The histone deacetylase complex MiDAC regulates a neurodevelopmental gene expression program

MiDAC is a recently identified histone deacetylase (HDAC) complex. While other HDAC complexes have been implicated in neurogenesis, the physiological role of MiDAC remains unknown. Here, we show that MiDAC constitutes an important regulator of neural differentiation. We demonstrate that MiDAC functions as a modulator of a neurodevelopmental gene expression program and binds to important regulators of neurite outgrowth. On the one hand, MiDAC upregulates gene expression by mediating the removal of H4K20ac on the promoters and enhancers of pro-neural genes such as those encoding the secreted ligands SLIT3 and NETRIN1 (NTN1). Conversely, MiDAC inhibits gene expression by reducing H3K27ac on promoter-proximal and -distal elements of negative regulators of neurogenesis. Furthermore, loss of MiDAC results in neurite outgrowth defects that can be rescued by supplementation with SLIT3 and/or NTN1. These findings indicate a crucial role for MiDAC in regulating the ligands of the SLIT3 and NTN1 signaling axes to ensure the proper integrity of neurite development.


Introduction
Epigenetic regulators often constitute chromatin-modifying enzymes which catalyze the addition and removal of posttranslational modifications on histones and are involved in the control of gene expression by modulating chromatin structure and function. Among them, the families of histone acetyltransferases (HATs) and histone deacetylases (HDACs) acetylate and deacetylate lysine residues on histones and other proteins, respectively (Shahbazian & Grunstein, 2007). Based on initial studies, a model emerged in which nuclear HDACs were thought to be recruited by transcription factors to facilitate transcriptional repression by creating a more condensed chromatin landscape at their target genes (Kadosh & Struhl, 1998, Rundlett, Carmen et al., 1998, Yang, Inouye et al., 1996, Yang, Yao et al., 1997. However, more recently it was shown that HDACs tend to localize to transcriptionally active loci including promoters, gene bodies and enhancers (Wang, Zang et al., 2009). Additionally, gene expression profiling studies also demonstrated that knockout of individual HDACs resulted not only in upregulation but also downregulation of a significant number of genes suggesting both activating and repressive roles for HDACs in regulating gene transcription (Bernstein, Tong et al., 2000, Harrison, Georgiou et al., 2011, Yamaguchi, Cubizolles et al., 2010, Zupkovitz, Tischler et al., 2006. HDAC1 and HDAC2 belong to the family of class I HDACs, are highly similar (83% identity) and are involved in the control of gene expression by modulating chromatin structure and function.
At least one of the scaffolding proteins within the NuRD, CoREST and MiDAC complexes contains an ELM2-SANT domain which is instrumental in recruiting and activating HDAC1/2 (Millard et al., 2017). While the molecular function of the SIN3, NuRD and CoREST complexes have been studied in greater detail little is known about the molecular function of MiDAC (Kelly & Cowley, 2013, Millard et al., 2017.
To decipher the physiological function of MiDAC, we employed mESCs as an experimental model system. We found that MiDAC is recruited to promoters and enhancers genome-wide, directly regulates a set of neurodevelopmental genes, and can act as both an activator and repressor of different gene sets to modulate gene expression by negatively regulating the repressive histone mark H4K20ac or the active histone mark H3K27ac, respectively. Specifically, MiDAC binds to promoters and enhancers of axon guidance ligands of the SLIT and NETRIN families and is required for their activation. Genetic deletion of MiDAC components results in inactivation of SLIT3 and NTN1 signaling during neural differentiation and severely impairs neurite outgrowth and network formation. These findings reveal a novel function for MiDAC in regulating neural gene expression programs to ensure proper neuronal maturation and/or neurite outgrowth during neurogenesis.

DNTTIP1 interacts with ELMSAN1 and the histone deacetylase HDAC1 to form MiDAC in mESCs
We began by investigating the function of DNTTIP1 and ELMSAN1, two scaffolding components of MiDAC, in mESCs ( Figure 1A). Firstly, we generated and validated CRISPR/Cas9-mediated knock outs (KOs) of Dnttip1 and Elmsan1 in mESCs ( Figure 1B; Figure S1A-F; Table S1). KO mESCs (Bantscheff et al., 2011, Choi et al., 2008, Hao et al., 2011. Elmsan1 and Hdac1 mRNA levels were unchanged, Hdac2 mRNA levels slightly reduced and Dnttip1 mRNA levels more strongly reduced in Dnttip1 KO versus WT mESCs (Figure S1C, E). However, ELMSAN1 protein levels were severely reduced with no significant effects on HDAC1 or HDAC2 protein levels in Dnttip1 KO versus WT mESCs ( Figure 1B). Compared to WT mESCs, Elmsan1 KO mESCs showed no significant differences in Elmsan1, Dnttip1, Hdac1, or Hdac2 mRNA levels (Figure S1D, F); at the protein level, HDAC1 and HDAC2 appeared unaltered, but DNTTIP1 levels were significantly lower ( Figure S1F). These results suggest that in mESCs MiDAC becomes destabilized upon loss of its core subunits, DNTTIP1 or ELMSAN1.

Indels in the identified
Immunoprecipitation (IP) of ELMSAN1 in WT and Dnttip1 KO mESCs confirmed the existence of an HDAC1-containing MiDAC complex ( Figure 1C). Furthermore, consistent with the ELMSAN1 IP in WT and Dnttip1 KO mESCs, IP of DNTTIP1 in WT and Elmsan1 KO mESCs corroborated the presence of an HDAC1-containing MiDAC complex in mESCs ( Figure S1G).
However, we did not observe an interaction between ELMSAN1 and HDAC2 in Dnttip1 KO mESCs and DNTTIP1 and HDAC2 in Elmsan1 KO mESCs, suggesting that DNTTIP1, ELMSAN1 and HDAC1 interact to form MiDAC in mESCs ( Figure 1C; Figure S1G).

