Advances in molecular technologies have helped to identify genetic causes in many cases of syndromic and non-syndromic forms of intellectual disability (ID) [57–59]. However, the molecular pathogenesis of ID still remains incompletely understood. It has been suggested based on known genetic etiologies that perturbed neuronal homeostasis altering synaptic outputs could be a key component of the cognitive impairment phenotype . This notion is strongly supported by the types of functions attributable to genes mutated in ID which include basic cellular functions, such as transcription, translation, RNA biogenesis, protein turnover, and cytoskeletal dynamics . One important emerging mechanism in ID is epigenetic dysregulation that ultimately affects transcription of multiple genes [1, 57]. KDM5C is one of more than 20 epigenetic regulators involved in ID . The identification of specific downstream targets exhibiting aberrant epigenetic marks in response to mutation of an epigenetic regulator will have an important impact on our understanding of the molecular pathogenesis of ID.
Previously, profiling of mRNA in lymphoblastoid cell lines of 12 males with KDM5C mutations compared to 5 controls identified 11 upregulated genes. These transcriptional changes were not very consistent among KDM5C mutations samples, and a combination of at least 6 genes was required to distinguish cases from controls . None of these 11 genes exhibited DNA methylation changes in our dataset. Thus, it is likely that these expression differences are more tissue and developmental stage specific, than the DNA methylation patterns, potentially reflective of disrupted binding of KDM5C specifically in lymphoblastoid cell lines. DNA methylation patterns can be maintained by DNMT1 through replication and multiple cell divisions , and thus if they occur early in development they could be represented in multiple lineages including peripheral blood, but not be reflective of gene expression patterns in all differentiated lineages. This has, in fact, been observed in neurodevelopmental syndromes such as Immunodeficiency–centromeric instability–facial anomalies (ICF), Fragile-X, Angelman and Prader-Willi syndromes . Further, the regions where we found loss of DNA methylation associated with KDM5C mutations, coincided with an enhancer mark H3K4me1 in ES and lymphoblastoid cell lines. In contrast, DNA sequences more proximal to the TSS were hypomethylated in both controls and cases and coincided with the active promoter mark H3K4me3 (Figure 4, Additional file 2: Figures S6-S9) . These data suggest that regions affected by loss of DNA methylation have enhancer-driven rather than basal promoter function. As enhancers are involved in the control of spatial and temporal gene expression , the relationship between loss of DNA methylation at identified sites and expression at downstream genes is likely to be more complex than a simple inverse correlation.
The mechanism of the observed loss of DNA methylation associated with loss of function mutations in KDM5C is not completely clear; however, it is unlikely to be the direct consequence of loss of KDM5C function, as KDM5C is not known to possess DNA methyltransferase activity. Based on the current literature, the most plausible mechanism is that a deficiency in H3K4 demethylase activity leads to increased H3K4 methylation, which protects DNA from de novo DNA methylation at KDM5C downstream target loci. In mouse ES cell, Dnmt3L recruits de novo methyltransferases to DNA associated with unmethylated forms of H3K4, and contact between Dnmt3L and the nucleosome is inhibited by all forms of H3K4 methylation . Biochemical assays have shown that the human de novo methyltransferase DNMT3A interacts with histone H3 unmethylated at K4, whereas di- and tri-methylation inhibit this interaction . There is also evidence from a yeast model system lacking endogenous DNA methyltransferases and ectopically expressing mouse Dnmt3a and Dnmt3L that depletion of H3K4 methylation results in increased DNA methylation . In humans, a correlation of increased tri-methylation in H3K4 with reduced DNA methylation at promoters has been shown in fibroblast cells from normal individuals .
