Regulation of CHD2 expression by the Chaserr long noncoding RNA is essential for viability

Genomic loci adjacent to genes encoding for transcription factors and chromatin remodelers are enriched for long non-coding RNAs (lncRNAs), but the functional importance of this enrichment is largely unclear. Chromodomain helicase DNA binding protein 2 (Chd2) is a chromatin remodeller with various reported functions in cell differentiation and DNA damage response. Heterozygous mutations in human CHD2 have been implicated in epilepsy, neurodevelopmental delay, and intellectual disability. Here we show that Chaserr, a highly conserved lncRNA transcribed from a region near the transcription start site of Chd2 and on the same strand, acts in concert with the CHD2 protein to maintain proper Chd2 expression levels. Loss of Chaserr in mice leads to early postnatal lethality in homozygous mice, and severe growth retardation in heterozygotes. Mechanistically, loss of Chaserr leads to substantially increased Chd2 mRNA and protein levels, which in turn lead to increased transcriptional interference by inhibiting promoters found downstream of highly expressed genes. We further show that Chaserr production represses Chd2 expression solely in cis, and that the phenotypic consequences of Chaserr loss are rescued when Chd2 is perturbed as well. Targeting Chaserr is thus a potentially viable strategy for increasing CHD2 levels in haploinsufficient individuals.


