Amelioration of Brain Histone Methylopathies by Balancing a Writer-Eraser Duo KMT2A-KDM5C

Abstract Histone H3 lysine 4 methylation (H3K4me) is extensively regulated by seven writer- and six eraser-enzymes in mammals. Nine H3K4me enzymes are associated with neurodevelopmental disorders to date, indicating their important roles in the brain. Opposing activities of writer-eraser enzymes highlight activity modulation as a therapeutic strategy. However, interplay among H3K4me enzymes in the brain remains largely unknown. Here, we show functional interactions of a writer-eraser duo, KMT2A and KDM5C, which are responsible for Wiedemann-Steiner Syndrome (WDSTS), and mental retardation X-linked syndromic Claes-Jensen type (MRXSCJ), respectively. Despite opposite enzymatic activities, the WDSTS and MRXSCJ mouse models, deficient for either Kmt2a or Kdm5c, shared similar brain transcriptomes, reduced dendritic spines, and increased aggression. Double mutation of Kmt2a and Kdm5c partially corrected altered H3K4me landscapes and transcriptomes from each single mutants, and clearly reversed dendritic morphology deficits and key behavioral traits including aggression. Thus, our study uncovers common yet mutually-suppressive aspects of WDSTS and MRXCJ and provides a proof of principle for balancing a single writer-eraser pair to ameliorate their associated disorders.


Introduction 39
Dysregulation of histone methylation has emerged as a major contributor of 40 neurodevelopmental disorders (NDDs) such as autism spectrum disorder and 41 intellectual disability (1-3). Histone methylation can be placed on a subset of lysines and 42 arginines by histone methyltransferases (writer enzymes) and serves as a signaling 43 platform for a variety of nuclear events including transcription (4). Reader proteins 44 specifically recognize methylated histones, thereby converting methylation signals into 45 higher-order chromatin structures (5). Histone methylation can be removed by a set of 46 histone demethylases (eraser enzymes) (6). All three classes of methyl-histone 47 regulators are heavily mutated in NDDs, indicating critical, yet poorly-understood, roles 48 of histone methylation dynamics in brain development and function (7-9). 49 50 Histone H3 lysine 4 methylation (H3K4me) is one of the most well-characterized histone 51 modifications. H3K4me is primarily found at transcriptionally active areas of the 52 genome. The three states, mono-, di-, and tri-methylation (H3K4me1-3), uniquely mark 53 gene regulatory elements and play pivotal roles in distinct steps of transcription. While 54 H3K4me3/2 are enriched at transcriptionally-engaged promoters, H3K4me1 is a 55 hallmark of transcriptional enhancers (10, 11). At promoters, H3K4me3 contributes to 56 recruitment of general transcription machinery TFIID and RNA polymerase II (12,13). 57 H3K4me1 has been shown to tether BAF, an ATP-dependent chromatin remodeling 58 complex, at enhancers (14). 59 60 Results 119 120 KMT2A and KDM5C co-exist broadly in the brain 121 We first examined expression patterns of KMT2A and KDM5C using publicly-available 122 resources, and found the two genes are broadly expressed throughout brain regions of 123 adult mice and humans ( Figure S1    H3K4me3 is a reaction product of KMT2A-mediated methylation (40), while a substrate 156 for KDM5C-mediated demethylation (41,42). We sought to determine the impact of 157 KMT2A-and KDM5C-deficiencies and double mutation on the H3K4me3 landscape 158 within the brain. We chose to examine the amygdala tissue, because it plays crucial 159 roles in social behavior and fear memory, which are impaired in . 160 In Western blot analyses, global H3K4me1-3 levels were not altered dramatically in any 161 mutant ( Figure S3A). We thus performed H3K4me3 chromatin immunoprecipitation 162 coupled with deep sequencing (ChIP-seq) to probe local changes genome-wide. To To obtain a global picture of H3K4me changes, we examined the H3K4me3 signals 171 between WT and the three mutants throughout the mouse genome partitioned into 1-172 kilobase (kb) bins ( Figure 2A). We found an overall similarity in H3K4me3 coverage 173 across Kmt2a-HET, Kdm5c-KO, and DM on a genome-wide scale, as well as at 174 promoter regions (Figure 2A, Figure S3C). We then broke down the genome into 175 promoter (± 1 kb from transcription start sites [TSS]), intergenic (between genes), and 176 intragenic (within a gene) regions, and asked if any areas are preferentially 177 dysregulated in any of the mutant animals. In WT, 61% of H3K4me3 fell within 178 promoters, consistent with H3K4me3 as a hallmark of promoters (11-13), while smaller 179 fractions, 18% and 21%, were found in intergenic or intragenic regions, respectively 180 ( Figure S3D). This H3K4me3 distribution pattern was largely consistent across the other 181 genotypes ( Figure S3D), except for Kdm5c-KO and DM amygdala which had slightly 182 higher proportions of intragenic (25% for both) and intergenic (30% and 29%, 183 respectively) methylation ( Figure S3D). 184

