香京julia种子在线播放

For the literature selection process, we utilized the databank PubMed and applied combinations of the following search terms: NuRD, chromatin remodel* (e.g., remodeler, remodeling), embryonic* or neural* (e.g., development, stem cell), and neurodevelopmental* or neuropsychiatric* (e.g., disorder, symptom, disease). Later search results were narrowed to intellectual disability (ID), autism spectrum disorders (ASDs), schizophrenia (SCZ), and bipolar disorder (BD). The search process covered the period 1980 to June 2019. Only studies in English that investigated NuRD or its core subunits were included. In addition, we followed up references from the identified publications, of similar articles indicated by PubMed, and of citatory publications by referring to Google Scholar^®.

The nuclear genome of eukaryotic cells is organized into chromatin in which DNA, RNA, and associated proteins are packaged together. Chromatin provides a large source of information that extends from the linear chromatin template to the basic building block, the nucleosome, and from there to three-dimensional (3D) structures. The nucleosome consists of ∼147 base pairs of double-stranded DNA coiled around core histones with chromatin-free DNA segments linking single nucleosomal units (Sitbon et al., 2017). Higher-level chromatin structures arise from further winding of the chromatin template and impose further compaction on the DNA. These 3D structures also contribute to the formation of interacting chromatin loops and of topologically associating domains that impact long-range gene regulation and the expression of functionally related gene groups. Core histones consist of highly conserved alkaline proteins, which serve as substrates for various PTMs. Well-studied PTMs include histone methylation, acetylation, phosphorylation, ubiquitination, and sumoylation. Together, these modifications produce modular signatures influencing chromatin organization (Sitbon et al., 2017). Chromatin remodelers, histone- and DNA-modifying enzymes, RNA, and a vast array of multifarious transcription factors (TFs), jointly read and shape a versatile chromatin landscape. At each structural level, chromatin modulation allows plastic responses to gene regulatory signals (Jaenisch and Bird, 2003) and places chromatin at the center of stable, yet adaptable, gene expression.

A number of mammalian chromatin remodeling complexes consume energy gained from ATP hydrolysis to shift nucleosomes relative to the DNA sequence. This process facilitates chromatin remodeling and TF access to DNA-binding sites (Hota and Bruneau, 2016). ATP-dependent chromatin remodeling complexes consist primarily of a single ATPase that contains a high-affinity substrate binding and a catalytic site. The catalytic activity of the ATPase is regulated through multiple associated subunits that direct additionally complex binding throughout the genome (Hota and Bruneau, 2016).

NuRD (alias NRD or Mi-2) is one such macromolecular protein complex that is unique in combining chromatin remodeling and protein deacetylase activity ( Figure 1 ). The remodeling subcomplex consists of one ATPase (chromodomain helicase DNA-binding protein 3/4/5; CHD3/4/5) that associates with one GATAD2A/B (GATA zinc finger domain containing protein 2A/B, alias p66α/β) protein, and the DOC1/CDK2AP1 protein (deleted in oral cancer/CDK2 associated protein 1). The associated deacetylase subcomplex consists of HDAC1/2 (class I lysine deacetylase1/2) proteins, two metastasis tumor antigen proteins (MTA1, MTA2, and/or MTA3), and the histone chaperones RBBP4/7 (retinoblastoma binding protein 4/7). In addition, the zinc finger proteins SALL1/4 (i.e., SAL-like 1/4) have been found to associate with the deacetylase subcomplex in a cell type and tissue-specific manner (Allen et al., 2013; Torchy et al., 2015). Lastly, one of the methyl-CpG-binding domain proteins 2/3 (MBD2/3) bridges the two subcomplexes in entire NuRD.

CHD, GATAD2, MBD, and MTA proteins are defining subunits of NuRD, whose differential assembly modulates its function in a cell-type-specific manner. For example, NuRD containing CHD3 or CHD4 has distinct although overlapping functions (Hoffmeister et al., 2017) and fulfills distinct roles in corticogenesis [see below (Nitarska et al., 2016)]. Likewise, the bridging proteins Mbd2 and Mbd3 are mutually exclusive in NuRD with Mbd3 required for early postimplantation development in mice, while Mbd2 is not [see below (Günther et al., 2013; Wood and Zhou, 2016)]. Mbd2 contains a bona fide methyl-CpG-binding domain thought to recruit NuRD to methylated DNA. By contrast, Mbd3’s methyl-CpG-binding like-domain is incapable of high-affinity methylated DNA binding (Zhang et al., 1999) and is dispensable in normal differentiation and development. Accordingly, Mbd3 operates solely to bridge remodeling and deacetylase subcomplexes within entire NuRD (Zhang et al., 2016). Although Mbd2/3 is necessary for entire NuRD, functional and genetic data raise the possibility that the CHD4 subcomplex can also function on its own during first lineage decisions in embryonic development (O’Shaughnessy-Kirwan et al., 2015). It remains to be clarified to what degree this behavior can be generalized to other developmental conditions and whether it applies to the deacetylase subcomplex as well.

