Cohesin-dependence of neuronal gene expression relates to chromatin loop length

Cohesin and CTCF are major drivers of 3D genome organization, but their role in neurons is still emerging. Here, we show a prominent role for cohesin in the expression of genes that facilitate neuronal maturation and homeostasis. Unexpectedly, we observed two major classes of activity-regulated genes with distinct reliance on cohesin in mouse primary cortical neurons. Immediate early genes (IEGs) remained fully inducible by KCl and BDNF, and short-range enhancer-promoter contacts at the IEGs Fos formed robustly in the absence of cohesin. In contrast, cohesin was required for full expression of a subset of secondary response genes characterized by long-range chromatin contacts. Cohesin-dependence of constitutive neuronal genes with key functions in synaptic transmission and neurotransmitter signaling also scaled with chromatin loop length. Our data demonstrate that key genes required for the maturation and activation of primary cortical neurons depend on cohesin for their full expression, and that the degree to which these genes rely on cohesin scales with the genomic distance traversed by their chromatin contacts.


Conditional deletion of cohesin in immature post-mitotic neurons
To explore the role of cohesin in post-mitotic neurons, we deleted the essential cohesin subunit RAD21 (Rad21 lox , Seitan et al., 2011) in immature cortical and hippocampal neurons using Nex Cre (Goebbels et al., 2006;Hirayama et al., 2012). Explant cultures of wild type and Rad21 lox/lox Nex Cre E17.5/18.5 cortex contained >95% MAP2 + neurons, with <1% GFAP + astrocytes or IBA1 + microglia ( Figure 1a). Immunofluorescence staining showed the loss of RAD21 protein specifically in GAD67neurons (Figure 1b and c). This was expected, as Nex Cre is expressed by excitatory but not by inhibitory neurons (Goebbels et al., 2006;Hirayama et al., 2012). Consistent with the presence of ~80% of GAD67and ~20% GAD67 + neurons in the explant cultures, Rad21 mRNA expression was reduced by 75-80% overall (Figure 1d, left). There was a corresponding reduction in RAD21 protein (Figure 1d,middle). To focus our analysis on cohesin-deficient neurons we combined Nex Cre -mediated deletion of Rad21 with Nex Cre -dependent expression of an epitopetagged ribosomal subunit (Rpl22-HA RiboTag; Sanz et al., 2009). We verified that Nex Cre -induced RPL22-HA expression was restricted to RAD21-depleted neurons (Figure 1-figure supplement  1a) and performed high throughput sequencing of Rpl22-HA RiboTag-associated mRNA (RiboTag RNA-seq, Supplementary file 1). Comparison with total RNA-seq (Supplementary file 2) showed that Nex Cre RiboTag RNA-seq captured excitatory neuron-specific transcripts, such as Slc17a7 and Camk2a. Transcripts selectively expressed in astrocytes (Gfap,Aqp4,Mlc1), microglia (Aif1), and inhibitory neurons (Gad1, Gad2, Slc32a1) were depleted from Nex Cre RiboTag RNA-seq (Figure 1figure supplement 1b). RiboTag RNA-seq enabled an accurate estimate of residual Rad21 mRNA, which was <5% in Rad21 lox/lox Nex Cre cortical neurons (Figure 1d, right). These data show nearcomplete loss of Rad21 mRNA and undetectable levels of RAD21 protein in Nex Cre -expressing Rad21 lox/lox neurons. Analysis of chromatin conformation by Chromosome-Conformation-Capture-Carbon-Copy (5 C) showed a substantial reduction in the strength of CTCF-based chromatin loops after genetic (Rad21 lox/lox Nex Cre , Figure 1e) and after proteolytic cohesin depletion (RAD21-TEV, Figure 1-figure supplement 2; Weiss et al., 2021). Of note, restoration of RAD21 expression rescued CTCF-based chromatin loop formation (Figure 1-figure supplement 2). These data show that cohesin is directly linked to the strength of CTCF-based chromatin loops in primary in post-mitotic neurons. .5-E18.5 cortices were dissociated and plated on poly-D-lysine. After 10 days, cultures were stained for pan neuronal (MAP2), astrocyte (GFAP), and microglia (IBA1) markers, and cell type composition was determined by quantitative analysis of immunofluorescence images. Based on 6 Rad21 +/+ Nex Cre and 8 Rad21 lox/lox Nex Cre different samples analyzed in four independent experiments. (b) Immunofluorescence staining of Rad21 +/+ Nex Cre and Rad21 lox/lox Nex Cre neuronal explant cultures for RAD21 and MAP2 (left) and distribution of RAD21 expression by MAP + neurons (right). Note the discontinuous distribution of RAD21 expression in Rad21 lox/lox Nex Cre neurons. Three independent experiments per genotype. DAPI marks nuclei. Scale bar = 60 μm. (c) Immunofluorescence staining for RAD21, MAP2, and the marker of GABAergic inhibitory neurons, GAD67 (left). Distribution of RAD21 expression in GAD67 + and GAD67neurons (right). Note that the discontinuous distribution of RAD21 expression in Rad21 lox/lox Nex Cre neuronal explant cultures is due to GAD67 + GABAergic inhibitory neurons. Three independent experiments for Rad21 +/+ Nex Cre and six independent experiments for Rad21 lox/lox Nex Cre . DAPI marks nuclei. Scale bar = 20 μm. (d) Quantitative RT-PCR analysis of Rad21 mRNA expression in Rad21 +/+ Nex Cre and Rad21 lox/lox Nex Cre cortical explant cultures (mean ± SEM, n=18). Hprt and Ubc were used for normalization (left). RAD21 protein expression in Rad21 +/+ Nex Cre and Rad21 lox/lox Nex Cre cortical explant cultures was quantified by fluorescent immunoblots (mean ± SEM, n=6, a representative blot is shown in Figure 1-figure supplement 1) and normalized to LaminB (center). Nex Cre RiboTag RNA-seq of analysis of Rad21 mRNA expression in Rad21 +/+ Nex Cre and Rad21 lox/lox Nex Cre cortical explant cultures (right, three independent biological replicates). (e) 5C heat maps of Rad21 +/+ Nex Cre and Rad21 lox/lox Nex Cre cortical explant cultures. Shown is a 1.72 Mb region covered by 5C analysis of chr2 107601077-110913077 . One of two independent biological replicates with similar results. CTCF ChIP-seq in cortical neurons (Bonev et al., 2017) and mm9 coordinates are shown for reference. Arrowheads mark the position of CTCF-based loops. Results were consistent across two replicates and three chromosomal regions. Histograms below show the quantification of representative CTCF-based loops (arrowheads) in two independent biological replicates for control and Rad21 lox/lox Nex Cre neurons.
The online version of this article includes the following source data and figure supplement(s) for figure 1: Source data 1.   (1028)  The online version of this article includes the following source data and figure supplement(s) for figure 2: Source data 1. Figure 2: Loss of cohesin from immature post-mitotic neurons perturbs neuronal gene expression.    activity-regulated by definition, and hence lowly expressed in basal conditions due to spontaneous synaptic activity. In Rad21 lox/lox Nex Cre neurons, previously defined inducible ARGs (Kim et al., 2010) were more frequently downregulated than constitutively expressed genes (Figure 2c, p=1.03e-16, odds ratio = 3.27). These data show that immature post-mitotic neurons require cohesin to establish and/or maintain the correct level of expression of genes that support neuronal maturation, including the growth and guidance of axons, the development of dendrites and spines, and the assembly, function, and plasticity of synapses.
A role for cohesin in the maturation of post-mitotic neurons We next set out to examine the impact of cohesin deletion on the maturation of immature postmitotic neurons. Rad21 lox/lox Nex Cre embryos were found at the expected Mendelian ratios throughout gestation, however postnatal lethality was evident (Figure 3-figure supplement 1a). Rad21 lox/lox Nex Cre cortical neurons did not show increased proliferation ( were comparable between Rad21 +/+ and Rad21 lox/lox Nex Cre neocortex. The neuronal transcription factors TBR1, CTIP2, and CUX1 were expressed beyond the boundaries of their expected layers in Rad21 lox/lox Nex Cre cortices, and deeper cortical layers appeared disorganized ( Figure 3a). While the total numbers of TBR1 + and CTIP2 + neurons were comparable in wild type and Rad21 lox/lox cortices, TBR1 + and CTIP2 + neurons were reduced in the subplate and increased in layers 6 and 7 Rad21 lox/lox Nex Cre cortices ( Figure 3a). These findings are consistent with the reported cohesin-dependence of neuronal guidance molecule expression (Kawauchi et al., 2009;Remeseiro et al., 2012;Guo et al., 2015) and migration (van den Berg et al., 2017). To assess the impact of cohesin on morphological maturation, we cultured cortical neurons in the presence of wild type glia (Kaech and Banker, 2006). Compared to freshly explanted E18.5 neurons (Figure 3b), neurons acquired considerable morphological complexity after 14 days in explant culture ( Figure 3c). We sparsely labeled neurons with GFP to visualize processes of individual neurons ( Figure 3d) and used Sholl analysis (Sholl, 1953) to quantitate the number of axonal crossings, the length of dendrites, the number of terminal points, the number of branch points and the number of spines in GAD67-negative Rad21 +/+ Nex Cre and GAD67negative Rad21 lox/ lox Nex Cre neurons. Cohesin-deficient neurons displayed reduced morphological complexity across scales, with reduced numbers of axonal branch and terminal points (Figure 3c and d). In addition, Rad21 lox/lox Nex Cre neurons showed reduced numbers of dendritic spines, the location of neuronal synapses ( Figure 3e). Taken together, these data show that the changes in neuronal gene expression that accompany cohesin deficiency have a tangible impact on neuronal morphology, and that cohesin is required for neuronal maturation.
The online version of this article includes the following source data and figure supplement(s) for figure 3: Source data 1. Figure 3: Cohesin contributes to the maturation of post-mitotic neurons.     6.48), and were preferentially downregulated ( Figure 4c). Thus, we conclude that activity-regulated genes and genes that facilitate neuronal maturation and homeostasis are directly affected by the acute depletion of cohesin, and not just as a result of impaired neuronal maturation.