MiDAC controls a neurodevelopmental gene expression program
To determine transcriptional targets of MiDAC, we performed RNA sequencing (RNA-seq) in WT, Dnttip1 KO and Elmsan1 KO mESCs. We identified 493 downregulated and 368 upregulated genes in Dnttip1 KO versus WT mESCs ( Figure S2A; Table S2). Differential gene expression analysis via RNA-seq revealed 467 downregulated and 306 upregulated genes in Elmsan1 KO versus WT mESCs ( Figure S2B; Table S2). The gene expression patterns between Dnttip1 KO and Elmsan1 KO mESCs were highly correlated (R 2 = 0.88) with many differentially expressed genes (DEGs) shared between Dnttip1 KO and Elmsan1 KO mESCs ( Figure 1D; Table S2). Of the DEGs in Dnttip1 KO and Elmsan1 KO mESCs a common set of 114 upregulated and 224 downregulated genes was identified ( Figure 1E; Figure S2C). Pathway analysis of this MiDACregulated gene set revealed an enrichment for axon guidance signaling and neural development related genes among the downregulated category ( Figure 1F). Similar results were also obtained when DEGs in only Dnttip1 KO or only Elmsan1 KO mESCs were analyzed independently of each other (Figure S2D-G). Downregulated gene classes were generally associated with promoting neural differentiation (e.g., nervous system, axon guidance and GABAergic neurogenesis), whereas upregulated genes showed enrichment for pathways associated with repression of neuronal development (e.g., antagonists of nerve growth factor (NGF) signaling) (Figure S2D-G; Table S2). Importantly, we identified a set of neurodevelopmental genes that is associated with neurogenesis, axon guidance, and neurotransmitter receptor signaling and coregulated by both ELMSAN1 and DNTTIP1 (Figure 1D, G). However, loss of MiDAC function did not affect self-renewal or pluripotency as evaluated by mRNA levels of the pluripotency factors Sox2, Pou5f1, and Nanog ( Figure S3A, B), nor did it alter the percentage of alkaline phosphatase (AP)-positive colonies ( Figure S3C). Furthermore, DNTTIP1 or ELMSAN1 loss in mESCs did not alter the proliferation rate or cell cycle profile ( Figure S3D, E), in contrast to the reported growth-promoting function of DNTTIP1 in oral and non-small cell lung cancer (Sawai, Kasamatsu et al., 2018, Zhang, Wang et al., 2018. Gene expression of Ccna2 and the cell cycle inhibitors Cdkn1a (p21) and Cdkn1b (p27) whose protein products were reported to interact with and/or be regulated by DNTTIP1 were also unaffected in Dnttip1 KO and Elmsan1 KO mESCs ( Figure   S3F) (Hein et al., 2015, Huttlin et al., 2015, Pagliuca et al., 2011, Sawai et al., 2018. Together, this suggests that loss of MiDAC function does not impair self-renewal, pluripotency or the cell cycle distribution in mESCs. Rather, our findings indicate that DNTTIP1 and ELMSAN1 regulate a highly similar set of target genes in mESCs and that MiDAC controls a neurodevelopmental gene expression program, including genes that are important regulators of neurite outgrowth and morphogenesis.

MiDAC regulates neurite outgrowth
To investigate the function of MiDAC in neuronal development, we differentiated mESCs into neuro-ectoderm (NE) (Figure 2A) (Li, Yu et al., 2009). We evaluated WT, Dnttip1 KO and Elmsan1 KO NE after 8 days of differentiation, a time point that corresponds to the initiation of differentiation into NPCs, which is accompanied by increased expression of nestin (Nes) and Pax6 as well as increased protein levels of the early NE differentiation markers, PAX6 and MASH1. No significant differences in mRNA and protein levels of the above-mentioned NPC markers were detected between WT and Dnttip1 KO or Elmsan1 KO cells at day 0 or day 8 ( Figure S4A-D).
Moreover, PAX6-positive cells from WT, Dnttip1 KO and Elmsan1 KO NE at day 8 displayed a similar cell cycle profile, indicating no differences in cell cycle distribution in NPCs ( Figure S4E).
These results demonstrate that loss of MiDAC function in mESCs does not affect the initiation of differentiation into the NE lineage. Upon extended differentiation through day 12, many WT cells underwent extensive morphological changes and developed into mature neurons with long neurites while the neurite length of Dnttip1 KO and Elmsan1 KO neurons was substantially shorter ( Figure   2B). Consistent with these observations, Dnttip1 KO and Elmsan1 KO neurons showed significantly reduced Tubb3 and Map2 mRNA levels (immature and mature neuron marker genes, respectively) compared to WT neurons, suggesting that loss of MiDAC function affects neuronal maturation and/or differentiation ( Figure 2C, D). Dnttip1 KO and Elmsan1 KO neurons also showed a significant decrease in the average neurite length per neuron as well as the total number of neurites per neuron compared to WT neurons as assessed by MAP2 staining at day 12 of differentiation ( Figure 2B, E, F). However, the percentage of neurons within the total differentiated cell population remained unchanged between WT and Dnttip1 KO or Elmsan1 KO NE ( Figure 2G; Figure S4F). We next performed qRT-PCR for a subset of NE genes involved in neurite outgrowth and morphogenesis at day 12. Similar to our findings in mESCs, Dnttip1 KO and Elmsan1 KO NE showed decreased expression of several positive regulators of neuronal development and axon guidance (e.g., Slit3, Ntn1, Ncam1) as well as increased expression of repressors of neuronal development (e.g., Spry4, Id1, Pacsin1) compared to WT neurons ( Figure   2H). Overall, our data suggest that MiDAC is required to transcriptionally control a gene set of regulators that are important for neuronal maturation or neurite outgrowth but that loss of MiDAC function does not affect initiation of neuronal differentiation per se.

