Neuronal genes deregulated in Cornelia de Lange Syndrome respond to removal and re-expression of cohesin

Cornelia de Lange Syndrome (CdLS) is a human developmental disorder caused by mutations that compromise the function of cohesin, a major regulator of 3D genome organization. Cognitive impairment is a universal and as yet unexplained feature of CdLS. We characterize the transcriptional profile of cortical neurons from CdLS patients and find deregulation of hundreds of genes enriched for neuronal functions related to synaptic transmission, signalling processes, learning and behaviour. Inducible proteolytic cleavage of cohesin disrupts 3D genome organization and transcriptional control in post-mitotic cortical mouse neurons, demonstrating that cohesin is continuously required for neuronal gene expression. The genes affected by acute depletion of cohesin belong to similar gene ontology classes and show significant numerical overlap with genes deregulated in CdLS. Interestingly, reconstitution of cohesin function largely rescues altered gene expression, including the expression of genes deregulated in CdLS.

T hree-dimensional (3D) genome organization into topologically associated domains (TADs), contact domains and chromatin loops spatially compartmentalises genes and enhancers and facilitates transcriptional control by gene regulatory elements [1][2][3][4] . 3D genome organization is achieved through the activity of architectural proteins, including the cohesin complex. Initially identified as essential for chromosomal integrity during the cell cycle, cohesin is now known to cooperate with the DNA binding protein CTCF in 3D chromatin contacts essential for transcriptional control [3][4][5][6][7][8] . Mechanistically, cohesin increases 3D contact probabilities of sequence elements, including enhancers and promoters, within boundaries marked by CTCF binding sites in convergent orientation 3,5,6 . In addition, a subset of promoters and enhancers are direct targets of CTCF, and genes that contact enhancers via CTCF-based cohesin loops are highly susceptible to deregulation when CTCF or cohesin are perturbed 3,9,10 . Mammalian genes have been classified into those that are controlled mainly by their promoters, and those that primarily depend on distal enhancers for transcriptional regulation 11 . This difference in regulatory 'logic' broadly separates ubiquitously expressed, promoter-centric 'housekeeping' genes from enhancer-controlled tissue-specific genes 12 . While the loss of cohesin affects the transcription of a limited number of genes, enhancer-associated 7,13 and inducible genes 14 , including neuronal activity-dependent genes 15 , are frequently deregulated when 3D organization is perturbed by the loss of cohesin.
Heterozygous or hypomorphic germline mutations of cohesin and associated factors such as the cohesin loading factor NIPBL result in a human developmental disorder known as Cornelia de Lange Syndrome (CdLS) [16][17][18] . All CdLS patients show a degree of intellectual disability, and 60-65% have autism spectrum disorder (ASD), often in the absence of structural brain abnormalities or neurodegeneration 17,19,20 . Experimental perturbations of neuronal cohesin or NIPBL during mouse development have shown changes in animal behaviour as well as abnormal neuronal morphology 21 , migration 22 and gene expression [21][22][23][24] .
NIPBL has historically been considered as separate from the core cohesin complex, and with distinct 22,[25][26][27] and sometimes even antagonistic activities 28 . However, current structural 29 and functional 30 evidence indicate that NIPBL is integral to the cohesin complex in its loading state 29 and contributes to ATPase activity 30 during loop extrusion, the process thought to form chromatin loops, TADs and contact domains 5 . Accordingly, the depletion of NIPBL and cohesin have similar or identical effects on 3D genome organization 7,8 . Consistent with the function of NIPBL as a loading factor for cohesin, CdLS patient-derived NIPBL +/− lymphoblastoid cells 31 , Nipbl +/− mouse embryonic fibroblasts 32 and fetal liver cells 33 show reduced global or local cohesin binding [31][32][33] and defective 3D chromatin contacts 32,33 . Deregulated genes show reduced cohesin binding in Nipbl +/− embryonic mouse brain 34 . Nevertheless, it remains unclear to what extent gene expression changes in cells with Nipbl mutations relate to cohesin or potential additional functions of NIPBL (ref. 22,[25][26][27]. Although aberrant gene expression is a likely cause of neuronal dysfunction in CdLS, studies of gene expression in CdLS patient cells have to date been limited to induced pluripotent stem cells (iPSCs) 35 , cardiac 35 and lymphoblastoid cells 31 . We, therefore, do not know whether CdLS neurons show aberrant gene expression, and how gene expression in CdLS relates to perturbations of neuronal gene expression in response to cohesin deficiency. In this study, we examine gene expression in neuronal nuclei isolated from post-mortem cerebral cortex of CdLS patients. We discover prominent downregulation of hundreds of genes enriched for important neuronal functions, including synaptic transmission, signalling processes, learning and behaviour, and neuroprotective sphingolipid metabolism. Deregulated genes show significant overlap with ASD. These findings support the idea that altered gene expression may contribute to neuronal dysfunction in CdLS. We establish experimental models that allow for the inducible degradation of the cohesin subunit RAD21 in post-mitotic primary cortical mouse neurons. These reveal that cohesin is continuously required to sustain neuronal gene expression. A significant number of cohesin-dependent genes are also deregulated in CdLS. Importantly, the expression of these genes is rescued by reconstitution of functional cohesin, indicating that at least some of these changes may be reversible.

