Different E-box binding transcription factors, similar neuro-developmental defects: ZEB2 (Mowat-Wilson syndrome) and TCF4 (Pitt-Hopkins syndrome)

Mowat-Wilson syndrome

Soon after its discovery as one of the first interacting TFs for Smad proteins (major intracellular effector proteins in transforming growth factor type β/bone morphogenetic protein (TGFβ/BMP) family signaling), hence its name Smad-interacting protein-1 (SIP1, renamed ZEB2)^[32], three teams of clinical geneticists independently identified de novo mutations in ZEB2 in patients with severe ID, typical facial dimorphism, and Hirschsprung disease (HSCR)^[33-35] as the cause of the well-defined, monogenic, and rare Mowat-Wilson syndrome (MOWS)^[36-38]. Mutant ZEB2 alleles have been determined for about 350 patients who display a great variety of clinical features^[39-44]. Typical in MOWS patients are delays in developmental milestones and motoric development, and anomalies in the eyes and teeth. Other features are typical craniofacial malformation, sensorineural deafness, and HSCR, which originate from defects in the ZEB2-positive (+) cells of the embryonic neural crest cell lineage. Excellent clinical reports have been published on various malformations in the central nervous system (CNS) of MOWS patients over a broad age range, including anomalies of the corpus callosum and/or hippocampus, which can be seen by neuroimaging. These reports have also started to document the longer-term evolution of electro-clinical defects such as focal seizures (by video-electroencephalograph monitoring, video-EEG)^[44-46].

The majority (~93%) of severe MOWS cases are caused by large deletions, nonsense mutations, or short deletions/insertions causing a frameshift leading to a premature stop codon. All of these mapped mutations thus far affect the protein-coding sequences and create haploinsufficiency. Rarer (~5%) are frameshift mutations and in-frame deletions affecting the C-terminal part of ZEB2 only. This raises the possibility that, depending on the location of such mutation, the C-terminally truncated protein is still produced and not subject to nonsense-mediated mRNA decay^[43,44,47]. Up to date, only few missense ZEB2 mutations (~1.5%) have been detected^[48,49]. These two smaller groups of ZEB2/MOWS-confirmed patients, in general, represent the milder and mildest cases, respectively.

The broad spectrum and relative numbers of severe, milder, and mildest cases make it difficult to firmly conclude on genotype-phenotype correlation. Normally, that conclusion would be based on larger panels of often subtly mutated and, therefore, an aberrant function of one or more of the (hitherto known) affected domains within the ZEB2 protein. For example, two mild MOWS cases have mutations in the NuRD interaction motif (NIM, see below), only causing a loss of interaction of this mutant ZEB2 with the NuRD co-repressor complex, and leaving the rest of ZEB2 intact. However, this has already detectable consequences at the level of ZEB2-dependent and direct target genes in the cellular systems where these mutant proteins were tested^[50,51].

To the list of mapped MOWS mutations that affect the protein-coding exons of ZEB2 (in patients now grouped as Class I), variation/mutation of non-protein-coding sequences of the ZEB2 locus have also begun to be added by the clinical geneticists and mapped exclusively to untranslated exons or (distal) REs (the patients being referred to as Class II). Co-operative enhancers active in NPCs were recently identified^[52] via mapping of chromatin conformational changes in the large genomic region encompassing the human ZEB2 and mouse Zeb2 locus (both on chr2q). These neural enhancers are located about 500 kilobases (kb) away from the transcriptional start site (TSS) of ZEB2/Zeb2, in a ~3.5 megabase (Mb)-long gene desert upstream of the gene^[52]. These add to recently identified, different Zeb2 enhancers in fetal and adult hematopoiesis in mice^[53]. Such findings may make clinical geneticists broaden the current characterization of sequenced MOWS alleles, in particular in mild but clinically convincing MOWS patients, wherein no mutation can be found by exon sequencing of ZEB2. In addition, such new non-protein-coding mutations will inspire establishing new cellular or animal models and documenting deviating ZEB2-dependent transcriptomes and ZEB2-containing GRNs in these.

Pitt-Hopkins syndrome

After the first report^[54] in 1978, it has taken quite some time to delineate the few similar syndromic cases and eventually claim the new syndrome Pitt-Hopkins syndrome (PTHS), based on three defects: (i) brain-related defects (including ID, often coming to attention in the first year of life); (ii) craniofacial malformation; and (iii) episodes of hyperbreathing that start in infancy, as well as breath holding and cyanosis. Several teams confirmed that de novo TCF4 haploinsufficiency causes PTHS; thus, not the phenotypically overlapping Angelman (OMIM #105830), Rett (OMIM #312750), alpha-thalassemia/mental retardation syndrome, and X-linked (ATR-X; OMIM #301040) syndromes^[55-57], as well as not MOWS, identified only around the year 2000^[33-35]. The situation is somewhat more complicated because of the existence of Pitt-Hopkins-like syndromes 1 and 2 (PTHLS1/2, OMIM #610042 and #614325) caused by autosomal recessive variants in CNTNAP2 and NRXN1, respectively, with patients who in general have milder delays in motor development.