MiDAC binds to and modulates the expression of genes that regulate neural differentiation and neurite outgrowth
To better understand how MiDAC regulates gene expression in mESCs, we performed ChIP sequencing (ChIP-seq) in WT, Dnttip1 KO and Elmsan1 KO mESCs to map the genome-wide occupancy of DNTTIP1. This analysis identified 61,505 DNTTIP1-specific peaks, of which 22% localized within +/-1 kb of a transcription start site (TSS) ( Figure 3A; Table S3). The remaining 78% of DNTTIP1 binding sites were located upstream of (24%), downstream of (12%), or within (43%) a gene. Of the 47,657 non-TSS DNTTIP1 peaks, a majority (84%) was found at a distance of > 5 kb from the TSS ( Figure 3A; Table S3). Interestingly, a significant number of DNTTIP1 binding sites overlapped with the active histone marks H3K4me1 and/or H3K27ac and TSSassociated DNTTIP1 peaks were strongly enriched for the active histone mark H3K4me3 along with H3K27ac and H3K4me1 ( Figure 3B; Figure S5A). Furthermore, low DNTTIP1 binding at TSS correlated with increased enrichment of the repressive histone mark H3K27me3 ( Figure   S5A). Non-TSS DNTTIP1 binding sites were enriched for the enhancer marks H3K4me1 and H3K27ac, hinting at a potential role for DNTTIP1 in enhancer-mediated processes ( Figure S5A).
We next correlated our DNTTIP1 ChIP-seq data with gene expression data from Dnttip1 KO mESCs. This analysis revealed that out of 493 downregulated genes in Dnttip1 KO mESCs, 358 genes (73%) were bound by DNTTIP1, while out of 368 upregulated genes in Dnttip1 KO mESCs, 314 genes (85%) were occupied by DNTTIP1 at either their promoters or other gene regulatory regions such as enhancers ( Figure 3C; Figure S5B, C; Table S4). Importantly, an even higher proportion of differentially expressed neurodevelopmental genes was also bound by DNTTIP1 (36 out of 39 (92%) in the downregulated and 22 out of 23 (96%) in the upregulated category) ( Table   S4). This overlap is highly significant and indicates that the majority of DEGs in Dnttip1 KO mESCs constitute direct targets of DNTTIP1, including important regulators of neuronal differentiation and neurite outgrowth. Moreover, motif analysis of DNTTIP1 peaks revealed that DNTTIP1 binding sites associated with down-or upregulated genes in Dnttip1 KO mESCs are selectively enriched for transcription factor (TF) binding motifs, including some that have been implicated in neurogenesis ( Figure S5D). These TFs include RBFOX2 (associated with genes upregulated in Dnttip1 KO mESCs) and ELK1, a member of the ETS TF family (associated with genes downregulated in Dnttip1 KO mESCs) ( Figure S5D) (Besnard, Galan-Rodriguez et al., 2011, Gehman, Meera et al., 2012. We further assessed the relationship between DNTTIP1 and ELMSAN1 by determining the genome-wide occupancy pattern of DNTTIP1 in Elmsan1 KO mESCs. While no DNTTIP1 binding to chromatin could be detected in Dnttip1 KO mESCs, DNTTIP1 association with chromatin was strongly reduced in the absence of ELMSAN1 ( Figure   3D). However, some DNTTIP1 enrichment in Elmsan1 KO mESCs could still be observed, suggesting that DNTTIP1 retains the ability to bind to chromatin when ELMSAN1 is depleted ( Figure 3D). To study the genome-wide MiDAC binding landscape in mESCs, we compared the occupancy patterns of DNTTIP1 and HDAC1. We found that many DNTTIP1-bound regions (58%) also displayed considerable enrichment for HDAC1, and that 40% of all HDAC1 peaks colocalized with DNTTIP1 ( Figure 3E). Considering that HDAC1 also exists within several other complexes, including SIN3, NuRD, and CoREST, this finding suggests that a significant number of HDAC1-occupied loci is bound by MiDAC. Collectively, our findings suggest that MiDAC localizes genome-wide to enhancers and promoters in mESCs, thereby transcriptionally activating or inhibiting neurodevelopmental genes. We next examined activators and repressors of neurite outgrowth more specifically. We investigated SLIT3 and NETRIN1 (NTN1), which are key ligands of the SLIT/ROBO and NETRIN/DCC/UNC signaling pathways, and are regulators of neuron maturation and neurite outgrowth but are also playing roles in angiogenesis, lung morphogenesis, mammary gland development and cancer progression involving processes such as cell migration, cell interaction and cell adhesion (Bashaw & Klein, 2010, Blockus & Chedotal, 2016, Lai Wing Sun, Correia et al., 2011, Seiradake, Jones et al., 2016. Both Slit3 and Ntn1 mRNA levels were decreased in Dnttip1 KO and Elmsan1 KO versus WT mESCs and NE (Figure 1G, 2H; Table S2). Furthermore, DNTTIP1 was directly bound to the promoters and putative intraand intergenic enhancers of Slit3 and Ntn1 in mESCs and many of these gene regulatory elements were also co-bound by HDAC1 ( Figure 3F; Figure S5E). Conversely, a repressor of neurite outgrowth and morphogenesis, sprouty4 (Spry4), and an inhibitor of neural differentiation, Id1, were both transcriptionally upregulated in mESCs and NE upon loss of DNTTIP1 or ELMSAN1 (Figure 1G, 2H; Table S2) (Alsina, Irala et al., 2012, Lyden, Young et al., 1999, Nam & Benezra, 2009). DNTTIP1 and HDAC1 co-occupied the promoters and putative enhancers of Spry4 and Id1 ( Figure 3G; Figure S5F). In summary, these results indicate that MiDAC binds to regulatory regions of activators and repressors of neural differentiation and neurite outgrowth, thereby positively or negatively modulating their transcriptional output, which is in accordance with the phenotypic defects observed in Dnttip1 KO and Elmsan1 KO neurons ( Figure 2B, E, F). We next sought to identify potential histone substrates of MiDAC by screening for global changes of several histone acetylation marks by performing western blots in WT, Dnttip1 KO, and Elmsan1 KO mESCs. Of the histone acetylation marks tested, only H4K20ac was found to be consistently increased in Dnttip1 KO and Elmsan1 KO mESCs, suggesting that H4K20ac could be a major MiDAC substrate (Fig. 3H). Interestingly, H4K20ac has been recently described as a repressive histone acetylation mark, and its targeting by MiDAC might explain why we observed a greater number of downregulated compared to upregulated genes in Dnttip1 KO and Elmsan1 KO mESCs versus WT mESCs (Kaimori, Maehara et al., 2016).

MiDAC directly targets positive and negative regulators of neurite outgrowth during neural differentiation
To further decipher the role of MiDAC in regulating select target genes during neurogenesis, we performed ChIP analysis of DNTTIP1, ELMSAN1, HDAC1 and the histone acetylation marks H3K27ac (active) and H4K20ac (repressive) in NE after 12 days of differentiation. We focused our attention on the four previously described neurodevelopmental genes-Slit3, Ntn1, Spry4, and KO compared to WT neurons and was partly restored upon treatment with CM of WT NE ( Figure   5A, B). Interestingly, classification of the neurite length into two categories, <50 µm or ≥50 µm, showed that only longer neurites (≥50 µm in length) were significantly affected in Dnttip1 KO and Elmsan1 KO neurons and could be selectively rescued by supplementation with WT CM ( Figure   5B). In addition, the reduced number of neurites per neuron in Dnttip1 KO and Elmsan1 KO versus WT neurons was also significantly restored upon treatment with CM of WT NE ( Figure 5C). To further validate these findings, we used a chamber assay to co-culture granule neuron progenitor cells (GNPs) in the lower chamber in parallel with differentiating WT, Dnttip1 KO or Elmsan1 KO NE in the upper chamber comprising a time window from day 7-12 of differentiation to induce neuronal network formation of GNPs ( Figure S8B). Co-culturing of GNPs with differentiating WT NE resulted in extensive neuronal network formation, while neuronal network formation was severely impaired when GNPs were co-cultured with Dnttip1 KO and Elmsan1 KO NE ( Figure   S8C, D). In summary, these results suggest that the loss of MiDAC function in NE results in a lack of certain secretome components that are otherwise required for proper neurite outgrowth. To further establish the specific effects of MiDAC on the SLIT3 and NTN1 signaling pathways, we investigated the ligand-receptor pairs SLIT3/ROBO3 and NTN1/UNC5B. Importantly, compared to lysates from WT NE, lysates from Dnttip1 KO NE showed strong reductions in SLIT3 and NTN1, whereas ROBO3 and UNC5B were not significantly affected ( Figure 6A). Furthermore, we also observed a significant reduction of SLIT3 and NTN1 in CM of Dnttip1 KO versus WT NE ( Figure 6A). The downstream effectors DAB1 of the NTN pathway and FAK of the SLIT/NTN signaling cascade were also deactivated, as shown by decreased phosphorylation of DAB1 and FAK in Dnttip1 KO versus WT NE, despite no change in total protein levels ( Figure   6A) (Zelina, Blockus et al., 2014). Having established that the SLIT3 and NTN1 signaling cascades are directly targeted by MiDAC during neurogenesis, we sought to test whether the loss of activity of these pathways is sufficient to explain the neurite outgrowth defects detected in neurons that had lost MiDAC function. To address this question, we differentiated Dnttip1 KO mESCs into NE for 12 days and supplemented the CM from day 7 onward with recombinant SLIT3, NTN1, or a combination of SLIT3 and NTN1 ( Figure S8A). To assess the specific effects of these ligands with their corresponding pathways, we combined the supplementation of CM with recombinant ligands with a blocking approach involving antibodies directed against the extracellular domains of either ROBO3 (SLIT3 pathway), UNC5B (NTN1 pathway), or a combination of ROBO3/UNC5B ( Figure S8A). We found that SLIT3, NTN1, and SLIT3/NTN1 were all able to largely restore the neurite outgrowth defects observed in Dnttip1 KO neurons and that the restoration of the phenotype was directly mediated by signaling through the ROBO3 and UNC5B signaling receptors (Figure 6B-D). We also obtained similar results using a modified version of the chamber assay described above, in which GNPs were treated with CM of WT or Dnttip1 KO NE and identical combinations of recombinant ligands and blocking antibodies ( Figure S8B). The reduced neuronal network formation, that was observed when GNPs were treated with CM of Dnttip1 KO NE, was largely rescued in the presence of SLIT3, NTN1, or SLIT3/NTN1, and the restoration of network formation was significantly abrogated by preblocking GNP-derived neurons with neutralizing antibodies against ROBO3 and/or UNC5B before ligand-containing CM of Dnttip1 KO NE was added (Fig. S8E, F). Taken together, these findings indicate that the loss of MiDAC function in NE during neural differentiation results in strongly reduced SLIT3 and NTN1 signaling which accounts for the majority of the observed neurite outgrowth defects.

