Altered Gene-Regulatory Function of KDM5C by a Novel Mutation Associated With Autism and Intellectual Disability

Intellectual disability (ID) affects up to 2% of the population world-wide and often coincides with other neurological conditions such as autism spectrum disorders. Mutations in KDM5C cause Mental Retardation, X-linked, Syndromic, Claes-Jensen type (MRXSCJ, OMIM #300534) and are one of the most common causes of X-linked ID. KDM5C encodes a histone demethylase for di- and tri-methylated histone H3 lysine 4 (H3K4me2/3), which are enriched in transcriptionally engaged promoter regions. KDM5C regulates gene transcription; however, it remains unknown whether removal of H3K4me is fully responsible for KDM5C-mediated gene regulation. Most mutations functionally tested to date result in reduced enzymatic activity of KDM5C, indicating loss of demethylase function as the primary mechanism underlying MRXSCJ. Here, we report a novel KDM5C mutation, R1115H, identified in an individual displaying MRXSCJ-like symptoms. The carrier mother’s cells exhibited a highly skewed X-inactivation pattern. The KDM5C-R1115H substitution does not have an impact on enzymatic activity nor protein stability. However, when overexpressed in post-mitotic neurons, KDM5C-R1115H failed to fully suppress expression of target genes, while the mutant also affected expression of a distinct set of genes compared to KDM5C-wildtype. These results suggest that KDM5C may have non-enzymatic roles in gene regulation, and alteration of these roles contributes to MRXSCJ in this patient.


INTRODUCTION
Intellectual disabilities (ID) affect 1.5-2% of the population world-wide (Leonard and Wen, 2002). Clinical features of ID include significant deficiencies in cognitive function and adaptive behaviors beginning before 18 years of age (American Psychiatric Association, 1994). ID syndromes are often accompanied with other comorbidities, including autism spectrum disorders (ASD), which are characterized by impaired speech, hindered social development, and repetitive behaviors. X-linked intellectual disability (XLID) has been thought to account for the higher frequency of ID in males compared to females (Chiurazzi et al., 2008). KDM5C is one of the most frequently mutated genes in XLID and estimated to explain approximately 0.7-2.8% of all XLID cases. KDM5C-deficiency is characterized by frequent autistic and aggressive behaviors (Jensen et al., 2005;Ropers and Hamel, 2005) and is currently referred to as Mental Retardation, X-linked, Syndromic, Claes-Jensen type (MRXSCJ: OMIM#300534). The KDM5C gene, located at Xp11.22-p11.21, encodes a histone demethylase, which specifically targets di-and tri-methylated histone H3 lysine 4 (H3K4me2 and H3K4me3) (Iwase et al., 2007;Tahiliani et al., 2007).
KDM5C is ubiquitously expressed, with the highest expression levels in human skeletal muscle and brain tissues (Jensen et al., 2005). Within the brain, KDM5C is also broadly expressed in key areas for cognitive function, such as the hippocampus, the cortex, and the amygdala, and both neurons and astrocytes contain KDM5C protein (Xu et al., 2008;Iwase et al., 2016). In mouse models, loss of Kdm5c led to defective development of dendrites and dendritic spines (Iwase et al., 2007(Iwase et al., , 2016, which are often observed in human individuals with ID/ASD. Kdm5c-deficient mice displayed impaired fear memory, spatial learning, increased aggression, and reduced social preference. In sum, both human genetics and mouse models have highlighted important roles of KDM5C in cognitive development. H3K4me2/3, the substrates of KDM5C, are generally associated with promoters of transcriptionally active or poised genes, and play important roles in gene transcription (Barski et al., 2007;Kouzarides, 2007;Vallianatos and Iwase, 2015). KDM5C has been reported to repress transcription in postmitotic neurons and breast cancer cells (Iwase et al., 2016;Shen et al., 2016), whereas KDM5C can promote gene expression when it acts on specific transcriptional enhancers in mouse embryonic stem cells (Outchkourov et al., 2013). In the Kdm5c-deficient brain, H3K4me2/3 levels were increased at the promoters of genes that encode key synapse-related genes, and some KDM5C-target genes were aberrantly expressed in these mutant mice. Despite this circumstantial evidence, it still remains elusive whether or not H3K4me2/3 demethylase activity is the sole mechanism of KDM5C-mediated gene regulation.
Both truncation and missense mutations of KDM5C have been found in MRXSCJ patients. The majority of missense mutations functionally tested to date result in reduced demethylation activity of KDM5C (Iwase et al., 2007;Tahiliani et al., 2007;Rujirabanjerd et al., 2010;Brookes et al., 2015). Thus, the predominant molecular mechanism underlying MRXSCJ appears to be loss of function of histone H3K4 demethylation. Here, we report a novel ID/ASD-associated KDM5C mutation, which compromises KDM5C's gene-regulatory function but not enzymatic activity or stability. Our results suggest non-enzymatic roles of KDM5C and a novel pathogenic mechanism contributing to MRXSCJ.