Results
Transcriptomic characterization of CdLS patient neurons. We sourced frozen post-mortem prefrontal cortex from four CdLS patients aged 19-48 years and six age-matched controls (Supplementary Fig. 1a). Three of the patients had heterozygous mutations in NIPBL, the gene most frequently mutated in CdLS (ref. 17,36 , Fig. 1a, Supplementary Fig. 1a, b). No mutations in NIPBL or other CdLS genes such as RAD21, SMC1A or HDAC8 were found in the fourth patient, consistent with the lack of identifiable mutations in~30% of CdLS patients 36 .
To characterise gene expression in CdLS patient neurons we isolated nuclei from the prefrontal cortex, stained with the neuronal marker NeuN (ref. 37 ), and sorted NeuN positive and NeuN negative nuclei by flow cytometry (Fig. 1b, Supplementary  Fig. 1c). We generated ATAC-seq, CAGE and RNA-seq data from neuronal nuclei. ATAC-seq and CAGE signals at enhancers did not show CdLS-specific features due to extensive inter-individual variation. However, CAGE analysis of promoters showed separation of CdLS patient and control samples along the major principle component PC1, and revealed significant deregulation of 766 gene promoters in CdLS neurons (adj P < 0.05; 358 upregulated, 408 downregulated; Supplementary Fig. 2a-c, Supplementary Data 1). Gene promoters that were downregulated in CdLS neurons were enriched for the gene ontology (GO) terms synaptic signalling and organization, ion and neurotransmitter transport, axon development, behavior and cognition, while upregulated promoters showed no striking functional enrichment ( Supplementary Fig. 2d, Supplementary Data 2). RNA-seq identified 617 differentially expressed genes (adj P < 0.05) in CdLS neurons, 310 genes were up-and 307 were downregulated (Fig. 1c, Supplementary Fig. 3a). In close agreement with CAGE, genes that were downregulated by RNA-seq were related to synaptic transmission, signalling processes, learning and behaviour, and sphingolipid metabolism, while upregulated genes showed no striking functional enrichment (Fig. 1d, Supplementary Data 4).
RNA-seq and CAGE of CdLS neurons showed highly significant overlap (Odds ratio = 11.86, P < 2.2e-16). The direction of gene deregulation by RNA-seq and CAGE was highly correlated (R = 0.68, P = < 2.2e-16 for genes found deregulated by RNA-seq or CAGE, and R = 0.95, P = < 2.2e-16 for genes found deregulated by both RNA-seq and CAGE; Supplementary Fig. 2e).
A majority of CdLS patients have ASD in addition to intellectual disability 17,20 . Clinical records relating to the brain samples examined here indicate that all four patients had cognitive impairment and three had confirmed ASD features (CDL380P, CDL744P, CDL764P, there was no information on ASD status for 2082). We, therefore, asked whether genes implicated in ASD were deregulated in CdLS. Genes found deregulated in the prefrontal cortex of patients with idiopathic ASD (ref. 38 ) showed strong overlap with genes deregulated in CdLS (Odds ratio = 6.25, P < 2.2e-16). The direction of gene expression changes in CdLS and idiopathic ASD was highly correlated (R = 0.56, P < 2.2e-16 for genes deregulated in idiopathic ASD, R = 0.85, P < 2.2e-16 for genes deregulated in both idiopathic ASD and CdLS; Fig. 1e, f). Strong overlap of deregulated genes and correlated changes in the direction of gene expression were also seen between CdLS and ASD caused by the duplication of chromosome 15q.11.2-13.1 (ref. 35 , R = 0.85, P < 2.2e-16; Fig. 1g). These data reveal similarities between the transcriptional programs of neurons in CdLS and in idiopathic and syndromic ASD.