Discussion
Here, we report that in mESCs, DNTTIP1 in association with ELMSAN1 and HDAC1, forms a chromatin-associated mitotic deacetylase complex called MiDAC. We provide evidence that MiDAC is enriched at promoters and enhancers genome-wide and can function as an activator and repressor of transcription in mESCs. More specifically, we uncover a crucial role for MiDAC in regulating a neurodevelopmental gene expression program that controls neurite outgrowth and network formation during neurogenesis. MiDAC achieves this by positively regulating a set of neurodevelopmental genes including genes of the axon guidance ligands SLIT3 and NTN1, while at the same time suppressing negative regulators of neurogenesis such as Spry4 and Id1. The results presented here support a model in which MiDAC is required to activate the promoters and enhancers of pro-neural genes such as Slit3 and Ntn1 while in parallel repressing negative regulators of neurogenesis by removing H3K27ac from promoters and enhancers of genes such as Spry4 and Id1 to allow neurons to attain their required neurite length during differentiation ( Figure   7).
MiDAC originally obtained its name from a chemoproteomic profiling study due to its increased association with certain HDAC inhibitors in protein extracts from mitotically arrested versus nonsynchronized proliferating cells that otherwise displayed comparable levels of DNTTIP1 (Bantscheff et al., 2011). Individual MiDAC subunits have been shown to interact with CCNA2 and/or the cell cycle-dependent kinase CDK2, and were reported to positively regulate the cell cycle inhibitors CDKN1A and CDKN1B and cell growth in various cancer cell lines, thus implying a potential role for MiDAC in cell cycle regulation (Gizard et al., 2005, Gizard et al., 2006, Hein et al., 2015, Huttlin et al., 2015, Pagliuca et al., 2011, Sawai et al., 2018, Zhang et al., 2018. However, we did not observe any effect on cell proliferation, cell cycle distribution, or any transcriptional changes in the above-mentioned cell cycle-related genes (Ccna2,Cdkn1a,Cdkn1b) in mESCs with a loss in MiDAC function, arguing in favor of a cell-and/or context-specific role for MiDAC in controlling cell proliferation and the cell cycle.
The C-terminus of DNTTIP1, subunit of MiDAC, has been reported to bind to a specific DNA binding motif following ChIP-seq studies utilizing a FLAG-tagged version of DNTTIP1 in HEK293 cells (motif: TGCAGTG-(14 bp)-CACTGCA flanked by AT-tracts) (Koiwai et al., 2015). In concordance with these results, our study confirms a chromatin-associated function and binding of DNTTIP1 to both promoter-proximal and -distal elements in mESCs. Although we did not find evidence of a clear DNTTIP1 consensus binding motif, we did detect a significant enrichment of several TF motifs at DNTTIP1-bound sites (data not shown). This implies that MiDAC is specifically recruited to chromatin by TFs and that its ability to bind nucleosomes might aid in MiDAC spreading from its initial recruitment site to carry out histone deacetylation.
Furthermore, motif analysis of differentially expressed genes that display DNTTIP1 enrichment at their promoters or enhancers revealed the binding motifs of several TFs with previously reported roles in neurogenesis. We postulate that MiDAC upregulates a transcriptional program that is driven by ELK1, a member of the ETS family of pioneer transcription factors, while downregulating genes through RBFOX2, which negatively affects neurodevelopmental processes (Besnard et al., 2011, Gehman et al., 2012. However, it remains to be determined whether these TFs affect MiDAC recruitment and/or function during neurogenesis. Although DNTTIP1 levels are strongly reduced in Elmsan1 KO mESCs, we still observed some DNTTIP1 recruitment to chromatin. With TRERF1 and ZFP541 (ZNF541 in humans) two ELMSAN1-related proteins have been described that could also potentially be incorporated into MiDAC (Choi et al., 2008, Gizard et al., 2001. While ZFP541 is only very tissue-specifically expressed during spermatogenesis and not in mESCs, TRERF1 expression can be detected in mESCs (data not shown) (Choi et al., 2008).
Thus, the remaining DNTTIP1 in Elmsan1 KO mESCs might form an alternative MiDAC complex with the ELMSAN1 paralog TRERF1 which could account for the residually detected DNTTIP1 recruitment in the absence of ELMSAN1. HDAC1/2-containing complexes are generally considered repressors of transcription (Kadosh & Struhl, 1998, Rundlett et al., 1998, Yang et al., 1996. However, accumulating evidence suggests that HDAC1/2 might also be involved in transcriptional activation (Bernstein et al., 2000, Harrison et al., 2011, Yamaguchi et al., 2010. Moreover, genome-wide studies have shown that HDAC1/2 localize to promoters, gene bodies, and enhancers of actively transcribed genes, thus implicating HDAC1/2 as positive regulators of transcription . Here, our finding that a high percentage of DNTTIP1-bound loci is associated with both down-and upregulated DEGs in Dnttip1 KO versus WT mESCs and NE shows for the first time a direct role for MiDAC in both transcriptional activation and repression. The exact mode of action whereby MiDAC achieves activation or repression has yet to be elucidated in greater detail, but our studies provide evidence for at least two possible scenarios. First, MiDAC could control gene activation and repression by deacetylating histones. Most histone acetylation marks, including H3K27ac, correlate with transcriptional activation, but recently H4K20ac has been reported as a mark that is associated with silenced genes (Kaimori et al., 2016). Our data indicate that MiDAC might function as a repressor of genes that are marked by H3K27ac and more broadly as an activator on repressed genes by mediating deacetylation of H4K20. As a second possibility, MiDAC might exert its function by deacetylating TFs or other chromatin-modifying proteins such as histone acetyltransferases (HATs). If specific TFs or HATs are negatively regulated by deacetylation through MiDAC at distinct target sites, this could also potentially explain the increase in H3K27ac (through CBP/p300) or other histone acetylation marks on enhancers and promoters of up-or downregulated genes between WT, Dnttip1 KO and Elmsan1 KO mESCs and NE.
HDAC1 and HDAC2 are important regulators of many developmental processes including neurogenesis (Kelly & Cowley, 2013). But their function remains enigmatic as they can be recruited into at least four independent HDAC1/2-containing complexes (Millard et al., 2017).
While the HDAC1/2-containing SIN3, NuRD and CoREST complexes have been implicated in neurogenesis, the physiological function of MiDAC has been unexplored to date (Andres et al., 1999, Knock et al., 2015, Nitarska et al., 2016, Wang et al., 2016. Here, we report that MiDAC directly controls a neurodevelopmental gene expression program and specifically targets enhancers and promoters of positive and negative regulators that are involved in neurite outgrowth during neural differentiation. While this study emphasizes the role of MiDAC in neural differentiation and neurite outgrowth, it is evident from our gene expression analyses that this neurodevelopmental gene expression program only constitutes a fraction of the MiDAC-regulated transcriptome in mESCs pointing to other biological roles of MiDAC. SLIT/ROBO, EPHRIN/EPH, NETRIN/DCC/UNC, and SEMAPHORIN/PLEXIN signaling represent the four major canonical axon guidance pathways (Bashaw & Klein, 2010). Integration of our gene expression and ChIP data showed that MiDAC directly activates the genes of the axon guidance ligands SLIT3 and NTN1 while repressing the genes of the negative regulators SPRY4 and ID1.
SLIT3 and NTN1 are ligands for their cognate receptors ROBO3 and UNC5B, respectively.
Recombinant SLIT3 and NTN1 either alone or in combination were able to largely rescue the neurite outgrowth defects in neurons that have lost MiDAC subunits. Interestingly, blocking only ROBO3 or only UNC5B signaling was sufficient to inhibit the rescue effects of SLIT3 in combination with NTN1, suggesting that a heteromeric ROBO3/UNC5B axis is required for efficient transduction of SLIT3/NTN1 signaling to ensure proper neurite outgrowth. While evidence exists that heteromeric receptor association and signaling of ROBO3/UNC5B occurs, it has yet to be determined whether this also applies to the role of MiDAC in an in vivo system (Zelina et al., 2014). In summary, while individual TFs have been reported to transcriptionally regulate select axon guidance ligands or receptors little is known about their regulation through chromatin-modifying components (Kim, Kim et al., 2016, Labrador, O'Keefe et al., 2005. Here, we describe how the axon guidance ligands SLIT3 and NTN1 are transcriptionally controlled by an epigenetic regulator, the MiDAC complex. Finally, neurite outgrowth defects are observed in several neurodevelopmental disorders such as autism, epilepsy, fragile X-syndrome and psychiatric disorders (Gilbert & Man, 2017, Huang & Song, 2019, Krejci, Legartova et al., 2017, Wen, Cheng et al., 2017. However, the underlying molecular mechanisms remain unclear. Our study implicates MiDAC as an essential regulator of neurite outgrowth which suggests that it might be playing an important role in certain neurodevelopmental disorders. Future studies of MiDAC in the context of neurodevelopmental disorders are thus likely to yield insightful mechanistic details about disease pathogenesis.

