Mutations in the intellectual disability gene KDM5C reduce protein stability and demethylase activity

Abstract

Mutations in KDM5C are an important cause of X-linked intellectual disability in males. KDM5C encodes a histone demethylase, suggesting that alterations in chromatin landscape may contribute to disease. We used primary patient cells and biochemical approaches to investigate the effects of patient mutations on KDM5C expression, stability and catalytic activity. We report and characterize a novel nonsense mutation, c.3223delG (p.V1075Yfs*2), which leads to loss of KDM5C protein. We also characterize two KDM5C missense mutations, c.1439C>T (p.P480L) and c.1204G>T (p.D402Y) that are compatible with protein production, but compromise stability and enzymatic activity. Finally, we demonstrate that a c.2T>C mutation in the translation initiation codon of KDM5C results in translation re-start and production of a N-terminally truncated protein (p.M1_E165del) that is unstable and lacks detectable demethylase activity. Patient fibroblasts do not show global changes in histone methylation but we identify several up-regulated genes, suggesting local changes in chromatin conformation and gene expression. This thorough examination of KDM5C patient mutations demonstrates the utility of examining the molecular consequences of patient mutations on several levels, ranging from enzyme production to catalytic activity, when assessing the functional outcomes of intellectual disability mutations.

Introduction

Intellectual disability (ID) is a clinically variable and genetically heterogeneous disorder characterized by limitations in intellectual functioning and adaptive behavior (1). ID affects 1–3% of the population, but remains poorly understood (2). Recent sequencing advances have enabled identification of numerous genetic mutations associated with ID, including mutations in many genes encoding chromatin regulators (3–5). Chromatin consists of the DNA helix spooled around octamers of histones H2A, H2B, H3 and H4. Histones are subject to a plethora of post-translational modifications (6), which are able to influence a variety of nuclear processes. Genes encoding histone methyltransferases, such as MLL (7), MLL2 (8) and SETD5 (9), and genes encoding histone demethylases, such as KDM6A (10), PHF8 (11) and KDM5C (12), are mutated in ID, suggesting that disruption of methylation dynamics may contribute to disease development.

KDM5C (also known as JARID1C and SMCX) encodes a histone demethylase that removes di- and tri-methylation of histone H3 lysine 4 [H3K4me2/3] (13,14), modifications associated with gene activation (6). Mutations in KDM5C account for 1–4% of X-linked intellectual disability [XLID] (15). Patients with KDM5C mutations suffer from mild to severe ID, often accompanied by symptoms such as behavioral disturbances and epilepsy (16) [OMIM 300534]. Nonsense and missense KDM5C mutations have been reported, and a number of missense mutations reduce protein demethylase activity (13,14,17), suggesting a loss-of-function disease mechanism. A KDM5C mutation has also been reported in autism spectrum disorder (18), and changes in KDM5C expression have been implicated in the pathology of ARX-related ID (19), Huntington's disease (20) and drug addiction (21), suggesting that KDM5C plays a critical role in neuronal function.

In this study, we derived primary skin fibroblasts from ID patients with four different KDM5C mutations: two missense mutations, one nonsense mutation, and an interesting mutation in the translation start codon. Using cell biological and biochemical approaches, we investigated the molecular consequences of these mutations. We demonstrate that mutation of the translation start codon of KDM5C results in translation re-start and production of a N-terminally truncated protein. Those patient mutations that are compatible with KDM5C protein production, including missense and truncation mutations, compromise KDM5C function through limiting protein stability and enzymatic activity. Reduced KDM5C function is intimated by target gene up-regulation, despite no changes in global histone methylation.

Results

Patient mutations in KDM5C

We obtained fibroblasts from seven patients suffering from XLID caused by four independent mutations in KDM5C (Table 1). The location of the mutations with respect to annotated features of the KDM5C protein is shown in Figure 1A. The mutations and clinical symptoms have been previously reported for p.D402Y (12), p.P480L (22) and c.2T>C (23). The patient with the c.3223delG (p.V1075Yfs*2) mutation exhibited severe ID, with developmental delay, short stature (5th–10th percentile) and hyperreflexia. Patient 3223delG-1 walked at 22 months, used only 6–8 words by the age of 13.5 years, and displays strabismus, a high narrow palate and constipation. We also examined fibroblasts from a female heterozygous carrier of the c.3223delG mutation (3223delG-HET; mother of the affected male), who required tutoring in school and has a high narrow palate. The patient's sister (not investigated) is also a carrier of the mutation and exhibits mild intellectual disability, developmental delay, hyperreflexia and a high narrow palate, suggesting that the mutation can be deleterious in the heterozygous setting. For controls, we obtained BJ foreskin fibroblasts from Dr George Q Daley (Boston Children's Hospital) and control male skin fibroblasts from the Coriell Cell Repository (Control-1: GM03348, 10 years; Control-2: AG06234, 17 years).

