Loss of Ezh2 in the medial ganglionic eminence alters interneuron fate, cell morphology and gene expression profiles

Enhancer of zeste homolog 2 (Ezh2) is responsible for trimethylation of histone 3 at lysine 27 (H3K27me3), resulting in gene repression. Here, we explore the role of Ezh2 in forebrain GABAergic interneuron development. Loss of Ezh2 increases somatostatin-expressing (SST+) and decreases parvalbumin-expressing (PV+) interneurons in multiple brain regions. We also observe fewer MGE-derived interneurons in the first postnatal week, indicating reduced interneuron production. Intrinsic electrophysiological properties in SST+ and PV+ interneurons are normal, but PV+ interneurons display increased axonal complexity in Ezh2 mutant mice. Single cell multiome analysis revealed differential gene expression patterns in the embryonic MGE that are predictive of these cell fate changes. Lastly, CUT&Tag analysis revealed differential H3K27me3 levels at specific genomic loci, with some genes displaying a relative increase in H3K27me3 indicating they may be resistant to epigenetic modifications. Thus, loss of Ezh2 in the MGE alters interneuron fate, morphology, and gene expression and regulation.


SUMMARY 28
Enhancer of zeste homolog 2 (Ezh2) is responsible for trimethylation of histone 3 at lysine 29 27 (H3K27me3), resulting in gene repression. Here, we explore the role of Ezh2 in forebrain 30 GABAergic interneuron development. Loss of Ezh2 increases somatostatin-expressing (SST+) 31 and decreases parvalbumin-expressing (PV+) interneurons in multiple brain regions. We also 32 observe fewer MGE-derived interneurons in the first postnatal week, indicating reduced 33 interneuron production. Intrinsic electrophysiological properties in SST+ and PV+ interneurons 34 are normal, but PV+ interneurons display increased axonal complexity in Ezh2 mutant mice. 35 Single cell multiome analysis revealed differential gene expression patterns in the embryonic 36 MGE that are predictive of these cell fate changes. Lastly, CUT&Tag analysis revealed differential 37 H3K27me3 levels at specific genomic loci, with some genes displaying a relative increase in 38 H3K27me3 indicating they may be resistant to epigenetic modifications. Thus, loss of Ezh2 in the 39 MGE alters interneuron fate, morphology, and gene expression and regulation. While performing cell counts in the CA2/3 region, we observed Tom+ cell bodies that were 167 too small to be interneurons, and we did not observe these cells in other brain regions (Fig. 4A). 168 Counting these cells separately, we found a very strong reduction of these CA2/3-specific cells in 169 the KO hippocampus (Fig. 4B). We stained WT hippocampal sections with various glia and 170 microglia markers and found that many of these small Tom+ cell bodies were Olig2+, indicating 171 that they are likely oligodendrocytes (Fig. 4C). This decrease in oligodendrocytes in Ezh2 KO 172 mice is consistent with findings demonstrating that loss of Ezh2 can block or delay gliogenesis 65,66 . 173 174

Normal intrinsic properties but altered morphology of Ezh2 KO interneurons 175
To characterize the intrinsic physiology of MGE-derived interneurons in KO mice, we 176 performed patch clamp recording of layer V/VI Tom+ cortical cells in acute brain slices. Cells were 177 classified as FS or NFS based on their intrinsic electrophysiological properties characterized 178 under current clamp-recording. NFS cells had larger half-width, input resistance and membrane 179 time constant/Tau, but smaller rheobase compared to FS cells. We analyzed action potential 180 shapes, resting membrane potential, spike adaptation ratio, afterhyperpolarization (AHP) 181 amplitude, input resistance and rheobase. There were no differences in intrinsic properties of FS 182 or NFS cortical interneurons between WT and KO mice (Supplementary Fig. 4). 183 However, reconstructions of recorded cells did reveal morphological changes in FS cells. 184 The axonal arbor of FS cells from KO mice were larger and more complex compared to WT cells 185 (Fig. 5A). Sholl analysis revealed a significant increase in axon intersections, axon length and 186 axon volume in FS KO cells, while there were no changes in dendritic arbors (Fig. 5B). This 187 increased axonal arbor is similar to what was observed when trkB signaling was blocked in PV 188 cells 67 . Thus, while intrinsic properties of Tom+ MGE-derived cortical interneurons were normal 189 in Ezh2 KO mice, FS cells displayed greater complexity in their axonal arbors compared to FS 190 cells from WT mice. 191 192