Introduction
The mammalian transcriptome is highly complex, and contains tens of thousands of noncoding RNA genes [1][2][3][4] . A significant subset of these, referred to as long noncoding RNAs (lncRNAs), are at least 200 nucleotides (nt) in length, 3' polyadenylated, and 5' capped, and are therefore structurally similar to mRNAs, but lack protein coding potential 5 . Only a small portion of lncRNAs have been functionally characterized, and only very few of these have been studied in the context of organismal development 6 . Multiple lines of evidence point to a strong link between lncRNA functions and those of chromatin modifying complexes 7,8 .
Numerous chromatin modifiers have been reported to interact with lncRNAs 8 . In addition, lncRNAs in vertebrate genomes are enriched in the vicinity of genes that encode for transcription-related factors 2,9 , including numerous chromatin-associated proteins, but the functions of the vast majority of these lncRNAs remain unknown.
We hypothesized that the proximity of some lncRNA genes to genes involved in chromatin biology may imply a functional connection. To explore the biology of such interactions, we focused on one of the most conserved lncRNAs in vertebrates, found in close proximity to Chromodomain Helicase DNA Binding Protein 2 (Chd2). CHD2 is an ATP-dependent chromatin-remodeling enzyme, which together with CHD1 belongs to subfamily I of the chromodomain helicase DNA-binding (CHD) protein family. Members of this subfamily are characterized by two chromodomains located in the N-terminal region and a centrally located SNF2-like ATPase domain 10 , and facilitate disassembly, eviction, sliding, and spacing of nucleosomes 11 . There are conflicting reports on the genomic occupancy of CHD2. According to one report, based on ChIP-seq data obtained with antibodies against CHD1 and CHD2, both proteins bind predominantly in the proximity of gene promoters and share up to 60% of their DNA binding sites in human cell lines 12 . Another study used MNase-ChIP-seq of endogenously tagged Chd1 and Chd2 in mouse embryonic stem cells (mESCs) 13 , and reported that the two proteins have different binding patterns -CHD1 binds predominantly to promoter regions, whereas CHD2 is associated with gene bodies of actively transcribed genes. CHD2 has also been linked to the deposition and incorporation of the H3.3 histone variant at transcriptionally activated genes 12,14,15 and at DNA damage foci, with the latter activity promoting repair of double strand breaks 16 .
Mice homozygous for a gene-trap stop cassette in intron 27 of Chd2 survive until E18.5 with a marked growth retardation, and no viable offspring of these mice can be recovered 17 .
Heterozygotes with this mutation show increased mortality at postnatal days 1-4, and in the long-term they exhibit growth retardation, shorter life spans, and altered morphology in various organs. However, a dominant negative effect of the truncated protein could not be Chaserr and Chd2 mRNA are tightly co-expressed across a panel of human and mouse tissues, and during mouse development according to ENCODE and FANTOM5 data (Fig. 1c and Extended Data Fig. 1d-e). Interestingly, both genes are expressed at appreciable levels in all studied samples, with particularly high expression in lymphocytes. The ratio between the expression levels of the two genes is also similar across the tissues, with notable   Fig. 1c and Extended Data Fig. 1d-e).
In order to study its function, we generated Chaserr null alleles in mice by injection of CRISPR/Cas9 with a pair of gRNAs targeting sequences flanking the promoter and the first exon of Chaserr (Fig. 1d). This resulted in two different Chaserr +/mouse strains that carry deletions of 1032 and 1142 bps (Extended Data Fig. 1f), which were sufficient to abolish expression of mature Chaserr (see below), and the two lines were phenotypically indistinguishable from each other, and so we used them interchangeably in subsequent experiments. Strikingly, out of 38 pups born following crosses between Chaserr +/mice, we observed no Chaserr -/pups. The numbers of Chaserr +/pups at weaning also deviated from expected Mendelian ratios (~37%, P < 10 -8 , Fig. 1e), and the surviving mice exhibited substantial growth retardation, occasional malocclusion, and neonatal lethality ( Fig. 1f and Extended Data Fig. 1g-i). Pathological analysis of Chaserr +/mice revealed a wide range of abnormalities, including fat depletion, kyphosis, and thymic depletion, neither of which was highly penetrant. Chaserr +/females very rarely became pregnant, therefore we focused on crosses between Chaserr +/+ females and Chaserr +/males for most of this study. Out of 136 pups born from such crosses, 35 were Chaserr +/-(~25%), significantly less than the expected 50% (P < 10 -5 , Fig. 1e), suggesting two copies of Chaserr are required for proper postnatal survival. In contrast, embryos at four different developmental stages up to E18.5 were recovered with the expected ~50% Chaserr +/vs. Chaserr +/+ ratios, suggesting that a single copy of Chaserr is sufficient for viable embryonic development (Fig. 1e).
Due to the close proximity and co-expression of Chaserr and Chd2, we tested whether Chd2 expression is affected in Chaserr +/embryos. Chd2 mRNA was significantly upregulated by ~1.5-fold during embryonic development at the examined time points (E9.5, E10.5, and E13.5), in mouse embryonic fibroblasts (mEFs) derived from E13.5 embryos, and in four adult tissues (Fig. 2a). Western blot (WB) in E13.5 embryos and in the adult tissues also demonstrated strong up-regulation of CHD2 protein (Fig. 2b-c).
As Chaserr -/mice could not be generated, in order to study regulation of Chd2 in Chaserr -/cells and efficiently compare different Chaserr perturbations we used CRISPR/Cas9 in mouse embryonic stem cell (mESCs) to engineer clones with either homozygous loss of Chaserr promoter and first exon (Chaserr p/p ), or deletion of the rest of the Chaserr gene body from the first intron to just downstream of Chaserr polyadenylation site (Chaserr b/b ) (Fig. 2d).
Importantly, the Chaserr b/b line has an intact Chaserr promoter, and thus helps address the possibility of a competition between the Chaserr and Chd2 promoters. Chaserr expression was abolished in both mutant lines, as evident in RNA-seq data (Fig. 2d), leading to significant Chd2 mRNA upregulation (Fig. 2e). This up-regulation is consistent with a previous study that examined Chaserr as part of a panel of lncRNAs (Chaserr was referred to as 'linc2025' in that study) and also observed an increase in Chd2 upon deletion of Chaserr promoter 27 . Chd2 was also significantly upregulated in neurons derived from Chaserr p/p and Chaserr b/b mESCs (Fig. 2e). Similarly, targeting of Chaserr in Neuro2a cells using antisense oligonucleotides (ASOs) or LNA Gapmers reduced Chaserr expression by ~75%, and led to a significant increase in Chd2 mRNA and protein levels ( Fig. 2e and Extended Data Fig. 2ab). We note that other perturbation methods, like insertion of polyA sites or ribozymes 28 are not effective for perturbing Chaserr ( 27 and data not shown). We conclude that an intact Chaserr represses Chd2 expression in all the systems we studied.