Table 1.

Patient fibroblast information

	Patient identifier	Family	Age at biopsy (year)	cDNA mutation	Protein mutation	ID	Clinical notes
D402Y-1	38505	A034	72	c.1204G>T	p.D402Y	Moderate to severe	DD, short stature, aggression
D402Y-2	38506	A034	67	c.1204G>T	p.D402Y	Moderate to severe	DD, short stature, rare seizures, aggression
P480L-1	13185	K9330	8	c.1439C>T	p.P480L	Moderate	DD, strabismus, mild scoliosis
P480L-2	13186	K9330	1	c.1439C>T	p.P480L	Mild	DD, short stature, tremor, seizures, aggression
2T>C-1	13008		23	c.2T>C	p.M1_E165del	Severe	DD, short stature, strabismus, seizures, ataxia, aggression
2T>C-2	13009		13	c.2T>C	p.M1_E165del	Severe	DD, short stature, ataxia,
3223delG-1	22986	K9607	17	c.3223delG	p.V1075Yfs*2	Severe	DD, short stature, strabismus, hyperreflexia,

Annotations based on Homo sapiens lysine (K)-specific demethylase 5C (KDM5C) transcript variant 1 (uc004drz.3/NM_004187) in the GRCh38/hg38 genome assembly.

ID, intellectual disability; DD, developmental delay.

Figure 1.

The c.3223delG mutation precludes KDM5C expression, but other patient mutations are compatible with transcript and protein production. (A) Schematic of KDM5C protein structure, with domains and patient mutations indicated. M166 is the predicted start codon for patients with the c.2T>C mutation. The KDM5C antibody is raised against the C-terminal 100 amino acids of the protein. (B) Analysis of KDM5C RNA amplified using either random primers (filled bars) or oligo dT primers (open bars) demonstrates that KDM5C RNA is detected in affected males exhibiting the p.D402Y, p.P480L or c.2T>C mutations, but not in the affected male with the c.3223delG mutation. Mean and standard deviation of 3–4 independent experiments are displayed (*P < 0.01 compared with Control-1, Student's t-test). RNA was detected using primers spanning exon–exon junctions, and normalized to housekeeping gene RSP18 and then Control-1 fibroblasts. (C) Whole cell lysates from patient skin fibroblasts shows KDM5C protein expression in affected males exhibiting the p.P480L and p.D402Y mutations. No expression is detected in the c.3223delG affected male. Males carrying the c.2T>C mutation produce a truncated KDM5C product. Lowest band, x, is non-specific. TUBULIN, loading control. (D) KDM5C shRNAs reduce KDM5C transcript in control-BJ and c.2T>C fibroblasts. Mean and standard deviation of two independent experiments are displayed (*P < 0.05 compared with FF, Student's t-test). RNA was detected using primers spanning exon–exon junctions, and normalized to housekeeping gene RSP18 and then control shRNA samples (FF, firefly luciferase). (E) Western blots show that the truncated product produced in c.2T>C fibroblasts is reduced upon KDM5C-shRNA treatment, suggesting that it is specific. Lowest band, x, is non-specific. ACTIN, loading control.

Patient mutations reduce KDM5C protein levels

We first analyzed the effect of the genetic mutations on the RNA expression level of KDM5C in patient and control fibroblasts using qRT-PCR (quantitative reverse transcription with PCR; Fig. 1B). We analyzed both total RNA amplified using random hexamers (filled bars) and polyadenylated mRNA amplified using oligo dT primers (open bars). Patient fibroblasts harboring p.D402Y, p.P480L and c.2T>C mutations expressed KDM5C. However, there were significant reductions in the levels of total and polyadenylated KDM5C RNA in fibroblasts with the p.P480L mutation, and in total KDM5C RNA in fibroblasts with the p.D402Y mutation. KDM5C mRNA levels in c.3223delG fibroblasts were dramatically reduced to ∼15% of control cell lines, consistent with the resultant mRNA being targeted for nonsense-mediated mRNA decay (NMD; Fig. 1B). KDM5C mRNA levels in the c.3223delG carrier female were slightly but significantly reduced compared with control skin fibroblasts.

