One of the first questions addressed was do patient mutations affect protein stability and function? Immunoblots of extracts from patient-derived EBV-transformed B-lymphocytes showed significantly reduced levels of ATRX protein from all patients tested (McDowell et al., 1999; Cardoso et al., 2000). Interestingly, in patients with early premature stop codons (e.g., p.Arg37X), translation was initiated from an internal methionine that produced a smaller truncated protein at ∼30% levels leading to a milder phenotype (Howard et al., 2004; Abidi et al., 2005; Basehore et al., 2015). Utilizing recombinant proteins, other studies demonstrated that mutations within the ATPase domain attenuated ATPase activity but did not reduce it, while mutations in the ADD domain or the PML-targeting domain reduced localization to chromocenters and PML nuclear bodies, respectively (Cardoso et al., 2000; Bérubé et al., 2008). Atrx-null mutations in mice show defective extraembryonic trophoblast development and die embryonically at ∼E9.5 (Garrick et al., 2006). Collectively, these studies indicate that ATR-X syndrome causing mutations are functional hypomorphs, while more severe mutations are not found and are presumably non-viable.

Several different ATRX-deficient mouse lines have been generated and used for functional characterization. The most widely used model is a floxed allele in which loxP sites flanked exon 18 which encodes the ATP-binding pocket (Bérubé et al., 2005; Garrick et al., 2006). These animals have been crossed with several different tissue-specific Cre driver lines to inactivate ATRX in skeletal muscle progenitors (Huh et al., 2012), Sertoli cells (Bagheri-fam et al., 2011), osteobalsts (Solomon et al., 2013), chondrocytres (Solomon et al., 2009), the retina (Medina et al., 2009; Lagali et al., 2016), and the developing forebrain (Bérubé et al., 2005) among others. A second transgenic line (Atrx^ΔE2) was developed by deleting exon 2 and replacing it with a SA-IRES-β-geo cassette (Nogami et al., 2011; Shioda et al., 2011). This mutation was meant to mimic the p.Arg37X mutation and make an N-terminally truncated ATRX protein by initiating translation from an internal methionine codon (Howard et al., 2004; Abidi et al., 2005). Both of these models will be discussed in more detail in the following sections. Finally, an overexpression transgenic line was created with the ATRX cDNA under control of a CMV enhancer/β-actin promoter which resulted in growth retardation, neural tube defects and a high incidence of embryonic lethality demonstrating the importance of ATRX dosage to normal development (Berube et al., 2002).

While each of these models has provided valuable insight into disease mechanisms (as highlighted below), the field still awaits a model whereby a single nucleotide variant is introduced into the ATRX gene to recreate a known patient mutation, such as the common p.Arg246Cys mutation within the ADD domain.

Microcephaly is common to many NDDs and has also been observed in mouse models that deleted other genes encoding chromatin remodeling proteins (Ronan et al., 2013). Most ATR-X patients develop postnatal microcephaly and, in instances where CT or MRI scans have been performed, mild cerebral atrophy was detected. Similarly, three patient autopsy reports also described that the brains were smaller in size (Gibbons, 2006).

The first indication that ATRX may be critical for cell growth came from co-culture experiments of embryonic stem cells (ESC) from control or Atrx-null cells. This experiment demonstrated that the Atrx-null cells were underrepresented after 4-days of co-culture. Flow cytometry was used to examine cell cycle distribution but no differences were observed suggesting that cells may have transitioned to a slower cycling, differentiated cell type (Garrick et al., 2006). Given that ATRX has high expression in the developing forebrain, the Atrx^fl/fl line was next crossed with the forebrain-specific Foxg1-Cre line (Atrx^Foxg1Cre) that initiates Cre expression in the developing telencephalon at ∼E8.5 (Hébert and McConnell, 2000). Loss of ATRX caused a 25–30% reduction in cell number with a noticeably smaller neocortex and hippocampus including almost a complete absence of the dentate gyrus that likely contributed to early postnatal lethality (Bérubé et al., 2005). Similar to ESC co-culture experiments, BrdU-pulse labeling experiments suggested no differences in the proportion of cycling cells. However, there was a dramatic increase in the number of TUNEL+ cells leading to a reduction in the number of neurons that reached the cortical layers (Bérubé et al., 2005). Similarily, the Atrx^ΔE2 mice were smaller and also showed brain hypocellularity, although to a milder extent (Nogami et al., 2011). Atrx inactivation in Sertoli and muscle cells, also showed a significant impact on the growth of the tissue (Bagheri-fam et al., 2011; Huh et al., 2012). However, a retina progenitor cell cKO only had a limited effect on the size of the mature tissue suggesting that defects in cell cycle progression lead to significant hypocellularity among tissues that require a rapid expansion over a narrow developmental timeframe (Medina et al., 2009).