Activity-regulated gene classes differ with respect to their reliance on cohesin
The disruption of baseline ARG expression suggested that cohesin-deficient neurons may be unable to induce the same activity-dependent gene expression program as wild-type neurons. To address this possibility, we performed RNA-seq of neuronal explant cultures treated either with tetrodotoxin +D-AP5 (TTX) alone to block neuronal signaling, or treated with TTX followed by 1 or 6 hr of sustained KCl exposure to induce neuronal depolarization. The fraction of constitutive genes deregulated in Rad21 lox/lox Nex Cre versus control neurons remained similar across conditions (baseline, TTX, 1 hr and 6 hr KCl, Figure 5a, top). By contrast, approximately 50% (154/305) of ARGs (Kim et al., 2010) were downregulated in Rad21 lox/lox Nex Cre versus control neurons under baseline conditions ( Figure 5figure supplement 1a). Of these, 76 were induced to control expression levels by KCl in Rad21 lox/lox Nex Cre neurons, while 29% failed to reach control levels, and 16% were expressed at increased levels ( Figure 5-figure supplement 1a). Multifactor analysis and hierarchical clustering ( Taken together, these data show that ARG expression is reduced in Rad21 lox/lox Nex Cre neurons under baseline conditions, but remains responsive to activation. Among neuronal ARG classes, IEGs and LRGs are known to differ in their 3D connectivity in that LRGs engage in longer-range chromatin contacts than IEGs in primary cortical neurons ; see methods for the definition of IEGs and LRGs). However, the functional consequences of this difference in 3D connectivity remain to be explored. We therefore asked whether IEGs and LRGs differ with respect to their reliance on cohesin. While IEGs were induced to at least wild-type levels in cohesin-deficient neurons by stimulation with KCl ( Figure 5a) or BDNF (Figure 5b), a substantial fraction of LRGs remained downregulated in cohesin-deficient neurons across conditions (Figure 5c). To test whether the expression of LRGs that remained downregulated in cohesin-deficient neurons across conditions was indeed cohesindependent, we restored RAD21 levels following transient proteolytic cleavage of RAD21-TEV. We found that the expression of LRGs was rescued by restoration of RAD21 ( Figure 5d). These data show that there are two classes of ARGs: (i) IEGs/LRGs that exhibit altered baseline expression but can fully regain expression in response to activation, and (ii) a subset of LRGs that remain deregulated in response to activation.