KDM5C is a member of the KDM5 demethylase family and KDM5C homologs might compensate for KDM5C loss-of-function. We therefore analyzed the mRNA expression level of KDM5 family members KDM5A, KDM5B and KDM5D (also known as RBP2, PLU-1 and SMCY, respectively), to see whether these were altered in patient fibroblasts with KDM5C mutations (Supplementary Material, Fig. S1A). Levels of KDM5A mRNA were comparable in all patient and control fibroblast lines. Levels of KDM5B RNA were slightly but significantly reduced in P480L-2, 2T>C-1, 3223delG-1 and 3223delG-HET. Y-linked SMCY was undetectable in the carrier female as expected, but present at similar levels in affected males and controls.

c.1204G>T and c.1439C>T are missense mutations predicted to give rise to full-length KDM5C protein with an amino acid substitution at a single residue: p.D402Y and p.P480L, respectively. As predicted, we detected KDM5C protein at the expected molecular weight (176 kDa) on western blots of whole cell lysates from fibroblasts with c.1204G>T and c.1439C>T mutations (Fig. 1C) using an antibody raised against the C-terminus of KDM5C (Fig. 1A). In contrast, c.3223delG is predicted to cause a frameshift mutation introducing a premature termination codon (PTC) at the position immediately after the first altered amino acid (p.V1075Yfs*2). This is consistent with the low levels of KDM5C RNA detected in 3223delG-1 fibroblasts (Fig. 1B) and we anticipated no protein production. We detected no KDM5C protein in c.3223delG fibroblasts (Fig. 1C), but the KDM5C antibody is raised to the portion of KDM5C beyond the PTC (Fig. 1A) and therefore would not detect a potentially truncated product. Interestingly, we detected KDM5C protein in fibroblasts with the c.2T>C mutation in the translation start codon of KDM5C (Fig. 1C). The c.2T>C KDM5C protein runs at a lower molecular weight than wild-type KDM5C, suggesting production of an in-frame, N-terminally truncated protein. Importantly, global levels of KDM5C protein are lower in all patient lines compared to controls (Fig. 1C).

To verify that the lower molecular weight band seen in KDM5C western blots from c.2T>C fibroblasts is an alternative version of KDM5C, we treated control and c.2T>C fibroblasts with shRNAs targeting KDM5C. qRT-PCR analysis confirmed that two KDM5C shRNAs reduced the levels of KDM5C transcripts in control and c.2T>C patient cells compared with uninfected and control shRNA infected (FF, firefly luciferase) samples (Fig. 1D). Western blot analysis showed that KDM5C shRNA constructs decrease the full-length KDM5C signal in control fibroblasts, and also decrease the lower molecular weight signal in c.2T>C fibroblasts (Fig. 1E) corroborating that the protein recognized by the KDM5C antibody in these patient cells is indeed truncated KDM5C.

C.2T>C patients express N-terminally truncated p.M1_E165del KDM5C protein

Translation initiation site prediction algorithms based on DNA sequence anticipate that when the c.2T>C mutation is present, KDM5C translation starts at the methionine (M) at position 166, suggesting that the truncated protein is p.M1_E165del. M166 is in frame with the canonical start codon, in keeping with recognition of the truncated form by an antibody raised against the C-terminus of KDM5C (Fig. 1A). p.M1_E165del is 20 kDa smaller than wild-type (156 versus 176 kDa), which is consistent with the molecular weight difference observed in westerns (Fig. 1C), and lacks the highly conserved JmjN and ARID (AT-rich interactive domain) domains (Fig. 1A).

To test the hypothesis that c.2T>C patient cells produce p.M1_E165del KDM5C, we generated C-terminally tagged KDM5C expression constructs encoding either the full-length coding sequence with the c.2T>C mutation, or a truncated coding sequence beginning with the M166 codon (c.A1_A495del). Overexpression of these constructs in fibroblasts lacking endogenous KDM5C (c.3223delG patient fibroblasts, Fig. 2A) or in BJ control fibroblasts (Supplementary Material, Fig. S1B) followed by western blots showed that proteins encoded by c.2T>C and c.A1_A495del KDM5C run at a similar molecular weight, supporting our hypothesis. Data from overexpression of N-terminally tagged KDM5C with other patient mutations were consistent with results from patient cells: the missense mutations p.P480L and p.D402Y produce full-length protein (Fig. 2A, Supplementary Material, Fig. S1B).

Figure 2.

p.M1_E165del KDM5C is produced in c.2T>C patient fibroblasts. (A) Overexpression of KDM5C-Flag-HA in fibroblasts lacking endogenous KDM5C demonstrates similar size of proteins from a c.2T>C KDM5C expression vector, and one starting at p.M166. Overexpression of HA-Flag-KDM5C harboring patient mutations confirm that p.P480L and p.D402Y are missense mutations. H514A is a catalytic-null mutation (13). ACTIN, loading control. (B) Overexpression of untagged KDM5C demonstrates that KDM5C starting at M166 is a similar molecular weight to endogenous KDM5C in c.2T>C patient fibroblasts. Lowest band, x, is non-specific.