185
Examining local differentially-methylated regions (DMRs), i.e. either hyper-or hypo-186 methylated for H3K4me3 compared to WT, we found fewer DMRs in Kmt2a-HET 187 (1,940) than in Kdm5c-KO (11,990) ( Figure 2B-G, Figure S3E-H). This difference is 188 likely due to the heterozygosity of Kmt2a which leaves one functional copy of Kmt2a, 189 versus the complete loss of Kdm5c. Consistently, complete loss of Kmt2a in 190 hippocampal neuronal nuclei was previously shown to reduce H3K4me3 in more than 191 four thousand loci (30). In the Kmt2a-HET amygdala, hypermethylated loci were 192 primarily found at intragenic regions (56%), while hypomethylated regions were found 193 mainly at promoters (76%, Figure 2C). Kdm5c-KO DMRs were biased towards an 194 increase in methylation signals (8,284 hypermethylated vs. 3,706 hypomethylated), 195 consistent with loss of a demethylase. In the Kdm5c-KO amygdala, the hypermethylated 196 loci showed a roughly even split between promoters, intragenic, and intergenic regions, 197 while the majority of the Kdm5c-KO hypomethylated regions were located at promoters 198 (89%, Figure 2F). Hypermethylation in non-promoter regions was also detected as an 199 appearance of additional H3K4me3 peaks in Kdm5c-KO amygdala ( Figure S3D). When We next asked if any of these DMRs were corrected in double mutants (DM). We 210 defined "rescued" regions as DMRs identified in single mutants that were no longer 211 categorized as a DMR in DM (therefore, no different from WT). We observed a rescue 212 of roughly half of single mutant DMRs in our DM animals: 42% (821/1,940) of 576/11,990)  indicating that the rescue effect was partial ( Figure 2N). If KMT2A and KDM5C simply 219 counteract, we should observe that hypermethylated regions in Kdm5c-KO are 220 hypomethylated in Kmt2a-HET, and normalized in DM. We indeed observed such cases 221 in a small fraction of rescued DMRs (solid bars in Figure 2N). Unexpectedly, some 222 regions showed reciprocal H3K4me3 changes between the single mutants in an 223 opposite way as expected; namely, hypomethylation in Kdm5c-KO and 224 hypermethylation in Kmt2a-HET (open bars in Figure 2N). The most prevalent pattern of 225 rescued DMRs was the hypermethylated regions of Kdm5c-KO that were still 226 moderately hypermethylated in DMs and unexpectedly in Kmt2a-HET as well (striped 227 bars in Figure 2N). Thus, simple counteractions between Kmt2a and Kdm5c are 228 relatively rare events, and rather deficiency of the single enzyme results in a complex 229 change of H3K4me3 homeostasis. Nonetheless, our analyses identified thousands of 230 genomic loci at which KMT2A and KDM5C fully or partially mediate aberrant H3K4me3 231 levels caused by loss of the opposing enzyme. 232