The original identification of the PTHS locus benefited from large (> 1 Mb) deletions on chr18q in patients, enabling the use of SNP arrays for molecular karyotyping, and was further facilitated by sequencing of many coding exons (and intronic flanking regions) of TCF4, one of the candidate genes in this deletion region. Subsequent selection of extra patients with at least two out of the three defects (see (i)-(iii) above), and exclusion of the other similar syndromes, then confirmed the 20-exon (ex) (with ex1 and ex20 considered non-protein-coding at that time) and at least 360 kb-long TCF4 gene as the cause of PTHS. Increased numbers of cases thereafter allowed clinically delineating PTHS, aided by the use of two different clinical tests. These diverged due to the difference in weight given to some of the applied criteria (such as facial features, ID, and gait), and the interpretation of the test score also varied with age^[58,59]. PTHS often includes severe psychomotor delay, seizures and epilepsy, and stereotypical and repetitive movements. The same holds for microcephaly, but minor brain abnormalities are revealed by imaging (dysplasia of cerebellum and corpus callosum, small hippocampus, and hyperintensity of temporal lobe white matter). Additional defects are eye defects (early-onset myopia and strabismus), wider spacing of teeth, anomalies of fingers and toes, chronic constipation, and intestinal malrotation^[58,60]. Thus, many of these defects overlap with those in MOWS^{[8,9,44-46,58,59]}.

The wide phenotypic spectrum and possible clinical inclusion bias, in general, have prompted a number of teams to join forces and apply ID-targeted high-throughput sequencing (HTS) of hundreds of genes in ~900 patients with moderate to profound ID, but without the firm mapping of their syndrome^[61]. Doing so, as part of this study, TCF4 variants were identified in eight of these patients and could be added to two other ones from a previous ID-HTS study. These ten patients were then evaluated via both aforementioned PTHS clinical tests, and five of them scored convincingly positive. These same studies revealed that the TCF4 mutation rate is 0.7% in individuals with undiagnosed ID, but also that mutations that cause ID are poorly suggestive of PTHS (and therefore had sometimes been referred to as nonspecific ID). Additional large patient cohorts (~280 patients) assembled from 58 publications with both clinical and sequence data yielded another 25 case reports that were re-investigated^[62].

Altogether, this well-coordinated and vast screening made clinicians aware that in PTHS, more precise diagnostic criteria as well as detailed phenotyping were needed. A survey of hundreds of children with PTHS allowed an estimation of missense (~15% of cases), nonsense (15%), and splice-site (10%) mutations. In addition, small insertions/deletions causing a frameshift (30%), larger deletions, and even translocations involving a part of or the entire TCF4 (see Ref.^[9], and references therein) have been detected. Most missense TCF4 mutations, which affect all isoforms, map to ex19 (encoding the bHLH domain, see below), although some are located in ex15 (middle part of AD2) or ex18 (Rep or RD), and a few cases of C-terminally elongated TCF4 in mild PTHS have been described.

ZEB2 protein/gene structure

Mouse Zeb2 [1215 amino acids (aa); human ZEB2 (1214 aa; ~140 kDa) was after mouse Zeb1, the second identified member of the small family of vertebrate ZEBs. ZEB1 and ZEB2 are structurally highly similar (for a schematic representation of ZEB2, see Figure 1A]: they have a NIM close to their N-terminus and, more downstream, a homeodomain-like domain (HD, but which does not bind to DNA) and a C-terminal binding protein (CtBP)-interacting domain (CID) with four interaction motifs for (the non-DNA-binding) CtBP1/2 co-repressors. ZEB proteins contain two separated clusters of (between ZEB1 and ZEB2) strongly conserved zinc fingers. The N-terminal (NZF, with four fingers) and C-terminal cluster (CZF, with three fingers) enable ZEB proteins to bind with their two hands (and two zinc fingers in each) to DNA. Binding occurs to two separated CACCT sequences^[63,64], often also found as CACCTG known as E2-boxes, and in few cases also presenting as CACANNTG^[8,52] (CANNTG being an E-box). This could make ZEB proteins putative competitors of bHLH TFs for the same E-type boxes, including for the TCF4-containing heterodimers that bind to the CANNTG E-box, with a preference for C or G in the N positions^[65]. In the cell nucleus, ZEB2 (not ZEB1) detectably binds to receptor-activated phospho(p)-Smad proteins that act downstream of liganded TGFβ/Nodal/BMP family receptors. In ZEB2’s short (defined as 14 aa-long) Smad-binding domain, a tandem repeat of QxVx is crucial for binding to all p-Smads^[32,66].