We then overexpressed either wild-type or c.A1_A495del KDM5C with no tags in KDM5C-null fibroblasts and compared the molecular weight of the resultant protein with endogenous KDM5C protein from control or c.2T>C patient fibroblasts (Fig. 2B). We observed that overexpressed wild-type KDM5C runs at a similar molecular weight to KDM5C in control fibroblasts, as expected. Overexpressed c.A1_A495del produces a KDM5C protein that runs at a similar molecular weight to the endogenous KDM5C in c.2T>C fibroblasts, suggesting that p.M1_E165del is produced in these patients.

P.M1_E165del KDM5C associates with chromatin

We first assessed whether the absence of the ARID DNA binding domain (24) prevents p.M1_E165del KDM5C associating with chromatin. To test this, we overexpressed KDM5C with patient mutations or KDM5C lacking either the JmjN or the ARID domain, and fractionated cells into cytoplasm, nucleoplasm and chromatin, before western blotting for KDM5C (Fig. 3). Loss of neither the JmjN nor the ARID domain prevents KDM5C being detected in the chromatin fraction. The truncated p.M1_E165del KDM5C is also found in the chromatin fraction (Fig. 3). Fractionation of c.2T>C patient cell lines confirmed endogenous association with chromatin (Supplementary Material, Fig. S2).

Figure 3.

Truncated c.2T>C KDM5C can associate with chromatin. Fractionation of KDM5C-null fibroblasts overexpressing KDM5C-Flag-HA demonstrates that KDM5C does not require the JmjN or ARID domains for chromatin association. TUBULIN (cytoplasm), P300 (nucleoplasm) and H3 (chromatin) loading controls. Right panel, fractionation controls.

Patient mutations in KDM5C reduce protein stability

Global levels of KDM5C protein are reduced in patient fibroblasts versus controls (Fig. 1C), while RNA levels are comparable (Fig. 1B), suggesting that patient mutations could affect KDM5C stability. To directly test this hypothesis, we carried out time-course analyses using the translation inhibitor cycloheximide on control and patient fibroblasts (Fig. 4). KDM5C protein is relatively stable in control fibroblasts, with a half-life of at least 12 h. Fibroblasts with the missense mutation p.D402Y have a KDM5C half-life of ∼4 h, while p.P480L KDM5C shows a more dramatic reduction to around 2 h. Fibroblasts with the c.2T>C mutation also show drastically reduced protein stability, with a half-life of ∼2 h (Fig. 4).

Figure 4.

Patient mutations destabilize KDM5C. Cycloheximide (CHX, 100 μg/ml) time-courses on control and patient fibroblasts demonstrate that KDM5C in patient fibroblasts with c.2T>C, p.P480L and p.D402Y mutations is less stable than KDM5C in control fibroblasts. TUBULIN, loading controls.

Similarly, cycloheximide time-courses on KDM5C-null fibroblasts overexpressing KDM5C with different mutations shows that c.2T>C and c.A1_A495del encoded proteins have substantially shorter half-lives than wild-type KDM5C (Supplementary Material, Fig. S3). Interestingly, a reduction in stability is also seen for KDM5C lacking the JmjN domain, but not for KDM5C lacking the ARID domain, suggesting that it is the absence of the JmjN domain that reduces the stability of c.2T>C patient KDM5C (Supplementary Material, Fig. S3). This is in keeping with the JmjN domain being important for stability in other JmjC-containing demethylases (25,26).

Patient mutations in KDM5C reduce its histone demethylase activity

Some patient missense mutations, including p.D402Y, have been shown to compromise KDM5C demethylase activity (13,14,17), but the effect of the p.P480L mutation has not been investigated. Moreover, the truncated p.M1_E165del KDM5C produced in c.2T>C patients contains the JmjC catalytic domain but is missing other highly conserved domains, and its demethylase activity is not known.

We therefore asked whether patient mutations decreased KDM5C demethylase activity in vitro. We purified wild-type KDM5C and KDM5C containing either a catalytic-null mutation [p.H514A] (13) or patient mutations (p.D402Y, p.P480L, p.M1_E165del) from insect cells (Fig. 5A). We used these purified proteins in demethylase assays on histone peptides (Fig. 5B) or on purified histones (Fig. 5C), and assessed the results using either mass spectrometry or western blot using histone modification antibodies, respectively. It should be noted that it was not possible to express and purify p.M1_E165del to the same level as the other constructs (Fig. 5A), likely due to the inherent instability of this protein, and this is a caveat in assessing its demethylase activity.

Figure 5.

Missense and translation re-start mutations reduce the demethylase activity of KDM5C in vitro. (A) Coomassie and KDM5C blots of KDM5C purified from Sf9 insect cells and used in demethylase assays on peptides (B) or histones (C). Wild-type KDM5C demethylates H3K4me3 and H3K4me2; the H514A catalytic-null form has no activity. p.D402Y, p.P480L and p.M1_E165del mutations reduce KDM5C demethylase activity.