Cohesin-dependent neuronal genes have longer chromatin loops
To explore features that may explain why a subset of ARG requires cohesin for full expression we analyzed ChIP-seq data for cohesin and CTCF binding to ARG promoters in wild-type neurons. ARG promoters are enriched for binding of CTCF (OR = 1.621, p=0.022, two-tailed Fisher Exact Test), the cohesin subunit RAD21 (OR = 1.866, p=0.005), and cohesin in the absence of CTCF (cohesinnon-CTCF, OR = 1.621, p=0.022) compared to non-ARGs expressed in cortical neurons. Within the ARG gene set, however, there were no significant differences in CTCF, RAD21, or RAD21-non-CTCF ChIP-seq binding at the promoters of IEGs (which as a group remained fully inducible in cohesindeficient neurons), the subset of LRGs that were downregulated across conditions in Rad21 NexCre neurons (both TTX and 6 hr KCl, adj p<0.05), and LRGs that remained fully inducible across conditions in Rad21 NexCre neurons (both TTX and 6 hr KCl, adj p>0.05, Figure 6-figure supplement 1a). Therefore, while ARG promoters are enriched for CTCF, RAD21, and cohesin-non-CTCF binding, this binding is not predictive of which ARGs remain fully inducible, and which are downregulated in cohesin-deficient neurons.
To test whether cohesin-dependence of ARG regulation might instead be linked to the genomic range of chromatin loops formed by these genes we analyzed high-resolution cortical neuron Hi-C data (Bonev et al., 2017). We found that the subset of LRGs that required cohesin for full expression   in TTX and full induction by 6 hr KCl (adj p<0.05) formed significantly longer Hi-C loops than IEGs and LRGs that were not deregulated across conditions (TTX and 6 hr KCl, adj p>0.05) in Rad21 NexCre neurons ( Figure 6a). Chromatin loops with CTCF binding at one or both loop anchors were also significantly longer for cohesin-dependent LRGs than for IEGs and cohesin-independent LRGs (Figure 6a, red). Chromatin loops that connect promoters with inducible enhancers also tended to span larger genomic distances at downregulated LRGs compared to IEGs or non-deregulated LRGs, even though due to the limited numbers of enhancer-promoter loops associated with each LRG class, these trends do not reach statistical significance (Figure 6a, blue). Overall, the degree to which ARGs depend on cohesin for their correct expression correlates with the length of chromatin loops they form. We next addressed whether cohesin/CTCF binding or chromatin loop length were important factors for the impact of cohesin on the expression of additional neuronal genes. We focused on the neuronal GO terms 'synaptic transmission' (GO:0007268) and 'glutamate receptor signaling pathway' (GO:0007215) because these gene ontologies were highly enriched among downregulated genes both in acute RAD21-TEV cleavage and genetic cohesin depletion, and remained enriched across conditions in Rad21 NexCre neurons (GO:0007268: 87 of 519 genes downregulated at 6 hr KCl, adj. p<0.05, p-value for enrichment = 2.99E-07; GO:0007215: 21 of 68 genes downregulated at 6 hr KCl, adj p<0.05, p-value for enrichment = 6.46E-06). The majority of synaptic transmission and glutamate receptor signaling genes are classified as constitutive (62.7% of expressed and 59.8% of downregulated GO:0007268 and GO:0007215 genes across conditions), rather than activity-regulated (1.8% of expressed and 1.2% of downregulated GO:0007268 and GO:0007215 genes cross conditions), thus complementing the analysis of ARGs. Of note, restoration of RAD21 after transient depletion in the RAD21-TEV system rescued the expression of 90% (70 of 77) downregulated synaptic transmission and glutamate receptor signaling genes (Figure 6b), confirming that their expression was indeed dependent on cohesin.
As described above for ARGs, the TSSs of neuronal GO term genes related to synaptic transmission and glutamate receptor signaling were enriched for binding of CTCF (OR = 1.409, p<0.0001, two-tailed Fisher Exact Test), RAD21 (OR = 1.745, p<0.0001), and cohesin-non-CTCF (OR = 1.585, p<0.0001) cCompared to the remaining expressed genes in cortical neurons. However, and again as described above for ARGs, the binding of CTCF, RAD21, or cohesin-non-CTCF to the promoters of neuronal GO term genes did not predict which neuronal GO term genes remained expressed at wildtype levels, and which were downregulated in Rad21 NexCre neurons ( Figure 6-figure supplement  1b).
Notably, however, synaptic transmission and glutamate receptor signaling genes that were downregulated in Rad21 lox/lox Nex Cre neurons engaged in significantly longer-range chromatin loops than genes that were either not deregulated or upregulated (Figure 6c, black, strip plots show the expression of the depicted GO term genes in control and Rad21 lox/lox Nex Cre neurons). As with ARGs, this pattern extended to Hi-C loops with CTCF binding at one or both loop anchors (Figure 6c, red). The subset of synaptic transmission and glutamate receptor signaling genes that were downregulated in Rad21 lox/lox Nex Cre neurons formed significantly longer Hi-C loops connecting promoters and enhancers than genes in the same gene ontologies that were not deregulated (p<0.0001) or upregulated (p<0.0001) in Rad21 NexCre neurons (Figure 6c, blue). These data extend the relationship p-values test the prevalence of deregulated genes in each class under each condition, two-sided Fisher exact test. (b) Pie charts show the expression of IEGs (top) and LRGs (bottom) in RAD21-TEV neurons under baseline conditions 24 hr after ERt2-TEV induction and in response to BDNF (120 min). RAD21-TEV cleavage led to the downregulation (adj. p<0.05) of 13 out of 18 expressed IEGs and of 23 out of 101 expressed LRGs. p-values test the prevalence of downregulated IEGs and LRGs with and without BDNF stimulation, two-tailed Fisher exact test. Note that BDNF stimulation reversed the downregulation of IEGs but not LRGs in cohesin-depleted neurons. (c) Strip plots depict the expression of IEGs, LRGs that are downregulated in Rad21 NexCre neurons compared to control across conditions (TTX and 6 hr KCl), and LRGs that are not downregulated in Rad21 NexCre neurons compared to control across conditions. (d) Transient cohesin depletion and re-expression as in Figure 1-figure supplement 2. Pie charts show the expression of LRGs in RAD21-TEV relative to control neurons. 24 out of 97 LRGs expressed in RAD21-TEV neurons were downregulated 24 hr after Dox-dependent TEV induction (adj. p<0.05). The downregulation of LRGs was reversible upon Dox washout and restoration of RAD21 expression.