Exome Sequencing, Validation and Analysis of the Variant
Written informed consent was obtained from all study participants in accordance with approved protocols from the Institutional Review Board at the University of Michigan. Clinical trio whole exome sequencing was performed through GeneDx (XomeDx) on genomic DNA from the proband and both parents. The Agilent SureSelect XT2 All Exon V4 kit was used to target the exon regions of the genomes. The targeted regions were sequenced using the Illumina HiSeq 2000 sequencing system with 100 bp paired-end reads. The DNA sequence was mapped to and analyzed in comparison with the published human genome build UCSC hg19 reference sequence. The targeted coding exons and splice junctions of the known protein-coding RefSeq genes were assessed for the average depth of coverage of 64X and data quality threshold of 95.9%. The XomeAnalyzer was used to evaluate sequence changes in the proband compared to other sequenced family members. All sequence variants in the proband and parental samples were confirmed by Sanger sequencing analysis.

X-Chromosome Inactivation
X-inactivation (XI) analysis using the well-characterized CpG methylation site and polymorphic CAG repeats within the Androgen Receptor (AR) locus was performed using standard protocol (Allen et al., 1992). In this assay, digestion with the methylation-sensitive HpaII restriction enzyme followed by PCR amplification was used to determine the ratio of methylation status between the maternal and paternal X chromosome. Parental samples were utilized to delineate allele status.

Plasmid DNA
The R1115H substitution was introduced into pENTR-KDM5C (human) (Iwase et al., 2007) using a PCR-based targeted mutagenesis. WT-and mutant KDM5C cDNA were then moved by LR recombination to a modified pHAGE, a lentivirus compatible mammalian expression plasmid (Murphy et al., 2006). In pHAGE plasmid, cDNA of interest is linked to a puromycin-resistant gene via the internal ribosome entry site (IRES), thereby allowing the selection of transduced cells via puromycin. The modifications of pHAGE are insertion of Gateway cassette (Invitrogen) and Strep-tag, and replacement of the CMV promoter with the PGK promoter. The entire KDM5C cDNAs were Sanger-sequenced to validate the single targeted mutagenesis.

Cell Culture, Cycloheximide Treatment, and Transduction
Lymphoblastoid cell lines from the proband and his father were isolated and cultured under identical conditions as described previously (Doyle, 1990;Burns et al., 2014). Briefly, buffy coat was isolated from citrate (yellow) blood by Ficoll density centrifugation and transformed with Epstein Barr Virus. Lines were maintained in RPMI-1640 media (Gibco) supplemented with 10% FBS, 2 mM GlutaMAX, 1% penn-strep, in an incubator set at 37 • C with 5% constant CO2. U2OS cells were cultured with DMEM media (Gibco) supplemented with 5% FBS, 2mM GlutaMAX, 1% penn-strep, in an incubator set at 37 • C with 5% constant CO 2 . Cycloheximide (Sigma) was resuspended in DMSO and used at a final concentration of 100 µg/ml. For in situ demethylation assays in U2OS cells, Strep-tagged KDM5C expression plasmids (pHAGE) were transfected into U2OS cells with Lipofectamine 2000 (Invitrogen) according to manufacturer's instructions. For in situ demethylation assays in neurons, cells were transduced after 1 day in vitro (DIV1) cells with lentivirus of equal titer containing either human KDM5C cDNA (WT, R1115H, or H514A) or vector alone.