Our data demonstrating DNA methylation alterations in individuals with mutations in the KDM5C gene further support the link between H3K4 methylation and DNA methylation. These data show, for the first time, a functional consequence of loss of function of H3K4 demethylase resulting in significant alterations of DNA methylation at specific gene targets at a genome-wide level. Furthermore, in agreement with cross-talk of H3K4 methylation and DNA methylation, there was significantly more loss than gain of DNA methylation resulting from mutations in KDM5C. In mammals several H3K4 demethylases have been described. Apart from four enzymes of the KDM5 family, KDM1A and B specifically act to demethylate di- and mono-methylated forms of H3K4, and KDM2B similarly to the KDM5 family specifically demethylates tri- and di-methylated forms of H3K4 [67, 68]. At this point it is not clear if the loss of DNA methylation associated with KDM5C mutations is specific to KDM5C loss of function or would also be observed in the context of loss of function of other H3K4 demethylases. In support of the concept that loss of DNA methylation could be a common association of loss of H3K4 demethylase activity, mouse oocytes deficient in the H3K4 demethylase Kdm1b demonstrated a global increase in H3K4 di-methylation and failed to generate normal DNA methylation marks at several imprinted loci . However, as we did not observe loss of DNA methylation at imprinted genes in patients with KDM5C mutations, the genomic sites demonstrating loss of DNA methylation for each H3K4 demethylase are likely to be specific, possibly reflecting the binding sites of these proteins.
Further support for the inter-dependence of histone methylation and DNA methylation in humans comes from cancer research. Mutations in IDH1 and IDH2, frequently found in gliomas and acute myeloid leukemias (AML), are characterized by enzymatic gain of function and subsequent production of hydroxyglutarate, which inhibits several histone demethylases, including H3K9, H3K27, H3K36 and H3K4 . The somatic mutations in IDH1/IDH2 are associated with genome-wide hypermethylation in AML compared either to normal bone marrow or to AML caused by mutations in other genes. However at this point it is not clear which specific histone marks contribute directly to this DNA hypermethylation phenotype . Interestingly, two genes, PABPN1 and ZNF532, demonstrating loss of DNA methylation in our study (Additional file 1: Table S4) were found to be hypermethylated in AML with IDH1 mutation . These data suggest that there could be some common mechanism regulating DNA methylation of these two genes in opposite directions in the context of loss of function of KDM5C and gain of function of IDH1.
The genes FBXL5, SCMH1 and CACYBP on which we focused in our downstream analysis have exhibited a surprisingly large degree of DNA methylation differences between cases and controls reminiscent of the DNA methylation alterations at imprinted loci in disorders affecting neurodevelopment, such as Prader-Willi and Angelman syndromes [71, 72]. Further, loss of DNA methylation at these sites was not observed in 946 population control blood samples from publically available datasets, suggesting that altered DNA methylation at these three genes could be used for establishing pathological authenticity of new missense mutations, as in silico predictions of effects on protein function are often inconclusive, and functional experiments are expensive and labor intensive. Thus, as KDM5C mutation cases are frequently indistinguishable from other genetic causes of ID based on clinical phenotype alone , DNA methylation analysis could complement KDM5C sequencing to provide more accurate molecular diagnosis leading to improved patient management.
Interestingly, the three top candidate genes are part of ubiquitin-ligase protein degradation pathways. Synaptic network remodeling, a vital part of central nervous system function, depends on ubiquitin-mediated protein degradation at the postsynaptic membrane . Genes involved in ubiquitination pathways have already been implicated in a number of other neurodevelopmental disorders, such as Angelman syndrome (loss of function of maternal copy of UBE3A) , and autism (copy number variants in PARK2, RFWD2, FBX040) . Furthermore, 7% of XLID genes identified to date are components of the ubiquitin pathway .
