Understanding the language of Lys36 methylation at histone H3

Abstract

Histone side chains are post-translationally modified at multiple sites, including at Lys36 on histone H3 (H3K36). Several enzymes from yeast and humans, including the methyltransferases SET domain-containing 2 (Set2) and nuclear receptor SET domain- containing 1 (NSD1), respectively, alter the methylation status of H3K36, and significant progress has been made in understanding how they affect chromatin structure and function. Although H3K36 methylation is most commonly associated with the transcription of active euchromatin, it has also been implicated in diverse processes, including alternative splicing, dosage compensation and transcriptional repression, as well as DNA repair and recombination. Disrupted placement of methylated H3K36 within the chromatin landscape can lead to a range of human diseases, underscoring the importance of this modification.

The packaging of DNA with basic histones and a vast array of additional factors mediates the formation of chromatin and thereby determines the outcome of virtually all of the DNA processes in eukaryotes. Acetylation and methylation of histones were originally identified in radiolabelling studies using cell extracts^1,2. Today, enzymes catalysing numerous histone post-translational modifications or ‘marks’ have been identified; these include phosphorylation, various modes of methylation at Arg and Lys side chains, acetylation and ubiquitylation. These marks are thought to exist in dynamic combinations to generate a ‘code’, or ‘language’, that can enforce the regulatory features of chromatin during nearly all of the aspects of cellular metabolism: a given modification or permutation of modifications dictates a distinct biological output. Given the critical biological roles of chromatin and the numerous pathologies related to its misregulation, understanding the role of histone modifications is a paramount, yet complicated task. In this Review, we focus on our current understanding of a key modification, the methylation of Lys36 at histone H3 (H3K36). Widely described to be associated with active chromatin, H3K36 methylation has also been implicated in transcriptional repression, alternative splicing, dosage compensation, DNA replication and repair, DNA methylation and the transmission of the memory of gene expression from parents to offspring during development. We highlight the specificity, regulation and functional role of the enzymes that methylate H3K36, but we refer readers to informative reviews for a detailed discussion of the counteracting demethylases^3–5.

Regulating the methylation of H3K36

Histone methyltransferase (HMTase) enzymes use S-adenosylmethionine to add methyl groups to specific histone Lys or Arg residues. To date, at least eight distinct mammalian enzymes have been described that methylate H3K36 in vitro and/or in vivo (FIG. 1). All of the H3K36-specific methyltransferases identified thus far have the catalytic SET domain in common, but they have varying preferences for Lys36 residues in different methylation states. In yeast, SET domain-containing 2 (Set2) performs all three methylation events at H3K36 (REF. ⁶), but in higher eukaryotes there is accumulating evidence that these events require a division of labour between the mono- and dimethylases and the SET2-type trimethylases (FIG. 1). Several of these enzymes also possess multiple chromatin-interacting domains, including those known to interact with methylated H3K36 itself (such as the PWWP domain) and with additional methylated histone residues (such as plant homeodomain fingers (PHD fingers))^7–9 (FIG. 1). Notably, SET2 proteins from multiple species have a carboxy-terminal domain that interacts with the large subunit of RNA polymerase II (RNAPII), which is known as RNAPII subunit B1 (RPB1)¹⁰.

Figure 1. Domain structures of enzymes that methylate H3K36.

a | A schematic of the enzymes that have been shown to promote the formation of methylated Lys36 on histone H3 (H3K36). With the exception of fly maternal-effect sterile 4 (MES-4), only human enzymes are shown. The SET domain is shown with its pre (AWS) and post domains; C5HCH is a zinc-finger (ZNF) domain; WW domains are known to interact with Pro-rich peptides; the PWWP domain is known to interact with trimethylated H3K36; and the AT hook is a DNA-binding domain. All domain assignments were derived from Ensembl. b | Depiction of the transitions between the multiple H3K36 methylation states, highlighting the enzymes that have been shown to function in changing a given methylation state. ASH1L, ASH1-like; BAH, bromo- associated homology; BROM, bromodomain; HMG, high mobility group; MYND, myeloid, Nervy and DEAF-1 ZNF; NIDs, nuclear receptor interaction domains; NSD, nuclear receptor SET domain-containing; PHD, plant homeodomain; RuBisCo, ribulose-1,5-bisphosphate carboxylase oxygenase; SETD, SET domain-containing; SETMAR, SET domain and mariner transposase fusion gene-containing; SMYD2, SET and MYND domain-containing 2.

Determining HMTase specificity

The most parsimonious scenario would predict that each H3K36 methyltransferase enzyme catalyses a transition between two distinct methylation states (for example, between unmethylated H3K36 and monomethylated H3K36 (H3K36me1)). However, this is not necessarily the case. There are many discrepancies as to the level of methylation (mono-, di- and tri-) imparted, as well as the residue (or residues) that is targeted, by the various H3K36-specific methyltransferases; these discrepancies may arise from multiple factors, including the nature of the substrate tested (for example, peptides and histones versus nucleosomes), the source of enzyme (for example, full length versus SET domain only) and the assay conditions themselves (for example, antibody specificity versus mass spectrometry). Physiologically relevant substrates, characterized using nucleosomes and/or loss-of-function experiments, have been determined for some H3K36 methyltransferases (TABLE 1); but for enzymes such as SETD3, the enzymatic activities reported so far are based on analyses of peptides and core histones^11,12.

Table 1.

