DNA methylation is an epigenetic modification essential for normal mammalian development. Initially associated with gene silencing, more diverse roles for DNA methylation in the regulation of gene expression patterns are increasingly being recognized. Some of these insights come from studying the function of genes that are mutated in human diseases characterized by abnormal DNA methylation landscapes. The first disorder to be associated with congenital defects in DNA methylation was Immunodeficiency, Centromeric instability, Facial anomalies syndrome (ICF). The hallmark of this syndrome is hypomethylation of pericentromeric satellite repeats, with mutations in four genes: DNMT3B, ZBTB24, CDCA7 and HELLS, being linked to the disease. Here, we discuss recent progress in understanding the molecular interactions between these genes and consider current evidence for how aberrant DNA methylation may contribute to the abnormal phenotype present in ICF syndrome patients.
DNA methylation as a regulatory epigenetic mark
DNA methylation, which generally refers to the direct chemical modification of DNA, was discovered more than 80 years ago  and is now one of the most studied and best characterized epigenetic modification. In the late 1970s, a role for DNA methylation in the regulation of gene expression was first demonstrated [2,3], and today its critical function in diverse biological processes, such as transcriptional regulation, silencing of transposable elements, genomic imprinting and X-chromosome inactivation, is well established . Not surprisingly, aberrant DNA methylation is a feature of many different types of cancer , and genetic defects in components of the DNA methylation machinery have been linked to human congenital disease .
In mammals, DNA methylation predominantly occurs on cytosine bases in the symmetrical CpG dinucleotide context, which enables the post-replicative maintenance of DNA methylation patterns. Specifically, the fifth carbon of the cytosine pyrimidine ring is covalently modified by the addition of a methyl group . DNA methylation is widespread in mammals, and approximately 70–80% of all CpGs are methylated. Depending on the genomic context in which it is placed, DNA methylation can be interpreted differently and play diverse regulatory roles. For example, transposable elements, which are dispersed throughout the genome and account for approximately 40% of the mammalian genome , are highly methylated sequences . In somatic cells, DNA methylation is required to suppress transposon expression and mobility , thereby protecting the genome from deleterious effects of these elements which when mobilized, can cause genome instability . In contrast, CpG islands (CGI), which are CpG-rich sequences often found in the vicinity of promoters, are usually devoid of DNA methylation . Promoter DNA methylation correlates with gene silencing and typically occurs in a tissue- or developmental stage-specific manner [11–13]. Historically, DNA methylation has been associated with gene repression, but it is now known that DNA methylation can also be associated with active transcription, when found across gene bodies . Gene body methylation has been suggested to prevent transcription from intragenic/cryptic promoters , to play a role in the regulation of splicing [15–17] and to function in transcript elongation . Nevertheless, further studies are required to fully understand the implications of heavily methylated gene bodies.
The DNA methylation machinery
DNA methylation patterns are established by the de novo methyltransferases, DNA methyltransferase 3 alpha (DNMT3A) and DNA methyltransferase 3 beta (DNMT3B) during early embryonic and germline development (Figure 1) [19,20]. The catalytic domains of DNMT3s lack DNA sequence specificity and have non-selective activity, but these enzymes can be targeted to or excluded from selected genomic regions by numerous mechanisms including post-translational modifications and via interaction with instructive partner proteins . For example, DNMT3A can form a complex with its catalytically inactive cofactor, DNA methyltransferase 3 like (DNMT3L), which stimulates DNMT3A activity during germline development [21,22], and in mouse embryonic stem cells (mESCs) . At the N-terminus, DNMT3 enzymes contain a Pro-Trip-Trip-Pro (PWWP) domain, which can specifically recognize histone H3 molecules trimethylated at lysine 36, a histone modification enriched at actively transcribed gene bodies [24–26]. Another important domain for chromatin interaction is the ATRX-DNMT3-DNMT3L (ADD) domain, which enables DNMT3s to specifically recognize unmodified histone H3 lysine 4 [27–29], while methylation of histone H3 lysine 4 is known to inhibit this recognition [27,29,30]. Consequently, histone H3 lysine 4 tri-methylated (H3K4me3), a mark associated with active transcription, is thought to protect promoter CpGs from gaining a methyl group and thus prevents gene repression . Once cell fate specification is completed, DNA methylation patterns are maintained after every mitotic cell division by the maintenance DNA methyltransferase 1 (DNMT1) . Through interaction with proliferating cell nuclear antigen (PCNA), DNMT1 can localize to replication foci during the S-phase in dividing cells [31,32]. Interaction of DNMT1 with ubiquitin-like, containing PHD and RING finger domains 1 (UHRF1) is necessary to guide DNMT1 to hemi-methylated DNA [33,34] (Figure 1).