The online version of this article includes the following source data and figure supplement(s) for figure 5: Source data 1. Figure 5: Activity-regulated neuronal gene (ARG) classes differ in their reliance on cohesin.  Neuron. GO term genes n = 583 Figure 6. The genomic distance traversed by chromatin contacts formed by neuronal genes predicts whether or not cohesin is required for their full expression. (a) The span of Hi-C loops (left), Hi-C loops with CTCF bound to at least one of the loop anchors (middle) and Hi-C loops between promoters and inducible enhancers (right) for Immediate early genes (IEGs) (n=18) and late response genes (LRGs) downregulated in Rad21 NexCre versus control neurons in both resting (TTX) and activation conditions (6 hr KCl, adj p<0.05 in both TTX and 6 hr KCl conditions, n=22, 'Downreg. LRG'), and LRGs not deregulated in either resting (TTX) or activation conditions (6 hr KCl, adj p>0.05 in both TTX and 6 hr KCl conditions) in Rad21 NexCre relative to control neurons (adj. p>0.05, n=43, 'Non-dereg. LRG'). Box plots show the longest loop for each gene rather than average loop length, as Hi-C loop calling at 10 kb resolution precludes detection of loops <40 kb . However, analysis of average loop length confirmed that downregulated genes form longer loops than non-deregulated genes among both ARGs and neuronal GO term genes (p=0.0056 and p<0.0001, respectively). Genes without loops are included except for analysis of enhancer loops. Box plots show the longest loop recorded for each gene. Cohesin is not essential for short-range loops between inducible enhancers and promoters at the activity-regulated Fos and Arc loci Typical IEG enhancers such as Fos and Arc enhancers are fully accessible prior to stimulation (Carullo et al., 2020), and show increase H3K27ac and eRNA transcription in response to neuronal activation (Malik et al., 2014;Kim et al., 2010;Beagan et al., 2020;Joo et al., 2016;Carullo et al., 2020).
To examine the contribution of enhancer activation and enhancer-promoter contacts to inducible ARG expression in Rad21 lox/lox Nex Cre neurons we focused on the immediate early response gene Fos. Fos expression in Rad21 lox/lox Nex Cre neurons was reduced at baseline and in the presence of TTX, but Fos expression remained fully inducible when Rad21 lox/lox Nex Cre neurons were stimulated with KCl ( Figure 7a) or with BDNF ( Figure 7-figure supplement 1). The transcription of neuronal genes is controlled by neuronal enhancers (Rajarajan et al., 2016;Yamada et al., 2019;Sams et al., 2016;Kim et al., 2010;Malik et al., 2014;Schaukowitch et al., 2014;Beagan et al., 2020). Neuronal Fos enhancers in particular have been extensively characterized (Joo et al., 2016;Beagan et al., 2020), and interference with Fos enhancers precludes full induction of Fos gene expression (Joo et al., 2016). Fos enhancers 1, 2, and 5 are known to undergo activation-induced acetylation of H3K27 (H3K27ac) and active eRNA transcription in response to KCl stimulation (Joo et al., 2016). We found that activation-induced transcription of Fos enhancers did remain intact in Rad21 lox/lox Nex Cre neurons ( Figure 7b) and activation-induced H3K27ac of Fos enhancers was also preserved (Figure 7c). Given that Fos can be fully induced, and Fos enhancers are activated in the absence of cohesin, we next set out to understand if Fos can form previously reported looping interactions with its activitystimulated enhancers . We conducted 5 C to generate 10 kilobase resolution maps of chromatin loops around key IEGs and LRGs. Consistent with previous data , Fos enhancers 1 and 2, which are located ~18 and ~ 38.5 kb upstream of the Fos TSS, showed inducible chromatin contacts with the Fos promoter that formed rapidly in response to activation of wild-type neurons (Figure 7d). Unexpectedly, Rad21 lox/lox Nex Cre neurons retained the ability to robustly and dynamically induce loops between the Fos promoter and Fos enhancers 1 and 2 (Figure 7d). Quantification showed that inducible enhancer-promoter contacts at the Fos locus were of comparable strength in control and Rad21 lox/lox Nex Cre neurons (Figure 7e, top and center), while a structural CTCF-based loop surrounding the Fos locus was substantially weakened (Figure 7e, bottom). Together these results reveal the surprising finding that the critical activity-stimulated IEG Fos can fully activate expression levels and form robust enhancer-promoter loops in the absence of cohesin.
We also examined the long-range regulatory landscape of the IEG Arc. The expression of the IEG Arc was reduced in Rad21 lox/lox Nex Cre neurons at baseline, but, like Fos, Arc remained inducible by stimulation with KCl (Figure 7-figure supplement 2a, top) and BDNF (Figure 7-figure supplement 2a, bottom). Stimulation of wild-type neurons is known to trigger the formation of chromatin contacts between the Arc promoter and an Arc-associated activity-induced enhancer located ~15 kb downstream of the Arc TSS  Figure 7-figure supplement 2b). As observed for Fos, Arc promoter-enhancer contacts were retained in Rad21 lox/lox Nex Cre neurons (Figure 7-figure  supplement 2b, c), while CTCF-based loops surrounding the Arc locus were weakened (Figure 7figure supplement 2b, c). In contrast to control neurons, Arc promoter-enhancer contacts became at promoters and constitutive or inducible enhancers (right) for genes in the neuronal GO terms synaptic transmission and glutamate receptor signaling. Gene expression in Rad21 NexCre versus control neurons was assessed in both resting and activation conditions: not deregulated in TTX or 6 hr KCl, n=401, Downregulated in both TTX and 6 hr KCl, n=85, Upregulated in both TTX and 6 hr KCl, n=41. Genes do not form loops are included except for analysis of enhancer loops. Boxes show upper and lower quartiles and whiskers show 1.5 of the interquartile range. p-values were determined by nonparametric Kolmogorov-Smirnov test. Strip plots show the expression of the depicted GO term genes in control and Rad21 NexCre neurons.
The online version of this article includes the following source data and figure supplement(s) for figure 6: Source data 1. Figure 6: The genomic distance traversed by chromatin contacts formed by neuronal genes predicts whether or not cohesin is required for their full expression.     . Black frames highlight interactions between the Fos gene and upstream enhancers 1 and 2. CTCF ChIP-seq in cortical neurons (Bonev et al., 2017) is shown for orientation and H3K27ac ChIP-seq in inactive (TTX-treated) and activated neurons is shown to annotate enhancer regions (Beagan and Phillips-Cremins, 2020;Beagan et al., 2020). RNA-seq in TTX-treated and 1 hr KCl-activated control and Rad21 lox/lox Nex Cre neurons shows KClinducible transcription of Fos enhancers in wild -type and cohesin-deficient neurons. Two independent biological replicates are shown in Figure 7figure supplement 3a. (e) Quantification of the interaction frequencies between the Fos promoter and Fos enhancer 1 (top), the Fos promoter and Fos enhancer 2 (middle), and CTCF-marked boundaries of the sub-TAD containing Fos (bottom, grey arrowhead). Two replicates per genotype and condition.
The online version of this article includes the following source data and figure supplement(s) for figure 7: Source data 1. Figure 7: Fos enhancer-promoter contacts are robustly induced in cohesin-deficient neurons.   least partially independent of activation in Rad21 lox/lox Nex Cre neurons (Figure 7-figure supplement  2b, c). Hence, cohesin is required for the correct baseline expression of ARGs, but largely dispensable for inducible transcription and for specific enhancer-promoter contacts at the IEGs Fos and Arc.