In summary, these data demonstrate that neurons from CdLS patients show deregulated expression of genes that have important neuronal functions, consistent with neuronal dysfunction in CdLS. They raise the question of whether deregulated gene expression in CdLS is related to reduced cohesin function and-if so-whether the deregulation of neuronal genes is a direct consequence of reduced cohesin function in post-mitotic neurons or secondary to a role for cohesin in neural development.
Temporal control over cohesin levels in neurons. Addressing the question of whether cohesin is directly required for neuronal gene expression requires experimental systems that enable acute cohesin withdrawal from post-mitotic neurons, ideally in a reversible manner. To control the expression of RAD21, a cohesin subunit associated with CdLS (ref. 16,17 ), we utilized Rad21 Tev , an allele that encodes RAD21 protein cleavable by tobacco etch virus (TEV) protease 14,40 . Rad21 Tev/Tev neurons express endogenous RAD21-TEV as their sole source of RAD21. We established explant cultures from Rad21 Tev/Tev cortices at e14.5 under conditions that enrich for post-mitotic neurons and promote neuronal maturation (Fig. 2a, Supplementary Fig. 5a, b).
We used lentivirus to transduce Rad21 Tev/Tev neurons with TEV protease fused to a tamoxifen-responsive estrogen receptor hormone binding domain (ERt2-TEV). Constructs were tagged by V5 and t2a-fused GFP (ERt2-TEV) for identification of ERt2-TEV-expressing cells (Fig. 2b, Supplementary Fig. 5c). ERt2-TEV remained cytoplasmic until nuclear translocation of ERt2-TEV was triggered, which occurred within minutes of 4-hydroxy tamoxifen (4-OHT) addition (Fig. 2b, Supplementary Fig. 5c). We monitored the cleavage of RAD21-TEV by western blotting and found that RAD21-TEV expression was reduced to~15% of pretreatment levels within 8 h of 4-OHT addition, and remained at this level for at least 24 h (Fig. 2c). In a second approach, we transduced Rad21 Tev/Tev neurons with a lentiviral construct that constitutively expresses Tet-On advanced transactivator (rtTA) and RFP. Addition of doxycycline leads to the expression of TEV protease with an exogeneous nuclear localization sequence (NLS-TEV, Supplementary Fig. 6a, b). Induction of NLS-TEV depleted~7 0% of RAD21-TEV within 24 h of doxycycline addition ( Supplementary Fig. 6c). These systems establish temporal control over cohesin levels in primary post-mitotic neurons.
Cohesin is continuously required for neuronal gene expression. We utilized temporal control of RAD21-TEV cleavage to quantify the impact of acute cohesin depletion on gene expression in postmitotic neurons. RNA-seq 24 h after 4-OHT-induced nuclear translocation of ERt2-TEV identified 750 deregulated (adj P < 0.05) genes, of which 463 were down-and 287 were upregulated (Fig. 2d, Supplementary Data 8). GO term analysis showed that downregulated genes were enriched for biological processes similar to those observed for downregulated genes in CdLS, in particular synapse and signalling, cell adhesion, neuron development and ion transport. As observed in CdLS neurons, no striking enrichment was observed for upregulated genes (Fig. 2e, Supplementary Data 9). Genes deregulated by ERt2-TEVmediated RAD21-TEV cleavage showed significant overlap with SFARI ASD risk genes 41 (Odds ratio = 2.19, P = 1.16e-08).
Reduced mRNA expression translated into reduced levels of protein in RAD21-TEV neurons, as illustrated for NLGN1, encoded by Nlgn1 ( Supplementary Fig. 8, Supplementary Data 12). Hence, acute depletion of RAD21-TEV in postmitotic neurons established that cohesin is continuously required to sustain the expression of genes that mediate important neuronal functions.