NuRD is as an abundant chromatin-associated complex that binds on a genome-wide scale to nearly all active enhancers and promoters in embryonic stem cells (ESCs) (Günther et al., 2013; Miller et al., 2016; Stevens et al., 2017). Current binding data support a global localization model and suggest that NuRD has a general affinity for open chromatin regions associated with transcriptional activity. Furthermore, a broad variety of TFs bound at specific DNA-binding sites interacts with NuRD to modify gene expression (e.g., Aguilera et al., 2011; Liang et al., 2017). Owing to NuRD’s genome-wide presence, such TFs are thought to increase NuRD’s local availability, rather than to recruit it ab initio. Intriguing as it is, this hypothesis still needs further experimental validation, e.g., by determining local NuRD availability dependent on transcriptional activation status and/or by kinetic analysis of NuRD/TF interactions.

Taken together, NuRD regulates gene expression by combining chromatin remodeling and protein deacetylase activity. Similar to other chromatin-modifying complexes (Ho and Crabtree, 2010), changes in NuRD’s subunit composition correlates, at least in part, with distinct changes in function to match the needs of specific cell types and developmental stages (see below).

ESCs offer a unique resource for the investigation of mitotic self-renewal and differentiation into virtually any cell type in the presence of appropriate signals. ESCs can be kept as fairly homogeneous population in unlimited amounts as opposed to heterogeneous and/or inaccessible tissues from living organism. Thus, ESCs provide a tractable model to gain mechanistic insight into NuRD’s basic functions that can also inform rodent and human studies referred to below.

To assess the role of the closely related Mbd proteins, Hendrich et al. (2001) generated a set of knockout mice: Embryos without Mbd3, but not without Mbd2, developed fatal postimplantation patterning defects, suggesting a need of Mbd3 for functional NuRD at this early stage. As a more tractable model, Kaji et al. (2006) subsequently established an Mbd3-null ESC line: although these cells showed under growth conditions high expression of pluripotency genes, they grew more slowly, yet without signs of spontaneous differentiation, relative to their wild-type counterpart. Importantly, NuRD complexes were no longer formed in Mbd3-deficient ESCs supporting Mbd3’s essential role as bridging factor. In contrast to wild-type ESCs, which efficiently formed all three germ layers under differentiation conditions, the differentiation potential of Mbd3-null ESCs was constrained, although not completely eliminated. At the same time, expression of Oct4 and Nanog, two genes with a pivotal role in pluripotency, was maintained in mutated ESCs and retained them in a state of self-renewal.

Collectively, this study strengthens the evidence that Mbd3 is required for NuRD formation in ESCs and assigns to NuRD/Mbd3 a role in self-renewal and lineage commitment.

Early studies on NuRD binding in ESCs (Hu and Wade, 2012) indicated a subtle balance between transcriptional activation and inhibition at NuRD-bound genes sharing an essential role in self-renewal and development. With the advent of genome-wide ChIP-seq (chromatin immunoprecipitation sequencing), Baubec et al. (2013) established genome-wide binding profiles for the family of methyl-CpG-binding proteins in ESCs and ESC-derived neuronal cells. In vivo binding of MBD2 required a functional MBD domain (absent in MBD3), and the existence of methyl-CpGs was largely proportional to the local methylation density, and thus mapped to inactive regulatory regions. By contrast, when MBD2 was present in a complex with NuRD, it mapped to a subgroup of methylation-free active regulatory sites. Notably, MBD3 likewise occupied these regions independent of canonical DNA methylation. These genomic regions contained a low number of unmethylated CpG residues, but were enriched in methylation and acetylation marks characteristic of active chromatin (histone 3 lysine 4 methylation and histone 3 lysine 27 acetylation). Moreover, these sites were also DNAse I hypersensitive, another indicator of an open chromatin configuration. Interestingly, a high percentage of these MBD3 sites and the subgroup of methylation-free MBD2 sites contained tissue-specific regulatory regions, active promoters, and enhancers.