Discussion
Given that mutations in cohesin and CTCF cause intellectual disability in humans (Deardorff et al., 2018;Gregor et al., 2013;Rajarajan et al., 2016), the extent to which deficits in cohesin function alter neuronal gene expression remains a critical underexplored question. To define the role of cohesin in immature post-mitotic neurons we use experimental deletion of the cohesin subunit Rad21 during a precise developmental window of terminal neuronal differentiation in vivo. We find impaired neuronal maturation and extensive downregulation of genes related to synaptic transmission, connectivity, neuronal development, and signaling in Rad21 lox/lox Nex Cre neurons. Such gene classes are central to neuronal identity, and their wide-spread downregulation is likely to contribute to the observed maturation defects of Rad21 lox/lox Nex Cre neurons. Acute proteolysis of RAD21-TEV corroborated a prominent role for cohesin in the expression of genes that facilitate neuronal maturation, homeostasis, and activation.
We have recently shown that cohesin loss in macrophages results in severe disruption of antimicrobial gene expression programs in response to macrophage activation (Cuartero et al., 2018). By contrast, cohesin loss only moderately affected genes constitutively expressed in uninduced and induced macrophages, supporting a model where cohesin-mediated loop extrusion is more important for the establishment of new gene expression than the maintenance of existing programs (Cuartero et al., 2018). Here, we extend this model to post-mitotic neurons in the murine brain. The two major ARG classes, IEGs, and LRGs, were broadly downregulated in cohesin-deficient neurons at baseline. However, IEGs and a subset of LRGs remained fully inducible by KCl and BDNF stimulation in the absence of cohesin. IEG-encoded transcription factors such as the AP1 members encoded by Fos, FosB, and JunB facilitate the induction of LRGs (Malik et al., 2014). Our data show that IEGencoded transcription factors remain fully inducible in cohesin-deficient neurons. The failure to induce a subset of LRGs to wild-type levels is therefore not explained by a lack of IEG-encoded factors. This is in marked contrast to inducible gene expression in cohesin-deficient macrophages. A substantial number of LRGs in macrophages rely on the expression of early-induced interferons (IFN), which act in an autocrine and paracrine manner to support LRG induction (Glass and Natoli, 2016). Expression of IFN-dependent LRGs can be partially rescued by provision of exogenous IFN to cohesin-deficient macrophages (Cuartero et al., 2018).
The reliance of ARGs on cohesin for full activity-induced expression was linked to the scale of chromatin interactions as quantified by the analysis of Hi-C loops. The subset of LRGs that exhibit defects in inducibility in the absence of cohesin is characterized by longer-range chromatin loops than either IEGs or LRGs that remain fully inducible in the absence of cohesin. In addition to defective chromatin architecture, deregulated expression of a particular gene may also arise from disturbances in upstream regulatory mechanisms, such as the activity of specific signaling pathways, or the expression of particular transcription factors. We are not aware of signaling pathways that would unambiguously distinguish the subset of neuronal LRGs that are fully induced in the absence of cohesin from the subset of neuronal LRGs that remain deregulated in the absence of cohesin. As discussed above, TFs encoded by IEGs are required for the induction of LRGs. Of note, these IEGs are fully induced in cohesin-deficient neurons.
The relationship between loop length and cohesin-dependence of neuronal gene expression extends to constitutively expressed cell type-specific neuronal genes. Neuronal genes related to synaptic transmission and glutamate receptor signaling that were downregulated in cohesin-deficient neurons also engaged in significantly longer chromatin loops than genes in the same GO terms that were not deregulated.
A subset of looping interactions made by ARGs and neuron-specific genes involve distal enhancers, suggesting that one role for cohesin in the expression of these genes may be to facilitate enhancerpromoter contacts. Our data indicate that specific enhancer-promoter loops at the key neuronal IEGs Fos and Arc can occur independently of cohesin in primary neurons. Fos enhancer-promoter loops remained responsive to environmental signals. Of note, Fos and Arc enhancer-promoter loops that were robust to cohesin depletion are relatively short-range ( <40 kb). By contrast, an inducible 1.68 Mb enhancer-promoter loop at the Bdnf locus, a constitutive chromatin loop between Bdnf promoter IV and an activity-induced enhancer, and all CTCF-based loops examined were substantial weakened in the absence of cohesin. While longer loops are inherently weaker than shorter loops, the observed differences in loop strength were robust to correction for the distance-dependent background signal. This indicates that our 5 C approach reliably quantifies the strength of both long and short loops.
In earlier studies, we found that contacts between the Lefty1 promoter and the +8 kb enhancer, and between Klf4 and enhancers at +53 kb remained intact in acutely cohesin-depleted ES cells (Lavagnolli et al., 2015). Synthetic activation of a Shh enhancer ~100 kb upstream of the TSS supported transcriptional activation of cohesin, while activation of a + 850 kb enhancer did not (Kane et al., 2021). Analysis of engineered enhancer landscapes in K562 cells indicates graded distance effects: Enhancers at ≥100 kb and 47 kb were highly and moderately dependent on cohesin, respectively, while loss of cohesin actually increased target gene transcription for enhancer distances ≤11 kb (Rinzema et al., 2021). Finally, promoter capture Hi-C in cohesin-depleted HeLa cells indicates ranges of 10 4 -10 5 bp for retained and 10 5 -10 6 bp for lost interactions (Thiecke et al., 2020). While these studies suggest that cohesin-dependence of chromatin contacts relates to genomic distance, they fail to link this observation to physiologically relevant gene expression. The new data described here demonstrate scaling of cohesin-dependence with genomic distance in primary neurons, and, importantly, link this finding to critical genome functions, specifically the implementation of cell type-specific gene expression programs during neuronal maturation and activation.
An open question concerns the mechanisms of enhancer-promoter contacts in the absence of cohesin. Current models of 3D genome folding posit competition between two forces, cohesinmediated loop extrusion and condensate-driven compartmentalization (Rao et al., 2017;Nora et al., 2017;Schwarzer et al., 2017;Beagan and Phillips-Cremins, 2020). RNAP2, Mediator, the transactivation domains of sequence-specific transcription factors, and the C-terminal domain of the chromatin reader BRD4 are thought to support the formation of molecular condensates enriched for components of the transcriptional machinery Boija et al., 2018;Rowley and Corces, 2018;Hsieh et al., 2020). The activation of inducible ARG enhancers involves dynamic H3K27ac , recruitment of RNA polymerases, and active transcription (Joo et al., 2016). Our data show that H3K27ac and transcription remain inducible at Fos enhancers in Rad21 lox/lox Nex Cre neurons, which could potentially contribute to loop formation. At the Arc locus, the persistence of an enhancer-promoter loop in unstimulated cohesin-deficient neurons would be consistent with a role for cohesin-mediated loop extrusion in chromatin state mixing, and the separation of enhancer contacts (Rao et al., 2017). Notably, enhancer-promoter contacts resist the inhibition of BET proteins (Crump et al., 2021), selective degradation of BRD4 (Crump et al., 2021), Mediator (El Khattabi et al., 2019;Crump et al., 2021), RNA Polymerase II (Thiecke et al., 2020), or inhibition of transcription (El Khattabi et al., 2019). These observations suggest that numerous components of the transcriptional machinery may redundantly support associations between active genes and regulatory elements Boija et al., 2018). Nevertheless, phase separation-like forces provide an attractive hypothesis for the manner by which cohesin-independent enhancer-promoter loops may form.
In summary, our data demonstrate that the extent to which the establishment of activity-dependent neuronal gene expression programs relies on cohesin-mediated loop extrusion depends at least in part on the genomic distances traversed by their long-range chromatin contacts.    Seitan et al., 2011), in combination with the Cre recombinase in the Nex locus (Goebbels et al., 2006) and where indicated Rpl22(HA) lox/ lox RiboTag (Sanz et al., 2009) were on a mixed C57BL/129 background. For timed pregnancies the day of the vaginal plug was counted as day 0.5. Genotypes were determined by PCR as previously reported (Seitan et al., 2011;Sanz et al., 2009). Rad21 tev/tev mice have been described (Tachibana-Konwalski et al., 2010;Weiss et al., 2021).

