 37 °C. Control plates underwent media exchanges without addition of isotonic components. For application of pharmacological activators or inhibitor, cultured neurons (DIV 8–11) were treated for 5 h with 2 μM Bisindolylmaleimide I (Cell signaling), 30 μM Forskolin (Cell signaling), 400 nM TPA (12-O-Tetradecanoylphorbol-13-Acetate; Cell signaling), 20 μM H-89 dihydrochloride (Cell signaling), 50 μM PD98059 (Cell signaling), 100 nM Calyculin A (Cell signaling), and 1 mM Ethylene glycol tetraacetic acid (Sigma). Five-hour stimulation protocols were used to assess the strength of Atrx maintenance at H3K9me3S10ph-marked loci, repetitive sequences that generally remain under tight regulatory control even during periods of cellular insult. Neurons were then scraped and flash-frozen in liquid nitrogen for further analyses.

Protein Expression Analysis.

Western blot analysis was carried out on acid-extracted histones, as previously described (41). Aliquots of protein were run on SDS/PAGE gels [4–20% (wt/vol)], transferred to a polyvinylidene difluoride (PVDF) membrane and probed with primary antibodies (anti-H3T3ph, 1:1,000, Millipore; H3K9me3S10ph, 1:1,000, Millipore; H4, 1:1,000, Millipore; ATRX, 1:1,000, BD Bioscience; GFP, 1:1,000, Millipore; Tubulin, 1:1,000, Millipore; H3K9me3, 1:1,000, Sigma; H3, 1:1,000). Membranes were washed, reacted with enhanced chemiluminescence reagent (ECL; Amersham Biosciences), and visualized.

Immunolabeling and Imaging.

Cortical neurons were fixed with 2% (wt/vol) paraformaldehyde in 1× PBS for 10 min, permeabilized with cold 100% methanol for 5 min at −20 °C, and blocked in 2% BSA for 30 min. Cells were incubated with primary antibodies against Atrx (1:200; Santa Cruz Biotechnology), HP1ɑ (1:200; Active Motif, and H3K9me3S10ph (1:200; Millipore) for 2 h, washed with 0.05% Tween 20 in PBS, and incubated with appropriate Alexa Fluor-conjugated secondary antibodies for 1 h. Nuclei were counterstained with 300 nM DAPI (4,6-diamidino-2-phenylindole; Molecular Probes) for 3 min and mounted in ProLong Gold Antifade Reagent (Molecular Probes). Immunofluorescent images were taken as Z-stacks with a DeltaVision image restoration microscope system and deconvolved with the SoftWoRx software (Applied Precision LLC). Exposure times and settings for deconvolution were constant for all samples to be compared within any given experiment. Projected images were generated by Image J (NIH). Colocalization was measured using ImageJ software (Colocalization Coloc 2, Pearson’s correlation coefficient), and background plus threshold corrections were applied for each region of interest measured.

ChIP.

ChIPs for H3K9me3S10ph and Atrx were performed with ∼2 × 10⁷ neurons per sample in biological triplicates per group. Primary cortical neurons ± isotonic KCl stimulation (5 h) were cross-linked with 1% paraformaldehyde (PFA) for 12 min at room temperature and then quenched for 5 min with 0.125 M glycine. Pelleted cells were washed 5× with 1× PBS containing protease inhibitors before being subjected to lysis, sonication, IP (7.5 μg of antibody per sample was used—see antibody above), and DNA purification, as previously described (42). Primer sequences for qChIP experiments are listed in Dataset S4.

ChIP-seq.

After DNA purification, ChIP-seq libraries were prepared according to the Illumina protocol and sequenced with the HiSEq. 2000. Reads were mapped to the mouse genome (build 37, mm9) using the Bowtie alignment software (43).

Data Analysis.

Peak calling, differential analysis, and genomic distribution.

Basal peaks for Atrx were identified using the MACS (44) program with default parameters, with the exception that model building was inhibited and the shift size was set to 150 bp. Differential peak detection was performed by diffReps (45) with default parameters and the mode was set as “peak.” Once peak and differential peak lists were obtained, they were analyzed for genomic distribution using a Python program called “region_analysis” (github.com/shenlab-sinai/region_analysis), which is part of the diffReps (45) package.

Correlational analysis of Atrx with basal gene expression.