How can these findings explain NuRD’s role in balancing ESC self-renewal versus differentiation? Part of the answer came from experiments conducted by Reynolds et al. (2012): Under self-renewal conditions, NuRD bound to a subset of pluripotency genes (Klf4, Klf5, and Tbx3) and confined their expression. Transcriptional heterogeneity describes the expression of one gene or a group of genes to varying degrees in a homogeneous cell population. Such variability in gene expression is thought to provide means by which stem cells can sort their progeny either to different lineages or to self-renewal. To test this hypothesis, Reynolds et al. quantified via immunofluorescence microscopy the abundance of short-lived proteins (such as Klf4 and Klf5) as a proxy to transcriptional output in single cells. In addition, the coding region of one Zfp42 allele was replaced by a destabilized (i.e., short-lived) GFP protein to measure output from the Zfp42 gene by flow cytometry. In support of the hypothesis, wild-type ESCs contained a mixture of cells with low and high expression of NuRD-regulated pluripotency genes, whereas NuRD-deficient ESCs expressed uniformly high levels of pluripotency genes and were impaired in differentiation. This result points to a role of NuRD in fine tuning gene expression rather than in categorical “on–off” switches. To strengthen this conclusion and as a more direct proxy to transcriptional regulation, single-cell sequencing rather than protein-based approaches remains desirable.

To examine this issue further, Bornelöv et al. (2018) established in ESCs an inducible NuRD system with tight temporal resolution. Induction of NuRD activity led to rapid genome-wide reorganization of nucleosome structure at enhancers and promoters and eviction of some chromatin-bound proteins and of RNA polymerase II from these sites. Shortly afterwards, the same or similar protein complexes rebound most genes that showed accordingly only transient changes in nascent mRNA production. By contrast, a subset of NuRD-bound genes showed sustained increases or decreases in nascent mRNA production that required NuRD’s nucleosome remodeling activity. This subset included various developmental genes that triggered lineage commitment during transition phases. Irrespective of this evidence, Mbd3-deficient ESCs still seemed to retain the appropriate differentiation trajectory, although they fail to reach a differentiated state (Kaji et al., 2006). A plausible explanation for this discrepancy is that Mbd3-null ESCs still express NuRD/Mbd2 that may replace NuRD/Mbd3 in gene regulation during early, but not during late stages of differentiation. Although Mbd2 and Mbd3 mapped to largely identical sites in undifferentiated ESCs when present in a complex with NuRD, even slight differences may become important to and/or increase during late differentiation.

Similar to MBDs, the presence of three MTA proteins for NuRD raises the question to which degree they differ in function. Although weak differences in terms of chromatin binding and protein interactions were detectable (Burgold et al., 2018), MTA proteins successfully replaced each other in NuRD activity in ESCs. These finding is in contrast to mature cells, where MTA proteins fulfill, at least in part, distinct functions. ESCs harboring knockouts of all three MTAs kept viable, but showed impairments in both lineage commitment and differentiation trajectories. This result indicates that NuRD ties together consecutive steps in early development.

Taken together, NuRD regulates genome-wide, gradual changes in gene expression in ESCs. NuRD-related transcriptional heterogeneity facilitates transition states, in which NuRD modulates transcription of developmental genes catalyzing lineage commitment and differentiation trajectory.

For clearness, we group studies on NuRD’s role in neural development, neuronal differentiation, and maturation by brain region (e.g., cerebellum) and cell type [e.g., neural progenitor cell (NPC)] rather than by the order of their appearance and summarize key points in a tabular format ( Table 1 ).

Despite early evidence for NuRD’s role in development (see above), a role of NuRD in neuronal maturation has remained largely unexplored until recently. The formation of neuronal circuits depends critically on the differentiation of synapses in brain development and beyond (Kandel et al., 2013) and is shaped by both cell-intrinsic and environmental signals. In 2014, Yamada et al. (Yamada et al., 2014) first reported that NuRD regulates the differentiation of presynaptic sites in rodent cerebellum. In vivo RNAi and conditional Chd4 knockout mice experiments showed that NuRD depletion in cerebellar cortex strongly impaired the development of granule neuron parallel fibers and of Purkinje cell synapses in vivo. Intersection of genome-wide RNA-seq and ChIP-seq data revealed a network of <200 repressed genes and decommissioned promoters at which NuRD turned off histone modifications associated with transcriptional activation during cerebellar development. A targeted in vivo RNAi screen of this network identified a subset of genes that encoded negative regulators of presynaptic differentiation: Nhlh1 (nescient helix loop helix) is a TF repressing the ubiquitously expressed bHLH factor TCF3 (alias E12/E47; E2A), and Elavl2 is an RNA-binding protein with a possible role in mRNA splicing and stability. Given the broad impact of these factors, they may serve as hub for NuRD’s effect on presynaptic development ( Figure 2 ). In further support of and consistent with a role in presynaptic connectivity, NuRD operated throughout sensitive time windows of postnatal neuronal plasticity.