Although not initially observed, delayed cell cycle progression through both S- and G2/M phases was later observed in other studies (Ritchie et al., 2008; Watson et al., 2013; Huh et al., 2016). For G2/M, the progression from prometaphase to metaphase was prolonged and associated with sister chromatid cohesion and congression defects that impaired proper separation at anaphase leading to DNA bridges and micronuclei (Ritchie et al., 2008). Evidence for DNA bridges and micronuclei in Atrx^Foxg1Cre mice were also detected by high magnification microscopy at the apical surface on cortical sections of the neuroepithelium where cortical progenitors complete mitosis. Interestingly, a recent study has also demonstrated that ATRX promotes telomere cohesion between sister telomeres to mediate the repair of DNA DSB (Lovejoy et al., 2020).

Defects in S-phase were observed using BrdU-pulse chase flow cytometry experiments where a delay from G1 to S-phase and also from G2/M to the following G1 phase was identified (Huh et al., 2012). Co-labeling experiments demonstrated that ATRX associated with replicating chromatin during mid-late S-phase and cytological analysis showed a high prevalence of genomic instability that was enriched at telomeres and pericentromeric heterochromatin (Huh et al., 2012; Watson et al., 2013). Moreover, treatment with a compound that binds and stabilizes G4 DNA increased the number of telomere dysfunction induced foci (TIFs) and decreased cell viability suggesting that G4 DNA formation was the main cause of replicative stress (Watson et al., 2013). Other studies indicate that replication stress at telomeres may be mediated by increased TERRA transcription (Nguyen et al., 2017). TERRA levels are tightly regulated and critical for both telomere formation, replication and maintenance (Bettin et al., 2019). However, when TERRA levels increase, as shown for ATRX-null cells, it enhances R-loop (RNA-DNA hybrid) formation and G4 DNA stabilization, each of which increase replication fork stalling and collapse that then induces homology directed repair (HDR) and TIFs (Nguyen et al., 2017). The regulation of R-loops has also been proposed for other proteins that interact with G4 DNA during replication and/or at telomeres (Zhou et al., 2014; Ribeiro de Almeida et al., 2018; Toubiana and Selig, 2018; Maffia et al., 2020). It should also be mentioned here that somatic ATRX mutations, and to a lesser extent H3.3 and DAXX mutations, are prevalent in cancers characterized by ALT (alternative lengthening of telomeres), a HDR mechanism to maintain telomere length that is normally suppressed by ATRX (Heaphy et al., 2011; Lovejoy et al., 2012; Schwartzentruber et al., 2012; Pickett and Reddel, 2015; Verma and Greenberg, 2016).

Another indicator of replicative stress as a major impediment to growth of Atrx-null cells was demonstrated by studies showing an increased sensitivity to hydroxyurea, enhanced DSBs, and the use of DNA fiber analysis to show increased stalled replication forks and reduced origin firing (Leung et al., 2013; Clynes et al., 2014; Huh et al., 2016). Mechanistically, ATRX physically interacts with the MRN complex where it is thought to block HDR at stalled replication forks to allow for fork restart after the G4 DNA is resolved (Clynes et al., 2014). Indeed, one group demonstrated that fork protection could be restored by treatment with an Mre11 exonuclease inhibitor (Huh et al., 2016). This study also suggested that hyperactivation of poly (ADP-ribose) polymerase-1 (Parp-1) during neurogenesis may function as a compensatory mechanism to protect stalled replication forks from collapse and HDR, thus dampening the extent of cell loss during neurogenesis.