Loss of Ezh2 in cycling progenitors is required for cell fate changes 193
We next wanted to determine at what stage of development loss of Ezh2 results in cell 194 fate changes. Ezh2 is enriched in cycling progenitors throughout the embryonic brain (Fig. 1A), 195 but Ezh2 may play a critical at other stages as well. To investigate this possibility, we generated 196 Dlx5/6-Cre;Ezh2 F/F ;Ai9 conditional KO mice in which loss of Ezh2 is restricted to postmitotic 197 neurons arising from the ganglionic eminences. We verified that Ezh2 is still expressed in MGE 198 ventricular zone cycling progenitors in Dlx5/6-Cre KO mice ( Supplementary Fig. 5A). There were 199 no differences in the densities or percent of SST+ and PV+ cells in the cortices of these Dlx5/  Cre KO mice ( Supplementary Fig. 5B), indicating that Ezh2 is required in cycling MGE progenitors 201 for proper interneuron fate and maturation. 202 A wave of programmed apoptosis occurs between the first and second postnatal weeks 203 that eliminates ~20-40% of cortical interneurons [68][69][70][71] . To determine if there were changes in the 204 overall production of MGE-derived interneurons during embryogenesis, we counted the number 205 of Tom+ cells in the cortex at P5, prior to programmed apoptosis. We found a significant decrease 206 in the number of Tom+ cells in the KO cortex compared to WT at P5 (Fig. 6A-B). This finding 207 supports the hypothesis that loss of Ezh2 in cycling MGE progenitors decreases the overall 208 production of MGE-derived cortical interneurons. The stronger decrease of Tom+ cells in the 209 superficial layers is consistent with preemptive depletion of the progenitor pool, which would (1)  210 primarily affect the later-born cells in the superficial layers, and (2) lead to more prominent loss of 211 PV+ interneurons due to their bias production at later embryonic timepoints compared to SST+ 212 Since there was a general increase in SST+ interneurons in the Ezh2 KO mouse, we 228 examined SST expression in this single cell dataset. While no obvious differences in SST 229 expression was apparent at E12.5, we did observe an increase in SST expression in the MGE of 230 E15.5 KO compared to WT (Fig. 7C). While PV is not expressed in the embryonic mouse brain, 231 two genes that are enriched in PV-fated interneurons and critical for their development are the 232 transcription factors Mef2c and Maf 21,73 . We found that both genes are strongly reduced in E15.5 233 KO MGEs compared to WT (Fig. 7D). Complementing this gene expression analysis, the motifs 234 for these transcription factors are enriched in accessible regions of the E15.5 WT MGE compared 235 to KO mice (Fig. 7E). Thus, our gene expression analysis reveals an apparent increase in SST 236 expression and decrease in Maf and Mef2c expression in the MGE of E15.5 KO mice, which is 237 consistent with the increase in SST+ and decrease of PV+ interneurons in Ezh2 KO brains. 238 Additionally, WT and KO cells appeared to display differential abundance (DA) in specific 239 regions of the UMAP plot, most notably in the E15.5 dataset ( Fig. 7F and Supplementary Fig. 6B). 