The different types of promoter regions from region_analysis were clamped to a single classification of “promoters” so that all genic regions were separated into two classifications: promoter and genebody. Using the fragments per kilobase of exon per million fragments mapped (FPKM) values derived from an independent RNA-seq experiment, genes were separated into three groups using arbitrarily chosen cutoffs: high (FPKM > 30), medium (FPKM > 0 and ≤ 30), and low (FPKM = 0). Gene lists derived from ChIP-seq peaks were then tested for significant overlaps with gene lists from RNA-seq expression analysis using the GeneOverlap (46) package.

Analysis of enrichment of Atrx and H3K9me3S10ph at repeat elements.

ChIP-seq enrichment at repeat elements was analyzed using in-house, open-source software referred to as diffRepeats (github.com/shenlab-sinai/diffRepeats). Analysis of stimulation-induced changes in ChIP-seq enrichment patterns followed a previously described procedure (37).

miR Construct Cloning and Lentiviral Packaging.

miR oligos against mouse Atrx were purchased (Invitrogen; see oligo sequences below), annealed, and ligated into pcDNA6.2-GW/EmGFP before being subcloned into a lentivirus vector (pRRLsin.PPTs.hCMV.GFP.Wpre). The negative control (miR_Neg) provided by the kit does not target any sequence in the vertebrate genome and was similarly cloned into lentiviral constructs. Lentiviral particles were prepared using common protocols. All viruses were titered using the QuickTiter Lentivirus Titer Kit (VPK-107; Cell Biolabs) and analyzed for GFP expression. For viral studies in cultured neurons, cells were infected with lentivirus at equal, high titer on DIV 4. On DIV 11, neurons were stimulated and harvested. Although we examined the impact of miR-mediated knockdown of Atrx using a single miR construct, it should be noted that off-target effects of small RNAs are possible. Sequences were as follows: miR_Atrx 1 top, TGCTGTTGACCTGCTGTCCACAAGCTGTTTTGGCCACTGACTGACAGCTTGTGCAGCAGGTCAA; miR_Atrx 1 bottom, CCTGTTGACCTGCTGCACAAGCTGTCAGTCAGTGGCCAAAACAGCTTGTGGACAGCAGGTCAAC; miR_Atrx 2 top, TGCTGTATTGTAGACAACTCCTTTCGGTTTTGGCCACTGACTGACCGAAAGGATGTCTACAATA; miR_Atrx 2 bottom, CCTGTATTGTAGACATCCTTTCGGTCAGTCAGTGGCCAAAACCGAAAGGAGTTGTCTACAATAC; miR_Atrx 3 top, TGCTGATTCACGGCTAACTGCTCTCTGTTTTGGCCACTGACTGACAGAGAGCATAGCCGTGAAT; miR_Atrx 3 bottom, CCTGATTCACGGCTATGCTCTCTGTCAGTCAGTGGCCAAAACAGAGAGCAGTTAGCCGTGAATC; miR_Atrx 4 top, TGCTGTGCATTAGTAGAACCATCCAAGTTTTGGCCACTGACTGACTTGGATGGCTACTAATGCA; miR_Atrx 4 bottom, CCTGTGCATTAGTAGCCATCCAAGTCAGTCAGTGGCCAAAACTTGGATGGTTCTACTAATGCAC.

FACS.

ESCs (C57BL/6J background) were cultured under standard conditions. miR construct-transfected ESCs were dissociated with trypsin/EDTA, and GFP⁺ cells were sorted on a BD FACSAriaII-2. After sorting, ESCs (∼10⁵ cells) were immediately lysed for Western blot analysis.

RT-qPCR.

Total RNA from primary cortical neurons was extracted and column-purified using the RNeasy kit (Qiagen). Genomic DNA contamination was removed by DNase (Ambion) treatment and 1 μg of RNA from each sample was used to generate cDNA using the SuperScript First-Strand Synthesis kit (Invitrogen). Repeat transcripts were measured using SYBR Green PCR Master Mix (ABI) with Gapdh as an internal control. No signal was detected in any of the controls that lacked reverse transcriptase [Threshold cycle (Ct) > 35]. Repeat primers were adapted as described (23), and primer sequences for RT-PCR are listed in Dataset S4. Average threshold values (Ct) were determined from a minimum of two PCR reactions for target and control genes relative to Gapdh using the comparative ΔΔCt method. Transcript levels were then normalized as fold changes relative to the transcript for each of the negative shRNA-infected control neurons.

Statistics.

Paired Student’s t tests were performed with GraphPad Prism 6.0 (GraphPad Software). Data were considered statistically significant if P < 0.05, as indicated.