In a subsequent study, Yang et al. (2016) sought to explore through a multifaceted approach whether NuRD-dependent synapse development relates to distinct cerebellar functions. Genome-wide ChIP-seq showed that NuRD/Chd4-bound promoters belonged by and large to actively transcribed signaling genes in mice cerebellum [note that only a small network of <200 NuRD target genes previously associated with repressive histone modifications (Yamada et al., 2014)]. This result prompted Yang et al. to reason whether NuRD utilizes mechanisms distinct from posttranslational histone modifications to regulate gene expression. One such mechanism is the exchange of histone variants that are known to modulate transcription (Buschbeck and Hake, 2017). In support of this hypothesis, 97% of Chd4-bound active promoters were enriched in the variant H2A.z in wild-type, but not in conditional Chd4 knockout mice. Combining RNA-seq with H2A.z ChIP-seq analyses revealed upregulation of >90% of the genes with reduced H2A.z occupancy, but little change in otherwise PTMs, in Chd4-null mice.

Together, these results suggest that NuRD mediates the replacement of the core histone H2A by the variant H2A.z in vivo at the promoters of a large group of neuronal-activity-dependent signaling genes in the cerebellar cortex and that deposition of H2A.z promotes gene deactivation ( Figure 3 ).

Activity-dependent transcription is well known to modulate neuronal connectivity (Kandel et al., 2013). What is less known are mechanisms and functional implications of resetting activity-dependent transcription. Remarkably, neuronal activity enhanced deposition of H2A.z at NuRD-bound genes during the inactivation phase of transcription. Conversely, absence of NuRD slowed deactivation, although it retained the response to renewed neuronal activation. Thus, NuRD regulates dynamically neuronal activity-dependent transcription. By means of a rotating rod motor learning task, Yang et al. (2016) further showed that behavioral activity induced neuronal activity-dependent transcription in vivo, whose deactivation was slowed in NuRD-deficient mice. During the synchronous development of granule neurons in vivo, NuRD/H2A.z disabled activity-dependent gene expression and promoted dendritic patterning and connectivity. Conversely, the total length and number of primary dendrites declined in the absence of NuRD during the pruning period. This finding indicates that NuRD-mediated deactivation of neuronal-activity-dependent genes impacts pruning activity during sensitive time windows of granule neuron development ( Figure 3 ). Such pruning processes are thought to fine-tune cerebellar circuit function and to enable sparse encoding of information. In support of this hypothesis, Chd4-null mice exposed to a sensorimotor stimulus (i.e., a treadmill task) showed hyperresponsive granule neuron activity in vivo. Such hyperresponsivity associated with impaired procedural learning capacity as evidenced in behavioral tests (i.e., accelerating rotarod and delayed eye-blink conditioning tests). By contrast, motor coordination was barely affected in Chd4-null mice.

Overall, these studies show NuRD-dependent chromatin remodeling in granule neuron synapse formation and connectivity during sensitive time windows. NuRD remodeling involves repression of a subgroup of inhibitory genes in synaptogenesis ( Figure 2 ) and deactivation of a vast array of neuronal-activity-dependent signaling genes ( Figure 3 ). While former process is more akin to NuRD’s silencer role, the latter process highlights NuRD’s role in constraining neuronal-activity-dependent gene expression changes that regulate granule neuron pruning activity and behavioral responses. In this respect, it would be interesting to know whether the deposition of the histone variant H2A.z at neuronal-activity-dependent genes is reversible and could respond to successive environmental exposures. While Yang et al. (2016) did not formally address this topic, it has been shown that the expression level and genomic deposition of histone variants are dynamically controlled (Buschbeck and Hake, 2017). Moreover, H2A.Z has been hypothesized to provide a means to dynamically increase chromatin accessibility and facilitate transitions between chromatin states. Along this line, H2A.Z has been assigned a role in nearly all functions of the chromatin template ranging from DNA repair and chromosome segregation to gene transcription and heterochromatin formation (Buschbeck and Hake, 2017). Taking these considerations into account, H2A.z deposition at neuronal-activity-dependent genes appears to be most likely reversible. Beyond PTMs, deposition and eviction of histone variants thus could add another layer of chromatin plasticity that impacts neuronal plasticity (Herre and Korb, 2019).