During mouse cortical development, the cortical layers are formed in sequential fashion from a pool of neural progenitor cells (NPC) that must continue to proliferate to maintain the pool size. Alterations in NPC proliferation depletes the pool often resulting in altered cell lamination typically observed as a reduction in upper layer neurons. For Atrx^Foxg1Cre mice, the most proliferative NPCs that ultimately would become upper layer neurons are more susceptible to incur replication-induced DNA damage. Frequently the resulting genomic instability will occur at telomeres and pericentromeric heterochromatin, but it could also occur at other genomic regions that can form G4 DNA or similar secondary DNA structures that induce replication fork stalling and collapse. Accumulation of sufficient damage further leads to their demise and decreases neuron production and brain size. Indeed, the Atrx^Foxg1Cre forebrain is reduced in size with a compromised production of upper layer neurons (Ritchie et al., 2014; Huh et al., 2016). Similar results have been observed in mice lacking CHD4 and SMARCA5 where the NPCs either fail to progress through the cell cycle or incur significant DNA damage, respectively, prior to undergoing apoptosis (Alvarez-Saavedra et al., 2014, 2019; Nitarska et al., 2016).

The SWI/SNF complex is also critical for brain development but utilizes different mechanisms than ATRX. The SWI/SNF complex is required during early neurogenesis for differentiation from radial glial progenitor cells into intermediate progenitor cells. This switch from a neural stem cell to an NPC is accompanied by the fundamental shift from the npBAF to nBAF complex, which involves the substitution of three subunits (BAF45, BAF53, and BAF55). Failure to switch leads to increased cell death, a small progenitor pool, and failure to further differentiate (Lessard et al., 2007; Wu et al., 2007; Bachmann et al., 2016). Interestingly, while SMARCA5 loss hampers NPC proliferation, loss of its ISWI ortholog SMARCA1 fails to repress expression of proliferation genes resulting in delayed neuronal differentiation and a larger brain (Yip et al., 2016). Taken together, these examples highlight the importance of chromatin remodeling proteins to NPC homeostasis and provide insight into the multitude of mechanisms at work often resulting in a similar phenotype.

Chromatin remodeling proteins were first identified as transcriptional coactivators and they continue to be implicated in the regulation of many genes. Since its identification, ATRX has also been presumed to regulate gene transcription. While there is a good level of understanding regarding how ATRX maintains genomic stability through the regulation of tandem repeats, telomeres and pericentromeric heterochromatin, the identification of direct transcriptional targets has proven more challenging. Initial ChIPseq experiments suggested that ATRX was bound at few promoters, gene bodies and regulatory elements (Law et al., 2010). Further work has suggested that ATRX binding may differ between tissues to ensure proper silencing of repetitive elements located near or within expressed genes in that particular tissue (Nguyen et al., 2017). Consistent with this idea, ATRX ChIPseq analysis of NPCs demonstrated a higher enrichment of binding sites at gene regulatory elements compared to what was observed in mouse ESCs suggesting that more genes may be under direct ATRX regulation within the brain (Law et al., 2010; He et al., 2015; Danussi et al., 2018). Another contributing factor to differential target gene expression is represented by ATRX effects on α-globin gene expression. Mutational analysis identified >15 ATR-X patients with the identical missense change (p.Arg246Cys), yet they showed a variable degree of hemoglobin H inclusions in blood samples, indicative of differing levels of α-globin expression (Gibbons et al., 1997). Repression of α-globin expression was dependent on the size of a GC-rich VNTR located within the globin gene cluster (Law et al., 2010). A second factor driving the tissue specificity and the variable effects was the formation of R-loops caused by the transcription of the GC-rich VNTR sequences (Nguyen et al., 2017). The larger sequences generate increased R-loops and G4 DNA structures that normally recruit ATRX to re-establish the normal chromatin structure. In the absence of ATRX the R-loop/G4 DNA is not resolved effectively which then impedes both replication and transcription processes (Nguyen et al., 2017). The slight stochastic nature of these effects likely also dampens readouts of differential expression from RNAseq experiments thereby raising the need for a scRNAseq approach in future studies.