240 To confirm this observation, we performed DA analysis using DA-seq 74 . DA-seq determines a DA 241 score for each cell, whereby a cell that is surrounded by KO cells in a k-nearest neighbor (KNN) 242 graph has a score closer to +1, and a cell surrounded by WT cells has a score closer to -1. This 243 DA score does not rely upon previously identified clusters, and it does not require similar cell 244 numbers between different conditions. Our DA-seq analysis revealed that most cells with a DA 245 score above +0.7 or below -0.7 were from the E15.5 MGE (Fig. 7G), whereas cells from the E12.5 246 MGE displayed little differential abundance. Furthermore, most cells with a DA score below -0.7 247 (blue, WT bias) were in the clusters enriched for Maf and Mef2c, which are putative PV+ 248 interneurons (Fig. 7G). The cluster containing SST+ cells contained numerous cells with DA score 249 above 0.7 (red, KO bias) (Fig. 7G). Thus, we observe a differential abundance of WT and KO 250 cells specifically in the E15.5 dataset, with an increase of KO cells in clusters expressing SST 251 and an increase of WT cells in clusters where Maf and Mef2c are strongly enriched. 252 To specifically focus on these clusters enriched for SST-and PV-fated interneurons, we 253 extracted the four clusters containing these cells from the dataset (clusters 3, 4, 5 and 7) (Fig.  254 7A,H). Using thresholds of a Log2 fold change (FC) > ± 0.2 and a false discovery rate (FDR) of 255 Log10P < 10 -6 , we identified 59 differentially expressed genes at E12.5 (46 downregulated and 15 256 upregulated in KO) and 176 differentially expressed genes at E15.5 (46 downregulated and 130 257 upregulated in KO) (Fig. 7I and Supplementary Tables 1-2). Notably, both Maf and Mef2c were 258 significantly enriched in the E15.5 WT MGE whereas SST was upregulated in the KO MGE, 259 consistent with the observations above (Fig. 7I). Additionally, the MGE-specific transcription 260 factors Nkx2.1 and Lhx6 were also upregulated in the KO MGE in these clusters. 261 Since differentially expressed genes of interest were restricted to E15.5, we reexamined 262 the integrated E15.5 RNA+ATAC dataset alone ( Supplementary Fig. 6B). In this dataset, SST 263 was strongly enriched in cluster 10 (Fig. 7J). The top gene expressed in this cluster was 264 Phosphodiesterase 1A, Pde1a, which is also significantly upregulated in the E15.5 KO MGE (Fig.  265 7I). Based on the Allen Brain Institute's single cell transcriptomic adult mouse brain dataset, 266 Pde1a is enriched in many SST+ interneuron subtypes while it's excluded from PV+ interneurons 267 H3K27me3 levels are strongly downregulated in the MGE of Ezh2 KO mice (Fig. 1). To 275 look at H3K27me3 changes at specific genes, we performed bulk CUT&Tag 75 with a H3K27me3 276 antibody in the MGE of WT and KO mice. We did not normalize total reads to a spike-in or E. coli 277 DNA control so that the global downregulation of H3K27me3 in the KO was intentionally ignored 278 from this analysis and instead we could focus on the relative changes of H3K27me3 levels at 279 specific loci between genotypes. We performed 3 biological replicates for each age and genotype, 280 This indicates that specific genomic loci are more susceptible (e.g., Foxp4) or resistant (e.g., 300 Nkx2.1) to H3K27me3 loss in the absence of Ezh2. The mechanism by which loss of Ezh2 301 generates these differential effects at genomic loci, and how these changes in H3K27me3 levels 302 relate to gene expression, require further investigation. 303

DISCUSSION 305
There is growing evidence that dysregulation of epigenetic mechanisms can lead to a 306 variety of human diseases and neurodevelopmental disorders [36][37][38]80,81 . For example, postmortem 307 tissue from schizophrenic patients displays alterations in genome organization and other 308 epigenomic characteristics 82,83 . Additionally, many genes associated with neurological and 309 psychiatric diseases are enriched in immature interneurons during embryonic 310 development 9,10,84,85 . Thus, advancing our knowledge of gene regulation mechanisms during 311 interneuron development is critical for understanding both normal development and disease 312 etiologies. 