Transcriptomic similarity between WDSTS and MRXSCJ models and rescue 234 effects in DM 235
We previously showed Kdm5c-KO mice exhibit aberrant gene expression patterns in the 236 amygdala and frontal cortex (31), and the hippocampus (32). Excitatory-neuron specific 237 conditional Kmt2a-KO mice were also characterized with altered transcriptomes in the 238 hippocampus and cortex (29,30). However, the global gene expression of 239 which is akin to the WDSTS syndrome genotypes, has not been determined. To  Figure 3E). 276 The rescue effects were visible when we analyzed all single-mutant DE genes as a 277 group ( Figure S4D). To better understand how transcriptomic similarity and rescue 278 effect can occur simultaneously, we plotted expression fold changes of the 118 rescued 279 genes ( Figure 3F). We observed that rescued genes were differentially dysregulated in 280 single mutants, e.g. upregulated in Kdm5c-KO but unchanged in Kmt2a-HET ( Figure  281 3F). Thus, these results indicate that the largely-separate sets of genes contribute to the 282 overall transcriptome similarity and the rescue effect in DM. 283

284
We then examined the relationship between the H3K4me3 landscape and transcriptome 285 alterations. If H3K4me3 changes drive the gene misregulation in mutants, we should be 286 able to observe a correlation between these two datasets. However, genes with altered 287 promoter-proximal H3K4me3 did not show significant changes in their expression as a 288 group ( Figure S6A). While H3K4me3 changes are the direct molecular consequences of 289 KMT2A-and KDM5C-deficiency, the steady-state mRNAs we captured in our RNA-seq 290 approach likely involve indirect and adaptive consequences of loss of these enzyme(s), 291 which can lead to an underwhelming correlation between H3K4me3 and transcriptome 292 data. Indeed, we observed a positive correlation between intergenic H3K4me3 levels 293 and spurious transcripts, which are generated at these regions yet likely unstable, 294 therefore, can reflect transcriptional activity more reliably than steady-state mRNA 295 levels ( Figure S6B). Such spurious intergenic transcripts were previously observed in 296 the Kdm5c-KO hippocampus (32). We also examined the H3K4me3 coverage at 297 promoter regions of DE genes ( Figure S6C). Across the different DE gene categories, 298 H3K4me3 levels did not differ between genotypes, with two exceptions: Kmt2a-HET 299 down-regulated and Kdm5c-KO up-regulated genes showed the expected changes in 300 median H3K4me3 levels ( Figure S6C). The correlation was also evident between 301 expression of rescued genes and H3K4me3 ( Figure S6D). Together, these observations 302 indicate that H3K4me3 changes are not sufficient, yet an important contributor, for gene 303 misregulation in single mutants and its correction in DM. 304 305 Gene-annotation enrichment analysis of these 118 rescued genes did not yield 306 statistically-significant enrichment of any functional pathways, however, we were able to 307 separate rescued genes into specific biological pathways that could potentially be 308 restored in DM (Supplementary Table 1

Memory alterations in Kdm5c-KO were reversed in DM 346
After observing the restorative molecular and cellular effects in DM mice, we next 347 sought to determine the effect of loss of Kmt2a and/or Kdm5c on mouse behaviors 348 through a battery of behavioral tests. In accordance with previous findings (31, 32), 349 Kdm5c-KO mice showed significant deficits in associative fear memory, as measured by 350 the contextual fear conditioning (CFC) tests ( Figure  were not attributable to differences in locomotor activity or shock responsiveness, as 362 none of these parameters showed significant differences among the genotypes (Figure 363 In the three-chambered social interaction test ( Figure 6A), we observed significant 368 differences between genotypes (F(3,61) = 4.314, p < 0.008). Kmt2a-HET mice showed 369 no differences from WT (p = 0.082), in accordance with previous tests in conditional 370 Kmt2a-KO mice (30). In contrast, Kdm5c-KO (p= 0.002), as previously shown (31), as 371 well as DM (p = 0.011) mice showed significantly less preference for the stranger 372 mouse compared with WT animals. These data suggest that Kmt2a heterozygosity does 373 not rescue deficits of social interaction in the Kdm5c-KO. 374