Wild-type KDM5C demethylates H3K4me2 peptides to H3K4me1, and H3K4me3 to H3K4me2 (Fig. 5B), as seen previously (13,14). The enzymatic-dead p.H514A protein shows no detectable demethylase activity on H3K4me2 or H3K4me3 peptides. Patient mutations p.P480L and p.D402Y strongly reduce demethylase activity on both H3K4me2 and H3K4me3 peptides compared with wild-type KDM5C, although they show residual activity compared with the catalytic-null mutant (Fig. 5B).

In addition to demethylase assays using histone peptides, we tested whether patient KDM5C mutations alter demethylase activity on purified histones, which are more physiologically relevant substrates. Wild-type KDM5C demethylates H3K4me2 and H3K4me3 on purified histones as previously reported, but results in no substantial changes in the level of H3K4me1, H3K9me3, H3K27me3, H3K36me2 or H3K36me3 (Fig. 5C). p.H514A catalytic-null KDM5C does not demethylate purified histones on H3K4me2/3, and neither does p.M1_E165del. p.D402Y and p.P480L KDM5C show limited demethylase activity on H3K4me2, whereas the demethylation of H3K4me3 by p.D402Y or p.P480L is comparable to that seen with wild-type KDM5C. The differences in mutant KDM5C activity on histones compared with peptides may be due to the presence of additional modifications on purified histones that may crosstalk with H3K4 methylation.

To complement these in vitro dememethylase assays, we overexpressed KDM5C with N-terminal (Fig. 6A) or C-terminal (Supplementary Material, Fig. S4) tags in control fibroblasts and assessed H3K4me3 at a single-cell level using immunofluorescence microscopy. In cells expressing high levels of wild-type KDM5C, H3K4me3 staining was markedly reduced (arrows, Fig. 6A, Supplementary Material, Fig. S4). This was not the case in cells expressing high levels of the p.H514A KDM5C catalytic mutant, where levels of H3K4me3 were equal across the population of cells regardless of their KDM5C level (Fig. 6A, Supplementary Material, Fig. S4). High levels of p.D402Y induced a reduction in H3K4me3, while H3K4me3 staining remained high in cells overexpressing the p.P480L, p.M1_E165del (Fig. 6A) and c.2T>C (Supplementary Material, Fig. S4) mutants, suggesting that these mutations reduce KDM5C activity in vivo. JmjN and ARID domains are both required for demethylase activity (Supplementary Material, Fig. S4).

Figure 6.

p.P480L and p.M1_E165del KDM5C have compromised demethylase activity ex vivo, and patient fibroblasts demonstrate local gene expression changes. (A) HA-Flag-KDM5C harboring different patient mutations was overexpressed in Control-BJ fibroblasts and the levels of KDM5C (Flag) and H3K4me3 were assessed by immunofluorescence. Cells with high WT or p.D402Y KDM5C showed reduced H3K4me3 staining, demonstrating demethylase activity. Overexpression of p.H514A, p.P480L and p.M1_E165del KDM5C did not reduce H3K4me3 levels. Arrows, cells expressing tagged KDM5C at high levels. (B) Acid histone extracts from patient skin fibroblasts show no consistent change in global H3K4 methylation. H3, loading control. (C) Changes in gene expression in patient fibroblasts compared with controls. Mean and standard deviation of three independent experiments are displayed (*P < 0.05 compared with Control-1, Student's t-test). RNA was detected using primers spanning exon–exon junctions, and normalized to housekeeping gene RSP18 and then Control-1.

Patient fibroblasts show localized changes in gene expression, but not global changes in histone methylation

Total cellular histone methylation levels are determined by multiple different methyltransferases and demethylases. To assess the effect of reduced KDM5C activity in patients (Figs 5 and 6A) on global histone methylation in vivo, we extracted histones from patient and control fibroblasts. Western blots showed no consistent change in H3K4 methylation in patient fibroblasts compared with controls, suggesting that KDM5C mutations do not determine the global levels of H3K4 methylation (Fig. 6B).

Altered gene expression patterns in XLID do not necessarily require a global change in histone methylation, but can result from localized changes at specific loci. This is consistent with increased expression (27), or reduced DNA methylation (22), at certain genes in lymphoblastoid cell lines and blood from patients with KDM5C mutations compared with controls. We tested these previously identified genes for expression changes in patient fibroblasts compared with controls using qRT-PCR (Fig. 6C). We found three genes which showed significant up-regulation in patient fibroblasts compared with controls: EMILIN2 (elastin microfibril interface 2), TNFSF4 [tumor necrosis factor (ligand) superfamily member 4] and ZMYND12 (zinc-finger MYND-type containing 12). These findings support the idea of XLID-associated local changes in H3K4 methylation patterns and gene expression in patient cells.