Neuronal cultures
For Nex Cre experiments, mouse cortices were dissected and dissociated from individual E17.5-E18.5 mouse embryos as described (Beaudoin et al., 2012) with minor modifications. Dissociated neurons were maintained in Neurobasal medium with B27 supplement (Invitrogen), 1 mM L-glutamine, and 100 U/mL penicillin/streptomycin for 10 days in vitro. Cells were plated at a density of 0.8 × 10 6 cells per well on six-well plates pre-coated overnight with 0.1 mg/ml poly-D-lysine (Millipore) and one third of the media in each well was replaced every 3 days. Cultures were treated with 5 μM Cytosine β-D-arabinofuranoside (Ara-C, Sigma) from day 2-4. For immunofluorescence staining neurons were plated on 12 mm coverslips (VWR) coated with poly-D-lysine at a density of 0.1 × 10 6 cells per coverslip. For cell-type-specific isolation of ribosome-associated mRNA, neurons from both cortices from each individual mouse embryo were seeded in a 10 cm dish. For RAD21-TEV experiments, mouse cortices were dissected and dissociated on E14.5-15.5 as described (Weiss et al., 2021). Neurons were maintained in Neurobasal medium with B27 supplement (Invitrogen), 1 mM L-glutamine, and 100 U/ml penicillin/streptomycin. Cells were plated at a density of 1.25 × 10 5 /cm 2 on 0.1 mg/ml poly-D-lysine (Millipore) coated plates, and half the media was replaced every 3 days. Cultures were treated with 5 μM Ara-C at day 5. For cleavage of RAD21-TEV, neurons were plated as described above and transduced at day 3 with lentivirus containing ERt2-TEV at a multiplicity of infection of one. 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 hr.
For KCl depolarization experiments, neuronal cultures were pre-treated with 1 μM tetrodotoxin (TTX, Tocris) and 100 μM D-(-)-2-Amino-5-phosphonopentanoic acid (D-AP5, Tocris) overnight to reduce endogenous neuronal activity prior to stimulation. Neurons were membrane depolarized with 55 mM extracellular KCl by addition of prewarmed depolarization buffer (170 mM KCl, 2 mM CaCl2, 1 mM MgCl2, 10 mM HEPES pH 7.5) to a proportion of 0.43 volumes per 1 ml volume of neuronal culture medium in the well. For BDNF induction experiments, neuronal cultures were treated with BDNF (50 ng/ml) for the indicated period of time at 10 days in vitro.
For Sholl analysis, dissociated cortical neurons were cultured as described (Greig et al., 2013). Astroglial monolayers were adhered to culture dishes and cortical neurons to coverslips, which were then suspended above the glia. Primary cultures of glial cells were prepared from newborn rat cortices. Four days before neuronal culture preparation, glial cells were seeded in 12-well plates at a density of 1 × 10 4 cells per well and one day before, the medium from the glial feeder cultures was removed and changed to neuronal maintenance medium for preconditioning. Mouse cortices were dissected from E17.5/E18.5 mouse embryos and kept up to 24 hr in 2 mL of Hibernate-E Medium (ThermoFisher) containing B27 supplement (Invitrogen) and 1 mM L-glutamine in the dark at 4 °C. Embryos were genotyped and the cortices from the desired genotypes were used to prepare neuronal cultures as described before. Neurons were plated on 24-well plates containing poly-D-lysine precoated 12 mm coverslips (VWR) at a density of 0.1 × 10 6 cells per well. Wax dots were applied to the coverslips, which served as 'feet' to suspend the coverslips above the glial feeder layer. 4 hr after neuronal seeding, each coverslip containing the attached neurons was transferred upside down into a well of the 12-well dishes with the glial feeder. Cultures were treated with 5 μM Ara-C from day 2-4 and subsequently one third of the media was replaced every 3 days. For sparse neuronal GFP labeling, cortical neurons were transfected using 1 μg of peGFP-N1 plasmid along with 2 μl per well of Lipofectamine 2000 (Invitrogen) after 12 days in culture. Cultures were maintained for 14 days in vitro before fixation.