Model of functions of the DNA methylation machinery in establishment and maintenance of DNA methylation patterns
For the correct placement of DNA methylation patterns and normal development, DNA methylation first needs to be erased during early preimplantation development and germline formation, a process often referred to as epigenetic reprogramming . DNA methylation can be removed by two distinct mechanisms. Passive demethylation refers to the dilution of DNA methylation through cell divisions and can be caused by the absence or inhibition of DNMT1 [36,37], absence of UHRF1  or delocalization of UHRF1 and DNMT1 to the cytoplasm [39,40]. Stella (Dppa3) is involved in the delocalization of UHRF1 in oocytes . Active demethylation involves the removal of the methyl group from 5-methylcytosine (5mC) and is carried out by the ten-eleven translocation (TET) family of proteins. The three family members, TET1, TET2 and TET3, exhibit oxidizing activity and can catalyse the conversion of 5mC to 5-hydroxymethylcytosine (5hmC), 5-formylcytosine (5fC) and 5-carboxylcytosine (5caC) [42,43]. In addition, thymine-DNA-glycosylase (TDG)-catalyzed base excision and the DNA base excision repair pathway can remove 5fC and 5caC so that unmodified cytosines can be incorporated  (Figure 1).
Immunodeficiency, centromeric instability, facial anomalies syndrome
Genetic defects in all three catalytically active DNA methyltransferases, DNMT1, DNMT3A, and DNMT3B, have been associated with human congenital disorders [6,45,46], emphasizing the importance of DNA methylation for normal mammalian development. One of the earliest reports of abnormal DNA methylation patterns in disease was in patients with Immunodeficiency, Centromeric instability, Facial anomalies syndrome (ICF; OMIM 602900) approximately 40 years ago [47–49]. ICF syndrome is a rare, autosomal recessive disorder and less than 100 patients have been reported worldwide . The main characteristics of the disease are unusual facial features, reduced levels or absence of serum immunoglobulins, and chromosome instability which is reflected in aberrant configurations of chromosomes 1, 9, and 16 in mitogen-stimulated lymphocytes . The mild facial dysmorphisms frequently include hypertelorism, epicanthic folds, a flat nasal bridge, and low-set ears . Most ICF patients are diagnosed at a young age, suffer from recurrent gastrointestinal and respiratory tract infections, sepsis and a failure to thrive, which often results in early childhood death . An early study on ICF patient-derived peripheral blood found a lack of memory B and plasma cells that could explain the hypo- and agammaglobulinemia phenotype , and hematopoietic stem cell transplantation has been used to treat ICF patients [51,54]. DNA hypomethylation of pericentromeric satellite 2 and 3 repeats  is the molecular hallmark of the disease and makes ICF syndrome the first example of a human disorder linked to a constitutive defect in DNA methylation. The finding that the centromeric α-satellite repeats are heterogeneously affected by the loss of DNA methylation among ICF patients has led to the early recognition of genetic heterogeneity of the disease .