We have shown that these three genes are ubiquitously expressed in human tissues at relatively low levels compared to the house-keeping gene GAPDH. Although it is currently not clear how loss of DNA methylation at these sites affect gene expression, we propose that abnormal expression of these genes at specific cell types/developmental stages causes disturbances in downstream pathways such as degradation of target proteins. FBXL5 has been recently discovered to be a component of an E3 ubiquitin ligase complex that targets IRP2, an iron regulatory protein 2 important in intracellular and plasma iron homeostasis [43, 76]. IRP2 regulates RNA stability and translation by binding to iron responsive - RNA stem loop structures, in a number of genes involved in iron uptake, storage and utilization . While FBXL5 loss of function leads to embryonic lethality in Fbxl5−/− mice, associated with aberrant iron accumulation and increased oxidative stress , Irp2 −/− knockout mice exhibit a neurological phenotype associated with locomotor abnormalities accompanied by iron accumulation in white and grey matter . Thus, abnormal expression of FBXL5 could result in abnormal iron accumulation, thereby contributing to seizures and/or ID phenotypes observed in males with KDM5C mutations. SCMH1 is a member of the Polycomb-group 1 complex, which not only is a transcriptional repressor, but also acts as an E3 ubiquitin ligase for the geminin protein involved in DNA replication and maintenance of undifferentiated cellular states. Specifically, SCMH1 has been shown to provide an interaction domain for geminin . Recently a genome-wide association study implicated SCMH1 in the regulation of human height , thus it is possible that loss of DNA methylation at SCMH1 is important for short stature associated with KDM5C mutations. CACYBP (Sip) is part of the SCF-like complex, involved in ubiquitin-mediated degradation of the transcriptional activator β-catenin [40, 41]. β-catenin is a signaling molecule playing an important role in neurodevelopmental processes such as neural crest development, development of cortical and hippocampal neuroepithelium, and dendrite spine morphogenesis [80–82]. In addition, it has been implicated in seizure susceptibility . Further, CACYBP was shown to dephosphorylate ERK1/2 , extracellular signal-regulated kinases, important in many aspects of early brain development and implicated in 16p11.2 and 22q11 deletion syndromes phenotypes .
As the described DNA methylation changes in our study were identified in blood samples, an important question that cannot be directly addressed by our data is the issue of whether parallel changes occur in brain. As de novo DNA methylation is an important process in epigenetic reprogramming occurring at early stages of embryonic development [22, 86], we suggest that loss of DNA methylation in the blood of patients with KDM5C mutations could at least in part result from abnormally high H3K4 di/trimethylation in the embryo, protecting DNA from de novo methylation. We expect that this state is maintained through differentiation into multiple lineages. In support of this, we found parallel sex-specific DNA methylation differences in both brain and blood at FBXL5 and CACYBP, whereas SCMH1 exhibited this difference only in blood, but not in brain. We suspect that these observed sex-specific differences are due to KDM5C/KDM5D dosage rather than the effects of sex hormones, as the DNA methylation at tested targets correlates better with sex chromosome constitution than with gonadal sex, e.g. the highest DNA methylation was observed in 47,XXX females, followed by 47,XXY males, 46,XX females, 46,XY males, 45,X females. The lowest DNA methylation is seen in males with KDM5C mutations (Figure 7). In addition we observed that female carriers of KDM5C mutations have DNA methylation levels similar to 45,X females, reflecting the fact that they have only one functional copy of KDM5C.
Based on DNA methylation comparison between brain and blood, we propose that the epigenetic status of FBXL5 and CACYBP is regulated by KDM5C in both brain and blood, and that their deregulation in brain can contribute to the intellectual disability and seizure phenotypes in individuals with KDM5C mutation. We propose as well that KDM5C contributes to sex-specific differences in brain function. In contrast, SCMH1 might be responsible for other aspects of the clinical phenotype associated with KDM5C mutation such as growth abnormalities. Furthermore, the KDM5C-mutation associated targets identified here could play a role in Turner syndrome. It has been previously suggested that X-linked genes escaping X-inactivation such as KDM5C are likely to be implicated in neurocognitive phenotypes of 45,X females with Turner syndrome, who in spite of normal cognitive abilities, frequently have problems in spatial reasoning and emotion recognition [87, 88]. Our observation of loss of DNA methylation at the FBXL5, SCMH1 and CACYBP promoters in 45, X females compared to XX females and XY males, but to a lesser degree than in males with KDM5C mutations, supports this hypothesis and suggests that deregulation of epigenetic targets of KDM5C could be relevant to the mild neurodevelopmental impairments found in females with Turner syndrome. Similarly, loss of DNA methylation at these three genes found in female carriers of KDM5C mutations could contribute to learning difficulties frequently observed in such individuals [5, 7].
In summary these data provide new opportunities to address the molecular basis, both genetic and epigenetic, of ID. An important area for future investigation would be to establish both spatial (tissue-specific) and temporal (developmental stage- specific) maps of KDM5C targets, and to annotate how loss of KDM5C function impacts expression of these targets through embryonic development and in diverse tissues. Validation of the affected molecular pathways, described here such as abnormal iron homeostasis or β-catenin dysregulation could also, in an animal model of KDM5C mutations, provide a framework for potential therapeutic developments for patients with KDM5C mutations.