The Pcdhb cluster is a classic example of CTCF-based cohesinmediated enhancer-promoter connections 3,39 , and 15 of 17 expressed Pcdhb genes were deregulated by inducible RAD21-TEV cleavage. Circular chromosomal conformation capture and sequencing (4C-seq) using the Pcdhb enhancers HS18-20 as the viewpoint showed that RAD21-TEV cleavage disrupted 3D contacts, and decreased interactions between Pcdhb promoters and enhancers (Fig. 2h). Genes that were downregulated in response to acute cohesin depletion in primary neurons were significantly enriched for the binding of RAD21 (Fig. 2i, Odds Ratio = 2.66, P = 2.32e-13) and CTCF (Fig. 2i, Odds Ratio = 2.38, P = 3.39e-11). Our finding that cohesin deficiency predominantly affects genes related to specialized neuronal functions is consistent with models that distinguish ubiquitously expressed, promotercentric 'housekeeping' genes from enhancer-controlled tissuespecific genes 11,12 , and with previous observations that enhancerassociated genes are preferentially deregulated by depletion of cohesin 7,13 . Accordingly, genes that were downregulated in response to acute cohesin depletion in primary neurons were significantly enriched for proximity to enhancers (Fig. 2i, Odds ratio = 3.78, P < 2.2e-16). These data suggest a mechanism by which RAD21-TEV cleavage disrupts cohesin-mediated chromatin contacts, including enhancer-promoter interactions. Upregulated genes showed no GO term enrichment at P < 1 × 10 −4 . f Enrichment of shared deregulated genes (adj P < 0.05) in response to acute cohesin depletion induced by ERt2-TEV and NLS-TEV. One-sided Fisher's exact test was applied for the odds ratio and P value. All comparisons P < 2.22e-16. g Scatter plot of gene expression, comparing log 2 fold-change of deregulated genes (adj P < 0.05) in response to acute cohesin depletion induced by ERt2-TEV and NLS-TEV (DE differentially expressed, R = Pearson correlation coefficient; P < 2.2e-16, two-sided F test).  (Fig. 3b, Supplementary Table 1).
There was slight, but statistically significant enrichment for binding of RAD21 and CTCF to genes that were downregulated in CdLS neurons (Fig. 3c). This enrichment increased considerably for genes that were deregulated both in CdLS and by the inducible cleavage of RAD21-TEV (compare Odds ratios in Fig. 3c). Genes that were deregulated both in CdLS and in acutely cohesin-depleted Rad21 Tev/Tev neurons showed significant enrichment for proximity to human neuronal enhancers 47 (Fig. 3c).
with Rad21 Tev/Tev neurons (Odds ratio = 2.06, P = 4.64e-10; Supplementary Fig. 9). In summary, a subset of genes deregulated in CdLS neurons bore hallmarks of cohesin target genes, and these genes were enriched in the overlap between CdLS and cohesin-dependent genes, as identified by acute depletion of cohesin in Rad21 Tev/Tev neurons.
Rescue of cohesin-dependent gene expression. Postnatal intervention can alleviate transcriptional changes and abnormal behaviour in mouse models of Rett Syndrome 51,52 . In light of this, we asked whether gene expression changes induced by acute cohesin depletion in Rad21 Tev/Tev neurons can be reversed when cohesin levels are restored. Doxycycline-induced NLS-TEV expression was reversible, and transient RAD21-TEV cleavage was followed by a return of RAD21-TEV protein to control levels after removal of doxycycline (Fig. 4a, b). Remarkably, comparison of RNA-seq during RAD21-TEV depletion (24 h) and after RAD21-TEV rescue (day 7) showed that the vast majority of genes that were found deregulated after 24 h (695 of 720 or 96.5%) were fully rescued upon restoration of cohesin levels (Fig. 4c, e; Supplementary Data 13). A small number of residual genes were refractory to rescue and remained deregulated. These included neuronal signalling (Gfra1, Spon1, Scg2, Cntnap3), transcription (Mycl1, Cited2, Maml3, Lzts1) and splicing factors (Rbfox1). To ask whether the observed rescue of gene expression reflected adaptation of neurons to reduced cohesin function we carried out RNA-seq experiments after prolonged depletion of RAD21-TEV. These experiments confirmed that rescue of gene expression required the restoration of RAD21 expression and was not due to the adaptation of neurons to reduced cohesin expression ( Supplementary Fig 10; Supplementary Data 14).