313 In this study, we find that loss of Ezh2 in the MGE decreases the density and proportion 314 of PV+ cells, often with a corresponding increase in SST+ cells. This shift in interneuron fate was 315 most prominent in the cortex and the CA2/3 region of the hippocampus, with an overall decrease 316 in total MGE-derived interneurons also observed in the cortex. A decrease in PV+ cells was 317 observed in the CA1, DG and striatum without a significant increase in SST+ cells  Supplementary Fig. 1-3). In the hippocampus, we also observe an increase in MGE-derived 319 nNos+ cells in the KO (Fig. 3). Unlike PV+ and SST+ interneurons, the spatial and temporal origin 320 of hippocampal nNos+ in the MGE is not well characterized, so it's unclear how loss of Ezh2 321 increased nNos+ cells. These phenotypes were due to Ezh2 function in cycling MGE progenitors, 322 as no changes were observed when Ezh2 was removed in postmitotic MGE cells (Supplementary 323 the top gene expressed in the SST-enriched cluster at E15.5 is Pde1a, which is expressed by 362 many mature SST+ interneuron subtypes but excluded from PV+ interneurons (Fig. 7J). In sum, 363 these transcriptional and cellular differences in the MGE are likely determinative for the shifts in 364 SST+ and PV+ interneurons in the adult brain of Ezh2 KO mice. 365 Despite the global downregulation of H3K27me3, we found that loss of Ezh2 had 366 differential effects on relative H3K27me3 levels at specific gene loci. One of the loci most 367 susceptible to Ezh2 loss was the Foxp4 locus, with a significant loss of H3K27me3 in the KO 368 MGE at E12.5 and E15.5 ( Fig. 8C-D). Surprisingly, a significant increase in Foxp4 expression 369 (predicted based on H3K27me3 downregulation) was not observed. Foxp4 is enriched in the 370 LGE 79 , but it has not been well-studied in neurodevelopment. In heterologous cell lines, FOXP4 371 can directly interact with the transcription factors SATB1, NR2F1 and NR2F2 89 , all of which are 372 critical for development of MGE-derived interneurons 90,91 . Whether these interactions occur in the 373 developing brain is unclear. A study on medulloblastoma found that Foxp4 and Ezh2 are both 374 targets of the microRNA miR-101-3p 92 , indicating the function of these genes might be linked in 375 some scenarios. Why the Foxp4 locus is extremely sensitive to Ezh2 loss requires further study. 376 In the Ehz2 KO MGE, we observed a significant relative increase in H3K27me3 levels at 377 several transcription factors critical for development of MGE-derived interneurons: Nkx2.1 and 378 Dlx1/2 locus at E12.5 and Lhx6 and Dlx5/6 locus at E15.5. This raises the possibility that some 379 genes playing critical roles in fate determination may be more resistant to epigenetic changes, in 380 this case loss of Ezh2. For example, the interaction between Sox2, a transcription factor essential 381 in the epiblast of pre-implantation embryos, and a critical enhancer downstream is maintained 382 even when artificial boundaries are introduced between these regions 93 . As Nkx2.1 is a 'master 383 regulator' of MGE fate, and the Nkx2.1 locus displays unique chromatin organization in the MGE 23 , 384 it could be more resistant to epigenetic modifications. 385 Similar to Foxp4, we did not observe a corresponding change in the global expression of 386 these transcription factors that is predictive of these H3K27me3 changes. In fact, we actually 387 observed a significant increase in Nkx2.1 and Lhx6 expression in the postmitotic SST-and PV-388 fated clusters (Fig. 7I), which is in contrast to the predicted decreased expression based on 389 H3K27me3 levels. While our previous results showed a strong relationship between H3K27me3 390 and gene repression in the embryonic mouse brain 23 , the correlation here is weaker. It will be 391 interesting to explore gene-specific changes in the resistance or susceptibility to epigenetic 392 changes going forward, both in terms of normal development and regarding neurodevelopmental 393 and psychiatric diseases. 394 395

METHODS 396