RNA extraction and RT-qPCR
RNA was extracted with QIAshredder and RNeasy minikit (Qiagen). Residual DNA was eliminated using DNA-free kit (Ambion) and reverse-transcribed using the SuperScript first-strand synthesis system (Invitrogen). RT-PCR was performed on a CFX96 Real-Time System (Bio-Rad) with SYBR Green Master Mix (Bio-Rad) as per the manufacturer's protocol and normalized to Ubc and Hprt mRNA levels. Relative level of the target sequence against the reference sequences was calculated using the ΔΔ cycle threshold method. RT-PCR primer sequences:

Gene
Forward (

RNAseq analysis
Total RNA was obtained in parallel from 10 day explant cultures of dissociated cortical neurons without stimulation (baseline); after overnight treatment with TTX and D-AP5 (TTX); and after overnight treatment with TTX and D-AP5 and depolarization with KCl for 1 hr (KCl1h) or 6 hr (KCl6h). RNA was extracted with QIAshredder and RNeasy mini kit (Qiagen). RNA-seq libraries were prepared from 600 ng of total RNA (RNA integrity number (RIN) >8.0) with TruSeq Stranded Total RNA Human/ Mouse/Rat kit (Illumina). For polysome immunoprecipitation experiments, 300 ng of total RNA was used for library preparation (RIN >9.0). RNA from Rad21-TEV neurons was purified with a PicoPure RNA Isolation kit (Applied Biosystems KIT0204), and 200 ng of total RNA was used to prepare libraries using the NEBNext Ultra II Directional RNA Library Prep Kit for Illumina (polyA enrichment), following the manufacturer recommendations. Library quality and quantity were assessed on a Bioanalayser and Qubit respectively. Libraries were sequenced on an Illumina Hiseq2500 (v4 chemistry) and at least 40 million paired end 100 bp reads per sample were generated per library and mapped against the mouse (mm9) genome. The quality of RNA-seq reads was checked by Fastqc (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/) and aligned to mouse genome mm9 using Tophat version 2.0.11 (Kim et al., 2013) with parameters 'library-type = fr-first-strand'. Gene coordinates from Ensembl version 67 were used as gene model for alignment. Quality metrics for the RNA-Seq alignment were computed using Picard tools verion 1.90 (https://broadinstitute.github.io/picard/) (Picard Toolkit, 2018). Genome wide coverage for each sample was generated using bedtools genomeCov-erageBed and converted to bigwig files using bedGraphToBigWig application from UCSC Genome Browser. Bigwig files were visualised using IGV. After alignment, number of reads on the genes were summarized using HTSeq-count (version 0.5.4;Anders et al., 2015). All downstream analysis was carried out in R (version 3.4.0). Differentially expressed genes between condition were determined using DESEq2 (Love et al., 2014). p-values calculated by DESeq2 were subjected to multiple testing correction using Benjamini-Hochberg method. Adjusted p-value of 0.05 was used to select the differentially expressed genes. Principal Component Analysis (PCA) and hierarchical clustering of samples were done on the normalized read counts (rlog) computed using DESeq2. KCl-inducible genes were defined as genes in Rad21 +/+ neurons with adjusted p<0.05 and log2 fold change ≥1 in KCl1h versus TTX or KCl6h versus TTX. As reference we used previously defined activity dependent genes (Kim et al., 2010). Constitutive genes were defined as expressed genes in wild-type neurons with adj. p≥0.05 in KCl1h versus TTX and KCl6h versus TTX.

Definition of ARGs and ARG classes
Our initial analysis used inducible ARGs described (Kim et al., 2010). The number of ARGs with assigned p-values across RNA-seq conditions was n=298 in the RiboTag RNA-seq of Rad21 NexCre neurons and n=305 in the RNA-seq analysis of Rad21 NexCre neurons. For the definition of ARG classes we used previously curated gene sets (Tyssowski et al., 2018). We refer to rapidly induced, translation-independent ARGs as early-induced IEGs (called rIEGs in Tyssowski et al., 2018) and late-induced LRGs (LRGs are called translation-independent delayed PRGs and translation-dependent SRGs by Tyssowski et al., 2018;Beagan and Phillips-Cremins, 2020;Beagan et al., 2020). The number of IEGs is n=19 (Tyssowski et al., 2018, their Supplementary Table 5). Of these, n=18 had assigned p-values in all conditions of our RNA-seq analysis, as well as informative Hi-C data, and were included in our analysis. The number of fully annotated LRGs as defined by Tyssowski et al., 2018 (their Supplementary Table 5) is n=149 (comprised of 113 delayed translation-independent PRGs and n=36 translation-dependent SRGs). Of these, the number of LRGs with assigned p-values across RNAseq conditions were n=107 in the RNA-seq analysis of Rad21 NexCre neurons, n=101 in the RNA-seq analysis of RAD21-TEV neurons, and n=99 in the RNA-seq analysis of transiently RAD21-depleted and subsequently reconstituted neurons. For inclusion in the comparison of chromatin loop length versus gene expression in cohesin-deficient neurons, ARGs had to meet the following criteria: (i) assigned p-values across RNA-seq conditions, (ii) downregulation or no deregulation across TTX and KCl conditions (ARGs that were downregulated in either TTX or KCl but not in both were excluded), and (iii) informative Hi-C data had to be available for the genomic region of each gene. These conditions were met by n=18 IEGs, n=22 downregulated LRGs, and n=43 non-deregulated LRGs.
5C library preparation 5 C primers which allow for the query of genome folding at ultra-high resolution but on a reduced subset of the genome were designed according to the double-alternating design scheme (Kim et al., 2018b;Beagan et al., 2020) using My5C primer design (Lajoie et al., 2009; http://my5c.umassmed.edu/ my5Cprimers/5C.php) with universal 'Emulsion' primer tails. Regions were designed to capture TAD structures immediately surrounding the genes of interest in published mouse cortex Hi-C data (Bonev et al., 2017). The following regions were analyzed in this paper: chr2 107601077-110913077, chr10 107002896-109474896, chr12 86201802-87697802, chr15 73376037-74580037, chr17 89772409-92148409, chrX 97543400-98835400. 5 C reactions were carried out as previously described . 600 ng (~200,000 genome copies) of 3 C template for each replicate was mixed with 1 fmole of each 5 C primer and 0.9 ug of salmon sperm DNA in 1 x NEB4 buffer, denatured at 95 °C for 5 min, then incubated at 55 °C for 16 hr. Primers which had then annealed in adjacent positions were ligated through the addition of 10 U (20 μL) Taq ligase (NEB M0208L) and incubation at 55 °C for 1 hr then 75 °C for 10 min. Successfully ligated primer-primer pairs were amplified using primers designed to the universal tails (FOR = CCTC TC T ATGG GCAG TCGG TGAT , REV = CTGC CCCG GGTT CCTC ATTC TCT) across 30 PCR cycles using Phusion High-Fidelity Polymerase. Presence of a single PCR product at 100 bp was confirmed via agarose gel, then residual DNA <100 bp was removed through Ampu-reXP bead cleanup at a ratio of 2:1 beads: DNA (vol/vol). 100 ng of the resulting 5 C product was prepared for sequencing on the Illumina NextSeq 500 using the NEBNext Ultra DNA Library Prep Kit (NEB E7370) following the manufacturer's instructions with the following parameter selections: during size selection, 70 μL of AMPure beads was added at the first step and 25 at the second step; linkered fragments were amplified using eight PCR cycles. A single band at 220 bp in each final library was confirmed using an Agilent DNA 1000 Bioanalyzer chip, and library concentration was determined using the KAPA Illumina Library Quantification Kit (#KK4835). Finally, libraries were evenly pooled and sequenced on the Illumina NextSeq 500 using 37 bp paired-end reads to read depths of between 11 and 30 million reads per replicate.