Relative expression
Chaserr Chd2  In order to test whether the mature Chaserr RNA is sufficient for repression of Chd2, we infected the Chaserr p/p mESCs with a lentivirus carrying a doxycycline (dox)-inducible Chaserr cDNA. Chaserr was overexpressed relative to the WT levels by ~3.5-fold following Dox addition, but Chd2 mRNA expression was not affected compared to no Dox control (Fig.   2e). We conclude that either the RNA product of Chaserr, or transcription from its endogenous locus (which might be affected by the ASO/Gapmer cleavage of the nascent transcript), are required for repression of Chd2 mRNA.
To evaluate if the broader genomic context is important for Chaserr function, we reconstituted the Chaserr locus on a plasmid, which contains a fragment of the mouse genome from the Chaserr promoter to the start codon of Chd2, excluding Chaserr introns, cloned immediately upstream of a firefly luciferase coding sequence (Extended Data Fig.   2c). Deletion of Chaserr promoter in this system led to a ~50% increase in luciferase expression (Extended Data Fig. 2c), consistent with an increase we observed in the endogenous setting. This suggests that the information required for at least part of Chaserr activity is encoded in the tested locus.
We next examined whether the early lethality of mice with Chaserr loss-of-function is mediated by increased Chd2 levels. We used CRISPR/Cas9 with a single gRNA targeted to the 3 rd exon of Chd2 ( Fig. 1d and Extended Data Fig. 2d) to generate Chd2 m/m mice that had a two bp deletion in the coding sequence of Chd2. This mutation had a mild effect on Chd2 mRNA expression, but dramatically reduced CHD2 protein levels, as expected from a frameshift in the main coding frame, with the residual protein possibly emanating from an alternative start codon (Fig. 3a, note that the monoclonal antibody we used recognizes a peptide in the C-terminal part of CHD2). Chd2 m/m mice were born at expected Mendelian ratios (Extended Data Fig. 2e) and had no gross developmental phenotypes, similarly to the Chd2 tm1b(EUCOMM)Hmgu mice generated by insertion of a LacZ cassette into the 2 nd intron and deletion of 3 rd exon of Chd2 29 . This enabled us to breed Chaserr +/mice with Chd2 m/m mice ( Fig. 3b). We intercrossed the resulting Chaserr +/-Chd2 m/+ offspring, and out of 44 pups born from such crosses, only 14 (~32%) were Chaserr +/-(Extended Data Fig. 2f), suggesting that one hypomorphic allele of Chd2 is not sufficient for compensating for loss of one Chaserr allele. Indeed, CHD2 protein levels were substantially higher in Chaserr +/-Chd2 m/+ mice when compared to their Chaserr +/+ Chd2 m/+ littermates (Fig. 3c), potentially explaining the neonatal lethality and growth retardation (Extended Data Fig. 2g).
The Chaserr +/-Chd2 m/+ offspring that survived allowed us to test with allele-specific qRT-PCR whether Chaserr affected Chd2 expression from the cis allele in vivo (Extended Data   Fig. 2h). Only Chd2 from the Chaserrallele was up-regulated in Chaserr +/-Chd2 m/+ mice when compared to Chaserr +/+ Chd2 m/+ littermates (Fig. 3d), suggesting that Chaserr represses Chd2 through cis-acting regulation. Interestingly, the Chd2 from the Chaserr + allele was repressed compared to WT levels, hinting at a possible feedback regulation resulting from excess CHD2 expression (see Discussion). In order to directly test whether the severe Chaserr loss-of-function phenotype is mediated by CHD2 over-expression, we used CRISPR/Cas9 to delete the Chaserr promoter and first exon in a Chd2 m/m mouse (Extended Data Fig. 3a), and thus generated a model in which loss of Chaserr increases the expression of a hypomorphic allele of Chd2. Out of 137 pups born from intercrosses of Chaserr +/-Chd2 +/m mice, 38 were Chaserr -/-Chd2 m/m (~27%), 58 Chaserr +/-Chd2 +/m (~42%), and 41 WT (~30%), which did not deviate significantly from normal Mendelian ratios (P =0.2; Fig. 3e and Extended Data Fig. 3b,c), despite