Gene expression analysis of control and Atrx^Foxg1Cre cortical samples at two timepoints (E13.5, P0.5) identified 202 and 304 differentially expressed genes (DEGs; ±1.5-fold change), respectively, with almost two-thirds of genes upregulated (Levy et al., 2008). Among these, 27 were common to both datasets including the downregulation of several ancestral pseudoautosomal region (aPAR) genes (Csfr2a, Dhrsxy, Cd99, and Asmtl) (Levy et al., 2008, 2014). In mouse, the aPAR genes are located in subtelomeric regions and contain potential G4 DNA sequences. Each gene analyzed had enriched histone H3.3 and ATRX binding within their gene body and showed reduced H3.3 levels when ATRX was absent (Levy et al., 2014). Interestingly, these intragene G4 DNA sequences also showed increased binding of RNA pol II in Atrx^Foxg1Cre samples suggesting that transcription becomes impeded at these regions within the gene leading to reduced expression. The authors extended this finding to Nlgn4, a gene encoding a post-synaptic cell adhesion molecule implicated in ASDs (Jamain et al., 2003; Laumonnier et al., 2004). This result conflicted with the study on R-loop formation which found no differences in RNA pol II loading or histone modifications across genes containing the GC-rich repeats (Nguyen et al., 2017).

Other downregulated genes from this analysis include Gbx2, NeuroD4, Wif1, Nxph1, Nxph2, and Mbp, each of which could contribute to cognitive deficits observed in patients but require further analysis to asses their contribution to the phenotype (Levy et al., 2008, 2014). In a similar experiment in the retina, 173 DEGs were identified with two-thirds upregulated (109 genes) and one-third (64 genes) downregulated (Lagali et al., 2016). Most of these genes were involved in the regulation of glutamate activity, ion channel regulation or encoded neuroprotective peptides, with four shown to be also dysregulated in the cortex (Csf2ra, Cbln4, Syt13, and Nlgn4). Each of these studies showed that the mutant samples had only small numbers of genes with large changes in gene expression and, while some downregulated genes may impede transcriptional elongation, this mechanism may not be universal, particularly as it pertains to upregulated genes.

However, other indirect mechanisms have been explored to explain transcriptional dysregulation, particularly a loss of repression. The ATRX/DAXX complex is critical for loading H3.3 at telomeres and pericentromeric heterochromatin (Goldberg et al., 2010; Wong et al., 2010). Research over the last few years has expanded this regulation to include H3.3 deposition at endogenous retroviral elements, regions associated with imprinted genes, and some CpG islands (Elsässer et al., 2015; He et al., 2015; Sadic et al., 2015; Voon et al., 2015). At the telomere, the loss of ATRX affected the transcription of telomeric DNA and the non-coding RNA TERRA, although studies conflict on whether levels increase or decrease (Eid et al., 2015; Nguyen et al., 2017). Surprisingly, TERRA was also shown to bind to an additional ∼4,000 binding sites aside from the telomere where it co-localized with ATRX (Chu et al., 2017). Many of these sites were within introns and comprised GA repeats, however, depletion of TERRA usually resulted in downregulation while ATRX depletion increased expression (Chu et al., 2017). While it remains to be determined how this might affect neuronal gene expression, ATRX has been shown to interact with other lncRNAs including Xist to facilitate PRC2 silencing and ChR01 that is required for heterochromatin reorganization in differentiating muscle cells (Sarma et al., 2014; Park et al., 2018). Indeed, ATRX binding to lncRNA or R-loops may be a key mechanism mediating transcriptional repression of specific target genes.