5C interaction analysis
5 C analysis steps were performed as described (Gilgenast and Phillips-Cremins, 2019;Beagan et al., 2020;Fernandez et al., 2020). Briefly, paired-end reads were aligned to the 5 C primer pseudo-genome using Bowtie, allowing only reads with one unique alignment to pass filtering. Only reads for which one paired end mapped to a forward/left-forward primer and the other end mapped to a reverse/left-reverse primer were tallied as true counts. Primer-primer pairs with outlier count totals, resulting primarily from PCR bias, were identified as those with a count at least 8-fold higher (100-fold for the lower-quality Arc region) than the median count of the 5 × 5 subset of the counts matrix centered at the primer-primer pair in question; outlier counts were removed. Primer-primer pair counts were then converted to fragment-fragment interaction counts by averaging the primer-primer counts that mapped to each fragment-fragment pair (max of 2 if both a forward/left-forward and a reverse/left-reverse primer were able to be designed to both fragments and were not trimmed during outlier removal). We then divided our 5 C regions into adjacent 4 kb bins and computed the relative interaction frequency of two bins (i, j) by summing the counts of all fragment-fragment interactions for which the coordinates of one of the constituent fragments overlapped (at least partially) a 12 kb window surrounding the center of the 4 kb i th bin and the other constituent fragment overlapped the 12 kb window surrounding the center of the j th bin. Binned count matrices were then matrix balanced using the ICE algorithm 69,71 and quantile normalized across all eight replicates (two per condition) within each experimental set (neuronal activation and cohesin-rescue experimental datasets were quantile normalized separately) as previously described , at which point we considered each entry (i, j) to represent the Observed Interaction Frequency of the 4 kb bins i and j. Finally, the background contact domain 'expected' signal was calculated using the donut background mode (Su et al., 2017) and used to normalize the relative interaction frequency data for the background interaction frequency present at each bin-bin pair. The resulting backgroundnormalized interaction frequency ('observed over expected') counts were fit with a logistic distribution from which p-values were computed for each bin-bin pair and converted into 'Background-corrected Interaction Scores' (interaction score = -10*log 2 (p-value)), which have previously shown to be informatively comparable across replicates and conditions .

Identification of neuronal enhancers
H3K27Ac ChIP-Seq and corresponding input datasets (Malik et al., 2014) were from NCBI GEO using accession GSE60192. The sra files were converted to fastq using sratoolkit and aligned to mouse genome mm9 using bowtie version 0.12.8 with default parameters (Langmead et al., 2009). Sequencing reads aligned to multiple positions in the genome were discarded. Duplicate reads were identified using Picard tools v 1.90 and removed from the downstream analysis. H3K27Ac peaks were identified using macs2 with 'broad' parameters (Zhang et al., 2008). Gene coordinates were obtained from Ensembl using 'biomaRt' R package (Durinck et al., 2009) and enhancers were assigned to nearest genes using 'nearest' function from GenomicRanges R package (Lawrence et al., 2013). Enhancers at Hi-C loop anchors were defined as reported previously  and are provided there in Table S11.

Hi-C loop span analysis
Loops were called on mouse cortical neuron Hi-C data (Bonev et al., 2017) as described , loop calls are provided as Table S16 in Beagan et al., 2020). Cortical neuron CTCF ChIP-seq (Bonev et al., 2017) reads were downloaded from GEO accessions GSM2533876 and GSM2533877, merged, and aligned to the mm9 genome using Bowtie v0.12.7. Reads with more than two possible alignments were removed (-m2 flag utilized). Peaks were identified using MACS2 version 2.1.1.20160309 with a p value cutoff parameter of 10 −4 . Loops were classified as 'CTCF loops' if a CTCF peak fell within at least one anchor of the loop. Loops were classified as enhancer-promoter loops if the TSS of a gene of interest fell within one loop anchor and the other anchor of the same loop contained an activity-induced enhancer (for ARGs) or a constitutive or activity-induced enhancer (for neuronal genes related to synaptic transmission and glutamate receptor signaling; Table S11 in Beagan et al., 2020). The span of loops with an anchor that contained the TSS of a gene of interest was quantified by calculating the genomic distance between the midpoint of the loop's two anchors.

Confocal image analysis and quantification
For quantification analysis of RAD21 negative neurons and inhibitory neurons (GAD67+) images (1024 × 1024 pixels) were acquired using TCS SP5 confocal microscope (Leica Microsystems), using a HCX PL APO CS 40 x/1.25 lens at zoom factor 1 (373 nm/pixel). Images were acquired with identical settings for laser power, detector gain, and amplifier offset, with pinhole diameters set for one airy unit. DAPIidentified nuclei that colocalized with the GAD67 signal, or without RAD21 signal were counted as inhibitory or RAD21 negative neurons, respectively; and were quantified using a processing pipeline developed in CellProfiler (version 2.2, Broad Institute, Harvard, Cambridge, MA, USA, https://cellprofiler.org/). For astrocytes (GFAP+) and microglia (IBA1+) quantification in dissociated cortical neuronal cultures, the entire coverslips were imaged using a IX70 Olympus microscope equipped with a 4 × 0.1 NA Plan-Neofluar lens (1.60 μm/pixel) and the number of astrocytes and microglia were counted. The analysis and quantification of different cell types were done for three different experiments, each experiment containing at least two different samples of each genotype.
For Sholl analysis of dissociated cortical neurons, samples were imaged using a TCS SP8 confocal microscope, a HC PL APO CS2 40 x/1.30 lens at zoom factor 0.75 with a resolution of 2048 × 2048 pixel (189 nm/pixel, 0.5 μm/stack). Images were acquired with identical settings for laser power, detector gain, and amplifier offset, with pinhole diameters set for one airy unit. Approximately 10 neurons were imaged per sample and at least two different samples per genotype in each experiment; each experiment was performed three times. GFP +neurons were traced using the FilamentTracer package in Imaris software (Bitplane AG). Statistical significance was assessed using a repeat measures ANOVA with a Bonferroni Post Test (Prism-GraphPad Software). and BDNF 120 min.

Data availability
RNAseq and 5C data generated in this study have been deposited at Gene Expression Omnibus under accession number GSE172429.
The following dataset was generated: