Evidence for a potential common gene network of Smith-Magenis and Potocki-Lupski syndromes, DAND (DEAF1-associated neurodevelopmental disorder) and ZEB1 -associated neurodevelopment disorder

Background: Neuropsychiatric disorders are highly heterogeneous and often display overlapping phenotypes, suggesting possible defects in common genetic networks. We previously reported pathogenic variants in deformed epidermal autoregulatory factor-1 (DEAF1) that contribute to DEAF1 -associated neurodevelopmental disorder (DAND) and predicted that DEAF1 regulates expression of retinoic acid induced 1 (RAI1) , the causative gene of Smith-Magenis syndrome (SMS) and Potocki-Lupski Syndrome (PTLS), through a putative DEAF1 binding sequence (DBS) in RAI1 intron 2, suggesting a common genetic network of these disorders that needs further study. Methods: In this study, we tested the DEAF1 binding and transcriptional regulation of RAI1 using luciferase reporter gene assay, EMSA and ChIP. We explored the upstream regulator of DEAF1 using bioinformatics analysis, qPCR and ChIP, and generated a potential network involving these genes based on patients’ phenotype and bioinformatics analysis. Results: We demonstrated that DEAF1 binds to the RAI1 DBS and regulates RAI1 expression in vitro , and a pathogenic variant in the DEAF1 SAND domain was defective in binding the RAI1 DBS. We also obtained evidence that DEAF1 expression is regulated by the transcription factor ZEB1 via predicted binding sites in DEAF1 intron 1. Microdeletions and variants in ZEB1 have been implicated in brain developmental disorders, including intellectual disability (ID), autism, agenesis of corpus callosum (ACC) and corneal dystrophy. Patients harboring deletions or pathogenic variants of Z EB1, DEAF1 or RAI1 display partially overlapped symptoms, including ID, autistic features, speech impairment, developmental delay and dysmorphologies. Conclusions: Together, these results provide evidence for a common molecular network in ZEB1 -associated neurodevelopment disorder, DAND, SMS and PTLS.


Plasmid construction
Construction of the wildtype DEAF1 expression plasmid in pcDNA3 (DEAF1-pcDNA3) and site-directed mutagenesis to introduce the c.737G>C (p.Arg246Thr) mutation in DEAF1 expression plasmids have been previously described [6]. A DNA segment encoding the DEAF1 open reading frame (GenBank accession number NM_021008) with an HA peptide inserted at the N-terminal after ATG translation start site was cloned into pcDNA3.1 (Invitrogen) to generate the HA-DEAF1 plasmid.
Luciferase reporter plasmids RAI1-Luc plasmid was generated by PCR amplified the All inserts were confirmed by Sanger sequencing. Sequences of oligonucleotide primers used for the construction of plasmids are listed in Table S1.

Cell transfection
Approximately 5x10 5 HEK293T cells were plated in 2 ml growth medium without antibiotics in each well of 6-well culture plates 24 h prior to transfection. A total of 4 µg total plasmid DNA was transfected into the cells in each well using 5 µl of Lipofectamine 2000 (Invitrogen). pUC19 plasmid was used as filler DNA to equalize the transfection efficiencies across all groups. Both the plasmids and Lipofectamine 2000 were diluted in 250 µl OptiMEM Reduced Serum Medium (Invitrogen), mixed and incubated for 20 min at room temperature before addition to wells. Following transfection, the cells were incubated at 37℃ under 5% CO 2 for 6 hours, at which time the culture medium was replaced. The cells were than incubated under the same conditions for an additional 24 h.
For siRNA transfection, 6 pmol siRNA (GenePharma Co. Ltd) were diluted in 100 µl OptiMEM Reduced Serum Medium (Invitrogen) for each well, after which 1 µl of Lipofectamine RNAiMAX (Invitrogen) was added and the mixture incubated at room temperature for 10-20 minutes. The mixture was then added to SH-SY5Y cells at a final concentration of 10 nM. Sequences of ZEB1 and control siRNAs are listed in Table S1.