Interestingly, all 57 genes that were deregulated both in CdLS and in response to NLS-TEV-mediated cohesin depletion were rescued by restoring cohesin expression. These included 14 genes implicated in ASD and shared between CdLS and NLS-TEV (Supplementary Table 1). Rescue was near-complete, regardless of the direction and the degree of the initial deregulation (Fig. 4d).
We identified 62 genes that were initially unaffected by acute cohesin depletion at 24 h, but were deregulated 7 days later, after cohesin levels had been restored to control levels (Fig. 4e). These de novo deregulated genes were associated with the development and cell communication and were highly enriched for genes that change expression during the maturation of control neurons in explant culture (day 10-17, adj P < 0.05). Of 62 de novo deregulated genes, 46 were maturation genes (Odds ratio = 6.69, P = 1.14e-12). Strikingly, the direction of deregulation of these genes was highly correlated with their regulation during neuronal maturation: genes that were upregulated during neuronal maturation in explant culture were also preferentially upregulated as a consequence of transient cohesin depletion (P = 0.0134, FDR = 0.0134), while genes that were downregulated during neuronal maturation in explant culture were also preferentially downregulated as a consequence of transient cohesin depletion (P = 1.51e-04, FDR = 3.02e-04) (Fig. 4f, g). Hence, while the majority of gene expression changes caused by loss of cohesin can be rescued by restoring cohesin levels, including the expression of genes deregulated in CdLS, the de novo deregulation of neuronal maturation genes illustrates that even transient cohesin depletion may have potentially damaging long-term secondary effects on neuronal gene expression.

Discussion
We characterized gene expression in primary cortical neurons from CdLS patients and found deregulation of hundreds of genes enriched for important neuronal functions, including synaptic transmission, signalling processes, learning and behaviour and sphingolipid metabolism. These findings provide experimental support for suggestions that CdLS pathologies, including intellectual disability, may be mediated at least in part by deregulated gene expression. The transcriptomic profile of CdLS showed extensive similarities to ASD, both in terms of shared deregulated genes and in the direction of deregulation. This similarity is of particular interest given the high prevalence of ASD in CdLS patients.
To explore the role of cohesin in neuronal gene expression we established in vitro models for the acute depletion of cohesin in primary cortical mouse neurons. Proteolytic cleavage of the essential cohesin subunit RAD21 disrupted 3D organization and perturbed the expression of neuronal genes. These experiments establish that cohesin is continuously required to sustain neuronal gene expression.
We found significant concordance between cohesin-dependent genes in mouse neurons and the neuronal transcriptome in human CdLS. Genes deregulated both in cohesin-depleted mouse neurons and human CdLS neurons were enriched for cohesin binding and enhancer proximity. Tissue-specific genes primarily depend on distal enhancers for transcriptional regulation 11,12 , and enhancer-associated genes are preferentially deregulated when 3D organization is perturbed by the loss of cohesin 7,13 . Our finding that cohesin deficiency predominantly affects enhancerassociated genes related to specialized neuronal functions is consistent with these models. Taken together with previous data demonstrating the impact of Nipbl heterozygosity on cohesin binding and 3D chromatin contacts 31-34 , our analysis supports the idea that deregulation both in CdLS in vivo and in in vitro models for acute cohesin cleavage identifies a core set of genes that are enriched for binding of CTCF and cohesin, and that show enhancer-dependent tissue-specific expression. These genes are susceptible to deregulation because they rely on 3D contacts with distal regulatory elements, and their 3D contacts are directly dependent on cohesin. These genes related to signaling, neurotransmitter and synapse components, ion channels, and transcription factors. Interestingly, the great majority of gene expression changes precipitated by acute cohesin depletion were reversible upon restoration of cohesin function, including all 57 genes deregulated in both CdLS and NLS-TEV. Hence, cohesin rescues neuronal genes that are deregulated in CdLS.