Figure 1. Schematic overview of ZEB2 and TCF4. (A) Representation of (human) ZEB2 gene exons/introns (top) and ZEB2 protein structure (bottom). The largest ex8 encodes for the largest protein segment (in mouse, this corresponds to ex7). NIM: NuRD-interacting motif; NZF and CZF: N- and C-terminal zinc finger clusters; SBD: Smad-binding domain; HD: homeodomain-like domain; CID: CtBP-interacting domain (see Ref.^[8]). (B) TCF4 gene protein-coding exons (20 in total), the thereafter identified 3’ untranslated exon (ex21) and introns (topl), and TCF4 protein structure (bottom). For reasons of huge complexity, the numerous other identified 5’-non-coding exons and various TSSs (indicated minimally as …) are not included here. AD: Transcription activation domain; NLS: nuclear localization signal; CE: conserved element; RD: repression domain (sometimes indicated as Rep); bHLH: basic helix-loop-helix DNA-binding domain; NE: nuclear export signal. (C) Alternative TCF4 transcripts as determined for brain. Untranslated regions are not included in this drawing, so only the used protein-coding exons are drawn (in black), together with the one 3’-untranslated ex21 (in grey). On the left, the number/letter combinations indicate that these transcripts also contain brain-specific exons (of which the numbers are given) that are still incorporated in the transcripts, but not coding for protein in these cases, while the letters indicate different mapped splice sites. Names of the transcripts are indicated on the right. The brown arrowhead at the bottom of the panel indicates the region of alternative splicing resulting in a full-length or a so-called Δ variant. The orange arrow at the bottom indicates the location of the splicing causing the generation of the so-called - and + isoforms. The images in (B, C) are based on the original report^[67] and inspired by reviews^[9,113].

In most assays, ZEBs act as transcriptional repressors of tested target genes or transfected promoter-reporter combinations, but, by interacting with histone acetyltransferase (HAT) containing P300 or the related CREB-binding protein (CBP), they can also activate (other) sets of genes (see review Ref.^[52], and references therein). The mouse Zeb2 locus contains upstream of the first translated exon altogether nine untranslated exons (named U1-U9), which via alternative splicing, are joined to the protein-encoding parts of the Zeb2 transcript, however, without resulting in Zeb2 protein variants. The longest protein segment encoding exon in mice is ex7, corresponding to ex8 in humans [Figure 1A].

TCF4 protein/gene structure

The structure of the TCF4 gene is very complex: through the subsequent mapping of more alternative 5’-exons [20 in total, not shown in Figure 1B], the demonstration of 20 coding exons and the confirmation of one previously unreported 3’-terminal non-protein-coding exon [Figure 1B], the total number of exons in TCF4 has now risen to 41^[67] (for the equal complexity of mouse Tcf4 alternative transcripts, see Ref.^[68]; for subsequent reviews, see Refs.^[9,69]; also note that Ensembl lists 48 transcripts) and the size of the human gene accordingly to > 400 kb. Together with alternative TSSs upstream of seven exons in total, this TCF4 human gene structure is proposed to yield at least 18 protein isoforms (TCF4-A to TCF4-R), which differ at their N-terminus; at least 13 of these are present in the brain. The numbers of different mRNA transcripts are even higher due to alternative splicing within the protein-coding exons [indicated in Figure 1C]. Alternative 5’-exons are used in TCF4, also resulting in transcripts that only include ex7-20. To our knowledge, it remains to be established firmly whether human iPSCs or differentiated organoids (or other ex vivo models) recapitulate this huge diversity in TCF4 transcripts, its heterogeneity between cell types, and whether this is also the case in the mouse. This mission announces difficulty, even when using single-cell RNA-seq, for full-length sequencing of transcripts combined with the highest depth of sequencing will be needed.

The first suggestion of the position of the mutation in TCF4 being relevant to part of the PTHS phenotype, in particular, those aspects being linked to mild non-syndromic ID rather than full PTHS, were mutations in the 5’ region of the gene and cassette exons and regions not affecting the important functional domains of TCF4^[70]. Subsequent mapping of TCF4 mutations in PTHS did suggest genotype-phenotype correlation: those within ex1-4 and ex4-6 would cause mild ID, within ex7-8 severe PTHS albeit without overt facial malformation, and those within ex9-19 typical PTHS^[71]. These have been suggested to correlate with preservation of the intact nuclear localization signal (NLS) and bHLH domains, and with mutation of the NLS and minimally the bHLH domain (encoded by ex18), respectively. Although a genotype-phenotype paradigm has been proposed based on these observations, this has remained rather unclear to molecular biologists. They likely expected extensive documentation of the function of the mutant TCF4 proteins, their heterodimerization with the Class II, cell-type specific bHLH TFs, and their stability vs. degradation. However, significant attempts at addressing their subnuclear localization and dimerization, including in quantitative terms, and activity on target promoters such as NRXN1β and CNTNAP2 promoters have been carried out. These assays indeed pointed out the multiple effects of PTHS missense mutations on TCF4 global function^[72].

HLH heterodimers composed of Class I (seemingly “ubiquitous”, so present in nearly all cell types) and II (clearly cell-type specific) TFs bind to E-boxes in promoters and enhancers (see review Ref.^[73]). Their affinity to E-boxes is co-determined by co-factors, as well as by Ca²⁺-dependent proteins such as calmodulin that interact with the basic (b) stretch in the (b)HLH domain. In addition, flanking DNA sequences influence the binding affinity of these TFs. Indeed, a subset of TCF4-bound sites has the expanded consensus binding specificity 5’-C(^A/_G)-CANNTG-3’, sometimes with an added downstream outer C(^A/_G). When present as CpG, both possible additions can, in the mammalian brain, be variably methylated and have positive as well as negative effects on TCF4-binding^[74].