375
In tests of social dominance ( Figure 6B), Kmt2a-HET mice won against WTs in 60.9% in 376 of the matches against WT (p = 0.091), and Kdm5c-KO mice won at least 68.4% of the 377 time (p = 0.008). Surprisingly, DM animals lost more than 80% of their bouts against WT 378 (p = 1.47 x 10 -5 ). Although DM mice were slightly smaller compared to single mutants 379 ( Figure 1D), this is unlikely to drive submissive behaviors, as body mass has been 380 shown to have minimal impact on social hierarchy unless excess difference (> 30%) is 381 present between animals (65-67), which is not the case in our study ( Figure 1D With any therapeutic intervention, careful assessments of side effects will be inevitable. 468 In our work, while a substantial fraction of H3K4me3 DMRs and gene misregulation in 469 single mutants were corrected in DM (Figures 2 and 3), combinatorial ablation of 470 KMT2A and KDM5C should reduce net regulatory action over H3K4me3, which may 471 lead to adverse consequences. Indeed, our genomics approaches identified H3K4me3 472 DMRs that are unique to the DM brain, and several genes uniquely altered in DM 473 animals ( Figure S9). It is still plausible that these gene and H3K4me3 changes in DM 474 can lead to phenotypic outcomes that were not examined in this study. Nevertheless, 475 we were encouraged that none of the neurological traits measured in this study showed 476 exacerbation in DM. 477 It is important to note that the double mutations introduced in our mice were constitutive, 479 and therefore a lifetime of adaptation to loss of these two major chromatin regulators 480 may occur from early developmental stages. A more realistic therapeutic strategy may 481 be acute inhibition of KDM5C and KMT2A in juvenile or mature brain. Previous work 482 characterizing mouse models with excitatory-neuron specific ablation of Kdm5c or 483 Total proteins from adult brain tissues were subjected to Western blot analysis using in-509 house anti-KDM5C (31), and anti-GAPDH antibodies (G-9, Santa Cruz). For histone 510 proteins, nuclei were enriched from the dounce-homogenized brain tissues using Nuclei 511 EZ prep Kit (Sigma, NUC-101). DNA were digested with micrococcal nuclease (MNase, 512 NEB) for 10 minutes at room temperature and total nuclear proteins were extracted by 513 boiling the samples with the SDS-PAGE sample buffer. The following antibodies were 514 used for Western blot analyses: anti-H3K4me3 (Abcam, ab8580), anti-H3K4me2 515 (Thermo, #710796), anti-H3K4me1 (Abcam, ab8895), and anti-H3 C-terminus (Millipore, 516 CS204377). 517 518

Brain histology 519
Mice were subjected to transcardial perfusion according to standard procedures. Fixed 520 brains were sliced on a freeze microtome, yielding 30 µm sections that were then fixed, 521 permeabilized, blocked, and stained with DAPI. Slides were imaged on an Olympus 522 SZX16 microscope, with an Olympus U-HGLGPS fluorescence source and Q Imaging 523 Retiga 6000 camera. Images were captured using Q-Capture Pro 7 software. Data were 524 collected in a blind fashion, where samples were coded and genotypes only revealed 525 after data collection was complete. 526 527