Luciferase assays
Transcription assays using the pDEAF1-pro-luciferase construct have been previously described [6]. Briefly, cells transfected with RAI1-Luc luciferase reporter plasmids were washed with 2 ml Dulbecco's phosphate-buffered saline (DPBS, Invitrogen), lysed, and scraped with 500 µl of Tropix lysis solution (ABI) in each well. Lysates were collected and centrifuged at 12,000 RPM for 2 min to pellet cellular debris. For each sample, 50 µl supernatant was transferred to each of the four wells in a 96-well luminometer plate. Two wells were treated with 70 µl diluted Tropix Galacton substrate (1:100, Tropix Galacton reaction buffer diluent with Tropix Galacton substrate, ABI), incubated for 30 min, after which 100 ml of Tropix light emission accelerator (ABI) was added. The other two wells were treated with 100 µl Steady-Glo luciferase substrate (Promega Corporation). Synergy 2 (Biotek) was used to quantify the luminescence. Relative luciferase activity was calculated by dividing the average number of light units from the wells containing the Steady-Glo luciferase substrate (Promega Corporation) by the average number of light units from the wells containing the Galacton substrate and Accelerator. Wells containing pUC19, psvb-Gal, and RAI1 luciferase constructs were used as baseline luciferase activity and normalized to 1. Each experiment was performed in triplicate biological replicates and p-values derived from ANOVA and post hoc Tukey's tests.

Quantitative polymerase chain reaction (qPCR)
The cDNA synthesis and quantitative PCR were carried out as described previously [18].
Sequences of DEAF1, ZEB1 and GAPDH primers are listed in Table S1. Gene expression levels of control samples were normalized to GAPDH and set to 1.

Electrophoretic mobility shift assays (EMSA)
Epitope FLAG-tagged WT protein was purified from transfected HEK293T cells and nuclear extracts was prepared as previously described [6]. EMSAs were performed using double-stranded (ds) DNA probes containing the putative DBS within RAI1 exon 2, a DEAF1 binding site in DEAF1 5' region [28], putative ZEB1 binding sites within DEAF1 5' region.
Sequences of the synthetic oligonucleotide DNA probes used in these experiments are listed in Table S1.

Chromatin immunoprecipitation (ChIP)
ChIP assays were performed following the Cold Spring Harbor protocol with minor modifications [29]. Briefly, HEK293T cells co-transfected with a HA-DEAF1 expression plasmid and RAI1 DBS plasmid were fixed in 1% formaldehyde and collected 48 h after transfection. Chromatin was digested using micrococcal nuclease (Pierce), as recommended by the manufacturer. Each immunoprecipitation reaction mix contained 50 μg chromatin and 5% of this amount (2.5 μg) chromatin was collected separately to quantify the amount of input target DNA present in each immunoprecipitation reaction mix. Five μg mouse anti-HA-Tag (6E2) monoclonal antibody (Cell Signaling Technologies, #2367; IgG1 isotype) was used for DEAF1 immunoprecipitation, 5 μg ZEB1 monoclonal antibody (Sigma, HPA027524) was used for DEAF1 immunoprecipitation. Immunoprecipitation with 5 μg mouse IgG (Sigma, I5318) used as a negative control. DNA was purified using MinElute reaction cleanup kits (Qiagen) and subjected to PCR using primers designed to amplify the DBS-containing fragment in RAI1 5' region. PCR products were quantified using qPCR and quantitative scanning of the bands in the gels using ImageJ program. The sequences of primers used to amplify the RAI1 DBS are listed in Table S1.