RNA-Sequencing
Mouse forebrain tissue was dissected from embryonic day 16 (E16) CD1 mouse embryos. Cells were dissociated, plated, and cultured as described previously (Iwase et al., 2016). After 1 day in vitro (DIV1) cells were infected with lentivirus of equal titer containing either human KDM5C cDNA (WT, R1115H, or H514A) or vector alone. Puromycin selection occurred at DIV4 with 0.2 µg/ml puromycin (Sigma). Control puromycin selection with untransduced cultures eliminated nearly all cells, indicating that majority of survived cells carry the KDM5C transgene. Cells were harvested at DIV10 by direct addition of TRI Reagent (Sigma). Samples were subject to total RNA isolation, and subsequent purification was performed using RNEasy Mini Kit (Qiagen). Ribosomal RNA was depleted using RiboMinus Eukaryotic Kit v2 (Life Technologies). Libraries were prepared using Direct Ligation of Adapters to First-strand cDNA as described previously (Agarwal et al., 2015). Multiplexed libraries were pooled in approximately equimolar ratios and purified from a 1.8% TBE-agarose gel. Libraries were sequenced on the Illumina HiSeq 4000 platform, with paired-end 150 base pair reads, according to standard procedures.
Reads were trimmed to 60 bp using BBDuk (35.51) and mapped to the mm9 mouse genome using STAR (2.5.3a) allowing zero mismatches, and only uniquely mapped reads were analyzed further. Due to low mapability of read2, only read1 was used for further analysis. BAM files were indexed and converted to BigWig files in a strand specific manner. BigWigs were normalized to 10 million non-rRNA and non-mitochondrial reads. DESeq2 (1.14.1.) was used to call differential gene expression between conditions, using a cutoff of p-value < 0.01. Cuffdiff (Cufflinks 2.2.1) was used to calculate FPKM. Differentially expressed genes were examined for functional annotation clustering using DAVID (6.8). To validate overexpression conditions, reads were mapped to a custom genome using STAR, allowing no mismatches, and only uniquely mapped reads were analyzed further. The custom genome contained human KDM5C cDNA (NM_001146702.1), where bases were masked with "N" at the H514 (c.1540-1541CA > NN) and R1115 (c.3344G > N) loci to allow for specific mapping under strict no-mismatch conditions. RNA-seq files can be found at Gene Expression Omnibus GEO:GSE104319.

Statistical Analyses
For histone demethylase assays in Figures 2A,B, error bars represent standard error of the mean (SEM) of a technical triplicate. For analysis of H3K4me levels in Figure 2D, H3K4me signals were normalized to pan-H3 signal and error bars represent SEM of technical triplicate. For RNA-seq data in

KDM5C p.Arg1115His Is Identified in Family UM1
The proband of family UM1 was first clinically examined at 26 months for developmental delay and microcephaly, after observing delays in developmental milestones from 1 year of age. He was diagnosed with ASD due to cognitive impairment and behavioral concerns, including severe tantrums, aggression, and anxiety (The full clinical description is available on MyGene2, family #1600) 1 . Clinical whole exome sequencing revealed a missense mutation in the KDM5C gene, c.3344G > A, in the proband ( Figure 1A). This variant results in an arginine-tohistidine substitution at amino acid position 1115 (R1115H), and was inherited from the maternal grandmother. The carrier mother and grandmother were phenotypically normal, and the variant was not present in the unaffected siblings or maternal uncle. We isolated genomic DNA from lymphoblastoid cell lines from father (UM1 II-1) and proband (UM1 III-3), and confirmed by Sanger sequencing that the variant is present specifically in the proband ( Figure 1B).
The R1115H variant lies in a region of high conservation and is itself highly conserved among vertebrates ( Figure 1C). KDM5C belongs to a family of histone H3K4 demethylases, in which KDM5A, KDM5B, and KDM5D are the additional members (Iwase et al., 2007). The R1115 residue in KDM5C is conserved in the protein family members, though it is substituted with lysine, another basic amino acid, in KDM5A, implicating a conserved functional role ( Figure 1D). Consistently, multiple genetic variant assessment algorithms predict this variant to be pathogenic (Supplementary Table S1). We did not find any homozygous or hemizygous substitutions of R1115 in the Genome Aggregation Database (gnomAD) 2 , indicating that variation in this amino acid residue is intolerant in the population. To date, 25 KDM5C mutations have been identified in MRXSCJ patients (Jensen et al., 2005;Santos et al., 2006;Tzschach et al., 2006;Abidi et al., 2008;Adegbola et al., 2008;Rujirabanjerd et al., 2010;Santos-Reboucas et al., 2011;Ounap et al., 2012;Brookes et al., 2015). Most of them tend to cluster around the JmjC catalytic core domain ( Figure 1E). The R1115H variant lies within the C-terminal segment of the KDM5C protein, making it the most distal missense mutation in KDM5C reported to date.