Discussion

Here we present characterization of four KDM5C mutations that cause ID (Tables 1 and 2). These mutations lead to a reduction in protein level, via NMD (c.3323delG) or through reduced protein stability (p.M1_E165del, p.P480L, p.D402Y) and a reduction in demethylase activity (p.M1_E165del, p.P480L, p.D402Y). Patient mutations do not lead to global changes in histone methylation, but expression changes are seen at specific genes, suggesting localized alterations in chromatin conformation.

Table 2.

Summary of the molecular consequences of KDM5C mutations

cDNA mutation	KDM5C RNA expression	KDM5C protein level	Protein mutation	KDM5C HDM activity
c.1204G>T	Unchanged	Reduced	p.D402Y	Reduced
c.1439C>T	Reduced	Reduced	p.P480L	Reduced
c.2T>C	Unchanged	Reduced levels of truncated protein	p.M1_E165del	Undetectable
c.3223delG	Severely reduced	Undetectable	p.V1075Yfs*2	—

HDM, histone demethylase.

KDM5C mutations reduce protein stability

We have shown that two KDM5C missense mutants, p.P480L and p.D402Y, and a N-terminally truncated protein, p.M1_E165del, exhibit reduced protein stability (Fig. 4). Changes in protein stability due to missense mutations have been described for various diseases (28), but protein stability had not been previously investigated with respect to KDM5C mutations that are compatible with protein production. Destabilization, and consequently reduced KDM5C levels, may be key to the common features of the KDM5C-associated XLID phenotype. It is also pertinent to studies examining other facets of KDM5C function, since reduced levels of endogenous or overexpressed mutant KDM5C may confound assay results.

KDM5C mutations affect catalytic activity to different extents

KDM5C encodes a histone demethylase and one postulated disease mechanism is that reduced KDM5C activity leads to aberrations in the chromatin landscape (3), namely an increase in the active modifications H3K4me2 and H3K4me3. Such changes are predicted to increase expression of KDM5C target genes, with resultant effects on neurological development or function. Assessing demethylase activity of specific patient mutations together with clinical outcomes (Table 3) is therefore informative.

Table 3.

Effect of KDM5C missense mutations on histone demethylase activity and clinical outcomes

DNA mutation	Protein mutation	Demethylase activity	ID	References
c.2T>C	p.M1_E165del	Undetectable	Severe	This study
c.260A>G	p.D87G	Slightly reduced	Mild to moderate	(14)
c.1162G>C	p.A388P	Strongly reduced	Moderate	(13)
c.1204G>T	p.D402Y	Slightly reduced	Moderate/severe	(14), this study
c.1439C>T	p.P480L	Strongly reduced	Mild/moderate	This study
c.1160C>A	p.P554T	Moderately reduced	Moderate	(17)
c.1924T>C	p.F642L	Moderately reduced	Severe	(13)
c.2092G>A	p.E698K	Strongly reduced	Severe	(14)
c.2191C>T	p.L731F	Moderately reduced	Severe	(13)
c.2252A>G	p.Y751C	Moderately reduced	Moderate	(13,14)

ID, intellectual disability.

In this study, we demonstrated that mutant p.P480L, p.D402Y and p.M1_E165del KDM5C show reduced demethylase activity (Figs 5 and 6A). Lack of detectable demethylase activity from p.M1_E165del is consistent with previous data showing loss of KDM5C demethylase activity when KDM5C lacks 69 N-terminal amino acids (14). Demethylase activity of p.D402Y was previously investigated using overexpression and immunofluorescence, which showed a moderate reduction in demethylase activity (14), consistent with our findings (Figs 5 and 6A). Interestingly, Tahiliani et al. (14), characterized several mutations, and suggested that the degree of activity reduction correlated roughly with the severity of ID in the corresponding patients. This is consistent with all patients with nonsense KDM5C mutations having severe intellectual disability (16). We have summarized the current literature regarding the effect of KDM5C mutations on demethylase activity and ID severity in Table 3. Together, these results suggest that while KDM5C demethylase activity is an important determinate of clinical phenotype, there are likely additional factors involved. An important limitation for a correlation of molecular effect and clinical outcome is that the evaluation of patient symptoms were performed by different clinicians for different studies, and is therefore very hard to compare directly.

Furthermore, KDM5C is known to target histones, but is likely to have additional uncharacterized non-histone substrates, as has been suggested for other JmjC-domain demethylases (29). Non-histone targets have been identified for several chromatin regulators, including NSD1 which can methylate NF-Kß (30) as well as H3K36 (31) and is mutated in Sotos syndrome, a disorder associated with ID (32) [OMIM 117550]. Thus, changes in methylation on non-histone targets may be responsible for some features of ID.