DEAF1 is a potential upstream regulator of RAI1
DEAF1 is a transcriptional factor recently implicated in ID, autism, and development delay [2]. It binds to specific sequences within gene regulatory regions and stimulates or represses mRNA expression [2]. To identify novel candidate target genes, we performed a genome-wide analysis of potential DEAF1 response elements in human gene transcriptional regions (~45 kb) in the Regulatory Sequence Analysis Tools (RSAT) database [30] using a preferred degenerate DEAF1 binding consensus sequence, 5' YtcggNNNNYtccg 3' [17,28], and identified 318 candidate genes, representing a subset of possible DEAF1-regulated genes (Table S2). Among these, Database for Annotation, Visualization, and Integrated Discovery (DAVID) database identified 66 genes related to human diseases (Table S3).
Further analyses revealed that 15 genes are implicated in psychiatric, neurological, and developmental abnormalities related to DAND phenotypes, including RAI1, the causative gene for SMS and PTLS, based upon information from the Online Mendelian Inheritance in Man database (OMIM) (http://www.omim.org/) and the NIH Genetic Association Database (http://geneticassociationdb.nih.gov/) ( Figure S1, Table S4) [31,32]. We have previously provided evidence that DEAF1 regulates RAI1 expression in human brain via a putative DEAF1 binding sequence (DBS) located within RAI1 intron 2 [18], suggesting that DEAF1 and RAI1 may function within a common molecular network.

DEAF1 binds and regulates RAI1 in vitro
To obtain evidence for DEAF1 regulation of RAI1 expression, we measured luciferase activity in HEK293T cells transfected with an expression plasmid (RAI1-Luc) containing a fragment from RAI1 intron 2 that includes the putative DBS positioned upstream from a minimal promoter firefly luciferase. A statistically significant increase in luciferase activity was observed in cells co-transfected with a WT DEAF1 expression plasmid, compared to cells transfected with RA1-luc alone. (Figure 1A).
To determine whether DEAF1 directly binds to the RAI1 intron 2 DBS, EMSAs were performed using a double-stranded (ds) DNA probe containing this sequence. A previously published dsDNA probe containing a DEAF1 binding sequence in the human DEAF1 promoter [33] was used as a positive control. As predicted, purified FLAG-tagged WT DEAF1 protein DEAF1 bound to the putative DBS in RAI1 intron 2 ( Figure 1B).
To confirm the binding of DEAF1 to RAI1 intron 2, chromatin immunoprecipitation (ChIP) assays were performed using nuclear extracts of HEK293T cells co-transfected with HA-DEAF1 expression and RAI1-DBS plasmid. DNA fragments co-immunoprecipitated with anti-HA antibodies were amplified by PCR ( Figure 1C). PCR products were quantified by quantitative scanning of the bands in the gels using ImageJ program ( Figure 1D).

DEAF1 variants are impaired in DNA binding of RAI1
DEAF1 comprises a N-terminal DNA binding domain containing Sp-100, AIRE, NucP41/75, and DEAF1 (SAND) motifs, a zinc-binding motif, a nuclear localization signal (NLS), a leucine-rich nuclear export signal (NES), and a C-terminal cysteine-rich protein interaction domain termed Myeloid translocation protein 8, Nervy, and DEAF1 (MYND) domain [4]. A SAND domain c.737G>C (p.Arg246Thr) has been reported to associate with ID, speech impairment, autism, and developmental delay [6]. We tested the influence of the SAND domain DEAF1 variant c.737G>C (p.Arg246Thr) on binding to the double-stranded DNA probe containing the RAI1 intron2 DBS. The results show that the c.737G>C (p.Arg246Thr) variant impairs the ability of DEAF1 to bind to the RAI1 intron 2 DBS ( Figure   S2).