In addition to directly cohesin-dependent genes, CdLS neurons also showed deregulation of genes that lacked hallmarks of direct cohesin targets and were not sensitive to acute cohesin depletion in mouse neurons. This may be due to cohesin-independent functions of NIPBL (ref. 22,[25][26][27] or to secondary effects of reduced cohesin function on the expression of other genes. Our data point to two mechanisms that may contribute to the indirect deregulation of genes in cohesin-deficient neurons. First, numerous genes related to neuronal maturation were deregulated late in response to acute cohesin depletion and remained deregulated after cohesin levels were restored. These genes showed a striking pattern of overshoot regulation, where genes upregulated during the maturation of control neurons were even more highly expressed after transient cohesin depletion and vice versa. Explant cultures contain >90% neurons, and the deregulation of these genes may therefore be driven by factors within or interactions between neurons. Second, analysis of NeuN negative nuclei from CdLS cortices showed highly significant upregulation of inflammatory genes, consistent with potentially damaging inflammatory responses by glial cells 53 . These data identify an inflammatory component to disrupted gene expression in CdLS, which may contribute to neuronal dysfunction via glia-neuron interactions. Interestingly, one of the gene ontologies Our data raise the possibility that aspects of the CdLS phenotype may be reversible by postnatal intervention. Precedent for such rescue comes from pioneering studies in mouse models of Rett syndrome, where postnatal correction of Mecp2 deficiency ameliorates deregulated gene expression and behavioural defects 51,52 . As the Rett protein MECP2, cohesin exerts its effects on transcription by interfacing with chromatin marking systems 55 , and the dynamics of cohesin on chromatin is controlled in part by acetylation and deacetylation of its SMC subunits 18 . The acetylation/deacetylation cycle is a prime target for pharmacological intervention, as are bromodomain proteins such as BRD4, which connect acetylation to transcription 56 . Mutations in the histone deacetylase HDAC8 and the bromodomain protein BRD4 can cause CdLS-like disease 18,57 . Epigenetic drugs 56,58 may therefore Inducible cleavage of RAD21-TEV. For cleavage of RAD21-TEV, neurons were plated as described above and transduced at day 3 with lentivirus containing either ERt2-TEV or NLS-TEV at a multiplicity of infection of 1. For ERt2-TEV dependent RAD21-TEV degradation, neurons were treated on culture day 10 with 500 nM 4-hydroxytamoxifen (4-OHT) or vehicle (ethanol). For 24 h depletion and rescue of RAD21-TEV, NLS-TEV dependent RAD21-TEV degradation, neurons were treated on culture day 10 with 6-hour doxycycline (100 ng/ml) pulse or vehicle (water). Cells were then rinsed, and media replaced with conditioned neuronal media. For 7-day long-term NLS-TEV dependent RAD21-TEV degradation, neurons were treated on culture day 10 with 24-hour doxycycline (1 μg/ml) pulse or vehicle (water). Cells were then rinsed, and media replaced with conditioned neuronal media.
Isolation of nuclei from post-mortem tissue. Anonymized post-mortem human tissue was obtained from approved tissue banks and used in accordance with the Human Tissue Act (UK) and with approval by the Imperial College London Research Ethics Committee. Isolation of nuclei was performed as previously described 62  Immunocytochemistry. Neurons plated on coverslips were fixed with PBS containing 4% paraformaldehyde and 4% sucrose warmed to 37 o C for 10 min at room temperature. Neurons were then permeabilized using 0.3% Triton X-100 for 10 min at room temperature, and blocked in blocking solution (10% normal goat serum, 0.1% Triton X-100 in PBS) for 1 h at room temperature. Samples were incubated with primary antibodies diluted in staining solution (0.1% Triton X-100, 2% normal goat serum in PBS) in a humidified chamber overnight at 4 o C. Primary antibodies were mouse monoclonal TUJ1 (1:500, Biolegend, 801213), mouse monoclonal V5 (1:250, Sigma-Aldrich, V8012) and mouse monoclonal NeuN (1:1000, Abcam, ab104224). Samples were then washed with PBS and incubated in secondary antibodies diluted 1:500 in staining solution (0.1% Triton X-100, 2% normal goat serum in PBS) in a humidified chamber for 1 h at room temperature. Secondary antibodies were, goat anti-mouse IgG (H + L) Alexa Fluor 488 (Ther-moFisher, A-1101) and goat anti-mouse IgG (H + L) Alexa Fluor 568 (Thermo-Fisher, A-11004). Cells were mounted in Vectashield medium containing DAPI (Vector Labs). Samples were visualized using a TCS SP5 Leica laser scanning confocal microscope using LAS X v2.7. Images were processed using Leica Confocal Software and FIJI. For quantification of NeuN-positive cells, DAPI-identified nuclei that colocalized with NeuN signal were counted as neuronal, and DAPIidentified nuclei without NeuN were counted as non-neuronal. Samples were quantified using a processing pipeline developed in CellProfiler (version 2.2, www. cellprofiler.org).