Formation of the highly organized cerebral cortex requires contributions from different classes of NPCs in a tightly spatiotemporally controlled sequence (Florio and Huttner, 2014). Key features of cortical progenitor cells and their role in cortical layer formation are schematically outlined in Figures 4A, B . Briefly, major groups of NPCs include apical progenitors (APs) and basal progenitors (BPs) that are classified according to cell polarity, presence of ventricular contact, location of mitosis, and the expression of astroglial markers. These NPCs undergo to different degrees self-renewing and differentiative cell divisions in the course of corticogenesis [interested readers are referred to Florio and Huttner (2014) for a comprehensive presentation].

Knock et al. (2015) first investigated a role of NuRD in the developing cerebral cortex in conditional Mbd3 knockout mice. Because deletion of both Mbd3 alleles is embryonic lethal, only heterozygous mice with unaffected viability were used. Macroscopically, heterozygous mice showed a reduction in cortical thickness that associated at the cellular level with an impaired specification of cortical projection neuron (PN) progenitors relative to wild-type mice: Pax6-positive APs were maintained in NuRD/Mbd3-mutant cortex but failed to respond to signals regulating symmetric versus asymmetric divisions and exited prematurely the cell cycle. Thereby, they produced insufficient amounts of Tbr2-positive BPs and neurons, and cortical plate neurons with deficits in terminal differentiation.

The mouse cortex develops between embryonic days 11 and 18 in a characteristic inside–out sequence: deep layers (designated 4–6) are formed first and upper, more superficial, layers (designated 2 and 3) are formed later. As a result, prospective upper layer neurons need to migrate through the deeper layer to attain their final destination (Florio and Huttner, 2014). Mbd3 was expressed in a subpopulation of cortical plate neurons that corresponded mainly to upper-layer neurons. In the absence of Mbd3, specification of upper layers (i.e., Satb2- and Brn2-positive neurons, see below) was compromised, and both deep and upper cortical layer markers became coexpressed. Expression profiling of microdissected tissues further revealed that NuRD/Mbd3 was necessary to guide neurodevelopmental differentiation programs by terminating the expression of proneural genes.

Briefly, this study indicates that NuRD/Mbd3 coordinates cerebral cell lineage choice and differentiation programs by terminating proneurogenic transcription in both progenitor and neural progeny.

Subsequent studies have sought to identify mechanisms regulating NuRD’s activity in this process. In this regard, the suppressor of Mek-null (Smek) was found to bind to Mbd3 (Moon et al., 2017) and to inhibit the recruitment of NuRD at neurogenesis-associated gene loci ( Figure 5A ). Smek proteins are evolutionary conserved across eukaryotes and regulate asymmetric cell division in invertebrate neuroblasts. Two orthologous proteins, Smek1 and Smek2, exist in mice, and especially Smek1 enhances neuronal differentiation of NSCs by inhibiting Par3, an essential regulator for asymmetric cell division and polarized growth. Consistent with this finding, double knockout Smek1/2 mice showed defects in cortical neurogenesis in vitro and in vivo ( Figure 5B ). To gain insight in the underlying mechanism, Moon et al. (2017) conducted a yeast two hybrid screen that led to the isolation of Mbd3 as interaction partner of Smek. Further genome-wide profiling revealed Smek binding at genes relevant to brain development, differentiation, and cell-fate determination. Both Smek and Mbd3 colocalized at proneural/neural genes, where Smek binding depended on the presence of Mbd3. Once bound, Smek triggered polyubiquitylation and degradation of Mbd3 and thus reduced formation of and repression by NuRD/Mbd3 ( Figure 5A ).

Taken together, both Smek and Mbd3 tilt the balance between self-renewing and neurogenic cell divisions in the same direction. Since both proteins colocalize in NSCs, in which Smek counteracts NuRD/Mbd3, this findings raises the question to what degree Smek- and NuRD/Mbd3-dependent corticogenesis differ from each other.

In contrast to Mbd3, a role of Mbd2 in corticogenesis is still uncertain. Unlike most other MBD proteins, Mbd2 may be dispensable for brain function (Wood and Zhou, 2016). Alternatively, Mbd2 may be required only in a small subpopulation of cells that do not manifest robust changes. In support of this hypothesis, proliferation and differentiation of olfactory receptor neurons were impaired in Mbd2-knockout mice (MacDonald et al., 2010), while olfaction-associated behavior was sustained. On the other hand, Lax et al. (2019) recently reported that Mbd2-knockout mice showed subtle deficits in cognitive, social, and emotional functions and downregulation of neuronal gene pathways in the adult hippocampus. Refined studies may help to define the contribution of NuRD/Mbd2 in neurodevelopment more precisely.

Apart from Mbd2/3, additional evidence suggests that distinct members of single NuRD subunits can serve as versatile regulatory mechanism during neurodevelopment. Of particular interest in this respect is NuRD’s ATPase activity, for which at least nine different genes have been identified so far. Most of these are expressed in neural cells through different stages of development (Micucci et al., 2015).

Unlike the cerebellum, Chd5, but not Chd4, was necessary for embryonic neocortical development in mice (Egan et al., 2013). While Chd5 was not expressed in rapidly proliferating progenitors, expression steadily increased in late-stage neuronal progenitors undergoing terminal differentiation. In utero knockdown of Chd5 resulted in a severe defect of progenitors to exit the germinal zones (ventricular, subventricular, and intermediate) and an accumulation of undifferentiated progenitors. In an ESC model of neurogenesis, Chd5-depleted cells failed to upregulate genes involved in late stage neural differentiation including synapse development, neuron projection, and neurotransmitter transport. At the same time, a subgroup of Polycomb target genes (Hoffmann et al., 2017b) underwent derepression in Chd5-depleted cells. Derepressed genes belonged to nonneuronal lineages including extraembryonic, mesodermal, and endodermal germ layers. Similar results were obtained in a cellular model of differentiated, Chd5-knockdown, neuroblastoma cells, which were analyzed by genome-wide ChIP-seq (Egan et al., 2013). Consistent with these findings, primary rat cortical neurons depleted of Chd5 showed alteration in the expression of neuron-specific genes and of other chromatin regulators such as the BAF complex (Potts et al., 2011).

Conclusively, terminal neural differentiation during embryonic corticogenesis requires Chd5’s two-sided function in gene regulation: to activate neuronal gene expression and to repress concurrently Polycomb-regulated genes controlling the expression of nonneuronal genes.

Nitarska et al. (2016) went on to define NuRD’s role in distinct stages of neocortical development by mass spectrometry of Hdac2 imunoprecipitates. The ATPases Chd3/4/5 were present as mutually exclusive subunits of NuRD and regulated distinct set of genes essential for brain development: Chd4 promoted the proliferation of BPs, while both Chd5 and Chd3 promoted stage-specifically cell migration: Chd5 facilitated early radial migration, whereas Chd3 guided late migration of cortical neurons and regulated aditionally laminar specification ( Figure 6 ).

NPCs depleted of Chd3 exited the cell cycle prematurely and led to a subsequent deficit in BPs. This event reduced specifically the formation of upper layer neurons (Satb2- and Cux-1-positive) and finally of cortical thickness, a finding resembling Mbd3-deficient mice (Knock et al., 2015).

Contrary to Chd4, Chd3/5 expression was very low in NPCs (Nitarska et al., 2016) and increased continuously during late stages of neurogenesis. At this stage, neurons migrate radially to form the cortical plate. Neurons that had populated the cortical plate expressed Chd3, while neurons residing in the SVZ expressed additionally Chd5. In utero knockdown of Chd5 resulted in an accumulation of neurons in the intermediate zone and a failure to reach the cortical plate. In contrast, knockdown of Chd3 caused a delay in late neuronal migration: cells lagged behind in the deeper cortical layers and only a reduced number of neurons immigrated into the upper layers. Additional microarray and ChIP experiments indicated that such a sequential switch of Chd3/4/5 underpins distinct NuRD activities and confers transcriptional competence for the selective regulation of genes involved in the proliferation of neural progenitors, early radial and late cortical migration, and the specification of cortical layers.

Overall, these studies support a regulatory role of NuRD through consecutive building blocks of corticogenesis that requires distinct ATPase subunits and assign to NuRD a more dynamic function than originally thought. Mechanistically, different classes of neurodevelopmental TFs may need to interact with specific Chd subunits to fulfill their roles and thus present the driving force behind the exchange of Chd subunits in the course of neurodevelopment.

High-level functions of the neocortex (e.g., cognition, articulation of language, sensory perception, and fine motor skills) are executed by excitatory PNs, presenting the largest portion of cortical neurons, and by inhibitory interneurons (Lodato and Arlotta, 2015).