up to 6-fold upregulation of Chd2 mRNA in different tissues of Chaserr -/-Chd2 m/m mice (Fig. 3f). Despite the mRNA up-regulation, CHD2 mutant protein levels in tissues from Chaserr -/-Chd2 m/m mice were not higher than the WT protein levels in WT mice (Fig. 3g). Chaserr -/-Chd2 m/m and Chaserr +/-Chd2 +/m mice showed no significant differences in weight ( Fig. 3h and Extended Data Fig. 3d), and no obvious phenotypes, in stark contrast to the common and pleiotropic phenotypes observed in Chaserr +/neonates on Chd2 +/+ background. A hypomorphic allele of Chd2 can thus rescue the lethality caused by loss of Chaserr, when the two occur on the same allele.
To further characterize the changes in levels of Chd2 mRNA in the double mutants, we isolated mEFs from E13.5 embryos from Chaserr +/+ , Chaserr +/-, and Chaserr -/genotypes on Chd2 m/m background. qRT-PCR analysis showed that Chaserr expression is completely abolished in Chaserr -/-Chd2 m/m mice, and that Chd2 mRNA was significantly overexpressed in a manner that was inversely dependent on Chaserr dosage (Fig. 3i).
Having established that the severe phenotype resulting from Chaserr loss is mediated by CHD2, we were next interested to understand the consequences on gene expression of Chaserr loss and CHD2 up-regulation. We isolated Chaserr -/-mEFs, which showed a >2-fold increase in Chd2 mRNA and protein levels (Fig. 4a), and used RNA-seq to characterize gene expression in Chaserr +/+ and Chaserr +/-E9.5 and E13.5 embryos, adult brains, as well as Chaserr +/+ , Chaserr +/-, Chaserr -/-, and Chaserr -/-Chd2 m/m mEFs (Fig. 4b). As expected from the difference in the severity of the phenotype, we observed limited differential expression in Chaserr +/embryos or mEFs compared to WT, whereas substantial numbers of differentially expressed genes were found in Chaserr -/-mEFs. The 1,493 significantly down-regulated genes (down by at least 25%, adjusted P<0.05) were enriched for GO categories such as nervous and connective tissue development and transcriptional regulation, and were very significantly enriched for genes whose loss-of-function in mice is associated with phenotypes such as "decreased length of long bones", "decreased body weight", "respiratory distress", and "premature death" (Supplementary Table 1), closely related to the phenotypes observed in our Chaserr loss-of-function model. Specifically, some of the most down-regulated genes, such as Ctsk, Cpt1c, and Dapk3, which were also validated by qRT-PCR (Fig. 4c), are known to be required for proper embryonic development 30 . or Chaserr -/-Chd2 m/m vs. WT mEFs and distance to the 3' end of the closest upstream gene transcribed from the same strand, for genes for which this distance is <10 Kb. f, Changes in gene expression in Chaserr -/or Chaserr -/-Chd2 m/m vs. WT mEFs for genes with the indicated distances from the closest upstream gene. Genes with distance <2 Kb were further subdivided based on the expression levels of the upstream gene. g, As in e, for the E13.5 whole embryo RNA-seq. The number of genes in each group is indicated below the distance threshold. Error bars show S.E.M. *P<0.05, **P<0.01 (two-sided t-test).
In contrast, the 616 significantly up-regulated genes (up by at least 25%, adjusted P<0.05) were enriched for various RNA-related GO categories (Supplementary Table 1). Notably, changes in E13.5 embryo were significantly anti-correlated with those in E13. Inspection of the loci of some of the most affected genes, such as Ctsk and Dapk3 (Fig. 4c), led us to suspect that promoters of down-regulated genes might be preferentially found in close proximity to transcription termination sites (TTSs) of genes transcribed on the same strand (and thus potentially susceptible to transcriptional interference), in a similar organization to the Chaserr-Chd2 locus. Indeed, we found a mild yet highly significant correlation between changes in gene expression in Chaserr -/-mEFs and the distance