The genetics of ICF syndrome
In 1999, the first mutation underlying the ICF syndrome was described and found to be located in the gene that encodes DNMT3B [20,56,57]. As mentioned above, DNMT3B is a de novo methyltransferase involved in the establishment of DNA methylation patterns early in life and during cell differentiation. The majority of ICF1 patients carry missense mutations in the C-terminal part of DNMT3B, where the catalytic domain is located. In addition, DNMT3B nonsense and splice-site mutations have been reported, and all patients described to date appear to have impaired methyltransferase activity [58,59]. DNMT3B nonsense mutations only occur together with missense mutations, and it has been suggested that complete loss of DNMT3B catalytic activity is not compatible with life . Consistent with this hypothesis, deletion of Dnmt3b in mice results in embryonic lethality , while mice carrying an ICF patient mutation in Dnmt3b are viable and resemble some of the phenotypic features of ICF syndrome patients [58,61]. Patients with DNMT3B mutations are designated ICF type 1 (ICF1) and account for approximately half of the known ICF patients. Three additional ICF syndrome-associated genes have been identified. Nonsense mutations in zinc-finger and BTB domain-containing 24 (ZBTB24) are found in approximately 30% of ICF patients, referred to as ICF2 . Twelve ICF cases with mutations in cell division cycle associated 7 (CDCA7) or in helicase, lymphoid specific (HELLS) have been reported [63,64], and are referred to as ICF3 and ICF4, respectively. A few patients remain who showed molecular and phenotypic features characteristic of ICF syndrome but do not have mutations in the four known genes, and were referred to as ICFX .
Functions for ICF genes in DNA methylation pathways and beyond
The previously reported differences in repeat hypomethylation in ICF patients [55,65] can now be associated with the different genetic defects. The α-satellite repeat hypomethylation phenotype is shared between ICF2, ICF3, and ICF4 types , while DNA methylation levels of subtelomeric repeats are not affected in these patients . In contrast, heavily methylated α-satellites  and the hypomethylation of subtelomeric repeats  are unique features of patients carrying DNMT3B mutations, and can be used to distinguish ICF1 from other ICF patients. Consistent with a function for DNMT3B in the establishment of DNA methylation in early development [11,68], ICF1 patients show promoter hypomethylation of germline genes  and loss of methylation at X-linked genes ; these sequences are not affected in ICF2, ICF3, and ICF4 patients . Indeed, in both mice and humans, it has been shown that ZBTB24, CDCA7, and HELLS are involved in the DNMT1-dependent DNA methylation maintenance pathway. In mouse embryonic fibroblasts (MEFs), siRNA-mediated depletion of Zbtb24, Cdca7, and Hells resulted in loss of DNA methylation at minor satellite repeats . Similarly, a progressive loss of DNA methylation at satellite repeats has been reported in human embryonic kidney (HEK) 293T cells knocked out for ZBTB24, CDCA7, or HELLS . Interestingly, transient depletion of each of the four ICF factors in normal human primary fibroblasts did not result in hypomethylation of satellite 2 or subtelomeric sequences . Biochemical studies have demonstrated that Hells can interact with Dnmt1 in vitro , and interactions between CDCA7 and HELLS [66,70,72], and CDCA7 and UHRF1 have also been reported . A recent study in Xenopus showed that Cdca7 is required for localization of Hells to chromatin and that Cdca7 can stimulate Hells nucleosome remodeling activity . Therefore, abnormal DNA methylation patterns in ICF3 patients could be the result of a defect in HELLS recruitment to chromatin . Of note, some hypomethylated loci are shared between ICF patients from all four types , and there is evidence that Hells participates in Dnmt3a/b-mediated de novo methylation in MEFs and during the differentiation of mESCs [73,74]. Thus, roles for ZBTB24, CDCA7, and HELLS in de novo DNA methylation cannot be excluded, but are as yet undetermined.