Histone H3.3 ChIPseq studies have also shown that it is enriched at the intracisternal A-particle endogenous retroviral elements (IAP/ERVs), which account for almost half of all mutation causing ERV insertions (Maksakova et al., 2006; Elsässer et al., 2015). Moreover, H3.3 deposition at these sites requires ATRX/DAXX to facilitate H3K9me3 and repression while depletion of ATRX, DAXX, or H3.3 results in reduction of the H3K9me3 mark and IAP/ERV derepression (Elsässer et al., 2015; He et al., 2015; Voon et al., 2015). In mouse ESCs, ERV derepression affected the expression of neighboring genes in a minority of cases with most genes neutral to ERV derepression. It raises the question of whether or not this type of derepression would affect many genes or occur rapidly within a post-mitotic neuron, and thus, function as a major effector in dysregulated gene expression in Atrx-null neurons. In this regard, a related study using cultured post-mitotic neurons demonstrated that the ADD domain can also bind the H3K9me3S10ph dual histone mark (Noh et al., 2015). This histone mark is rapidly induced by neuronal depolarization where it appears at centromeric and pericentromeric heterochromatin co-localized with ATRX to repress transcription of non-coding centromeric minor satellite sequences (Noh et al., 2015). While it is unclear what the impact of increased centromeric minor satellite transcription would have on disease pathology, it remains to be determined whether this dual mark affects activity-dependent transcription of genes mediating learning or memory.

It was also demonstrated that ATRX was bound to 56 CpG islands which was unexpected since they are often associated with active chromatin, typically promoters (Voon et al., 2015). However, these CpG islands were associated with H3K9me3, almost half were methylated and many corresponded to imprinted loci often residing in intragenic regions within a transcriptional unit (Voon et al., 2015). Indeed, in all cases examined ATRX was bound to the silenced imprinted allele which became reactivated in ATRX KO cells (Voon et al., 2015). This study contrasted somewhat with an independent report in which ATRX was recruited by MeCP2 to silence the active allele of several imprinted genes in the developing telencephalon (Levy et al., 2014). The difference in these studies may reflect differential regulation of imprinting loci in ESCs versus differentiating NPCs. Perhaps the most compelling example of derepression came from a study with the Atrx^ΔE2 mice (Shioda et al., 2018). In this model, the authors identify a small list of 31 DEGs in the adult hippocampus but with most genes (23/31) downregulated (Shioda et al., 2018). Among the upregulated genes was an imprinted gene from the lymphocyte-regulated gene family, Xlr3b. Although Xlr3b had widespread expression across many tissues, it was only upregulated in the brain. Further work showed that ATRX bound to a G4 DNA sequence within the CpG island of the Xlr3b gene where it normally interacted with DAXX and H3.3 and recruited DNMT1 and DNMT3 to silence the gene. The subsequent overexpression of Xlr3b in the Atrx^ΔE2 mice was shown to produce a protein that localized to dendritic RNA granules where it interacted with ribonucleoproteins, dynein proteins and the RNA-binding protein, TIA1, to regulate mRNA transport (Shioda et al., 2018). One of the targets identified was the mRNA for CAMK II-α which they had previously shown to be deregulated in these animals. Excitingly, they also showed that the G-quadruplex-binding ligand 5-aminolevulinic acid (5-ALA) was able to decrease RNApol II occupancy and Xlr3b expression in the Atrx^ΔE2 mice, although methylation of the G4 DNA sequence within the CpG island was not affected. It seems that formation or stabilization of this G4 DNA sequence is required to activate the Xlr3b gene and that ATRX normally prevents this by facilitating heterochromatin formation. In this regard, mapping of G4 DNA sequences have shown an enriched number in gene regulatory elements where many function to increase transcription when stabilized (Hänsel-Hertsch et al., 2016). While G4 DNA stabilization occurs in the Atrx^ΔE2 mice, further work is required to explain how 5-ALA represses Xlr3b transcription when it should stabilize the G4 DNA. Regardless, the derepression of G4 DNA within CpG islands and/or other regulatory elements is an exciting mechanism that can explain DEG upregulation, particularly when coupled with the finding that ATRX binding is increased at regulatory elements in NPCs compared to ESCs. Collectively, the derepression of tandem repeats, retrotransposable elements and G4 quadruplexes can all function to impinge on neuronal function.