ZEB1 regulates DEAF1 expression
To further explore the genes involved in the DEAF1-RAI1 network, we used BrainSeq Consortium database (http://eqtl.brainseq.org) to look for potential upstream regulators of DEAF1. We identified 106 expression quantitative trait loci (eQTLs) in the dorsal lateral prefrontal cortex (DLPFC) of a 412 subjects cohort, and 66 eQTLs in a 237 subjects cohort which are significantly associated with DEAF1 mRNA expression based on RNA sequencing (RNA-seq) and genotype data in BrainSeq Consortium database. The eQTL modeling in the database was adjusted for multiple covariates, including sex, age (>13 years), and ancestry (multidimensional scaling components). Significant eQTLs were defined as SNP/DEAF1 mRNA expression pairs with a false discovery rate (FDR) less than 1% that replicated with the same allelic direction and p<0.01 in two independent sets of DLPFC samples from the CommonMind Consortium (https://www.nimhgenetics.org/resources/commonmind) and the GTEx project (https://www.gtexportal.org) databases. Among the 52 overlapped significant eQTLs identified from the two cohorts, we screened for SNPs that had Regulome database scores ≥2b, which denotes evidence for transcription factors (TF) binding, a known TF motif, a DNase footprint and location within a DNase-sensitive site. Three SNPs (rs10751662, rs6597991, rs71464105) had Regulome database scores of 2b, suggesting possible transcriptional factors that regulate DEAF1 mRNA expression (Table S5). Low-and high-expression alleles of these SNPs for DEAF1 mRNA expression in human dorsolateral prefrontal cortex (DLPF) are shown in Figure S3. Pairwise linkage disequilibrium (LD) coefficients D' and r 2 calculated using the LDlink database (https://ldlink.nci.nih.gov/) showed that all three SNPs are in high LD in the Caucasian population (Table S6). We compared the locations of histone III lysine 27 acetylation (H3K27Ac) and DNase sensitive clusters near these eQTLs using UCSC Genome Browser (http://genome.ucsc.edu/) and found strong signals for H3K27Ac and DNase sensitive clusters near SNP rs6597991, consistent with chromatin remodeling and active transcription in this region. Based on the Regulome database, SNP rs6597991 is located within a predicted ZEB1 binding site, suggesting that the ZEB1 transcription factor may contribute to the regulation of DEAF1 transcription ( Figure   2). ZEB1 (zinc finger E-box binding homeobox 1; OMIM 189909) is strongly expressed in early development of the central nervous system (CNS) [27], and has been implicated in patients with autism, intellectual disability (ID), agenesis of corpus callosum abnormality (ACC) [63,65], posterior polymorphous corneal dystrophy 3 (PPCD 3, OMIM 609141) [34,35] and Fuchs' endothelial corneal dystrophy (FECD6, OMIM 613270) [36]. ZEB1 recognizes a single or bipartite motif comprising one CACCT sequence and one CACCTG E-box on its target gene with various orientations and spacing [37]. We analyzed the sequence in the DEAF1 promoter region using the Find Individual Motif Occurrences (FIMO) program in the MEME suite (meme-suite.org/), and identified a potential bipartite ZEB1 binding sequence, CATGT and CAGGTG (ACATG and CACCTG in the complementary DNA strand) separated by 30 nt within DEAF1 intron 1, with SNP rs6597991 located within the CAGG(T/C)G sequence (Table S7). DNA binding activity to a dsDNA probe containing the putative rs6597991-centered ZEB1 binding site in DEAF1 intron 1 was detected using HEK293T nuclear extracts in EMSAs ( Figure 3A). We carried out chromatin immunoprecipitation (ChIP) assays in HEK293T cells using mouse monoclonal antibodies against human ZEB1 and PCR to amplified a 118 bp DNA fragment centered on rs6597991 ( Figure 3B). Quantitative PCR confirmed statistically significant differences between the amounts of targeted DNA precipitated by anti-ZEB1 antibodies or by IgG ( Figure 3C).
We noticed that the sequence CATGT is less conserved than the canonical AGGTG sequence in the middle two nucleotides. Scanning a wider region in the 5' regulatory sequence of DEAF1 identified a cluster of potential ZEB1 binding sites within DEAF1 intron 1, located ~900 bp upstream of SNP rs6597991 (chr11:693,829-695,048, 1219 bp, Figure 4).
Using the FIMO program in MEME suite, we identified clusters of possible bipartite ZEB1 binding sites with more conservation in the consensus sequence [37]. E-box sequences in this region predicted by the FIMO program are listed in Table S8. We carried out ChIP assays and confirmed that ZEB1 binds to this 5' region of the DEAF1 gene ( Figure 5A). Quantitative PCR detected significant differences between immunoprecipitation of the target DNA fragment compared to chromosomal DNA immunoprecipitated by IgG. Examination of the literature also confirmed that these sites were detected in a genome-wide, ChIP-based scan of potential ZEB1 binding sites [38]. And an independent study from the ENCODE Transcription Factor Targets dataset by ChIP-seq also confirmed that ZEB1 binds to DEAF1 [39].
Knock down of ZEB1 mRNA expression in neuronal SHSY5Y cells using ZEB1 siRNA increased DEAF1 mRNA expression by ~50%, compared to cells treated with control siRNA ( Figure 6). Because ZEB1 encodes a zinc finger transcription factor that is likely to play a role in transcriptional repression of its target genes when bound to its consensus binding site [33], our data suggest that ZEB1 may function as a potential upstream repressor of DEAF1 in neuronal cells by binding to multiple bipartite ZEB1 binding sites within DEAF1 intron 1.
To identify possible networks common to RAI1, DEAF1, ZEB1 and mTOR, we explored  (Table S9, S10). It's worth mentioning that all the 18 genes have high confidence with autism, with average confidence of 0.605. Tissue-specific expressions of these genes were demonstrated using the HumanBase database ( Figure S4). And we further tested if there's tissue-specific gene network by investigating co-expression-based molecular interactions among these genes (Figure 8). In neurons, DEAF1, ZEB1 and GSK3β have more high-confidence interactors than other genes inquired, suggesting that these three genes play crucial roles in neuronal growth and differentiation. In all tissues, CTNNB1 and AKT1, but not DEAF1, ZEB1 and RAI1 have more high-confidence interactors than other genes inquired, implies that ZEB1, DEAF1 and GSK3β comprise a neuronal functional network compared to other tissues, that possibly contributes to psychiatric and development disorders.
Evidence for a common network in neurons suggests that pathogenic variants in these genes may produce overlapping defects in neurogenesis and development. To test this hypothesis, we compared the phenotypes of RAI1-associated SMS/PTLS, DEAF1-associated DAND and ZEB1-associated neurodevelopment disorder ( and RAI1 function within common networks and/or interact with each other.