to the TTS of the closest tandem upstream gene (Spearman R=0.2 P=2×10 -151 ), with the effect observed mostly when the distance was shorter than ~6 Kb (Fig. 4e). In contrast, no such effect was observed in Chaserr -/-Chd2 m/m mEFs (R=0.0055 P=0.45, Fig. 4e). Further, downregulation was stronger when the expression of those upstream neighboring genes was higher (Spearman R=−0.17 P=3.4×10 −7 between the change in expression of the downstream gene and the absolute expression of the upstream gene for intergenic distances <2Kb). Genes with a close and abundant upstream neighbor were significantly repressed in Chaserr -/-mEFs, and in stark contrast, much smaller differences, and in opposite direction, were observed for neighboring genes transcribed on opposite strands, when considering distances between 5' ends ( Fig. 4f). Notably, in most cases of repression of close tandem downstream gene, both the upstream and the downstream gene were protein-coding (190 of the 194 pairs with intergenic distance <2 Kb and a reduction of at least 25% in the expression of the downstream gene in Chaserr -/-mEFs). Similar trends were also found in data from E13.5 embryos, with much smaller effect sizes (Fig. 4g).
To further characterize the regulatory dysregulation, we performed ATAC-seq 31 using WT and Chaserr -/-mEFs. Consistently with the increase in Chd2 production, we observed a mild increase of ~7% in accessibility of Chd2 promoter in Chaserr -/-mEFs compared to WT. In order to analyze the changes in accessibility throughout the genome following Chaserr depletion, we merged accessibility peaks from all samples, assigned them to the nearest gene, and classified the peaks as "Core promoter" (<300 nt from TSS), "Extended promoter" (>300nt and <2 kb from TSS), "Gene body", or "Intergenic", based on RefSeq annotations.
Comparing WT and Chaserr -/-mEFs, we found 84 and 348 regions with reduced or increased accessibility, respectively (t-test P<0.05), with changes observed in gene bodies and in promoters (Fig. 5a). Promoter peaks whose accessibility was significantly decreased in Chaserr -/-mEFs were associated with concordant changes in gene expression (Fig. 5b).
Strikingly, these peaks were predominantly found at promoters of genes separated by short intergenic regions from TTSs of other genes with the same relative orientation (Fig. 5c, median distance of 2.6 Kb compared to 43.9 Kb for unaffected promoters). All nine genes with significantly reduced core promoter accessibility, a tandem upstream neighbor within <5 Kb, and robust expression in the RNA-seq data were repressed by at least 25% (all with adjusted P<0.05, Fig. 5d,f). Notably, in these cases, the upstream gene was expressed at dramatically higher levels than the affected downstream gene (15-fold on average, Fig. 5e), and the expression of the upstream genes was not substantially affected by loss of Chaserr ( Fig. 5f).  13 . Bottom -read coverage in ATAC-seq data from mEFs isolated from three embryos from each background. e, RNA-seq expression levels for the indicated pairs of genes in WT cells. f, Differential expression in Chaserr -/-mEFs compared to WT mEFs for the indicated gene pairs. g, Same as c left, but using data from Chaserr -/-Chd2 m/m mEFs, compared to WT mEFs. h, As in d, for the Chaserr/Chd2 locus. i, Metagene of CHD2 occupancy 13 on genes in mESCs, divided into four groups based on their expression FPKM.
To test if these effects stem from CHD2 over-expression, we used the same methodology to compare Chaserr -/-Chd2 m/m mEFs with WT mEFs matched from the same pregnancies, and observed fewer changes in accessibility (67 and 26 peaks with decreased and increased accessibility, respectively, based on the same criteria as above), and no preferential association of decreased promoter accessibility with distance to tandem upstream genes (Fig.   5g), suggesting that the transcriptional interference in Chaserr -/cells is driven by excess CHD2 levels.