Neural Differentiation
Directed differentiation into neuro-ectoderm WT, Dnttip1 KO and Elmsan1 KO mESCs were dissociated by trypsinization, resuspended in differentiation medium 1 and pelleted by centrifugation. 1 x 10 6 mESCs were resuspended in 3 ml differentiation medium 1 and then transferred into one well of 6-well ultra-low attachment plate (Corning,3471). Cells were grown in differentiation medium 1 for 3 days. The differentiation medium 1 was changed every day. After 3 days the emerging embryoid bodies (EBs) were cultured in differentiation medium 2 containing 2 µM retinoic acid (Santa Cruz Biotechnology, sc-200898) for an additional 3 days. The differentiation medium 2 was changed every day. After 6 days the EBs were picked up gently with a 200 µl large orifice tip and transferred into a 1.5 ml tube. The EBs were allowed to settle down by gravity followed by careful aspiration of the supernatant. 1 ml differentiation medium 3 containing 20 ng/ml FGF2 and 20 ng/ml EGF was carefully added to the 1.5 ml tube containing the EBs, and the EBs were gently picked up with a 200 µl large orifice tip and transferred onto a 0.1% gelatin-coated well of a standard 6-well plate (Corning, 3516).
Subsequently, 1 ml of additional differentiation medium 3 was carefully added to the well containing the EBs. The EBs were allowed to differentiate for the next 6 days, with daily changing of differentiation medium 3. The EBs will attach slowly over time and aster-shaped neurons will emerge from the periphery of the EBs forming increasingly longer neurite extensions and networks. Neurite outgrowth was quantified and visualized after 12 days of differentiation.
Experiments to rescue the neurite outgrowth defects of Dnttip1 KO and Elmsan1 KO neurons with CM from WT NE were conducted as above with the following alterations. After 6 days of differentiation all Dnttip1 KO and Elmsan1 KO CM was removed and replaced daily with 50% fresh differentiation medium 3 and 50% CM from WT NE for the next 6 days. Experiments to rescue the neurite outgrowth defects of Dnttip1 KO neurons with axon guidance ligands were carried out as described above except that recombinant SLIT3 (R&D Systems, 9296-SL-050), NTN1 (R&D Systems, 6419-N1-025) or a combination of SLIT3/NTN1 were added daily to differentiation medium 3 (at a concentration of 0.5 and 0.25 ug/ml for SLIT3 and NTN1 respectively) from day 6 of differentiation onward. The rescued neurite outgrowth defects of Dnttip1 KO neurons via SLIT3, NTN1 or SLIT3/NTN1 supplementation were blocked by using antibodies directed against the extracellular domain of the axon guidance receptors ROBO3 (R&D Systems, AF3155), UNC5B (R&D Systems, MAB1006), a combination of ROBO3/UNC5B or IgG (Millipore, 12-370) as a negative control. Blocking was performed daily from day 7 onward by adding 0.5 µg of each antibody to differentiation medium 3 two hours before addition of the ligands SLIT3, NTN1 or SLIT3/NTN1. Differentiation medium 2: Differentiation medium 1; 2 µM retinoic acid.