Excitatory PNs are classified into numerous subtypes based on three major criteria: first, by the location within six cortical layers that are defined by histological measures; second, by their axon projections to different intracortical, subcortical, and subcerebral regions; and third, by the expression of genes signifying specific subtypes of neurons ( Figure 7A ). In the context of NuRD’s role in cortical development, we focus here particularly on latter criteria.

The transcriptional logic underpinning specification of major classes of PNs including callosal PNs (CPNs), subcerebral PNs (SCPNs), and corticothalamic PNs (CThPNs) is schematically summarized in Figure 7A [interested readers are referred to Lodato and Arlotta (2015) for a comprehensive presentation]. In essence, PN identity evolves progressively by a fine-tuned transcriptional balance between genetic programs guiding the development of alternative types of PNs. Mechanistically, this balance involves cross-repression and cross-activation of key developmental regulators and extends also to their regulatory feedback loops ( Figure 7B ). The aggregated effects from this transcriptional logic are thought to sort postmitotic PNS into corticothalamic, subcerebral, or callosal fates. When discussing the role of NuRD in PN specification, we will adhere to the molecular logic of this transcriptional circuitry rather than the temporal order of the referred publications.

Fezf2 is required for SCNP specification ( Figure 7A ), while Tbr1, and also SOX proteins (SRY box), including repressors Sox4 and Sox5, and activator Sox11, are regulators of Fezf2 expression (Kast and Levitt, 2019). Additionally, LHX2 (LIM homeobox TF 2) has been recently found to contribute to this regulatory pathway (Muralidharan et al., 2017). Superficial layer neurons retain Lhx2 expression from their birth date through maturity. By contrast, Lhx2 expression is rapidly downregulated in deep layer 5 and 6 neurons. Conversely, cortex-specific Lhx2 knockout mice showed a strong increase in layer 5 neurons expressing high levels of Fezf2 and Ctip2, indicating SCNP fate. Concurrently, Tbr1 expressing layer 6 neurons were reduced and led to cortex thinning in Lhx2 knockout mice. Further ChIP-seq experiments revealed that Lhx2 bound at distal regulatory elements present in Fezf2 and Sox11. Mass spectrometry of Lhx2 immunoprecipitates revealed the presence of NuRD subunits, including Rbbp4, Hdac2, and Lsd1, as Lhx2 binding partners. Together, these results suggest that Lhx2 regulates subtype specificity in deep layer 5 corticofugal PNs through enhanced recruitment of NuRD-repressor activity to the central regulator Fezf2 and its upstream transactivator Sox11 ( Figure 8 ).

Ctip2, a downstream target to Fezf2, encodes a transcriptional repressor, which bind sequence specifically to DNA or interacts with other promoter-bound members of the COUP-TF family. Topark-Ngarm et al. (2006) showed that CTIP2 complexes from human neuroblastoma cells contained GATAD2A/B, MTA1/2, RBBP4/7, or HDAC1/2. These NuRD subunits were recruited in a CTIP2-dependent manner to a plasmid harboring the promoter region of p57KIP2, encoding a cyclin-dependent kinase inhibitor, and conferred repression ( Figure 7C ). Consistent with this finding, knockdown of CTIP2 in neuroblastoma cells enhanced expression of a number of genes, including p57KIP2.

In mice, deletion of p57kip2 led to cortical hyperplasia during late embryogenesis and postnatal life (Mairet-Coello et al., 2012). Cell cycle re-entry of RGs and IPs was increased during early corticogenesis but decreased at middle stages. Consequently, deletion of p57kip2 enhanced primarily layer 5–6 neuron production. Taken together, these findings indicate a role of Ctip2/NuRD in negative feedback control of deep layer neuron production via p57kip2 repression.

In addition, NuRD targets also directly Ctip2: expression of the transcriptional coregulator Ski (Ski sarcoma viral oncogene homologue) was high in postmitotic cells of the developing cortical plate in superficial layers (Baranek et al., 2012). Thereby, the expression of Ski closely resembled the one of Satb2. While Ski knockout mice showed relative to controls comparable thickness and cell numbers in cortical layers, Ski-deficient callosal neurons lost their identity and ectopically expressed Ctip2, albeit not Fezf2, indicating that they acquired some but not all characteristic of wild-type SCPNs. Since Ski-deficient callosal neurons phenocopied in part Satb2-deficient mice (Britanova et al., 2008), both factors seemed to operate in a shared genetic pathway. In support of this hypothesis, Ski and Satb2 interacted in vitro and in situ in upper layer neurons. Furthermore, in vivo ChIP experiments showed that Ski was recruited by Satb2 to previously identified Satb2-binding sites (known as matrix attachment region) in the Ctip2 locus ( Figure 9A ). Consistent with previous findings (Britanova et al., 2008), Satb2 repressed Ctip2 by recruiting NuRD via interacting with Hdac1 and Mta2. In the absence of Ski, the interaction of Satb2 with Mta2, but not with Hdac1, was retained, indicating Ski’s role to bridge Satb2 to Hdac1 ( Figure 9B ).