Pol2
Chaserr +/+  The presence of a TTS closely upstream of the promoters most affected by loss of Chaserr ( Fig. 5d) resembles the organization of the Chaserr and Chd2 locus (Fig. 5h). We observed substantial occupancy of CHD2 protein (using tagged CHD2 MNase-ChIP-seq data from mESCs 13 ) in the short intergenic regions between the upstream TTS and the promoters of the repressed genes, as between Chaserr TTS and Chd2 (Fig. 5d,h), suggesting that the effect seen on the accessibility of these promoters and gene expression might be a direct result of increased CHD2 activity in the intergenic regions. Concordantly with these individual cases, a metagene analysis of tagged CHD2 occupancy across the gene body showed that CHD2 preferentially associates with ~2Kb regions immediately downstream of TTSs (Fig. 5i).

ATAC-seq
CHD2 thus appears to function in regions downstream of TTSs in WT cells, and the CHD2 hyperactivity caused by loss of Chaserr leads to repression of promoters adjacent to those regions.

Discussion
Decoding lncRNA functions is a formidable challenge, in particular because of the substantial heterogeneity in their biology and modes of action. A prominent group of vertebrate lncRNAs, including some of the most conserved ones, are found in regions flanking genes involved in transcription, including numerous chromatin modifiers. We hypothesized that such co-localization may imply a connection between the biology of the chromatin modifier and the mode of action of the lncRNA, potentially through a feedback loop. By focusing on one such pair that is particularly highly conserved in evolution, we uncovered a lncRNAmediated circuit that regulates the expression of Chd2.
Negative autoregulatory feedback loops are prevalent in RNA biology, and help tune levels of factors involved in mRNA splicing, polyadenylation, editing, and modifications, and in processing of small RNAs [32][33][34][35][36][37][38] . There are also examples of such feedback loops regulating transcription-related genes 39 , but to the best of our knowledge, there are no known autoregulatory loops involving chromatin modifiers in mammals. We propose a model in which the Chaserr-Chd2 tandem organization forms a negative feedback loop that tunes CHD2 levels. The genomic occupancy of CHD2 in regions immediately downstream of TTSs ( Fig. 5i), as well as the enrichment of CHD2-mediated accessibility changes in gene bodies (Fig. 5a), suggest that CHD2 acts in TTS-proximal regions. Accordingly, we show that under conditions of excess levels of CHD2 protein, this TTS-proximal occupancy is associated with transcriptional interference on downstream neighbors of highly-expressed genes, especially when the intergenic regions between them are particularly short (Fig. 4e-g, Fig. 5c-g). Due to the short intergenic region between Chaserr and Chd2, we suggest that such transcriptional interference serves as the basis of a negative feedback loop that leads to repression of Chd2 production when CHD2 protein levels are high, thus maintaining a tight control on its expression. In cells with Chaserr loss-of-function, this loop is compromised, and CHD2 is upregulated. Supporting this model, in Chaserr +/-Chd2 m/+ tissues, increase in the CHD2 protein leads to repression of transcription of the Chd2 m allele, which is found next to an intact Chaserr (Fig. 3d).
The exact mechanism by which Chaserr represses Chd2 expression remains unknown. The simplest feedback model suggests that Chaserr transcription, or perhaps its termination and cleavage and polyadenylation, are important. Such a mechanism would resemble the activity of the SRG1 noncoding RNA in the yeast S. cerevisiae, that represses the transcription of SER3, which is found immediately downstream of SRG1 and on the same strand, by deposition of nucleosomes at the SER3 promoter that prevent binding of transcription activators [40][41][42][43][44] . The genomic arrangement of SRG1 and SER3 resembles that of Chaserr and Chd2, though the 3' end of SRG1 overlaps SER3 promoter 41 , and the distance between their promoters is much shorter (~500 bp between SRG1 TSS and SER3 TSS vs. ~17 kb between the TSS of Chaserr and TSS of Chd2). However, there is evidence that the RNA product of Chaserr is important for its function. The exon-intron architecture and sequence of Chaserr are highly conserved, suggesting that production of a particular RNA species, or splicing at particular locations, is important either because of the RNA product itself, or because they modulate the amount of transcription or Pol2 velocity upon encounter with the ~2 Kb intergenic region between Chaserr and Chd2. The similar effects we observed when deleting Chaserr promoter or gene body, or when targeting the RNA with ASOs or Gapmers, also point to the importance of the RNA product, though we cannot rule out that ASOs and Gapmers may affect transcription by cleaving the nascent transcript.
Both Chaserr and Chd2 appear to be under the shared control of several enhancer regions found in the ~200 Kb gene desert upstream of Chaserr (Extended Data Fig. 4), and those enhancers potentially regulate both genes together, which can explain the tight co-expression between Chaserr and Chd2 mRNA (Fig. 1c). We propose that shared regulation of both genes is potentially important for maintaining the integrity of the feedback mechanism, as we found that expression levels of the upstream gene appear to be consequential for transcriptional interference resulting from CHD2 hyperactivity (Fig. 4f).

Interestingly, in contrast to previous reports based on a gene trap inserted in intron 27 of
Chd2 17,45,46 , nearly complete loss of CHD2 protein appears to be largely compatible with normal viability in mice in our hands, and in the similar mouse model generated by the EUCOMM consortium 18 . The previous observations of embryonic lethality of CHD2 loss-offunction can potentially be attributed to a dominant negative effect of the truncated protein product created by the gene trap. In contrast to the CHD2 loss models, we report here that increase in CHD2 levels brought about by loss of even a single copy of Chaserr is toxic and leads to perinatal lethality. We note that this is one of the most severe phenotypes reported so far for loss of a lncRNA 6,47 . The severe, CHD2-mediated phenotype is consistent with the high conservation of Chaserr-Chd2 genomic organization and sequence during 500 million years of vertebrate evolution. Of the lncRNAs annotated in human and mouse, only ~5% (~100 genes) have evidence of conservation in fish 26,48 , and we were able to identify homologs of Chaserr in every vertebrate species we examined (Fig. 1a).
Our results have important implications from the therapeutic perspective. Individuals that bear mutations in the CHD2 gene exhibit epilepsy and neurodevelopmental disorders 19 . In all described cases, these individuals are haploinsufficient for CHD2, and so bear an intact WT copy of CHD2. Therefore, increase of CHD2 expression through perturbation of Chaserr, e.g., using antisense oligonucleotides, might have a therapeutic benefit. Importantly, Chaserr is highly conserved between human and mouse (Fig. 1a), and targeting of the human Chaserr using Gapmers in human MCF7 and SH-SY5Y cells leads to an increase in CHD2 mRNA and protein levels (Extended Data Fig. 5), suggesting that the results we observe in the mouse model are of direct relevance to human Chaserr.
Similarly to many other chromatin remodelling factors, that are increasingly implicated in human disease, the precise molecular function of Chd2 and the direct consequences of its dysregulation are poorly understood. Here, we could leverage the similarity between the genomic arrangements of Chaserr and Chd2 and those of the genes most affected by Chaserr perturbation to highlight the potential role of chromatin remodelling that CHD2 plays downstream of TTSs. Decoding lncRNA functions can therefore provide important insight into other layers of gene regulation. As chromatin modifiers are often flanked by lncRNAs, some as highly conserved as Chaserr, we expect that further research into the functions of these lncRNAs may uncover new paradigms in chromatin biology.