Identification of human mutations. NIPBL-mutation screening was performed by PCR of genomic DNA and Sanger Sequencing. Genomic DNA was isolated from brain cortical tissue using DNeasy blood & tissue extraction kit (Qiagen), according to the manufacturer's protocol. The entire NIPBL coding region (exons 2-47) was screened for mutations. Primer pairs were designed using ExonPrimer to amplify c Volcano plots of gene expression log 2 fold-change versus adjusted P value in RAD21-TEV + Dox treated neurons before and after rescue of RAD21-TEV expression (n = 3). Left: 570 genes were down-and 150 genes were upregulated following RAD21-TEV depletion (adj P < 0.05, Wald Test, Benjamini-Hochberg adjusted). Right: 43 genes were down-and 44 genes were upregulated after rescue of RAD21-TEV expression (adj P < 0.05, shown in red and orange, Wald Test, Benjamini-Hochberg adjusted). Genes in red are significantly deregulated in both RAD21-TEV depletion and rescue, genes in orange are significantly deregulated only after rescue (DE differentially expressed). d Log 2 fold-change of significantly deregulated genes (adj P < 0.05) following RAD21-TEV depletion (left) and after rescue (right). e The number of significantly deregulated genes (adj P < 0.05) after RAD21-TEV depletion (in green) and after rescue of RAD21-TEV expression (in orange). DE differentially expressed. f Heatmap of de novo deregulated genes following RAD21-TEV rescue (adj P < 0.05) and their maturation trajectory in control (left) and +Dox treated samples (right). g Barcode plot of de novo deregulated genes following RAD21-TEV rescue (adj P < 0.05) and their enrichment for directionality in maturation. Top: significantly de novo upregulated genes following rescue are enriched for genes upregulated during neuronal maturation. Bottom: de novo downregulated genes after cohesin rescue are enriched for genes that are downregulated during neuronal maturation. ATAC-seq. ATAC-seq was performed on 5 × 10 5 nuclei isolated from control (n = 5) and CdLS patients (n = 4) per sample, following the omni-ATAC protocol 66 . Omni-ATAC is an improvement of the original ATAC protocol, with the inclusion of several additional detergents, NP-40, Tween-20 and digitonin in the nuclei digestion. After 30 min of incubation, samples were purified using the Zymo ChIP DNA Clean and Concentrator (Zymo Research) and libraries prepared as previously described 67 .
4C-seq. 4 C was performed as previously described 39 with modifications. Briefly, neuronal cells were cross-linked in PBS with 1% formaldehyde for 10 min at RT and nuclei were isolated in lysis buffer (10 mM Tris-HCl pH 7.4, 150 mM NaCl, 0.5% NP-40, 1 x protease inhibitors). After digestion using HindIII, digested products were ligated by T4 DNA ligase. 3 C templates were purified and then digested again using a second enzyme, NlaIII. Final 4 C sequencing libraries were purified using a Roche High-Pure PCR Product Purification Kit, and then sequenced using an Illumina HiSeq 2500 platform. Sequence data was analysed using the 4 Cseqpipe software suite 68 75,76 was performed with GseaPreranked tool using Hallmarks gene set. For RAD21-TEV RNAseq data, polyadenlylated genes were annotated using PolyA_DB (ref. 77 ) and non-polyadenlylated genes were removed from the analysis. Glial genes were defined by control NeuN-negative versus control NeuN-positive RNAseq data with adj P < 0.05 and fold-change >20. We processed CdLS NeuN-positive RNAseq data after removing glial genes. In both human CdLS NeuN-positive and NeuNnegative RNAseq dataset, we only focused on autosomes to mitigate sex specific differences. Unwanted batch effects were controlled in CdLS NeuN-positive and NeuN-negative RNAseq by using R Bioconductor package RUVseq (ref. 78 ) with RUVg function and k = 2 for both. For CdLS NeuN-negative RNAseq, sample CDL-744P was excluded. Heatmaps were generated using R Bioconductor package ComplexHeatmap (ref. 79 ) and GraphPad Prism 8.0.