Some consequences and extra complexity

Interestingly, the 5’-C(^A/_G)-CANNTG-3’ motif has also been identified as a binding site for ZEB1 and, due to the high conservation of DNA-binding zinc fingers between ZEBs, been accepted for both ZEB1 and ZEB2^[8,63]. For such motif, the influence of methylation (of a 5’-CG) for binding of ZEBs has not been documented yet. In reverse, it is tantalizing that, via the bHLH domain, TCF4 may be involved in sensing cytosine modification, and that this might also be so for ZEB proteins. Similar to ZEB proteins, TCF4 recruits HATs such as P300^[75-77]. Additionally, in mouse NPCs, Tcf4 binds Mediator and thus co-defines super-enhancers that maintain cell identity^[78].

TCF4 is expressed in many cell types/organs, which classified TCF4 readily as a “ubiquitous” bHLH factor, despite the subsequent identification of cell- and/or tissue-specific presence of the many isoforms [for a schematic representation of TCF4, see Figure 1B]. Furthermore, the aforementioned complexity of its transcripts means that different TCF4 isoforms [for a selection, see Figure 1C; for recent update, see Ref.^[9]] contain or do not contain domains such as AD1, NLS1, and CE (then referred to as the Δ isoforms), or that the longer isoforms are the only ones that have a complete AD1 [Figure 1B]. The same holds for alternative incorporation or exclusion of an RSRS-tetrapeptide sequence (named as + are - isoforms, respectively) due to two alternative donor splice sites in ex18. In any case, all TCF4 transcripts have ex10-20, hence encompass AD3 (ex10-12), AD2 (ex14-16), Rep/RD (ex18), and HLH, NLS2, and NES1/2 (ex19).

All this makes the precise mapping of spatial expression and the estimation of the levels of each of the various TCF4 transcripts (and Δ and +/- isoforms) very complex and technically very difficult. This is also the case in the brain, but in general, the majority of transcripts are expressed here [these are specifically shown in Figure 1C], except for those encoding protein isoforms TCF4-J to -N, as these are only present in the testes.

ZEB2 and TCF4 gene regulation

Many studies involving in vitro, ex vivo (e.g., embryonic brain slices, which can also be subjected to focal electroporation^[79]), and in vivo perturbations using single- or two-allele Zeb2 genetic inactivation (often cKOs) have been carried out (in mouse, about 50 in total). These KOs were initially based on the expression domain of Zeb2 in mouse embryos and early post-natal stages (see review Ref.^[8]). These domains include the forebrain cortex, where Tcf4 is also expressed in all cells (be it at different intensities depending on the cortex layers, see below), and with Zeb2 transcripts and Zeb2 protein present in post-mitotic neurons of the upper layers, and both TFs are present in hippocampus. The KO-based studies in cultured cells and mice have also been combined with Zeb2 cDNA-based (again single- or two-allele) transgene cell-type specific rescue (involving cell-type specific Cre-controlled, from the Rosa26 locus with inserted Zeb2 cDNA). This allowed for the creation of various levels of Zeb2 in the model and obtaining detectable ranges of severity in cellular phenotype. Such experiments have contributed significantly to the conclusion that Zeb2 dosage, at the transcriptional level, needs precise control. Such control is necessary at every state/stage of Zeb2+ cells, for example, during exit from primed pluripotency (in ESCs^[80]), but also in NPCs and other differentiating cell lineages in vivo, and/or terminal maturation of these (e.g., natural killer cells^[81] and hematopoietic cell differentiation^[82]).

ZEB2 expression is controlled through promoter-proximal (where TFs such as Smad2, ETS1, HIF1α, POU/Oct and NF-κB, E2F1, FoxQ1 and FoxA2, and Fra1/AP1 have been shown to bind) as well as distant regulations located upstream in the ~3.5 Mb-long gene desert (see Ref.^[52], and references therein). In fact, ZEB2 ranks at the very top of proposed super-enhancer containing genes^[83]. Interestingly, subtle differences in ZEB2 expression, in spatial-temporal terms and strength, and across species (e.g., human, gorilla, and chimpanzee, with its levels experimentally perturbed in the respective cerebral organoids), influence neuroepithelial transition (hence, also cell shape) and have been proposed to co-determine brain expansion in evolution^[84]. Positive autoregulation of ZEB2 via a ZEB-binding site in the promoter-proximal region is an important driver of its upregulation during neural differentiation of pluripotent cells in vitro^[64]. In addition, an adult-specific remote enhancer 165 kb upstream of the Zeb2 TSS that binds the bHLH TF E2A/Tcf3 via four E-boxes has been identified and experimentally proven to be critical with regard to the structural integrity of the Zeb2 locus and its regulation in adult (vs. fetal) hematopoiesis. In adults, this enhancer is essential for the production of monocytes, plasmacytoid dendritic cells (pDCs), and splenic B-cells^[53].

Irrespective of the different TCF4 mRNA variants and protein isoforms and the resulting complexity, a detailed mapping of TCF4 presence in the brain remains critically needed to start correlating TCF4 dysfunction with pathophysiology in PTHS. Kim and Berens^[85] constructed and validated a green fluorescent protein (GFP) reporter mouse model that produces a tagged TCF4-GFP protein under control of the otherwise unperturbed TCF4 locus. They specifically achieved this through the insertion of a splice acceptor-loxP-P2A-GFP-stop-loxP cassette between introns 17 and 18 (upstream of the bHLH domain encoding exons). In doing so, they documented TCF4 presence in the pallial region and cerebellum of postnatal brain. In the forebrain cortex, TCF4 is present in both glutamatergic and GABAergic neurons, as well as in astrocytes and mature oligodendrocytes. These observations, highly similar to those with ZEB2, strongly suggest that these latter cell types, and not only neurons, should also be studied when assessing TCF4 pathophysiology. Intronic SNPs in the TCF4 locus have been identified by GWAS, and one combination (rs9960767 with rs9960767) was found to correlate strongly with schizophrenia (SCZ) and-remarkably-improvement of verbal memory^[86,87]. Jung et al.^[88] systematically documented Tcf4 expression in the developing and adult mouse brain and provided strong support for the role of Tcf4 in the development and plasticity of cortical and hippocampal neurons. In doing so, they confirmed the structurally similar brain defects between Tcf4 mutant mice and PTHS patients. They also expanded their TCF4 localization studies to fetal and/or adult rhesus monkey and human brains and reported highly similar localization between these and mouse brains.

iPSC-derived PTHS patient NPCs and neurons and PTHS patient fibroblasts can also be used as models for studying PTHS^[89]. Such cells also allow investigating how hypomorphic or dominant-negative mutations could cause PTHS. Genetic variation within the TCF4 locus or intronic SNPs may also cause ID in PTHS or associate with SCZ, respectively. Additionally, more detailed insight in TCF4 locus control may feed into pharmacological approaches to alter TCF4 expression in the CNS (and other tissues) and in the cellular models experimentally and document the effects via RNA-sequencing (RNA-seq).

Zeb2

Many reviews on ZEB2’s dynamic mRNA expression and ZEB2’s structure-function and role(s), including in the development of the nervous systems, have been published^[8,90-93]. In short, in mice, Zeb2 cKOs (Δex7, targeted by Cre-recombinase and the critical and largest ex7^[94]) have clearly confirmed Zeb2’s role(s) in early CNS development in mammals^[79,95]. Most studies had their focus on the forebrain, but some also included mid- or hindbrain, or spinal cord, and neural crest-derived peripheral/enteric nervous systems (PNS/ENS; see review Ref.^[8], and references therein). In the forebrain, studies expanded to early post-natal stages^[79,96], adult neurogenesis^[97], and injured brain^[98], as well as in the PNS to (re)myelination^[51,99] and pain sensing (see below).

In more recent studies, adult challenged mice have started to provide important roles of Zeb2 both as a marker and critical protective or causal gene (e.g., ulcerative colitis^[100], heart infarct^[101], liver steatosis and fibrosis^[102], and melanoma^[103]), thus, often in mouse systems/organs where no embryonic role or homeostatic role has been found, Zeb2 is not detectable or only at very low level expressed there or then. The recent increasing numbers of phenotypic descriptions of these challenged Zeb2-cKO mice, and in their organs or systems, may also urge clinical geneticists to follow up ontheir MOWS patients in the future in one or more of these extra directions.

Tcf4

Excellent overviews of the molecular mechanisms of TCF4 in PTHS, sometimes also addressing how these patients are treated and how TCF4 links to other psychiatric diseases, have been published (see^[109], and references therein). Here, we focus on Tcf4 mutant mice and the consequences for embryonic and adult neurogenesis.

Mapping of genome-wide binding sites

No satisfying anti-Zeb2 antibodies are available for efficient ChIP-seq. Recently, mouse ESCs were edited (using CRISPR-Cas) that produce tagged Zeb2-V5 (V5 heterologous epitope) from one of its two endogenous alleles^[64]. ChIP-seq (using anti-V5 antibody) mapped ~2400 Zeb2 binding sites in ESC-derived NPCs, with ~2300 of these annotated to ~2000 protein-encoding genes. Zeb2 is mainly recruited to the promoter-proximal region of transcription regulator or protein modifying enzyme encoding genes, but it also binds to distant REs. The ChIP+ genes include genes classified in Wnt signaling, which is in line with previous observations of DEGs of the Wnt system in Zeb2-cKO (CNS) mouse models. A second group included Dnmt3 family genes, which is in line with previous observations in Zeb2-KO ESCs submitted to neural differentiation^[80]. Of the aforementioned 2000 genes, about 1200 are differentially expressed during neural differentiation in vitro on Day 4 (epiblast-like stem cells) and Days 6 and 8 (NPCs) as compared to Day 0 (ESCs), including Tcf4. Importantly, Zeb2 was also found to potentiate its own expression during NPC production via a promoter-proximal binding site. This autoregulation, as shown by homozygous deletion of the ChIP+ DNA segment (again using CRISPR-Cas), is crucial for proper control of many of its key ChIP+ target genes in ESC-based neural differentiation^[64].

ChIP-seq for tagged V5-Tcf4 in mouse NSCs showed Tcf4 binding to enhancers and super-enhancers^[83], including loci encoding neurogenic bHLH factors, but also-albeit not yielding a strong signal in NSCs-its own promoter^[78]. Tcf4 was found to bind to its targets together with its interaction partners such as Smad4, Sox2, and Chd7^[14]. Gene ontology analysis on Tcf4 target genes, i.e., bound and activated by Tcf4, showed that Tcf4 predominantly controls genes involved in the cell cycle and M-phase of the latter. This could fit with the microcephaly phenotype that is part of PTHS.

Interactomes

Transitions of stem cells into different lineages, and progression of cell differentiation and maturation, including neural cell types, are driven by changes in gene expression regulated by specific TFs that often bind to evolutionary conserved DNA motifs in TSS-proximal sequences (e.g., promoters and first intron) and to distant REs. Transcriptome, epigenome, and higher-order chromatin do not stand by themselves but are intertwined and influence each other herein. This, in collaboration with the actions of chromatin-modulating complexes, results in the proper activation or repression of target genes. Chromatin modulation involves re-modelers and modifying enzymes including methyltransferases, HATs and kinases (often collectively called chromatin writers), demethylases and histone deacetylases (HDACs, erasers), and complexes such as the switch/sucrose nonfermenting (SWI/SNF) complex [also known as BRG1/BRM-associated factor (BAF complex); named readers]. Enhancers are often located in-cis kb to sometimes Mb away from the specific genes they regulate, requiring TF- and Mediator (complex)-assisted DNA-looping. Thus, correct gene expression also requires epigenetic reorganization via chromatin modifications and chromatin remodeling.

Already from the earliest studies with Zeb1, ZEBs have been proposed as putative competitors in DNA-binding for E-box-binding transcriptional activators (bHLH TFs) simply by target-site occupancy, soon followed by the first reports of ZEBs being active transcriptional repressors in most assays. For this latter activity, co-repressor binding has proven important, e.g., with NuRD. It is composed of, among others, at least the ATPases CHD3/4/5, HDAC1/2, metastasis-associated protein (MTA)-1/2/3, retinoblastoma binding protein and histone chaperone RbBP-4/7 (also named RbAp48/46), methyl-CpG-binding domain protein (MBD)-2/3, and zinc-finger TFs GATA2A/2B^[127,128]. For repression, this complex thus combines deacetylation of histone-H3 by HDAC1/2 with demethylation of H3K4 (see review Ref.^[129]). NuRD-mediated chromatin remodeling in cerebral cortex development, neurodevelopmental disorders, and myelination was recently reviewed^{[6,29,130,131]}.

Various subunits of the NuRD complex, including CHD4, were identified in affinity-purified Zeb2 complexes isolated from Zeb2-overproducing heterologous cells^[50]. ZEB2 associates with NuRD via its NIM (see above), but the subunit/s of NuRD that mediates this direct interaction is/are not known. In two aforementioned patients with mild MOWS caused by mutations affecting the ZEB2 NIM, one being a subtle R22G missense mutant, the only direct molecular detectable consequence is that ZEB2 interaction with the NuRD complex is lost. Such aberrant ZEB2 is indeed unable to recruit NuRD (subunits) but, as a result, displays reduced transcriptional repression activity on, e.g., the XBMP4 gene promoter (in Xenopus (X) embryos), a target of XZeb2. By using a dominant-negative variant of the ATP-binding NuRD component CHD4/Mi-2 (its in vitro made sense RNA was injected in 1-4 cell stage Xenopus embryos), it was shown that this subunit and thus NuRD is actively involved in repression of the ZEB2 target gene Cdh1 (encoding E-cadherin, cadherin-1) and in XZeb2-controlled neural induction^[50]. This work also showed that the absence of NuRD binding to human ZEB2 can be the cause of mild forms of MOWS. Similar functional consequences (and inability to rescue Zeb2-KO) of NIM-mutant Zeb2 were documented in myelination and adult neurogenesis^[51,97]. Cellular systems that enable documenting GWBS for a NIM-defective human ZEB2 are not available yet; hence, target genes that may co-depend or depend exclusively on NuRD-ZEB2 interaction remain unknown.

Tcf4 interacts with more than 100 partners^[14], including chromatin remodeling complexes such as SWI-SNF, NuRD, and TRRAP, but also many TFs. Known partners, including HLH factors Ascl1 (also named Mash1), Id1, and Id3/4, were also identified in this study. Other bHLH factors (Tcf12 and E2-A/Tcf3) bind to Tcf4 as well. Interestingly, Zeb2 was detected in tagged Tcf4, Sox2, and Npas3 interactomes^[14]. A major functional finding here was that Tcf4 binds and activates multiple genes involved in primary microcephaly, such as Mcph1 and Wdr62. It co-localizes on primary microcephaly genes with its interaction partners Smad4, Sox2, Chd7, and Ep300. Smad4 also (co-)activates transcription of Mcph1 and Wdr62. Mutations in SMAD4, EP300, CHD7, and SOX2 cause microcephaly in at least some patients. Therefore, a picture has emerged that TCF4 is part of a protein interaction network regulating human primary microcephaly genes, in which network members cause microcephaly as part of the human phenotype they cause.