Animals
The study was conducted in accordance with the guidelines of the Weizmann Institutional Animal Care and Use Committee (IACUC). C57black6 Ola HSD mice were purchased from Harlan Laboratories (Rehovot, Israel). All other mouse strains were bred and maintained at the Veterinary Resources Department of the Weizmann Institute.

Generation of KO mice
The CRISPR KO mice were generated as in (Wang et al., 2013). All KO mice were generated by standard procedures at the Weizmann transgenic core facility. To generate Chaserr KO mice we used four single guide RNAs (sgRNAs 1-4, see Supplementary Table 2), two targeting regions before the Chaserr transcription start site and two targeting regions in the first intron, which resulted in two founder lines. Chaserr mutant mice were genotyped using primers flanking the targeted region followed by Sanger sequencing. To generate Chd2 mutant mice we used one sgRNA targeted to the third exon of Chd2 mRNA. Chd2 mutant mice were genotyped by amplicon sequencing which identified a founder with an ORF shift.
Amplicon library preparation was as follows: Chd2 mutant locus was amplified using Chd2 genotype nesting primers (Supplementary Table 2 All other cell lines were routinely cultured in DMEM containing 10% fetal bovine serum and 100 U penicillin/0.1 mg ml -1 streptomycin at 37˚ C in a humidified incubator with 5% CO2.
Cell lines were routinely tested for mycoplasma contamination.
Primary mEFs were isolated as previously described 49 .

Neuronal differentiation
Neuronal differentiation was performed as previously described (Ying et al., 2003). mESCs were first grown in the absence of mEFs for two passages and then seeded on gelatin-coated plates at a density of 1.

RNA and RT-qPCR
Total RNA was extracted from different cell lines and mouse tissues, using TRIREAGENT (MRC) according to the manufacturer's protocol. cDNA was synthesized using qScript Flex cDNA synthesis kit (95049, Quanta). Fast SYBR Green master mix (4385614) was used for qPCR.

RNA-seq
Strand-specific mRNA-seq libraries were prepared from 500-4000 ng total RNA using the TruSeq Stranded mRNA (Illumina) or SENSE mRNA-Seq (Lexogen) library preparation kits, according to the manufacturer protocols. Libraries were sequenced on a NextSeq 500 to obtain 38-50 nt paired-end reads. Coverage tracks for the UCSC genome browser were prepared by aligning reads to the mm9 genome assembly using STAR 51 . Gene expression levels were quantified using RSEM 52 and a RefSeq gene annotation database which was manually edited to correct the annotation of the last exon of Chaserr. Differential expression was computed using DESeq2 with default settings 53 . Genomic context was also analyzed using the RefSeq gene annotations. RNA-seq and ATAC-seq datasets are deposited in GEO database under the accession GSE124375 (reviewer token wtkleowgbhotjgn). RNA-seq data from previous studies were downloaded from the SRA database, and quantified using RSEM with the same annotation file.

Western Blot
Total protein was extracted from tissues and cell lines by lysis with RIPA supplemented with protease inhibitors and DTT 1mM. Proteins were resolved on 8-10% SDS-PAGE gels and transferred to a Polyvinylidene difluoride (PVDF) membrane. After blocking with 5% non-fat milk in PBS with 0.1% Tween-20 (PBST), membranes were incubated with the primary antibody followed by the secondary antibody conjugated with horseradish peroxidase. Blots were quantified using Image Lab software. Primary antibodies as follows: anti-Chd2 (Millipore, MABE873), anti-β-tubulin (Sigma, T4026). Secondary antibodies as follows: antirat (AP136P), anti-mouse (115-035-003).