Depending on the developmental stage, targeted genomic location, protein complex composition, cell cycle, or post-translational modifications, ZBTB24, CDCA7, and HELLS could also have functions that differ from DNA methylation regulation but are relevant to ICF syndrome pathology. For instance, both CDCA7 and HELLS have recently been linked to the classical non-homologous end joining (c-NHEJ) pathway . CDCA7 was shown to interact with members of the c-NHEJ pathway, and depletion of both CDCA7 and HELLS in HEK293T cells was associated with c-NHEJ defects and delayed accumulation of Ku80 at sites of DNA damage . Unresolved Holiday junctions have been suggested to underlie multiradial chromosome formation in ICF patients , and the findings by Unoki et al.  provide supporting evidence for this hypothesis.
In the case of ZBTB24, its function as a transcription factor and interaction with CDCA7 is now beginning to be understood. The functional connection between ZBTB24 and CDCA7, whereby ZBTB24 positively controls CDCA7 expression, was initially demonstrated in a mESC-based Zbtb24 knockout model and in fibroblasts and T cells derived from ICF2 patients . Subsequent studies have confirmed these results [66,70,77–79]. Recent genome-wide ChIP-seq studies reported that ZBTB24 is mainly enriched at CpG-rich promoters [77,78] and that its C2H2 zinc fingers direct sequence-specific binding . Down-regulation of cdca7 has also been observed in a zbtb24 knockout zebrafish model , demonstrating that CDCA7 is probably a highly conserved ZBTB24 target.
ICF syndrome patients have an altered DNA methylation landscape
Pericentromeric satellites 2, 3 and centromeric α-satellites are classical targets of DNA hypomethylation in ICF syndrome patients. In addition, abnormal DNA methylation of subtelomeric repeats is a characteristic feature of ICF1 patients [67,81,82]. Initially ICF-related DNA hypomethylation appeared to be confined to the repetitive compartment of the genome [65,67,83–87], and only a few single copy loci exhibiting DNA hypomethylation in ICF patients were known [69,88–92]. Rapid developments in next-generation sequencing technologies and the identification of ZBTB24, CDCA7, and HELLS as ICF disease genes have stimulated research in this area, allowing a more detailed view of the ICF-related methylome. The first genome-wide DNA methylation map, generated from ICF1 patient-derived cultured lymphoblastoid cell lines (LCLs), has revealed widespread hypomethylation and a 41% reduction in global methylation levels when compared with a control individual . All autosomes were equally affected and loss of methylation was not restricted to repetitive regions but detected along all genomic features including promoters, gene bodies, and intergenic regions  (Figure 2). A later study, using independent ICF1 patient-derived cultured LCLs and reduced-representation bisulfite sequencing (RRBS), reported similar findings with respect to global DNA methylation alterations . In mESCs, it has been reported that Dnmt3b is the main enzyme responsible for genic DNA methylation , and mESCs lacking Dnmt3b lost DNA methylation across exons and introns . Consistent with this, ICF1 patients show hypomethylation of gene bodies . It has also been shown that most DNMT3B binding occurs at intragenic positions of genes and that DNMT3B binding is not impaired upon the introduction of an ICF1 mutation in the catalytic domain . Intriguingly, DNMT3B patients with a homozygous missense mutation close to the PWWP domain have been reported , demonstrating that the ICF syndrome phenotype can be recapitulated by mutations in the regulatory N-terminal part of DNMT3B. Patients carrying the PWWP mutation show DNA hypomethylation of satellite 2 repeats , and it will be interesting to determine whether they have additional methylation defects that are similar or different from the ones that have been described for ICF1 patients with impaired DNMT3B catalytic activity. Of note, introduction of the PWWP mutation into mESCs resulted in loss of Dnmt3b recruitment to histone H3 lysine 36 tri-methylated (H3K36me3)-decorated gene bodies .