X-Chromosome Inactivation Skewing Predicts a Pathogenicity of KDM5C R1115H
In female mammals, one of the two X chromosomes is randomly chosen to be silenced during early embryogenesis; therefore, females are chimeras of cells that silenced either the paternally-or the maternally-inherited X (Augui et al., 2011). Female carriers of recessive X-linked disorders often display skewed X-inactivation, i.e., a higher percentage of silencing in either the paternal or the maternal X chromosome over the other (Brown, 1999). Skewed X-inactivation might be due to (dis)advantage of carrying the X-linked mutation in cell proliferation and/or survival. Multiple MRXSCJ cases are reported to have highly skewed X-inactivation in carrier mothers [(Ounap et al., 2012) and Table 1]. When we tested the X-inactivation status of the carrier mother (UM1 II-2), we observed highly skewed inactivation (95:5, Table 1), and moderate skewing in the grandmother (UM1 I-2). In the carrier mother, the X-chromosome passed on to the proband is dominantly CpG-methylated at the androgen receptor (AR) locus. AR and KDM5C are located nearby (∼14.5 Mb), flanking the centromere, and no meiotic recombination site is known between the two loci (Kong et al., 2002); therefore, KDM5C is very likely mutated on the inactive X-chromosome of the carrier mother. These results indicate that the KDM5C-R1115H mutation might have an impact on cell proliferation/survival during development (also see Discussion).

Enzymatic Activity Is Largely Retained in KDM5C R1115H
A majority of KDM5C-MRXSCJ mutations that have been tested for enzymatic activity exhibit a decrease in histone demethylase activity, suggesting a loss-of-function pathogenic mechanism (Iwase et al., 2007;Tahiliani et al., 2007;Rujirabanjerd et al., 2010;Brookes et al., 2015). We set out to test whether the R1115H variant affects histone demethylation capabilities of the KDM5C protein. WT and R1115H mutant KDM5C proteins containing an N-terminal Strep-tag were expressed at equal levels in insect cells and affinity purified (Supplementary Figure S1A). We performed in vitro demethylation assays using purified KDM5C and two synthetic histone H3 N-terminal tail peptides carrying the following modifications: K4me3, the direct substrate for the KDM5C catalytic subunit; or K4me3 and K9me3, which is recognized by the first PHD domain of KDM5C (Iwase et al., 2007). Our results showed that the R1115H mutation did not dramatically reduce H3K4me3 demethylase activity on peptides carrying either H3K4me3 alone or both H3K4me3 and K9me3 (Figures 2A,B). These data indicate intrinsic demethylation ability of the KDM5C R1115H protein is largely retained. We next tested demethylation activities in cells. We over-expressed Strep-tagged human KDM5C cDNA in U2OS FIGURE 2 | KDM5C R1115H has largely intact enzymatic activity. (A,B) In vitro demethylation assay. Full-length KDM5C wildtype (WT) or mutant (R1115H) were affinity-purified from Sf9 or Hi5 insect cells infected with baculoviruses and subjected to demethylation assay using synthetic histone peptide carrying (A) H3K4me3 or (B) H3K4me3 + K9me3. Relative fluorescence values of formaldehyde, produced by the demethylation reaction, are normalized by the amount of purified proteins in the reaction and plotted. Error bars represent SEM of a technical triplicate. (C) In situ demethylation assay. Expression constructs of strep-KDM5C WT, mutant R1115H, or catalytically inactive H514A were transiently transfected into U2OS cells and stained with antibodies for Strep (red) and H3K4me3 (green). Compared to untransfected cells, significant decrease of K4me3 was found in WT-and R1115H over-expressing cells. Demethylation activity was completely abrogated by the H514A mutation. Nuclei were stained with DAPI. (D) H3K4me levels in lymphoblastoid cell lines from proband (UM1-III-3, KMD5C R1115H) and father (UM1-II-1, KDM5C WT) were measured by quantitative Western blot analysis. H3K4me signals were normalized to pan-H3 signal (n = 3, Mean ± SEM). No noticeable change was found between the two cell lines.
cells and examined H3K4me3 levels using immunofluorescence ( Figure 2C). Cells expressing KDM5C WT showed dramatically reduced H3K4me3 staining. Similar to WT, KDM5C R1115Hexpressing cells exhibited a marked reduction in H3K4me3 signal. In contrast, cells expressing the catalytically-inactive KDM5C H514A mutant (Iwase et al., 2007) retain high H3K4me3 levels. While we observed cytoplasmic signal more frequently in R1115H-expressing cells than WT, this potential mislocalization does not seem to impair demethylation activity of KDM5C R1115H. We also over-expressed Streptagged human KDM5C cDNA in primary mouse neurons and examined H3K4me1/2/3 levels using immunofluorescence (Supplementary Figure S1B). Cells expressing KDM5C WT often showed reduced H3K4me2 and H3K4me3 staining, and no effect on H3K4me1, as predicted. Similar to WT, KDM5C R1115H-expressing cells frequently exhibited a marked reduction in H3K4me2/3 signal. In contrast, cells expressing the catalytically inactive KDM5C H514A mutant (Iwase et al., 2007) retain high H3K4me2/3 levels. When we examined global H3K4me levels of lymphoblastoid cell lines from father (WT) and proband (R1115H) by quantitative Western blot analysis, proband-derived cells were indistinguishable from father-derived cells (Figure 2D, Supplementary Figure S1C). Taken together, these data strongly suggest that KDM5C R1115H substitution does not lead to substantial reduction in histone demethylation activity.