Animals 397 All experimental procedures were conducted in accordance with the National Institutes of 398 Health guidelines and were approved by the NICHD Animal Care and Use Committee (protocol 399 #20-047). The following mouse lines were used in this study: Nkx2.1-Cre (Jax# 008661) 60 , Ezh2 F/F 400 (Jax# 022616) 94 , Dlx5/6-Cre (Jax# 008199) 95   Sections were restricted to the anterior and middle hippocampus; the posterior hippocampus was 489 excluded due to greater variability in interneuron density in this region. Hippocampal sections 490 were divided into CA1, CA2/3 and DG regions using DAPI staining (Fig. 3A). Small Tom+ cell 491 bodies (identified as oligodendrocytes) in CA2/3 (Fig. 4)

In vitro electrophysiology 501
Slice preparation: Mice were anesthetized with isoflurane (5% isoflurane (vol/vol) in 100% 502 oxygen), perfused transcardially with an ice-cold sucrose solution containing (in mM) 75 sucrose, 503 87 NaCl, 2.5 KCl, 26 NaHCO3, 1.25 NaH2PO4, 10 glucose, 0.5 CaCl2, and 2 MgSO4, saturated 504 with 95% O2 and 5% CO2 and decapitated. Brain was rapidly removed from the skull and 505 transferred to a bath of ice-cold sucrose solution. Coronal slices of 300 µm were made using a 506 vibratome (Leica Biosystems) and were stored in the same solution at 35°C for 30 min and at 507 room temperature (RT) for an additional 30-45 min before recording. 508 Electrophysiology: Whole-cell patch clamp recordings on tdTomato+ cells in cortical layers 509 V/VI cells were performed in oxygenated ACSF containing (in mM) 125 NaCl, 2.5 KCl, 26 510 NaHCO3, 1.25 NaH2PO4, 10 glucose, 2 CaCl2 and 1 MgCl2. The ACSF was equilibrated with 95% 511 O2 and 5% CO2 throughout an entire recording session which typically lasted between 30 minutes 512 to 1 hour to ensure sufficient permeation of neurobiotin. Recordings were performed at 30°C-513 33°C. Electrodes (3-7 MΩ) were pulled from borosilicate glass capillary (1.5 mm OD). The pipette 514 intracellular solution contained (in mM) 130 potassium gluconate, 6.3 KCl, 0.5 EGTA, 10 HEPES, 515 5 sodium phosphocreatine, 4 Mg-ATP, 0.3 Na-GTP and 0.3% neurobiotin (pH 7.4 with KOH, 280-516 290 mOsm). Membrane potentials were not corrected for the liquid junction potential. During 517 patching, cell-attached seal resistances were >1 GΩ. Once whole-cell configuration was 518 achieved, uncompensated series resistance was usually 5-30 MΩ and only cells with stable series 519 resistance (<20% change throughout the recording) were used for analysis. Data were collected 520 using a Multiclamp 700B amplifier (Molecular Devices), low-pass filtered at 10 kHz and digitally 521 sampled at 20 kHz, and analyzed with pClamp10 (Molecular Devices). To characterize the 522 intrinsic membrane properties of neurons, hyperpolarizing and depolarizing current steps were 523 injected at 0.1 Hz under current-clamp configuration. 524 Data analysis: All intrinsic properties were measured in current-clamp configuration and 525 calculated from 800 millisecond-long current injections unless noted otherwise. The resting 526 membrane potential (in mV) was measured with 0 pA current injection a few minutes after entering 527 whole-cell configuration. All other properties were measured holding the cell at -70 mV. Input 528 resistance (in MΩ) was calculated using Ohm's law from averaged traces of 100 ms long negative 529 current injections of -20 pA. Action potential (AP) threshold was calculated as the potential when 530 voltage change over time was 10 mV/ms using the first observed spike. AP amplitude (in mV) 531 was calculated as the time difference in potential from the spike peak to spike threshold. AP/spike 532 half-width (in ms) was calculated as the difference in time between the ascending and descending 533 phases of a putative spike at the voltage midpoint between the peak of spike and spike threshold. approximately 20 Hz firing. Afterhyperpolarization (AHP) amplitude was calculated as the 537 difference between AP threshold and the most negative membrane potential after the AP, 538 measured on the response to the smallest current step evoking an AP (Rheobase). Membrane 539 time constant (in ms) was determined from a monoexponential curve best fitting the falling phase 540 of the response to a small hyperpolarizing current step. 541