Chamber assay to assess neuronal network formation from granule neuron progenitor cells (GNPs)
GNPs were isolated from mouse cerebella from postnatal day 6-7 (a gift from Martine Roussel) (Vo, Wolf et al., 2016). GNPs were grown as neurospheres in GNP medium in 6-well ultra-low attachment plates (Corning, 3471). GNP neurospheres were dissociated by trypsinization, resuspended in GNP culture medium and pelleted by centrifugation. 10 5 GNPs were resuspended in 600 µl GNP culture medium and seeded into one well of a 24-well ultra-low attachment plate (Corning, 3473) and allowed to form neurospheres for 3 days. GNP culture medium was changed every day. From this point onward GNPs were differentiated in the same fashion as described in "directed differentiation into neuro-ectoderm" by sequential culturing in differentiation medium 2 and 3 and adjustment of media volumes to a 24-well format (one fifth of a 6-well format) with the following alterations. After culturing in differentiation medium 2 GNPs were plated into one well of a 24-well transwell chamber (pre-coated with 1% gelatin) in differentiation medium 3 (Corning, 354480) and differentiated for another 6 days. WT, Dnttip1 KO and Elmsan1 KO mESCs were differentiated according to the "Directed differentiation into neuro-ectoderm" protocol for the first 6 days but experiments were scaled down so that EBs were grown in one well of a 24-well ultralow attachment plate (Corning, 3473) and all media volumes were adjusted to one fifth of the 6well protocol. The EBs were carefully transferred into a matrigel-coated insert and allowed to differentiate for the next 6 days, with daily changing of differentiation medium 3 in the insert and GNP medium in the bottom chamber. In parallel, GNP medium and MEF CM medium was added to separate inserts as positive and negative controls, respectively and media changes were carried out in the same way as for differentiation medium 3. Experiments to rescue the network formation defects of GNP-derived neurons caused by culturing in CM of Dnttip1 KO NE were carried out as described above for the first six days of differentiation. However, instead of co-culturing GNPs with Dnttip1 KO NE, GNPs were supplemented daily with CM of Dnttip1 KO NE (50%), 50% fresh differentiation medium 3 and recombinant SLIT3 (R&D Systems, 9296-SL-050), NTN1 (R&D Systems, 6419-N1-025) or a combination of SLIT3/NTN1 for 6 days from the day of GNP seeding. Network formation of GNP-derived neurons grown in CM of Dnttip1 KO NE and differentiation medium 3 supplemented with SLIT3, NTN1 or SLIT3/NTN1 was blocked by using antibodies directed against the extracellular domain of the axon guidance receptors ROBO3 (R&D Systems, AF3155), UNC5B (R&D Systems, MAB1006), a combination of ROBO3/UNC5B or IgG (Millipore, 12-370) as a negative control. Blocking was performed daily for 6 days from the day of GNP seeding by aspirating the old medium and adding 0.1 µg of each blocking antibody in fresh differentiation medium 3 two hours before adding CM of Dnttip1 KO NE (50%), 50% fresh differentiation medium 3 with either the ligands SLIT3, NTN1 or SLIT3/NTN1. Network formation of GNP-derived mature neurons was quantified and visualized 6 days after GNPs were seeded. A neuronal network was scored when all neurite projections of an individual neuron were well connected with the neurites of neighboring neurons and formed a closed local circuit as assessed by TUBB3 staining (Doetsch & Alvarez-Buylla, 1996, Shepherd, 1998. The percentage of network formation was scored as follows: =total number of complete networks per TUBB3positive neuron/total number of DAPI-positive cells) x 100. The different groups were statistically analyzed using ONE-way ANOVA t-test.

Alkaline Phosphatase Staining and Quantification
Alkaline phosphatase staining of mESCs was carried out with the Alkaline Phosphatase Detection Kit (Millipore, SCR004) following the manufacturer's instructions. 100 mESC colonies were assessed per genotype.

Chromatin Immunoprecipitation (ChIP)
ChIPs were performed according to a modified version of Lee et al., 2006(Lee, Johnstone et al., 2006. For ChIP-seq applications of non-histone proteins ChIP was carried out with 5 x 10 7 cells and for histone ChIPs with 2-2.5 x 10 7 mESCs. For manual ChIP applications of non-histone proteins ChIP was carried out with 2 x 10 6 NE cells and for histone ChIPs with 10 6 NE cells. For non-histone ChIPs dual crosslinking was performed at RT for 30 min with 2 mM disuccinimidyl glutarate (DSG) in DPBS followed by addition of paraformaldehyde to a final concentration of 1% and further crosslinking for 15 minutes. For histone ChIPs crosslinking was performed at RT for 15 minutes with 1% paraformaldehyde. Crosslinking was quenched with 150 mM glycine for 5 minutes at RT. After quenching the cells were pelleted, the supernatant aspirated and the cell pellet washed once in DPBS and then snap-frozen and stored at -80°C or immediately further processed.
The crosslinked cell pellet was carefully resuspended in Lysis Buffer 1 with a transfer pipet and the cell suspension incubated for 10 minutes on a nutator at 4°C. Nuclei were then pelleted and the supernatant aspirated. The cell pelleted nuclei were carefully resuspended in Lysis Buffer 2 with a transfer pipet and the cell suspension incubated for 10 minutes on a nutator at 4°C. Nuclei were then pelleted again and the supernatant aspirated. The nuclear pellet was then resuspended in Lysis Buffer 3 and the suspension sonicated with a probe sonicator (Fisher Scientific Model 705 Sonic Dismembrator) at output setting 55 (27-33 W) for 12 cycles with each cycle constituting a 30 second sonication burst followed by a 60 second pause. After sonication 1/20 volume of 20% Triton X-100 was added and the sample was mixed. The chromatin was cleared by centrifugation for 10 minutes at 20,000 g at 4°C and the supernatant was transferred to a fresh tube. The cleared chromatin was then stored at 4°C. For each ChIP 100 µl Protein A/G-Plus agarose (Santa Cruz Biotechnology, sc-2003) slurry was used. The beads were washed in 1 ml Blocking Solution, pelleted and the supernatant was removed. The bead wash step was repeated another time. The beads were resuspended in 1 ml Blocking Solution followed by addition of the appropriate antibody and were incubated on a nutator at 4°C overnight. For ChIP-seq applications of nonhistone proteins and histones 10 µg of antibody was used. For manual ChIP applications of non-histone proteins and histones 3 µg of antibody was used. On the next day the stored chromatin was cleared another time by centrifugation for 10 minutes at 20,000 g, 4°C and the supernatant was transferred to a fresh tube. A small chromatin aliquot of 50 µl was reserved as input control and was kept on ice until further processing (see below). The beads were washed twice with 1 ml Blocking Solution as on the previous day and resuspended in 100 µl Blocking Solution. The cleared chromatin was added and incubated with the beads on a nutator for 3 hours at 4°C. After the incubation the beads were pelleted and the supernatant aspirated. 1 ml Wash Buffer was added and the sample was mixed gently by inverting the tube several (8-10) times to ensure that all beads were fully resuspended. The beads were then pelleted and the supernatant aspirated. This wash step was repeated another 4 times (5 washes total). A final wash with 1 ml TE buffer (10 mM Tris Experiments were conducted in triplicate for cell cycle analysis of mESCs.
Following aspiration of the Permeabilization Solution the cells were rinsed three times with 1 ml Wash Buffer and aspiration of the Wash Buffer after each rinse. Blocking was performed with 1 ml/200 µl (6-well/24-well) Blocking Buffer for one hour at room temperature. Blocking Buffer was aspirated and 500 µl/100 µl (6-well/24-well) Wash Buffer containing the primary antibody was added and the sample was incubated in a humidified light-tight chamber at 4°C overnight. On the next day Wash Buffer containing the primary antibody was aspirated. The sample was then washed 3 times for 5 minutes with 2 ml/400 µl (6-well/24-well) Wash Buffer with gentle shaking on a rotator at room temperature and aspiration of the Wash Buffer after each step. 500 µl/100 µl (6-well/24-well) Wash Buffer containing the fluorochrome-conjugated secondary antibody and DAPI (1 µg/ml final concentration) was added and the sample was incubated with gentle shaking on a rotator in a humidified light-tight chamber for one hour at room temperature. The sample was again washed 3 times for 5 minutes with 2 ml/400 µl (6-well/24-well) Wash Buffer with gentle shaking on a rotator at room temperature and aspiration of the Wash Buffer after each step. 1 ml/200 µl (6-well/24-well) DPBS was added and the sample was imaged with an inverted widefield microscope platform (Leica, DMi8).