Protein Stability of KDM5C R1115H Is Largely Unaffected
Some KDM5C-MRXSCJ mutations have been shown to destabilize KDM5C protein, representing another potential loss-of-function effect. To test the stability of KDM5C R1115H, we treated the lympoblastoid cells from the father and the proband with cycloheximide, an inhibitor of protein synthesis, for 0-28 h. Reduction kinetics of endogenous KDM5C levels were not dramatically different in proband compared to father throughout the time course, suggesting protein stability is largely unaffected by the R1115H mutation (Figure 3 and Supplementary Figure S2).

RNA-Seq Reveals Impact of the R1115H Variant on Gene Expression
We previously reported that Kdm5c regulates neurodevelopment genes in cultured mouse forebrain neurons and multiple brain regions in vivo including the cortex and the amygdala (Iwase et al., 2016). To test consequences of the R1115H substitution on KDM5C gene regulatory behavior, we over-expressed KDM5C WT and KDM5C R1115H in primary mouse forebrain neuron culture and performed RNA-seq analysis on three biological replicates. To assess the impact of H3K4 demethylation activity on the transcriptome, we also overexpressed KDM5C H514A as a representative of other mutations that reduce enzymatic activity (Iwase et al., 2007). As outlined in Figure 4A, transduced neurons were selected by puromycin and subjected to RNA-seq. Expression of intended mutants was confirmed by examining the nucleotide sequence of reads mapped to the human KDM5C cDNA (Supplementary Figure S3A). Mapping the sequencing reads to the human KDM5C cDNA also allowed us to validate similar levels of human KDM5C WT and R1115H mutant cDNAs, while we noted moderately higher level of H514A mutant expression compare to the other two KDM5C cDNAs (Supplementary Figures S3B,C). We compared the sequencing reads mapped to human KDM5C cDNA and mouse Kdm5c gene without allowing any mismatches, and estimated that each KDM5C cDNA was expressed approximately 15-to 30-fold higher than endogenous Kdm5c mRNA ( Supplementary  Figures S3B,C). The expression of endogenous Kdm5c does not appear to be largely affected by human KDM5C overexpression (Supplementary Figure S3C).
Compared to the vector control, overexpression of KDM5C-WT resulted in 42 differentially-expressed (DE) genes (7 up-regulated, 35 down-regulated). KDM5C R1115H expression resulted in 64 DE genes (16 up-regulated, 48 downregulated) compared to WT, while KDM5C H514A gave 99 DE genes (45 up-regulated, 54 down-regulated) (p < 0.01; Figure 4B). Consistent with its role as a transcriptional repressor in primary neuron culture (Iwase et al., 2016), we observe a majority (83%) of these DE genes are down-regulated by KDM5C-WT overexpression. KDM5C R1115H expression similarly led to predominant down-regulation of genes (75% down, 25% up), while KDM5C H514A expression resulted in a near even split between up-and down-regulated genes (54% down, 46% up). To test if R1115H and/or H514A mutations affect gene-regulatory function of KDM5C, we plotted the fold change and P-values of 42 DE genes, whose expression was altered by WT overexpression, in the R1115Hor H514A-overexpression datasets ( Figure 4C). Both KDM5C R1115H and H514A failed to fully repress this group of genes, suggesting that the two mutations interfere with KDM5C's gene-regulatory function (Figure 4C). The partial deficiency of H514A mutant despite the higher level of KDM5C-H514A mRNA implicates that demethylation activity is not the sole mechanism for KDM5C-mediated gene control. It is noteworthy that KDM5C R1115H mutant showed a similar level of deficiency compared to H514A mutant in regulating the 42 genes.
We then compared the identity of DE genes upon overexpression of KDM5C-WT, R1115H, and H514A. Interestingly, overlap between DE genes in each condition were limited, suggesting that R1115H and H514A mutants can regulate different sets of genes compared to KDM5C-WT and the two mutants are functionally distinct from each other ( Figure 4D). Representative genes for the following four expression patterns across conditions are shown in
Having observed both common and unique impacts of R1115H and H514A substitutions, we sought to gain biological insights associated with the mutations by using the database for annotation, visualization, and integrated discovery (DAVID) (Huang da et al., 2009a,b). Given that KDM5C primarily acts as a transcriptional repressor, we reasoned that down-regulated genes reflect direct impact of KDM5C overexpression more likely than up-regulated genes; therefore, here we primarily analyzed down-regulated genes. Genes down-regulated by WT were enriched in developmental processes and cell signaling gene ontology (GO) terms ( Figure 5A). These GO terms are absent from both R1115H ( Figure 5B) and H514A ( Figure 5C) downregulated genes, pointing to the inability of the mutant KDM5Cs to regulate the relevant genes as WT (see Supplementary  Table S2 for complete list of GO terms). Meanwhile, downregulated genes by R1115H and H514A were enriched in largely distinct sets of GO terms compared to WT, and these GO terms are also different between the two mutants. For example, KDM5C-R1115H mutant down-regulated and "membranebound organelle" and "muscle structure development, " while KDM5C-H514A down-regulated "gliogenesis" and "protein binding." Distinct sets of genes appear to contribute the diverse GO term enrichments between the conditions (Supplementary Table S2). Taken together, these data suggest that KDM5C R1115H is less potent in suppressing WT-regulated genes, yet the mutant gains unique gene regulatory roles, which may lead to deleterious and distinct consequences in neuronal development and functions.