ChIP-seq 528
Brains from adult (6-8 months) male mice were microdissected to enrich for the 529 amygdala. N=2 animals were used for WT, and N=3 animals were used for Kmt2a-HET, 530 Kdm5c-KO, and DM as biological replicates. Nuclei were isolated using Nuclei EZ prep 531 Kit (Sigma, NUC-101), and counted after Trypan blue staining. 20,000 nuclei for each 532 replicate were subjected to MNase digestion as previously described (79). We 533 essentially followed the native ChIP-seq protocol (79) with two modifications. One was 534 to use a kit to generate sequencing libraries in one-tube reactions (NEB, E7103S). 535 Another modification was to spike-in the panel of synthetic nucleosomes carrying major 536 histone methylations (EpiCypher, SKU: 19-1001) (43). For ChIP, we used the rabbit 537 monoclonal H3K4me3 antibody (Thermo, clone #RM340). 538 539 Libraries were sequenced on the Illumina NextSeq 500 platform, with single-end 75 540 base-pair sequencing, according to standard procedures. We obtained 20 to 59 million 541 reads per sample. Reads were aligned to the mm10 mouse genome (Gencode) and a 542 custom genome containing the sequences from our standardized, synthetic 543 nucleosomes (EpiCypher) for normalization (80), using Bowtie allowing up to 2 544 mismatches. Only uniquely-mapped reads were used for analysis. Range of uniquely 545 mapped reads for input samples was 38-44 million reads. All IP replicates had a mean 546 of 9.1 million uniquely mapping reads (range: 7.4 to 13.9 million). The enrichment of 547 mapped synthetic spike-in nucleosomes compared to input was calculated and used as 548 a normalization coefficient for read depth each ChIP-seq replicate (80). 549 550 Peaks were called using MACS2 software (v 2.1.0.20140616) (81) using input BAM files 551 for normalization, with filters for a q-value < 0.1 and a fold enrichment greater than 1. 552 Differentially-methylated regions (DMRs) were called using the MACS2 bdgdiff 553 command with default parameters and incorporating the synthetic nucleosome 554 normalization into the read depth factor. Bedtools was used to calculate coverage 555 across individual replicates. We also used Bedtools to intersect peaks of interest with 556 mm10 promoters (defined here as ±1 kb from annotated transcription start site [TSS]), 557 intragenic regions (as defined by annotated mm10 gene bodies, but excluding the 558 previously defined promoter region), and intergenic regions (regions that did not overlap 559 with promoters or gene bodies). DMRs from single mutants (2a-HET or 5c-KO) were 560 considered "rescued" in DM animals if that single-mutant peak was not called as a DMR 561 in the DM analysis. For the global H3K4me3 analysis, the Bedtools multicov command 562 was used to calculate coverage over 1 kb windows throughout the genome, as well as 563 at each promoter (±1 kb from annotated TSS). HOMER (v4.10) was used to carry out 564 motif enrichment analysis (82). We selected the top 5 motifs, and only motifs from 565 known mammalian ChIP-seq experiments were represented in our data. Normalized 566 bam files were converted to bigwigs for visualization in the UCSC genome browser. 567 Genes near peaks were identified by Bedtools and RefSeq genomic accession number 568 were converted to official gene symbol using bioDBnet (83).