Schematic of the mammalian DNA methylation landscape and the putative influence of ICF genes on DNA methylation
While ICF1 patients show prominent loss of DNA methylation in CpG-rich regions (CGI and CpG shores) and gene bodies, it has become evident that in addition to the satellite repeats, CpG-poor, heterochromatic regions that are often referred to as ‘open sea’, are predominantly affected in ICF2, 3, and 4 patients (Figure 2). This was only recently revealed in a comprehensive study comparing ‘primary’ whole blood methylomes generated from 15 patients comprising samples from all four ICF types and ten gender- and age-matched controls . Many of the hypomethylated CpGs in ICF2, 3, and 4 patients mapped to late-replicating regions and were found to be located in the open sea regions of the genome , which are characterized by low CpG density . Heterochromatin is usually characterized by the presence of the repressive histone modifications histone H3 lysine 9 di/tri-methylated (H3K9me2 and H3K9me3), and binding of HP1 . DNA methylation can contribute to heterochromatin formation through interaction with components of the chromatin remodeling and histone modification machinery, since DNMTs cannot bind and efficiently methylate nucleosomal DNA [98,99]. For instance, DNMTs can interact with the chromatin remodeler HELLS (also known as LSH)  or histone-modifying enzymes, such as suppressor of variegation 3-9 homolog 1 (SUV39H1) , and the H3K9me3 reader, UHRF1 . In addition, DNMT3B can be directly recruited to centromeric heterochromatin by centromere protein C (CENPC)  or the linker histone H1 [103,104]. How ZBTB24, CDCA7, and HELLS contribute to DNA methylation and possibly heterochromatin formation at (peri)centromeric repeats and open sea regions, and why these sites are specifically affected in ICF2, ICF3 and ICF4 patients, remains an open question. Gene regulatory elements are also located in open sea regions, and it has recently been suggested that short-term depletion of Hells in MEFs is associated with increased chromatin accessibility at selected genes and enhancers . Hypomethylation of large chromosomal domains has been observed in MEFs derived from Hells−/− embryos, and a function for Hells in controlling DNA methylation in a nuclear compartment partly defined by lamina-associated domains (LADs) has been proposed . Given the functional connections between the three disease genes and the overlapping hypomethylation phenotype in ICF2–4 type patients it will be interesting to test whether ZBTB24, CDCA7, and HELLS may also have roles in LAD and/or enhancer regulation.
Members of big gene cluster families including the olfactory receptor genes and the clustered protocadherins (PCDH) exhibit aberrant DNA methylation in all four ICF patient types . Epigenetic mechanisms play important roles in the regulation of gene clusters  and their dysregulation could be associated with ICF syndrome phenotypes. For example, the clustered PCDH genes, which are predominantly expressed in the nervous system, are key components of neuronal diversity . In mice, it has been shown that Dnmt3b is essential for the de novo methylation of Pcdh cluster genes in early embryonic development . Defects in Pcdh cluster regulation have also been reported upon neuronal ablation of the H3K9me3 histone methyltransferase, Setdb1, in mice , suggesting cross-talk between DNA methylation and histone modification pathways. In addition, Setdb1 is important for preserving a large topologically associated domain (TAD) encompassing the Pcdh locus . Since intellectual disability is a common feature in ICF syndrome patients , it will be interesting to determine whether defects in PCDH gene cluster regulation can contribute to aspects of the disease.
Finally, it is worth mentioning that abnormal DNA hypermethylation of different genomic features including promoters has been observed in ICF patients [66,93,94]. Although its relevance is not yet understood, given the well-established function of DNA methylation in gene silencing, it can be envisioned that DNA hypermethylation may also contribute to the phenotypes observed in ICF syndrome patients.
Linking aberrant DNA methylation to ICF syndrome phenotype
Indeed, the challenge now is to link DNMT3B-, ZBTB24-, CDCA7-, and HELLS-related genome-wide DNA methylation alterations to gene expression changes that could explain ICF patient phenotypes. For a large part, progress in this area has been hampered by the scarcity of ICF patient-derived tissues or cell lines. Consequently, an integrated analysis of genome-wide DNA methylation and gene expression data has so far only been reported for cultured LCLs from ICF1 patients . That study showed that aberrant gene body methylation in ICF1 patients was associated with alternative transcription start site (TSS) usage, the regulation of sense-antisense transcription and alternative exon splicing . A similar function for gene body methylation in preventing aberrant transcript initiation has been shown in mESCs and could be ascribed to the catalytic activity of Dnmt3b . While it is challenging to directly translate observations from model systems or cultured cells to the ICF patient situation, it is feasible that spurious transcript initiation could critically affect phenotype, when associated with the production of aberrant proteins.