DISCUSSION
MRXSCJ has been primarily attributed to either reduced enzymatic activity and/or stability of the KDM5C protein by associated mutations. In the present study, we describe a novel KDM5C missense mutation, R1115H, which was identified in an individual with a typical MRXSCJ phenotype. While the R1115H substitution does not result in appreciable changes in enzymatic activity or protein stability, this substitution appears to alter the transcriptional regulatory function of KDM5C. Another KDM5C mutation, D87G, was shown to not interfere with the demethylation activity (Tahiliani et al., 2007). However, the functional consequence of the D87G substitution remains elusive. Thus, our study implicates a novel mechanism underlying MRXSCJ.
We provide several lines of evidence that support contribution of the KDM5C-R1115H mutation to developmental and behavioral phenotypes of the proband. First, the clinical phenotypes including short stature, aggressive behavior, and ASD associated with ID align well with previously described individuals with MRXSCJ (Jensen et al., 2005;Santos et al., 2006;Tzschach et al., 2006;Adegbola et al., 2008;Santos-Reboucas et al., 2011;Ounap et al., 2012). Second, the mutation clearly segregates with the cognitive impairment, as both unaffected brother and maternal uncle did not carry the R1115H mutation (Figure 1). Third, KDM5C-R1115 is highly conserved among vertebrates and the R1115H substitution is predicted to be functionally damaging by multiple prediction algorithms (Figure 1 and Supplementary Table S1). Forth, the carrier mother showed highly skewed X-inactivation (Table 1). KDM5C was originally discovered as an X-linked gene that escapes X-inactivation both in human and mouse, although escaping is not complete -KDM5C is expressed at a lower level from the inactive X compared to that of the active X (Wu et al., 1994a,b). We recently reported that KDM5C is necessary and sufficient for initiating X-chromosome inactivation by inducing expression of the Xist non-coding RNA (Gayen et al., personal communication). In KDM5C heterozygous knockout female embryos, cells that chose the KDM5C mutantcarrying X as the active X chromosome failed to inactivate one of the two Xs, and these cells were quickly lost during development (Gayen et al., personal communication). We speculate that a similar cell selection process took place in the carrier mother, as we determined that KDM5C is very likely mutated on her inactive X-chromosome. These data, together with the observations in mouse models, support the deleterious impact of the R1115H substitution on the function of KDM5C. The moderate bias of X-inactivation in the carrier grandmother may imply that additional genetic event(s) in the carrier mother might have contributed to the MRXSCJ-like symptoms in her son. Finally, our RNA-seq data indicate that R1115H alters KDM5C's gene-regulatory function in neurons. Generation and characterization of knock-in mice carrying KDM5C-R1115H will allow us to understand the causal roles of this mutation in cognitive development in the future.