ATAC-seq
ATAC-seq was performed as previously described (Buenrostro et al., 2013). 5×10 4 mEFs derived from WT, Chaserr -/-, or Chaserr -/-Chd2 m/m backgrounds. Libraries were sequenced with paired-end sequencing on Illumina NextSeq 500. Reads were aligned to the mm9 genome assembly using Bowtie2 54 and peaks were called using MACS2 55 . Normalized read coverage files were computed by MACS2. Peaks form individual samples were merged using bedtools 56 and read coverage in the peaks was computed using bigWigAverageOver bed UCSC utility 57 . Differential peaks were then called using two-sided t-test. Peaks were assigned to the closest RefSeq gene and annotated as "Core promoter" if they fell within 300 nt of a TSS; "Extended promoter if they fell within 2,000 nt of a TSS; "Gene body" is they overlapped a transcription unit; or "Intergenic" otherwise.

Single-molecule FISH
Stellaris probe libraries targeting Chaserr introns or exons were designed using the Biosearch Technologies server (Supplementary Table 3) and ordered from Biosearch Technologies.
Duodenum sections were hybridized according to a previously published protocol 58 with the exception of using 25% formamide and 6µM sections. DAPI (Sigma-Aldrich, D9542) and a FITC-conjugated antibody against E-Cadherin (BD Biosciences, 612131) were used to identify the nucleus and cell-membrane, respectively. Imaging was performed on a Nikon-Ti-E inverted fluorescence microscope with a 100 × oil-immersion objective and a Photometrics Pixis 1024 CCD camera using MetaMorph software as previously described 59 . The analysis was carried on ImageM, a custom Matlab program 60 to compute single-molecule mRNA concentrations in the nucleus or cytoplasm by segmenting each cell manually according to the cell borders and the nucleus.

Cycloheximide treatment
mESCs and mEFs were treated with DMSO (vehicle) or CHX (Sigma #C7698) 100µg/mL and collected at the indicated time points for RNA/protein analysis.

Extraction of cytoplasmic and nuclear RNA
Cells were washed twice with ice-cold PBS, then scraped with ice-cold buffer A (EGTA 15µM, EDTA 10µM, protease inhibitor cocktail (Sigma, P8340) and RNase inhibitor (ERX-E4210-01)) and centrifuged at 400g, 4˚C for 5' min. Supernatant was discarded, fresh buffer A was added and the pellet was mechanically pipetted with 21G followed by 27G syringe.
Cells were then centrifuged at 2000g, 4˚C for 5 min and the syringe step was repeated. The cells were then centrifuged at 2500g, 4˚C for 5 min. The pellet was then kept as the nuclear fraction and the supernatant was centrifuged again at 6000g, 5 min followed by another supernatant collection (clean cytosolic fraction). Nuclear pellet was then washed three times with buffer A. RNA was extracted TRIREAGENT (MRC).
Chaserr cloning and lentiviral production cDNA from Source BioSicence clone C130076G01 was amplified with PCR adding restriction sites for NheI at 5' end and AgeI at 3' end (Supplementary Table 2) and cloned into pLIX_402 vector (a gift from David Root, Addgene #41394) using restriction ligation.

Micro-CT scanning and analysis
Prior to Micro-CT scanning the mice were anesthetized using IP injection of a mix of Xylazine (10mg/Kg) and Ketamine (100mg/Kg). Mice were scanned using a micro-CT device TomoScope ® 30S Duo scanner equipped with two source-detector systems. The scanner uses two x-ray sources and a detector system that are mounted on a gantry that rotates around a bed holding the animal. The operation voltages of both tubes were 40kV. The integration time of protocols was 90ms (360 rotations) for 3cm length and axial images were obtained at an isotropic resolution of 80µm. Due to the maximum length limit, to cover the whole mouse body, imaging was performed in two-three parts with overlapping area and then all slices merged to one dataset representing the entire ROI. The radiation dose range was 0.9Gy. All micro-CT scans were reconstructed using a filtered back-projection algorithm using scanner software. Then the reconstructed data sets for each mice were merged to one data set using ImageJ software. 3D volume rendering images were produced using Amira Software.