In addition to the ICF1-specific anomalous gene body methylation, ICF patient methylomes are characterized by hypomethylation of the heterochromatic compartment of the genome, as discussed above. By definition, such regions are transcriptionally inert and enriched for repressive epigenetic marks . Genome-wide expression profiling in ICF1 patient-derived LCLs identified modest changes in gene expression outside gene bodies, although global DNA hypomethylation was observed . Indeed, histone modification ChIP-seq revealed increased histone H3 lysine 27 three-methylated (H3K27me3) levels predominantly at hypomethylated CpGs and a compensatory function for this repressive mark has been suggested . Re-distribution of H3K27me3 at gene cluster regions and surprisingly mild changes in gene expression have also been reported in Hells−/− MEFs, consistent with a redundancy in epigenetic silencing pathways upon loss of DNA methylation . Whether these findings reflect convergent functions of Dnmt3b and Hells in de novo DNA methylation, or are a unifying feature in all ICF patients, remains to be determined.
Described as a disorder of ‘variable immunodeficiency, facial anomalies and centromeric heterochromatin instability’ approximately 40 years ago, the ICF syndrome has fascinated scientists ever since. Aberrant DNA methylation is probably involved in the causation of the disease, but what types of changes in DNA methylation patterns lead to the phenotypic aspects of the disorder remains an open question. A great deal has been learned about the genetic defects underlying ICF syndrome and the genome-wide epigenetic consequences thereof. In addition, the molecular interactions between the four disease genes and their functions in cellular processes are beginning to be understood. Gaining further insights into these mechanisms through increasingly sophisticated genetics, biochemistry and genomics approaches will enable us to progress toward a fuller understanding of how the dysfunction of four different proteins can lead to the same disease phenotype.
ICF is a rare primary immunodeficiency characterized by multiradiated chromosomes 1, 9 and 16, and hypomethylation of pericentromeric satellite repeats 2 and 3, and has been linked to defects in four genes, namely, DNMT3B, ZBTB24, CDCA7, and HELLS.
Genomic regions that are commonly affected in all four ICF types include members of big gene cluster families such as the clustered PCDH.
Anomalous gene body methylation is specific to ICF1 patients and has been associated with aberrant transcript initiation.
In addition to α-satellite repeat sequences, CpG-poor regions enriched for heterochromatin marks are predominantly affected by DNA hypomethylation in ICF2, 3, and 4 patients.
M.V. and L.D. both wrote the review.
This work was supported by the Leiden University Medical Centre (LUMC Fellowship) (to L.D.); and the Dutch Research Council [grant number NWO-ZonMw-Vidi 91718350 (to L.D.)].
The authors declare that there are no competing interests associated with this manuscript.
cell division cycle associated 7
centromeric heterochromatin by centromere protein C
classical non-homologous end joining
DNA methyltransferase 1
DNA methyltransferase 3 alpha
DNA methyltransferase 3 beta
DNA methyltransferase 3 like
helicase lymphoid specific
human embryonic kidney
heterochromatin protein 1
histone H3 lysine 27 tri-methylated
histone H3 lysine 36 tri-methylated
histone H3 lysine 4 tri-methylated
histone H3 lysine 9 di/tri-methylated
Immunodeficiency, Centromeric instability, Facial anomalies syndrome
lymphoblastoid cell line
mouse embryonic fibroblast
mouse embryonic stem cell
proliferating cell nuclear antigen
reduced-representation bisulfite sequencing
SET domain bifurcated histone lysine methyltransferase 1
developmental pluripotency-associated 3
suppressor of variegation 3-9 homolog 1
topologically associated domain
transcription start site
ubiquitin-like, containing PHD and RING finger domains 1
zinc-finger and BTB domain-containing 24