Discussion
In this study, we demonstrated that the transcription factor DEAF1 binds to a DBS within intron 2 of RAI1 and stimulates RAI1 mRNA expression in vitro, providing experimental confirmation of previous bioinformatic-based predictions of RAI1 regulation in human brain [18]. These observations provide the first experimental evidence that DEAF1 and RAI1 function within the same intracellular network. Our study also provides bioinformatic and experimental evidence that DEAF1 expression is regulated by the transcription factor ZEB1, Deletion or mutation of HDAC4, a histone deacetylase eraser associated with brachydactyly mental retardation syndrome (BDMR, OMIM 600430) and 2q37 deletion syndrome [44], also results in reduced expression of RAI1 [45]. Reporter studies revealed that RAI1 directly regulates BDNF expression, a gene associated with neurodevelopment and behavior problems [22]. RAI1 positively regulates CLOCK, a key component of the mammalian circadian oscillator, and was associated with mood problems. Haploinsufficiency of RAI1 in SMS fibroblasts and in mouse hypothalamus results in dysregulation of multiple circadian genes including CLOCK [21,46].
Whole exome sequencing identified several pathogenic variants in DEAF1 producing a spectrum of symptoms recently designated DEAF1-associated neurodevelopmental disorder (DAND) [5,[7][8], and a subset of these symptoms are also observed in SMS/PTLS. The SAND domain of DEAF1 is crucial for multimerization and DNA-binding as well as protein-protein interactions [47]. It represses the DEAF1 promoter through autoregulation [48], and recognizes core-binding motifs (TTCGGGNNTTTCCGG or flexibly spaced TTCGGN 3-8 TTCGG) in other gene promoters [17,33]. It also bind with lower affinity to a single half-site motif sequence TTCG in 5-HTR1A promoters, a gene associated with major depression, anxiety, suicidal tendencies, and panic disorder [49]. We tested the DNA binding affinity of pathogenic DEAF1 variant, and showed that the SAND variant c.737G>C (p.Arg246Thr) [6,9,14] had reduced binding affinity for the putative DEAF1 consensus sequence in RAI1. Exploring how RAI1 expression is regulated by DEAF1 helps to elucidate the genetic mechanisms underlying these disorders, and may prove helpful information for identifying potential drug targets for their treatment or prevention.
We also showed evidence that DEAF1 expression is regulated by the transcription factor  [37]. It binds with relatively low affinity to a single E-box [51], and with high affinity to a bipartite motif as a monomer with its N-terminal zinc finger cluster attached to one E-box and its C-terminal zinc finger cluster attached to another E-box/E-box-like sequence [37] It also regulates target gene expression indirectly by interacting with transcription factors that bind chromosomal DNA at sites unrelated to the canonical ZEB1 binding site [52]. ZEB1 is a downstream gene in the AKT/mTORC1/GSK3β pathway, involved in the TGF-β1-induced epithelial-to-mesenchymal transition (EMT) [40], a physiologic process occurs largely during embryonic development but is aberrantly reactivated in pathologic situations [53].
Mechanistic target of rapamycin (mTOR), a member of phosphoinositide 3-kinase (PI3K)-related kinase family, plays important roles in neurogenesis, cellular proliferation, apoptosis, metabolism and development. Altered activity of mTOR has been implicated in many neurological and neuropsychiatric disorders, including autism, intellectual disability and epilepsy [42]. In the AKT/mTORC1/GSK3β pathway, AKT3 are predominantly expressed in the CNS, and regulates axon regeneration in CNS, including in retinal ganglion cells, via interactions between mTOR and GSK3β [54]. Depletion of RPTOR, a component of mTOR1 complex, significantly decreased AKT3-induced axon regeneration [54]. AKT1 also interacts with GSK3β to mediate astroglial autophagy [43]. Interestingly, DEAF1 has been identified as an interactor and in vitro substrate of glycogen synthase kinase-3 (GSK3α and GSK3β) which also interact with the PI3K/mTOR pathway [55]. Yeast two-hybrid assays demonstrated that DEAF1 and LDB1 binds with LMO4 through LIM domains [56]. LDB1 interacts with CHD8, which is regulated by CTNNB1, a key downstream gene of GSK3B. These results implied that DEAF1 may act downstream of the mTOR/GSK3β pathway. RAI1 was also predicted to interact with GSK3 β based on yeast two-hybrid assays [57]. Circadian and mTOR signaling pathways were significantly altered in both MBD5 and RAI1 siRNA-mediated knockdown SH-SY5Y cells in microarray analysis, and levels of mTOR mRNA levels were significantly increased in SMS patient-derived lymphoblastoid cells (LCLs) compared to LCLs derived from controls [24]. These results suggest that RAI1 may function in the feedback loop of the mTOR network in neuropsychiatric disorders.
Through network analysis together with the above literature integration, our results show that DEAF1 binds and transcriptionally regulates RAI1, and may function downstream of ZEB1. The common molecular network comprised these genes may help to explain overlapping symptoms in ZEB1-associated neurodevelopment disorder, DAND, SMS, PTLS, as well as other neuropsychiatric disorders.