Gene expression signatures of patient cells with KDM5C mutations

Despite the reduced catalytic activity of mutant KDM5C, we do not see changes in the global levels of H3K4 methylation in patient fibroblasts (Fig. 6B), consistent with the presence of a myriad of histone methytransferases and demethylases targeting H3K4 (including other KDM5 family members). This suggests that localized changes in H3K4 methylation at specific genomic locations could be responsible for XLID, through altering expression of specific genes. Several studies have looked for local changes in gene expression (27) or DNA methylation (22) in patient lymphoblastoid cell lines or whole blood, and identified specific target genes. We demonstrated that some of these targets are conserved in fibroblasts, showing up-regulation in fibroblasts from patients with KDM5C mutations. The genes that were significantly altered (EMILIN2, TNFSF4, ZMYND12) have no known links with neurodevelopmental disorders, but are expressed in the brain as well as in other tissues. EMILIN2 encodes an extracellular matrix glycoprotein, which may be important in elastin-rich tissues including smooth muscle, blood vessel and skin (33). TNFSF4 encodes a ligand with a role in cell death, most closely studied in T cells, suggestive of a role for inappropriate necrosis in KDM5C-related XLID (34). TNFSF4 is also implicated in narcolepsy (35), intimating neurological importance. Very little is known about ZMYND12, but MYND domains often mediate protein–protein interactions, and so it may exacerbate changes in gene activation patterns. Of note, a ZMYND12 family member, ZMYND8, is known to directly interact with KDM5C (36). The lack of up-regulation of other previously identified KDM5C-target genes may be due to tissue-specificity of epigenetic changes. Genome-wide analysis of H3K4me2/3 and gene expression changes in patient cells, together with KDM5C binding sites, would be highly informative in understanding XLID disease mechanisms. This would be best assessed in neuronal cell types, which are most relevant for the disease.

Translational re-start may mitigate the clinical outcomes of truncating mutations

The c.2T>C mutation was postulated to prevent translation (23), thereby causing loss of KDM5C protein. However, we have demonstrated that an N-terminally truncated KDM5C, p.M1_E165del, is produced (Figs 1 and 2). c.2T>C patients exhibit a severe clinical phenotype, with severe ID, profound speech impairment, short stature, ataxia and a constantly smiling facial expression (23). The presence of almost all reported symptoms of KDM5C-related XLID in c.2T>C patients is consistent with them being functionally null for KDM5C, despite the presence of low levels of p.M1_E165del KDM5C. Interestingly, severe truncating mutations have been found to cause a surprisingly milder outcome due to translational re-start and partial rescue by the truncated protein for disease mutations in ARX (37), ATP7A (38), ERG (39) and DMD (40). Further direct comparison of the c.2T>C family with patients with nonsense or frameshift mutations, and investigation of whether p.M1_E165del KDM5C retains low levels of catalytic activity or other functions, will help elucidate whether the presence of p.M1_E165del KDM5C ameliorates the clinical symptoms seen in c.2T>C patients.

Predicting the effect of KDM5C mutations on biology

It is particularly fascinating to consider the effects of missense alterations on protein function in the context of benign versus pathogenic alterations. This is pertinent as deeper and more routine sequencing enables identification of new variants, whose effects on biology are unknown. We identified many KDM5C missense variants using EVS [Exome Variant Server; (41); 35 variants] and ExAC [Exome Aggregation Consortium; (42); 172 variants]. We used in silico prediction software to anticipate the effects of these variants on protein function (Supplementary Material, Table S1). All previously described patient mutations are considered to be damaging by a range of prediction software, including Polyphen-2 (43) and SIFT (44). The majority of missense variants found in the general population are not expected to be deleterious (SIFT: 142/207 tolerated), but damaging variants are predicted. This is hard to interpret for several reasons. Firstly, there are likely to be female heterozygotes who may not show disease symptoms. Secondly, without having detailed clinical information it is impossible to assess whether these variants are benign or pathogenic. However, it is interesting to note that of the 207 missense variants, there are none in the JmjN domain and only five in the JmjC domain, of which four are predicted to be damaging by SIFT. This suggests intolerance to, or strong selection against, variation within domains critical for KDM5C demethylase activity.

In summary, this study emphasizes the value of dissecting the molecular consequences of individual patient mutations to further understand protein stability and function, translation re-start and the contribution of KDM5C-dependent demethylation to the etiology of ID.

Materials and Methods

All human subjects research was carried out in accordance with approved ethics guidelines at all locations (Boston Children's Hospital IRB-P00002276).

Cell culture

Fibroblasts were cultured in DMEM supplemented with 10% fetal bovine serum, 2 mm l-glutamine, 1% MEM non-essential amino acids and 100 U/ml penicillin/streptomycin. Cycloheximide (Sigma) was re-suspended in DMSO and used at a final concentration of 100 μg/ml.