Mediator was not picked up in Tcf4 co-purifications^[14], but vice versa Tcf4 did show up prominently in purification of the Mediator complex from NSCs^[78]. Identification of enhancers and super-enhancers in NSCs by Mediator ChIP-seq showed that they nearly completely overlap with genome-wide binding of Tcf4. Moreover, Tcf4 KD lowered Mediator binding to common binding sites, showing that Tcf4 exerts genome-recruitment activity on Mediator^[78]. This co-localization of TCF4 and Mediator is striking, and in fact, goes beyond enhancers, making both Tcf4 and Mediator mark so-called broad promoters^[132,133]. Broad promoter containing genes are strongly linked to cell identity. This suggests that the identified Tcf4-Mediator axis plays a major role in NSCs^[78].

Impact on or by Wnt signaling

The most affected deregulated genes resulting from Zeb2 deficiency, certainly in the development of the nervous systems, but also in cell culture, often map to the Wnt signaling system. This is the case in mouse hippocampal development^[134], embryonic CNS myelinogenesis, and post-natal Schwann cell (SC)-mediated (re)myelination in the PNS^[51,99,121]. Myelination (see also above) is absolutely required for optimal conduction velocity of nerve impulses, and defects may cause disrupted motor function. It occurs by the formation of myelin sheets generated by oligodendrocytes/SCs, which derive from OPCs. The differentiation of these OPCs toward mature, myelinating cells is inhibited by BMP, Wnt, and Notch signaling^[135]. Zeb2 protein is present in cells that accompany motor neuron axons in the trunk of early mouse embryos at the level of the limbs^[136]. Zeb2 was proposed as a strong direct target for the upstream TF Olig2, and Zeb2 mRNA requires normal levels of Olig1. Olig1 and Olig2 (including their overproduction) also impact neurodevelopmental disorders, even in Down Syndrome and autism spectrum disorders^[137]. Furthermore, Olig1 and Olig2 bHLH TFs are essential for CNS myelinogenesis^[137]. Analysis of Olig1 and Olig2 KO mice revealed that in P14 and E14.5 spinal cords, respectively, Zeb2 mRNA levels become extremely low to undetectable. In cultures of adult rodent early-OPCs, overproduction of Olig1 and/or Olig2 activates Zeb2. High levels of Zeb2 protein are present in mature oligodendrocytes, and only very low levels are detected in OPCs. These observations provided a strong basis for generating Olig1-Cre;Zeb2 cKOs and documenting the effects of Zeb2’s removal on CNS myelinogenesis. Such mice are viable, but two weeks after birth, they develop generalized tremors, limb paralysis, and seizures, similar to the symptoms observed in other genetic mouse models with defective myelinogenesis.

Intact Zeb2 was further shown to be dispensable in platelet-derived growth factor receptor-alpha (PDGFRα)+ OPCs, but its presence is critical for their differentiation. Subsequent experiments demonstrated that Zeb2 downstream of Olig1 and Olig2 is needed during CNS myelinogenesis for differentiation of OPCs to myelinating cells, and that this requires both Zeb2’s repressor and activator activities^[8]. Importantly, Zeb2 represses sets of BMP-Smad-activated genes that normally inhibit myelinogenesis, and Zeb2 directly activates Smad7, which (co-)downregulates BMP signaling with Zeb2 (via BMP-pSmad interaction). In a Smurf-dependent manner, Smad7 also sends Wnt-activated β-catenin (βcat), with the latter normally inhibiting myelinogenesis also, into degradation^[121]. Thus, in a cell-intrinsic manner, Zeb2 is needed to generate anti-BMP-Smad activity and anti-Wnt-βcat effects in CNS myelinogenesis.

Based on this, a number of predictions could be tested and confirmed. One is that, in neonatal OPCs, whose proliferation is stimulated by PDGF-AA, Zeb2 overproduction would enhance their differentiation to mature oligodendrocytes, as detected by downregulation of genes encoding negative regulators of differentiation, and a strong increase of Sox10 and Olig2 expression. Furthermore, other experiments demonstrated that Zeb2-P300 co-operation is responsible for direct activation of the pro-myelination genes. Another prediction was that Smad7 overproduction would rescue the myelination defect in Zeb2 KO OPCs, which was the case^[121]. These results are the start of new investigations aiming at addressing the Zeb2 function in SCs of the adult PNS, where Zeb2 is present^[51]. Sciatic nerve injury already rapidly upregulates Zeb2 (in a few hours), and Zeb2 levels remain high in remyelinating SCs for a long time after the injury^[99].