Limitations
In this study, molecular mechanism for the potential common gene network of SMS, PTLS, DAND and ZEB1-associated neurodevelopment disorder are mostly in vitro evidence.
To understand how the dysregulation of this gene network contributes to the overlapped phenotypes observed in these disorders in vivo, future studies using genetic engineering animals and/or recruiting patients with multiple variants in these genes need to be carried out.
Moreover, of a potential upstream regulator, mTOR regulation of RAI1, DEAF1 or ZEB1 need to be explored in future studies. These studies would help to elucidate the roles of these genes in neurogenesis and defects in general, including autism, ID and rare neuropsychiatric syndromes.

Conclusions
In this study, our results show that DEAF1 regulates RAI1 mRNA expression by binding to the RAI1 DBS and may function downstream of ZEB1. Pathogenic variants in RAI1, DEAF1 or ZEB1 produce overlapped phenotypes in SMS, PTLS, DAND and ZEB1-associated neurodevelopmental disorder, suggest that these genes may function in neuronal intracellular signaling networks associate with multiple neuropsychiatric disorders.
Our findings help to elucidate the genetic mechanisms underlying the overlapping symptoms in these disorders and provide evidence for a common molecular network across neurodevelopmental and psychiatric syndromes.    and a possible weaker, second binding site that is required for high-affinity ZEB 1 binding as predicted by FIMO program are underlined.