CUBIC clearing and streptavidin staining 543
After performing electrophysiological recordings, brain slices were fixed in 4% PFA in 0.1M 544 PB and kept overnight at 4°C and then kept in 20% sucrose (in PB). The brain slices were 545 processed for CUBIC (Clear, Unobstructed Brain/Body Imaging Cocktails and Computational 546 analysis) clearing 98 . Slices were first washed with 0.1M PB (3 times for 10 min) at RT, followed 547 by immersion in CUBIC reagent 1 for two days at 4°C. After two days of incubation, slices were 548 10 Tris-HCl (Ph.7.4), 10 NaCl, 3 CaCl2, 1 DTT, with 0.1% Tween-20, 1.5% BSA and 1 U/μL in 591 nuclease-free water, 5 mL per sample). Transfer lysed nuclei suspension through pre-wetted filter, 592 then rinse dounce with 1 mL Multiome Wash Buffer and transfer through filter. Divide nuclei 593 suspension into 2 pre-chilled 2 mL tubes and spin at 500 g for 5 minutes at 4 o C. AGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT -q 20 --minimum-length 25. Trimmed reads 619 were mapped to mouse reference genome (GRCm38/mm10) using bowtie2 100 v2.4.2 with the 620 following parameters: --no-unal -N 1 --no-mixed --no-discordant --very-sensitive-local -local --621 phred33 -I 10. Aligned reads in sam files were further processed to remove multimappers if MAPQ 622 was less than 10 (-q 10) and then sorted using samtools 101 v1.12. Aligned reads that intersected 623 blacklist regions 102 were removed and saved to bam files using bedtools 103 v2. The Cell Ranger ARC (v2.0.0) pipeline was used to process sequenced libraries with 672 default parameters unless otherwise noted. Demultiplexed FASTQ files were generated by 673 cellranger-arc mkfastq from BCL files. Reads were aligned to custom-built mouse 674 (GRCm38/mm10) reference genome modified to include tdTomato using cellranger-arc count. 675 Reads with de-duplicated and valid cell barcodes were used to build gene-by-barcode (scRNA-676 seq) and peak-by-barcode (scATAC-seq) matrices by cellranger-arc count per genotype. 677 Individual matrices were aggregated to a single feature-barcode matrix file containing every 678 genotype using cellranger-arc aggr without depth normalization (--normalize=none). 679 680 scRNA-seq data analysis 681 Seurat: An aggregated feature-barcode matrix was used as input to Seurat 110 (v4.0.5, 682 https://satijalab.org/seurat/) in R (v4.1.1, https://cran.r-project.org/). After imputing missing values 683 to zero in metadata, outlier removal was performed on the number of counts per gene and percent 684 reads mapping to mitochondrial genome (mitochondrial percentage). Lower limits for the number 685 of counts per gene and mitochondrial percentage were set to 100 counts per gene and three 686 standard deviations (SD) below the mean, respectively. Upper limits were set to three SD above 687 the mean for both metrics. Negative datapoints created by subtraction of three SD from the mean 688 were reset to 1, while datapoints that exceeded the upper limits were reset to the maximum 689 datapoint. Cells were removed if they were more extreme then the upper/lower limits, or if they 690 were eliminated from the scATAC-seq dataset during QC. The numbers of remaining cells were 691 following: WT E12.5: 6,391; WT E15.5: 11,477; Het E12.5: 6,608; Het E15.5 10,027; KO E12.5: 692 8,546; KO E15.5: 8,607. Remaining cells were proceeded to the normalization workflow using 693 Seurat::SCTransform using default parameters. For integration of scRNA-seq datasets from 694 E12.5 and E15.5, 3,000 variable features were found using Seurat::SelectIntegrationFeatures on 695 SCTransformed data. Prior to integration, anchors were identified using 696 Seurat::FindIntegrationAnchors with the parameters dims, anchors.features, and 697 normalization.method set to 1:30, the 30,000 variable features, and SCT, respectively. The 698 integration was performed using Seurat::IntegrateData with identical dims and 699 normalization.method to those from Seurat::FindIntegrationAnchors, along