Antibodies

Anti-KDM5C antibody was raised by immunizing rabbits with the C-terminal segment of human KDM5C (1459–1559 amino acid, NP_001269551), which was expressed and purified as a histidine-tagged protein in E. coli. Resultant serum was affinity purified using the immunized protein as the ligand. For western blotting, we use anti-histone H3 C-terminus (ab1791), anti-H3K4me0 [CMA301, kind gift from Hiroshi Kimura (45)], anti-H3K4me1 (Active Motif 39297), anti-H3K4me2 (Millipore 07-030), anti-H3K4me3 (Abcam ab8580), anti-H3K9me3 (Abcam ab8898), anti-H3K27me3 [CMA323, kind gift from Hiroshi Kimura (46)], anti-H3K36me2 (Abcam ab9049), anti-H3K36me3 [13C9, kind gift from Hiroshi Kimura, (47,48)], anti-TUBULIN (Sigma T9026) and anti-ACTIN (Sigma A3854) antibodies. For immunofluorescence, we used anti-Flag M2 (Sigma F1804) and anti-H3K4me3 (Abcam ab8580).

Western blotting

Whole cell lysates were made using RIPA buffer. Histones were extracted as described previously (49). Lysates were run on 8% (KDM5C) or 15% (histones) SDS–PAGE gels and transferred to nitrocellulose membranes. Membranes were blocked (1 h), incubated overnight with primary antibody, washed and incubated (1 h) with HRP-conjugated secondary antibodies, all in 0.05% Tween/PBS with 5% non-fat dry milk. HRP-conjugated antibodies were detected with ECL western blotting detection reagents according to the manufacturer's instructions.

qRT-PCR

Total RNA was isolated using TRIzol reagent (Life Technologies) according to the manufacturer's instructions. Total RNA (1 μg) was retrotranscribed with 10 U Superscript III reverse transcriptase (Life Technologies). Expression levels were normalized to the RSP18 house-keeping gene. Primer sequences are in Supplementary Material, Table S2.

Overexpression or shRNA knockout of KDM5C

Fibroblasts were plated in 12 well plates (2.5 × 10⁴ cells/well), and treated with retrovirus (pMSCV) carrying KDM5C overexpression constructs or shRNA against firefly luciferase or KDM5C.

Cellular fractionation

We fractionated fibroblasts into cytoplasm, nucleoplasm and chromatin fractions using the Pierce Subcellular Fractionation kit according to the manufacturer's instructions.

Demethylase assays

Wild-type and mutant KDM5C proteins were cloned into a baculovirus expression vector, pFastBacHT A (Life Technologies) and then expressed in Sf9 insect cells using the Bac-to-bac baculoviral expression system (Life Technologies). Cells were lysed in 10 mm HEPEs pH7.6, 3 mm MgCl_2, 310 mm KCl, 5% glycerol, 0.5% NP40, 1 mm PMSF and Protease Inhibitor Cocktail (Roche). Recombinant proteins were immobilized on Ni-NTA agarose (Qiagen), washed with Wash buffer (50 mm NaH₂PO₄, 300 mm NaCl, 20 mm imidazole) and eluted in Wash buffer containing 250 mm imidazole. Ten to twenty micrograms of purified KDM5C were incubated with 2 μg of histone peptides or 3 μg of calf histones for 4 h at 37°C in the demethylase reaction buffer (50 mm Tris–HCl, 50 mm NaCl pH7.5, 1 mm MgCl₂, 1 mm α-ketoglutarate, 100 μm [NH₄]₂Fe[SO₄]₂, 1 mm α-ketoglutarate, 2 mm ascorbic acid) as described previously (50).

Immunofluorescence

Fibroblasts were plated in 12-well plates (2.5 × 10⁴ cells/well), and treated with retrovirus carrying HA-Flag-KDM5C or KDM5C-Flag-HA with different patient mutations. Cells were allowed to reach confluence and then split onto coverslips. Cells were fixed with 2% formaldehyde for 10 min, and then with 100% ice-cold methanol for 10 min. Cells were permeabilized with 0.1% NP-40 in PBS, and blocked (1 h) with 20% FBS in PBS. Coverslips were incubated overnight at 4°C with the appropriate primary antibodies, washed and incubated with the corresponding secondary antibodies for 30 min at room temperature. Alexa 594-conjgated donkey-anti-rabbit and Alexa 488-conjugated goat-anti-mouse secondary antibodies (Molecular Probes) were used at 1:1000. Coverslips were then washed, mounted with Vectashield containing DAPI (Vector Laboratories) and analyzed by a fluorescence microscopy (Nikon E600) using a ×60 objective.

Supplementary Material

Supplementary Material is available at HMG online.

Funding

This work was supported by the European Molecular Biology Organisation (Long-term postdoctoral fellowship to E.B., 23-2012), the Jérôme Lejeune Foundation and the National Institutes of Health [MH096066 to Y.S. and Jun Xu (Washington State University)].