Taken together, these Zeb2 studies of the cellular and molecular phenotypes in CNS myelinogenesis and PNS myelination are comparable in many respects. They show that intact Zeb2 is needed to control signaling pathways that impact myelinating cells, i.e., Zeb2 is needed to generate and maintain anti-Notch/Hey2 and anti-Sox2, as well as (in both CNS and PNS) anti-BMP and anti-Wnt activities in these cells. Rescue experiments with the NIM-deficient Zeb2^R22G mutant (see above) highlighted the importance of NuRD-interaction. Importantly, these results in the mouse models recapitulate the phenotypes observed in MOWS patients presenting a delay in myelination and defects in white matter, proposed to cause motor deficiencies such as a delay in walking.

Modulation of Wnt/βcat active signaling and epigenetic status also caught attention by the TCF4 field^[138]. In the case of signaling, the approaches used Wnt pathway activation via the addition of Wnt3a-containing conditioned media or CHIR-99021, a potent inhibitor targeting GSK3, wherein GSK3 normally negatively regulates canonical Wnt signaling. Both treatments resulted in 1.4-6-fold upregulation of TCF4 depending on the specific transcripts analyzed, but these are experimentally difficult to allocate via Western blotting, due to the different TCF4 protein isoforms (see above). Further analysis, also involving the annotation of potential TCF/LEF binding sites in TSS-proximal sequences of Wnt-responsive genes, revealed that the highest Wnt-responsive TCF4 transcripts [encoding a short form of TCF4, TCF4-A; see Figure 1C] contained multiple TCF binding sites, which could-albeit indirectly-be confirmed for binding (i.e., by ChIP-seq of TCF7L2).

A next approach by the same authors used two different pharmacological inhibitors of Wnt canonical signaling in NPCs stimulated with Wnt-containing conditioned medium (CM) (i.e., the tankyrase inhibitor XAV-939, causing enhanced degradation of βcat, and ICG-001, preventing βcat from binding to the HAT CBP, a member of the P300 family). These treatments were found to downregulate TCF4 mRNA expression. They also paved the way to test selective HDAC inhibitors, investigate, in particular, the exclusive synergism of Class I HDAC inhibitors in Wnt-CM conditions only, and document these experiments by RNA-seq and TCF4 binding site analyses, in particular for DEGs^[138].

Many of these DEGs are known as critical to nervous system development and function and act in synaptogenesis, post-synaptic density scaffolding and signaling, and axon growth and guidance. In addition, many of these DEGs and other ones have been associated with autism, ID, and/or psychiatric disorders. They represent interesting downstream markers and are part of signatures that can be compared between PTHS patient-derived cells and neurons. Such work is also inspiring for ZEB2 research into pharmacological modulation that targets ZEB2 expression regulation from the single remaining intact ZEB2 allele in MOWS.

ZEB2 and TCF4, each other’s modifier for neural differentiation in vitro, but also of other genes in normal brain development?

As illustrated above for ZEB2 and TCF4, but also applicable to other neurodevelopmental TFs, the combination of genetic analysis and mutation mapping, the phenotypic and transcriptomic analyses of appropriate cellular and animal models, and the mapping of genomic TF binding-sites and TF interactomes give insight into TF actions (also in concert with chromatin remodeling complexes) at the level of individual disease genes, as well as GRNs and PRNs, including these steering neural differentiation.

The first example of new insight is that a number of Zeb2 ChIP+ target genes, at least as mapped thus far in mouse ESC-derived NPCs, could be associated with human neurodevelopmental disorders (such as ID, mental disorders, eye defects, seizures, and speech impairment). For instance, a strong Zeb2 binding site has been mapped to the PTHS Tcf4 gene proximal region^[64]. In addition, in the same set of experiments, perturbation of Zeb2 autoregulation via mutating the Zeb2-binding site in the promoter-proximal region of Zeb2 results in strong downregulation of not only Zeb2 but also Tcf4. This DNA segment, however, remains to be tested firmly for Tcf4 occupancy. Preliminary observations suggest this signal is not strong, at least in NSCs, possibly as a result of the combination of Zeb2 being expressed at a low level and Tcf4 having a preference for enhancer binding, including its own enhancer located 200 kb within the gene^[78]. Thus, by extension, in NSCs and early neurogenesis, the aforementioned possibility of competition between Zeb2 and a Tcf4-bHLH heterodimer for E-boxes is unlikely, although this still has to be investigated in depth for competition at the Zeb2 autoregulatory site.

Obviously, one still needs details on enhancers as well as meaningful SNPs for both TCF4 and ZEB2, and how variation in those affects the levels of their mRNA and protein levels including in, e.g., human NPCs, as well as how this affects for each factor its interactions. In more general terms, this also means that, in neurodevelopmental iPSC-derived models, the effect of a perturbation of a given TF or chromatin remodeler will have to be documented in terms of its epistasis or GRN/PRN. This can occur via KD or, alternatively, via mutation in a domain or allelic genetic inactivation. The choice of KD vs. KO may influence epistatic relations and compensatory pathways, which may differ between choices. In any case, such sophisticated biochemical and functional studies are expected to confirm ZEB2 and/or TCF4 as important hubs in GRNs/PRNs, but also as modifier gene(s). They could very well be modifiers not only of one another (empowered by similarities of defects between MOWS and PTHS), but also of other causal genes in phenotypically overlapping neurodevelopmental syndromes (discussed in Ref.^[64]).