The previously reported missense mutations in KDM5C all fall within the N-terminal half of KDM5C, which harbors a JmjC domain, the catalytic core for histone H3K4 demethylation (Figure 1). The N-terminal half of KDM5 family demethylases, encompassing JmjN, Bright, PHD1, JmjC, and zinc finger domains, was shown to be sufficient for its catalysis (Johansson et al., 2016;Vinogradova et al., 2016). By contrast KDM5C-R1115 resides outside this catalytic segment, and indeed, the KDM5C-R1115H substitution did not interfere with enzymatic activity. The C-terminal half of KDM5C contains a PHD finger domain (PHD2) with unknown function, and R1115 is located 73 amino-acid upstream of PHD2. Given that KDM5C-PHD1 recognizes H3K9me (Iwase et al., 2007), it is plausible that this region recognizes specific histone modification(s). Alternatively, the C-terminal segment may interact with other transcriptional regulators to recruit them to KDM5C-target genes. Interestingly, four truncation mutations that remove this C-terminal segment have been identified in MRXSCJ patients (Jensen et al., 2005;Abidi et al., 2008;Rujirabanjerd et al., 2010;Brookes et al., 2015). These observations suggest the important roles of the C-terminal segment for KDM5C's gene-regulatory roles. The R1115H substitution may interfere with these uncharacterized roles of the C-terminal regions of KDM5C. A minor population of cells show cytoplasmic signal of KDM5C-R1115H mutant in addition to nuclear signals ( Figure 2C). Given the intact H3K4 demethylation activity of KDM5C-R1115H and low frequency of its cytoplasmic presence, the mislocalized protein may not have a major impact. However, we cannot rule out the possibility that KMD5C-R1115H could demethylate ectopic target(s) at the cytoplasm. Future studies are needed to determine the roles of the C-terminus in KDM5C-mediated transcriptional regulation, impact of disease mutations in this segment, and the roles of enzymatically active mutants that can be present outside nuclei.
Our RNA-seq data suggest that the R1115H mutation may not only lead to loss of function of KDM5C but also acquisition of a new gene-regulatory function. It is tempting to speculate that KDM5C neofunctionalization might be involved in the pathophysiology of this specific MRXSCJ-like patient. However, limitations of interpreting the RNA-seq results should be noted. The RNA-seq was carried out upon overexpression of WT-and mutant-KDM5C, 10 days after introducing KDM5C transgenes into neurons. Gene expression changes may therefore involve both direct impact of on KDM5C's target genes as well as indirect consequences, such as alteration in neuronal maturation processes. In addition, gene expression changes upon KDM5C overexpression may not simply reflect the dysregulation of bona fide KDM5C target-genes. Nonetheless, KDM5C-R1115H illuminates a novel gene regulatory function of KDM5C, which is independent of its enzymatic activity, and potentially represents a novel molecular etiology contributing to MRXSCJ.

AUTHOR CONTRIBUTIONS
CNV, CEK, and SI conceived the study and designed the experiments. CEK performed the clinical evaluation of the proband and coordinated genetic testing of the proband and family members. MJF oversaw the X-chromosome inactivation analysis. MB oversaw the generation of lymphoblastoid cell lines. CNV performed all experiments and analysis. All authors contributed to the writing and editing of the manuscript.