SLIC-CAGE analysis. Sequenced libraries were demultiplexed using CASAVA allowing zero mismatches for barcode identification. Demultiplexed CAGE tags (47 bp) were mapped to a reference GRCh37/hg19 human genome using Bowtie2 (--phred33-quals -U) 80 with default parameters allowing zero mismatches per 22 nucleotide seed sequence. The mapped reads were then sorted using Samtools v 1.10 (ref. 81 ) and uniquely mapped reads kept for downstream analysis in R graphical and statistical computing environment (http://www.Rproject.org/). The mapped and sorted unique reads were imported into R as bam files using the standard workflow within the CAGEr package v1.20 (ref. 82 ). All 5'ends of reads are CAGE-supported transcription start sites (CTSSs) and the number of each CTSS (number of tags) reflects the transcript expression levels. Raw tags were normalized using a referent power-law distribution (alpha = 1.48) and expressed as normalized tags per million (TPMs) 83 . The highest expressed CTSS is termed the 'dominant CTSS'. The correlation of samples on CTSS level was estimated based on TPM values. This analysis led to the removal of sample CDL-744P due to its low correlation with the rest of the samples. CTSSs with sufficient support (at least 0.1 TPM in at least one sample) were further used to construct promoter regions. First, sample-specific clustering of CTSSs was performed to define tag clusters, allowing a maximum distance of 20 bp between any 2 CTSSs in the same tag cluster. In order to allow for between-sample expression profiling, tag clusters with sufficient support (at least 5 TPM) within 100 bp of each other were aggregated across all samples to define promoter regions (consensus clusters).
To allow comparison with RNA-seq data, genomic coordinates of consensus clusters were converted into GRCh38/hg38 coordinates using the liftOver tool 84 . Consensus clusters were annotated using ChIPseeker v.1.22.1 (ref. 85 ) with the transcript annotation TxDb.Hsapiens.UCSC.hg38.knownGene to define TSSs of known genes and the org.Hs.eg.db annotation to allocate Ensembl IDs to gene symbols. A consensus cluster was assigned to the promoter of a known gene if it was within a region 500 bp upstream and 100 bp downstream of the annotated TSS or within its 5' UTR. This resulted in a set of 18,867 promoter-associated consensus clusters. To reduce the effect of sex differences, we excluded consensus clusters located on chromosomes X and Y. Consensus clusters were further filtered to exclude those associated with glialspecific genes, resulting in a final set of 18,103 consensus clusters and corresponding Ensembl IDs. DESeq2 v.1.26.0 (ref. 72 ) was used to define differentially expressed genes, accounting for treatment and age effects. Genes with BH-adjusted P < 0.05 and log 2 fold-change values of > 0 or < 0 were considered as significantly up-or down-regulated. Principal component analysis was implemented in DESeq2 on rlog-transformed read counts. PCA plots were created using the top 10,000 consensus clusters with the highest row variance. Variance associated with age group was removed from the rlog-transformed data with limma v.3.42.0 (ref. 86 ) for visualization and plotted as described above. Functional enrichment analysis of significantly up-or downregulated genes was performed using GO over-representation test with clusterProfiler v.3.14.3 (ref. 87 ). Enriched GO terms associated with biological processes were determined based on a BH-adjusted P value cutoff = 0.05 and a q value cutoff = 0.05 against a background of genes expressed in control samples (mean CAGE signal > 2 TPM, n = 11,634). CAGE data were visualized using the Integrative Genomics Viewer (Broad Institute) 88 .
Cohesin binding and enhancer proximity. The binding of genes by cohesin and CTCF was defined by RAD21 (ref. 21 ) (ENCODE ENCSR198ZYJ) or CTCF ChIPseq (ENCODE ENCSR677HXC, ENCODE ENCSR822CEA) peaks overlapping with gene bodies. The proximity of genes to enhancers was defined by whether genes were the nearest neighbour to a neuronal enhancer 47,93 . R Bioconductor package GeneOverlap 94 was applied for one-tailed Fisher's exact test and reported the odds ratio and P value.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data that support this study are available from the corresponding author upon reasonable request.