Immunoprecipation (IP)
2 x 10 7 mESCs were used for each IP. mESCs were scraped off with a cell scraper and pelleted by centrifugation. The cell pellet was resuspended in DPBS and pelleted again. Cytoplasmic lysis was performed in 1 ml Lysis Buffer 1 by resuspending the cell pellet followed by incubation on a nutator for 5 minutes at 4°C. Nuclei were centrifuged for 5 minutes at 1500 g at 4°C, the supernatant (cytoplasmic fraction) aspirated and the nuclear pellet resuspended in 500 µl Lysis Buffer 2. The protein concentrations of the nuclear extracts were determined by a protein assay (BioRad, 5000006) with a spectrophotometer (Eppendorf, 6133000010) using disposable cuvettes (Eppendorf, 0030079353). Equal protein amounts were used for all samples of the same IP experiment and volumes were adjusted to 545 µl with Lysis Buffer 2. 45 ul of each sample was removed and used as Input. For each IP 50 µl Dynabeads Protein G (Thermo Fisher Scientific, 10004D) slurry was used. The beads were washed in 1 ml Lysis Buffer 2, collected with a magnetic stand (Thermo Fisher Scientific, 12321D) and the supernatant was removed. The bead wash step was repeated another time. The beads were resuspended in 500 µl Lysis Buffer 2 followed by addition of 5 µg of the appropriate antibody and were incubated for two hours on a nutator at 4°C.
The beads were collected with a magnetic stand and the supernatant was aspirated followed by two wash steps in 500 µl Lysis Buffer 2. The nuclear extracts were added to the beads and incubated with the beads on a nutator at 4°C overnight. After the incubation the beads were collected with a magnetic stand and the supernatant aspirated. 1 ml Lysis Buffer 2 was added and the sample was mixed gently by inverting the tube until all beads were fully resuspended. The beads were then collected with a magnetic stand and the supernatant aspirated. This wash step was repeated another 2 times (3 washes total). The beads were resuspended in 50 µl of 1 x SDS Laemmli Buffer and the inputs with 15 ul 4 x SDS Laemmli Buffer and all samples were boiled for 5 minutes at 95°C on a heating block (Thermo Fisher Scientific, 88870003). Inputs were centrifuged in a table top centrifuge for 5 minutes at full speed while bead IP samples were centrifuged for 2 minutes at 850 g, RT. Equal volumes of Input and IP samples were loaded and separated on a 4-20% gradient SDS-PAGE gel (Bio-Rad, 4561096 and 4561093) in running buffer (Bio-Rad, 1610772) using a Bio-Rad electrophoresis and blotting system (Bio-Rad, 1658033).

Western Blotting
Whole cell lysates from mESCs or differentiated neuroectoderm were obtained according to the following protocol. Medium was aspirated, DPBS was added and cells were detached by resuspension with a pipet or a cell scraper and transferred into a fresh tube. After centrifugation the supernatant was aspirated and the pellet resuspended in 1 ml ice cold DPBS. After repelleting in a cooled table top centrifuge and aspiration of the supernatant the cell pellet was resuspended in RIPA buffer followed by incubation on a nutator for 30 minutes at 4°C. The resulting cell lysate was centrifuged in a cooled table top centrifuge for 5 minutes at full speed and transferred to a new tube. The protein concentration of the supernatant was determined by a protein assay (BioRad, 5000006) with a spectrophotometer (Eppendorf, 6133000010) using disposable cuvettes (Eppendorf, 0030079353). The protein concentration was adjusted to a desired final protein concentration of 1-2 µg/µl with 4 x SDS Laemmli Buffer and additional RIPA Buffer to achieve a final 1 x SDS Laemmli Buffer concentration. 1 x SDS Laemmli Buffer samples were boiled for 5 minutes at 95°C on a heating block (Thermo Fisher Scientific, 88870003) and centrifuged in a table top centrifuge for 5 minutes at full speed, RT. From the supernatant equal amounts of protein (usually 10-30 µg) were loaded and separated on a 4-20% gradient SDS-PAGE gel (Bio-Rad, 4561096 and 4561093) in running buffer (Bio-Rad, 1610772) using a Bio-Rad electrophoresis and blotting system (Bio-Rad, 1658033). Proteins were transferred to either a PVDF or nitrocellulose membrane (Santa Cruz Biotechnology, sc-3723 and sc-3724) with the same Bio-Rad system using Western Transfer Buffer (Bio-Rad, 1610771 and 20% methanol) for 90 minutes at 400 mA. The membranes were blocked for 1 h at room temperature with gentle rocking in 5% dry milk in TBST buffer. Primary antibody incubation was performed overnight at 4°C with gentle rocking in 5% dry milk in TBST Buffer. On the next day the membranes were washed three times for 5 minutes with 10 ml TBST Buffer with gentle rocking and then incubated with horse radish peroxidase coupled secondary IgG-specific antibodies in TBST Buffer for 1 h at room temperature with gentle rocking. After an additional three washes for 5 minutes with 10 ml TBST Buffer with gentle rocking the membranes were developed using Immobilon Crescendo Western HRP Substrate (Millipore, WBLUR0500) and imaged on an Odyssey Fc imaging system (LI-COR, Model: 2800).
Buffer compositions and antibodies that were used and their concentrations are listed below. For histone western blots cells were trypsinized, resuspended in growth medium and pelleted by centrifugation. The pellet was resuspended in DPBS and cells were counted with an automated cell counter (Thermo Fisher Scientific, Countess II FL Automated Cell Counter). Equal numbers of cells were pelleted, resuspended in 1 x SDS Laemmli Buffer, boiled for 10 minutes at 95°C on a heating block and centrifuged in a table top centrifuge for 5 minutes at full speed, RT. From the supernatant equal volumes (usually 5-15 µl) were loaded and separated on a 4-20% gradient SDS-PAGE gel. The remainder of the workflow was identical to the one for the RIPA buffer extracted samples described above.
Briefly, 400,000 C57/BL6 Bruce-4 mESCs grown in feeder free culture conditions (DMEM KO medium) were transiently co-transfected with 500 ng of gRNA expression plasmid (Addgene, 43860), 1 µg Cas9 expression plasmid (Addgene, 43945), and 200 ng of pMaxGFP via nucleofection (Lonza, 4D-Nucleofector™ X-unit) using solution P3 and program CA137 in small (20 µl) cuvettes according to the manufacturer's recommended protocol. Cells were single cell sorted by FACS to enrich for GFP-positive (transfected) cells, clonally selected and verified for the desired targeted modification via targeted deep sequencing. Three clones were identified for each modification, assessed in relevant assays and then transitioned to chemically defined naïve culture conditions (2iL). Two clones for each gene knock-out were further used for more specific assays in this paper. The sequences for each gRNA and relevant primers are listed in Table S1.

Mapping reads and visualizing data
ChIP-seq raw reads were aligned to the mouse and Drosophila melanogaster hybrid reference genomes (mm9+dm3) using BWA (version 0.7.12; default parameter) and duplicated reads were then marked with Picard (version 1.65), with only nonduplicated reads kept by samtools (version 1.3.1, parameter ''-q 1 -F 1024''). Mapped reads were then split into two bam files (mapped to mm9 and dm3 respectively). For data quality control and to estimate the fragment size, the nonduplicated version of SPP (version 1.11) was used to calculate the relative strand correlation value with support of R (version 3.3.1). To visualize ChIP-seq data on the integrated genome viewer (IGV) (version 2.3.82), we utilized genomeCoverageBed (bedtools 2.25.0) to obtain genome-wide coverage in BEDGRAPH file format and then converted it to bigwig file format by bedGraphToBigWig. The bigwig files were scaled to 15 million reads to allow comparison across samples.