Briefly, Ski and Satb2 both target Ctip2 in cortical CPNs and both proteins are required to maintain repression by NuRD in the Ctip2 locus. Absence of either factor abrogates transcriptional repression and triggers partial loss of upper layer identity. As a whole, the picture emerging from these studies is that of a well-calibrated transcriptional logic of mutual “check and balances,” modulated by NuRD, in early specification of major classes of neocortical PNs.

On the other hand, our understanding of how PNs attain their final features during postnatal stages is still less advanced. In this regard, Harb et al. (2016) suggested recently an additional role of NuRD in the specification of PN subtypes. Although Ctip2 and Satb2 direct early specification of subcerebral and callosal PNs, respectively, their coexpression increases progressively in the postnatal somatosensory cortex. This coexpression of Ctip2 and Satb2 established two subtypes of layer 5 neurons, whereby one type projected to the contralateral cortex and the other type to the brainstem. These neuronal subtypes differed in their morphological and electrophysiological features from those neurons expressing solely Ctip2.

Ski expression was retained in Ctip2/Satb2 coexpression PNs, suggesting a mechanism independent of Ski to control the postnatal increase in Ctip2/Satb2 double positive somatosensory neurons. In support of this hypothesis, Lmo4 (Lim domain only), a transcriptional adaptor known to interact with several NuRD subunits, showed increasing peri- and postnatal expression in Ctip2/Satb2 coexpressing neurons. Overexpression and knockdown experiments demonstrated that Lmo4 acted in the specification of Ctip2/Satb2-double positive neurons primarily by modulating Ctip2 expression in layer 5: Lmo4 sequestered Hdac1 before its interaction with Satb2/NuRD and thus interfered progressively with Satb-2-dependent repression of Ctip2 ( Figures 9C, D ).

In short, regulators such as Ctip2 and Satb2 with opposite function during embryonic corticogenesis can colocalize postnatally and contribute to the generation of diverse PN subtypes. Thereby, transcriptional adaptor Lmo4 targets Satb2/NuRD complexes in the Ctip2 locus and promotes Ctip2 expression by interfering with NuRD-mediated deacetylation.

Overall, NuRD is a critical modulator of the molecular circuitry underlying specification of neocortical PNs in embryonic corticogenesis. Key regulators, such as Lhx2, Ctip2, and Satb2, increase the availability of NuRD at their target sites in a time- and area-specific manner to provide negative feedback control. Additionally, NuRD is a critical regulator of postnatal PN subtype specification through variation of molecular signatures that are shared with early corticogenesis. Similar connectivity or molecular code among neocortical areas has been hypothesized to originate, at least in part, from variations on a “common theme” rather than from the activity of many independent and region-specific genetic programs (Harris and Shepherd, 2015). Along this line, specific Chd subunits within NuRD have been previously shown to coordinate consecutive building blocks in neurogenesis. Similar variation in NuRD subunits, including and beyond Chds, may contribute as well to the specification of neocortical PNs. Future studies are needed to define NuRD subunit composition and the interaction with accessory subunits more precisely in this paradigm. In this respect, the aggregate effect from the interaction of tissue- or cell-type-specific TFs and NuRD subunits may present a critical determinant of molecular variation in postnatal specification of cortical neurons. This prompts also the question whether TFs and NuRD subunits may intersect in the regulation of their expression levels as to establish coherent transcriptional programs.

NDDs are complex conditions that result from anomalous brain development. Frequently, they present with impairments in multiple domains including cognition, language, social behavior and communication, and/or motor skills. ID, ASDs, attention deficit/hyperactivity disorder (ADHD), communication disorders, and SCZ fulfill the criteria of NDDs (American Psychiatric Association, 2013). Here, we consider genetic variation in NuRD subunits as risk factor for common polygenic forms of NDDs (with a focus on ID and ASD) and neurodevelopment-related psychiatric disorders (with a focus on SCZ and BD).