We have discussed global effects on DNA replication and transcription that occur in the absence of ATRX in the previous two sections. In this section, we review the morphological and functional repercussions of these deficits. Aside from being reduced in size, the Atrx^Foxg1Cre mice had a normal cortical morphology with proper lamination although a reduction of upper layer neurons (Bérubé et al., 2005; Ritchie et al., 2014; Huh et al., 2016). The reduction in upper layer neurons may also contribute to the partial agenesis of the corpus callosum observed in some patients (Gibbons, 2006). The hippocampus was also reduced in size while the dentate gyrus consisted of only a few disorganized cells. Behavior analysis was not performed due to the early postnatal lethality, although female heterozygous mice showed impairment in spatial, contextual fear, and novel object recognition memory (Tamming et al., 2017).

The Atrx^ΔE2 mice also had smaller brains but no differences in cell density within layers II/III of the prefrontal cortex (PFC) or hippocampus (Shioda et al., 2011). Examination of dendritic spines in the PFC showed that the Atrx^ΔE2 mice had similar numbers but fewer mature spines and many more, thin, long immature spines (Shioda et al., 2011). Behavioral analysis indicated that the mice have impaired contextual fear memory (fear conditioning test), spatial memory (Y-maze), but not anxiety behaviors (Nogami et al., 2011; Shioda et al., 2011). Electrophysiology studies in hippocampal slices demonstrated reduced NMDAR-dependent long term potentiation (LTP) evoked by high stimulation frequency in hippocampal CA1 neurons which was mediated by increased CAMK2A and GluR1 phosphorylation (Nogami et al., 2011). This was in contrast to a later article by the same group that showed phosphorylated CAMK2A levels were reduced in the Atrx^ΔE2 mice while 5-ALA restored the levels at the synapse and the phosphorylation levels (Shioda et al., 2018). A recent article examining hippocampal function using CAMKII-Cre mice to inactivate Atrx in postnatal excitatory forebrain neurons demonstrated reduced paired-pulse facilitation and LTP in proximal and distal apical dendrites of CA1 synapses (Gugustea et al., 2019). This represented the first study of mice in which Atrx has been inactivated after neurogenesis and it will be interesting to ascertain the full characterization of these mice.

Studies of the retina have also provided useful information into ATRX function. Many ATR-X patients have visual problems although this has been an under-appreciated aspect of the phenotype (Medina et al., 2009). Inactivation of ATRX in retinal progenitors Atrx^Pax6Cre resulted in a slight reduction in retina size and a specific reduction of interneurons, namely amacrine and horizontal cells (Medina et al., 2009). Surprisingly, Atrx^Pitf1aCre mice that ablates ATRX in a bi-potential progenitor that generates amacrine or horizontal cells did not recapitulate the phenotype, while inactivation with a bipolar cell specific Cre driver (Atrx^Vsx2Cre) did not affect bipolar cell survival but did result in reduced amacrine and horizontal cells suggesting that interneuron survival was a cell non-autonomous effect (Lagali et al., 2016). Additional characterization of these mice showed that the bipolar axons were mislocalized within the inner plexiform layer and many had axonal swellings or tortuous paths to their targets. Gene expression analysis identified alterations in the glutamate pathway, ion channel regulation and altered expression of neuroprotective peptides. Altered axonal pathfinding was also observed in Drosophila XNP mutants, the homolog to the ATPase domain of ATRX (Sun et al., 2006). It will be important to further explore in greater detail whether axonal pathfinding is also altered within forebrain or hippocampal neurons.