RNA-seq 576
Brains from adult (3 to 6 months) male mice were microdissected to enrich for the 577 amygdala. N=3 animals were used per genotype. Tissue was homogenized in Tri 578 Reagent (Sigma). Samples were subjected to total RNA isolation, and RNA was purified 579 using RNEasy Mini Kit (Qiagen). ERCC spike-in RNA was added at this stage, 580 according to manufacturer's instructions (Life Technologies). Ribosomal RNA was 581 depleted using NEBNext rRNA Depletion kit (New England Biolabs). Libraries were 582 prepared using the Click-seq method, using primers containing unique molecular 583 identifiers (UMIs), as described previously (44)  Prior to behavioral testing, mice were acclimated to the animal colony room for one 644 week single-housing in standard cages provided with lab diet and water ad libitum. A 645 12-hour light-dark cycle (7:00AM-7:00PM) was utilized with temperature and humidity 646 maintained at 20 ±2 ºC and >30%, respectively. mice were allowed to explore on two identical objects (jar or egg, counterbalanced 659 across animals) for two, 10-minute trials spaced three hours apart. All animals were 660 returned to the arena tested 24 hours after the first training session and presented with 661 one training object ("familiar" object: jar or egg) and one "novel" object (egg or jar). 662 Exploration of the objects was defined as nose-point (sniffing) within 2 cm of the object. 663 Behavior was automatically measured by Ethovision XT9 software using a Euresys 664 Picolo U4H.264No/0 camera (Noldus, Cincinnati, OH). Preference was calculated as 665 the time spent exploring novel object/total time exploring both objects. One-sample t-666 tests against 50% (no preference) were used to establish whether animals remembered 667 the original objects. 668 669 Contextual Fear Conditioning: Context fear conditioning was assessed as previously 670 described (98). Mice were placed into a distinct context with white walls (9 ¾ × 12 ¾ × 9 671 ¾ in) and a 36 steel rod grid floor (1/8 in diameter; ¼ spaced apart) (Med-Associates, 672 St. Albans, VT) and allowed to explore for 3 minutes, followed by a 2-second 0.8 mA 673 shock, after which mice were immediately returned to their home cages in the colony 674 room. 24 hours later, mice were returned to the context and freezing behavior was 675 assessed with NIR camera (VID-CAM-MONO-2A) and VideoFreeze (MedAssociates, St 676 Albans, VT). Freezing levels were compared between genotypes using a between-677 groups analysis (one-way ANOVA) with genotype as the between-subjects factor. to habituate for 10 minutes. 24 hours later, mice were returned to the apparatus that 684 now included a 2-3 month old stranger male mouse (C57BL/6N) on one side of the box 685 ("stranger"), and a toy mouse approximately same size and color as stranger mouse on 686 other ("toy"). Exploration of either the stranger or toy was defined as nose-point 687 (sniffing) within 2 cm of the enclosure and used as a measure of social interaction (99). 688 Behavior was automatically scored by Ethovision XT9 software as described above, and 689 social preference was defined as time exploring stranger/total exploration time. Social 690 preference was analyzed using one-sample t-tests for each genotype. A repeated 691 measures analysis was used for each aggression (genotype x aggression measures 692 ANOVA) and submissive behaviors (genotype x submissive) to analyze aggressive 693

behaviors. 694
Social Dominance Tube Test: 24 hours prior to testing, mice were habituated to the 696 plastic clear cylindrical tube (1.5 in diameter; 50 cm length) for 10 minutes. During test, 697 two mice of different genotypes were placed at opposite ends of the tube and allowed to 698 walk to the middle. The match concluded when the one mouse (the dominant mouse) 699 forced the other mouse (the submissive mouse) to retreat with all four paws outside of 700 the tube (a "win" for the dominant mouse) (100-102). Each mouse underwent a total of 701 three matches against three different opponents for counterbalancing. Videos were 702 recorded by Ethovision XT9 software as described above, and videos were manually 703 scored by trained experimenters blind to genotype. The number of "wins" was reported 704 as a percentage of total number of matches. Data were analyzed using an Exact 705 Binomial Test with 0.5 as the probability of success (win or loss). 706 707 Resident-intruder aggression: Resident-intruder tests were used to assess aggression. 708 Tests performed on consecutive days, where the resident mouse was exposed to an 709 unfamiliar intruder mouse for 15 minutes (103, 104). A trial was terminated prematurely 710 if blood was drawn, if an attack lasted continuously for 30 seconds, or if an intruder 711 showed visible signs of injury after an attack. Resident mice were assessed for active 712 aggression (darting, mounting, chasing/following, tail rattling, and boxing/parrying), as 713 well as submissive behaviors (cowering, upright, running away). Intruder mice were 714 assessed for passive defense (freezing, cowering, and digging). Behavior was recorded 715 and videos scored manually by experimenters blind to genotype. Data were analyzed 716 using a between groups analysis (one-way ANOVA) with genotype as the between-717 subjects factor. 718

Acknowledgements 720
We thank Dr. Ken Kwan, Mandy Lam, and Own Funk for their assistance with the Click-721 seq library preparation protocol and use of their microscope; Chris Gates for his 722 assistance with RNA-seq analyses; and Clara Farrehi, Jordan Rich, and Demetri 723 Tsirukis for their assistance with experiments for transcriptome analyses, global histone 724 methylation Western blots, and brain histology, respectively. We also thank Drs. Sally  between group means all aggressive (A-E) and submissive (F-H) behaviors (mean ± 960 95% confidence intervals, *p<0.05, **p<0.01 in Least Significant Difference (LSD) test). 961