with the computed 700 anchors. Dimensionality reduction was performed using Seurat:: RunPCA  was computed by Seurat after the multimodal integration of age-integrated scRNA-seq and 745 scATAC-seq datasets. Cells were projected onto 2D space based on UMAP coordinates, which 746 were computed by Seurat after multimodal integration of age-integrated scRNA-seq and scATAC-747 seq datasets. DA cells were determined by running DAseq::getDAcells, with the k.vector 748 parameter set to every 50 between 50 and 500, and DAseq::updateDAcells with the pred.thres 749 parameter set to +/-0.7. DA-seq conducted a random permutation test on abundance scores, 750 using a threshold of +/-0.7, to identify cells with an abundance score greater than 0.7 or less than 751

Statistics and reproducibility 771
Cell counts: All cell counts were performed by hand and blind to genotype. Number of 772 brain sections per mice and mice per genotype are described above and in figure legends. One-773 way ANOVA was used to compare WT, Het and KO for all brain regions, followed by Tukey's 774 Multiple Comparison Test to identify significant differences between conditions. All statistical 775 analysis was performed on GraphPad Prism (version 9.4.1). All raw cell counts, ANOVA F-and 776 P-values, and results of Tukey's multiple comparison tests related to Figures 2-4, Figure 6 For Western Blots, we desired n > 2 biological replicates for each genotype to determine the 787 percent of decreased H3K27me3 signal in the Ezh2 KO. For cell counts with Nkx2.1-Cre mice, 788 we wanted n > 5 mice per genotype from at least 3 different litters. For cell counts with Dlx5/6-789 Cre mice, we wanted n > 4 WT and KO mice per genotype from at least 3 different litters. For 790 electrophysiological recordings, we wanted to record from > 10 neurons for each condition. For 791 the single cell Multiome experiments, we wanted a minimum of 5,000 high-quality sequenced 792 nuclei per condition (age and genotype), which should be sufficient to identify significant 793 differences between conditions. This goal required ~15,000 nuclei input for each sample (with the 794 expectation of recovering ~30-70% of nuclei/cells for each reaction based on 10x Genomics 795 recommendations and our previous experience). Viable nuclei that passed QC ranged from 796 6,391-11,477 nuclei per condition (see above). Per standard single cell sequencing protocols, 797 nuclei that did not pass stringent QC measurements (nCount_RNA and % mitochondria reads for 798 snRNA; nCount_ATAC, nucleosome_signal and TSS enrichment for snATAC) in the Multiome 799 datasets were excluded from analysis (as detailed in Supplementary Figure 6). For the CUT&Tag 800 experiments, we strove for 100,000 nuclei for each reaction (actual range from 90,000-120,000 801 per reaction), with n = 3 biological replicates for each condition (age and genotype). All 802 computational and statistical analysis are discussed in detail above and/or  28. Zhao Z, et al superficial (layers I-III) and deep (layers IV-VI) cortical layers defined by differential DAPI 1219 densities. D = dorsal, V = ventral. B. Graphs displaying the density of Tom+, SST+ and PV+ cells 1220 in WT, Het and KO mice. C. Graphs displaying the percent of Tom+ cells expressing SST or PV 1221 in WT, Het and KO mice. For all graphs, statistics are one-way ANOVA followed by Tukey's 1222 multiple comparison tests: * = p < .05, ** = p < .005, *** = p < .0005, **** = p < .0001. n = 5 WT, 1223 5 Het and 6 KO brains from a total of 4 different litters. 1224 Nkx2.1-Cre;Ezh2;Ai9 WT and KO mice depicting axons (blue) and dendrites (red). Scale bar = 1254 20 µm. B. Sholl analysis reveals increased axon intersections, axon length and axon volume in 1255 FS cortical interneurons from KO mice compared to WT littermates. No significant differences 1256 were found in the dendritic arbors. All statistics are two-way ANOVA followed by Holm-Sidak's 1257 test: ** = p < .005, *** = p < .0005; n = 6 cells from 4 WT mice and 7 cells from 4 KO mice. 1258 1259 1260