Peak calling, annotation and motif analysis
MACS2 (version 2.1.1 20160309) was used to call narrow peaks (DNTTIP1, HDAC1, H3K27ac and H3K4me3) with option "nomodel" and "extsize" defined as fragment size estimated by SPP and a FDR corrected p-value cutoff of 0.05. For broad peak/domain calling (H3K4me1 and H3K27me3). SICER (version 1.1, with parameters of redundancy threshold 1, window size 200 bp, effective genome fraction 0.86, gap size 600 bp, FDR 0.00001 with fragment size defined above) was used for broad peak/domain calling (H3K4me1 and H3K27me3). Enriched regions were identified by comparing the ChIP library file to input library file. Peak regions were defined to be the union of peak intervals from two ChIP replicates of WT, Dnttip1 KO or Elmsan1 KO mESCs, respectively. Promoter regions were defined as ± 1000 bp from a TSS based on the mouse RefSeq annotation. Genomic feature annotation of peaks was carried out by annotatePeaks.pl, a program from HOMER suite (v4.8.3, http://homer.salk.edu/homer/). HOMER software was used to perform de novo motif discovery and to check for enrichment of known motifs from a set of DNTTIP1 peaks (associated with up-or downregulated genes in Dnttip1 KO versus WT mESCs) or all DNTTIP1 peaks.

Spike-in normalization and differential analysis
ChIP-seq raw read counts were reported for each region/each sample using bedtools 2.25.0. Spikein normalization was performed by counting Drosophila reads and mouse reads in each ChIP sample and corresponding input sample and using those counts to generate a normalization factor for each sample, which was calculated as (ChIP_dm3.reads/ChIP_mm9.reads)/(Input_dm3.reads/Input_mm9.reads). Raw read counts were voom normalized and statistically contrasted using the pipeline limma in R (version 3.3.1). The normalization factor defined above was used to modify the mouse library size in edgeR (version 3.16.5) for CPM calculation and differential analysis. An empirical Bayes fit was applied to contrast Dnttip1 KO and Elmsan1 KO samples to WT samples and to generate log2 fold changes, p-values and false discovery rates for each peak region. Histograms showing average ChIP-seq intensity over gene bodies were generated using ngsplot (v2.61).

RNA-seq analysis
Total stranded RNA sequencing data were generated and mapped against mouse genome assembly NCBIM37.67 using the StrongArm pipeline described previously (Wu, Barnhill et al., 2016). Gene level quantification values were obtained with HT-seq based on the GENCODE annotation (vM20) and normalized by the TMM method with edgeR (version 3.16.5). Differential expression analysis was performed with the voom method applying the limma pipeline in R (version 3.3.1).
Significantly up-and down-regulated genes were defined by at least a 1.5 fold change in gene expression and a p-value < 0.01. Reactome and gene set enrichment analysis (GSEA) were carried out using GSEA or EnrichR, respectively. Gene expression log2CPM (counts per million) values were computed for heatmap and box plot visualization. Log2 FPKM gene expression values were applied for bar plot diagrams.

Neurite and neuron analysis
The neuron-specific markers TUBB3 and MAP2 were used to identify neurons by IF as described under "Neural Differentiation". NeuriteTracer, a plugin of ImageJ (version 1.52a), was used to manually trace the length and number of neurites per neuron and to automatically detect nuclei based on DAPI staining using the IF raw image data for each genotype. The length of neurites was Statistical analysis of qPCR and qRT-PCR data qPCR signals for manual ChIPs were calculated as % of input (% Input) from technical duplicates.
Error bars represent the standard deviation from technical duplicates. For qRT-PCR relative gene expression levels were calculated applying the delta-delta Ct method (2 -ΔΔCt ). Error bars depict the standard deviation from technical triplicates. Statistical significance was determined by applying the Student's unpaired t-test.

Public datasets and additional bioinformatics analysis
HDAC1 ChIP-seq data from WT mESCs was obtained from gene expression omnibus (GEO): GSM1335482. Data was analyzed as described under "Quantification and statistical analysis" within the "ChIP-seq analysis" section if applicable.

Data availability
The accession number for the RNA-seq and ChIP-seq datasets reported in this paper is GEO: GSE131062. G RNA-seq heatmaps depicting down-and upregulated genes from a gene set of neurodevelopmental genes that is mutually regulated by DNTTIP1 and ELMSAN1 (FC < -1.5 or > 1.5, p < 0.05). The color scale depicts the z-score of normalized gene expression (log2CPM). Unpaired t-test was performed throughout where ***, p ≤ 0.001; **, p ≤ 0.01; *, p ≤ 0.05; and ns, p > 0.05 is not significant.   Figure 3F and G. IgG was used as a control antibody. Unpaired t-test was performed throughout where **, p ≤ 0.01; and ns, p > 0.05 is not significant. One-way ANOVA was performed throughout where ***, p ≤ 0.001; **, p ≤ 0.01; and *, p ≤ 0.05.  and/or UNC5B. MAP2 IF staining was performed after 12 days of differentiation. Nuclei were stained with DAPI. To facilitate analysis the neuronal cell body (blue) and its neurites were manually traced with ImageJ software and for each sample one traced neuron is displayed in the inlet. The white scale bar represents 50 µm.

Figure Legends
C, D Quantification of C neurite length and D the total number of neurites per neuron from the MAP2 IF staining in Fig. 6e using ImageJ. C Neurite length was divided into two categories of short neurites < 50 µm (green box plots) and longer neurites ≥ 50 µm (white box plots). C, D The neurites of 200 neurons were assessed per sample.
F WB for the indicated MiDAC components from total cell lysates of WT, Elmsan1 KO1 and Elmsan1 KO2 mESCs. Actin is the loading control.
G IPs were carried out with IgG and DNTTIP1 antibodies from nuclear extracts of WT and        Figure S7C and E are highlighted by orange boxes.
C qPCR from manual ChIP experiments against DNTTIP1, ELMSAN1 and HDAC1 from WT, Dnttip1 KO1 and Elmsan1 KO1 NE targeting the promoter of the Robo3 locus as highlighted in Figure S7A. IgG was used as a control antibody.
E qPCR from manual ChIP experiments against DNTTIP1, ELMSAN1 and HDAC1 from WT, Dnttip1 KO1 and Elmsan1 KO1 NE targeting the promoter of the Unc5b locus as highlighted in Figure S7B. IgG was used as a control antibody.  A Experimental outline of assay to rescue neurite outgrowth defects in Dnttip1 KO1 or Elmsan1 KO1 NE as described in Figure 5A-C and 6B-D.
B Experimental outline of the chamber assay to rescue the neuronal network formation defects of granule neuron progenitor (GNP)-derived neurons co-cultured with Dnttip1 KO1 or Elmsan1 KO1 NE as described in Figure S8C, D or CM of Dnttip1 KO1 or Elmsan1 KO1 NE as described in Unpaired t-test was performed throughout where ***, p ≤ 0.01. Tables   Table S1. List of guide RNA and deep sequencing primer sequences used for CRISPR/Cas9mediated genome editing of Dnttip1 and Elmsan1 in mESCs.