Reported substrate specificities for enzymes that methylate H3K36

Enzyme*	Assay conditions	Histone substrates identified	Refs
NSD1 (KMT3B)	Nucleosomes	H3, H3K36, H3K36me1, H3K36me2, H4K20	14, 16–18
	Cell based and/or in vivo	H3K36me1, H3K36me2, H4K20me3	17–19, 78
NSD2 (WHSC1, MMSET)	Nucleosomes	H3, H3K36, H3K36me1, H3K36me2	16,17, 19,30
	Cell based and/or in vivo	H3K27me2‡, H3K36me2	17,28, 36
NSD3 (WHSC1L1)	Nucleosomes	H3, H3K36	16
	Cell based and/or in vivo	H3K36me2	40
ASH1; ASH1L	Nucleosomes	H3, H3K36, H3K36me2, H3K36me3§	41,48, 49
	Cell based and/or in vivo	H3K4, H3K4me3	49
SMYD2	Core histones (octamers)	H3, H3K4, H3K36me2	42,43
SETMAR (METNASE)	Core histones (octamers)	H3K4, H3K36me2	12
	Cell based and/or in vivo	H3K36me2	96
SETD2 (KMT3A, HYPB)	Nucleosomes	H3, H3K36me1, H3K36me2, H3K36me3	34,64
	Cell based and/or in vivo	H3K36me3	25,33,34
SETD3	Core histones (octamers)	H3K4, H3K36	11

Table entries for the knockdown (cell based) or knockout (in vivo) analysis indicate a reported decrease in a given methylation state. ASH1, Absent, small and homeotic discs 1; ASH1L, ASH1-like; H, histone; NSD, nuclear receptor SET domain-containing; me1, monomethylated; me2, dimethylated; me3, trimethylated; SETD, SET domain-containing; SETMAR, SET domain and mariner transposase fusion gene-containing; SMYD2, SET and MYND domain-containing 2.

^*Alternative names are provided in brackets.

^‡RE-IIBP isoform of NSD2.

^§Gln2265Ala mutant of ASH1L.

NSD1 as a mono- and dimethylase for H3K36

Although it was originally shown to bind steroid nuclear receptors¹³, nuclear receptor SET domain-containing 1 (NSD1; also known as KMT3B) was subsequently reported to use its SET domain to methylate H3K36 as well as H4K20 (REF. ¹⁴). However, NSD1 also methylates non-histone substrates, such as the p65 subunit of nuclear factor-κB (NF-κB)¹⁵. Indeed, several HMTases have multiple substrates, including non-histones, and it is important to consider this when interpreting phenotypes derived from loss-of-function experiments.

NSD1 has specific mono- and dimethylase activity for H3K36 (REFS ^16–18), generating H3K36me1 and dimethylated H3K36 (H3K36me2), and there are also mixed reports as to whether it has specificity for H4K20 (REFS ^14,19,20). Enzymatic assays using recombinant nucleosomes containing unmethylated H3K36 or a mimic of H3K36me1 (REF. ²¹) were shown to serve as specific substrates for NSD1 (REF. ¹⁶). However, when histone octamers were used as substrates, NSD1 was found to additionally methylate histone H4 as well as histones H2A and H2B, suggesting that nucleosomes contribute to H3K36 specificity. Structural data have also confirmed the specificity of NSD1 as a mono- and dimethylase that targets H3K36 (REF. ¹⁸). Intriguingly, this study showed that the substrate-binding channel of NSD1 is blocked by an autoinhibitory loop that resides between the SET and post-SET domains. As Lys36 resides near the core of the nucleosome, interactions between the enzyme and DNA may allosterically relieve inhibition by this loop, a result that is consistent with the preference of NSD enzymes for nucleosome substrates, as opposed to octamers. This furtherunderscores the necessity for using physiologically relevant substrates in order to properly characterize enzyme specificity towards H3K36.

NSD1 controls the levels of methylated H3K36 within and surrounding the body of the bone morphogenic protein 4 (BMP4) gene in human HCT116 colorectal cancer cells¹⁷. In the absence of NSD1, BMP4 expression is significantly impaired and the levels of all three forms of methylated H3K36 are reduced within the body of the gene. These data could support a model in which NSD1 catalyses H3K36 trimethylation (H3K36me3), but a more likely scenario is that the loss of mono- and dimethylation at H3K36 also inhibits trimethylation by failing to provide another HMTase, SETD2 (also known as KMT3A and HYPB), with its substrate. This is an important distinction to make when considering which HMTases regulate the formation of a trimethylated residue.

The role of NSD1 as a mono- and dimethylase also seems to be conserved in other metazoans. Worms and flies each have one NSD1 orthologue called maternal-effect sterile 4 (MES-4)^22,23. In Caenorhabditis elegans, MES-4 has been reported to be a dimethylase that is specific for H3K36 (REF. ²²). Similarly to NSD1, Drosophila melanogaster MES-4 catalyses global mono- and dimethylation of H3K36 in vivo, but SET2, the fly orthologue of human SETD2, regulates all trimethylation at this site; again, MES-4 is probably required to provide the proper dimethylated H3K36 substrate for the SET2 enzyme.

In worms, mutation of the SETD2 orthologue histone methyltransferase-like 1 (met-1) significantly reduced global H3K36me3 levels²⁴. This supports the notion that the worm MES-4 and MET-1 proteins coordinate dimethylation and trimethylation, respectively²⁴. These experiments are generally consistent with studies of mammalian NSD proteins and SETD2 (REFS ^16,25). By contrast, separate studies showed by immunostaining that worm embryos individually deficient in mes-4 or met-1 retain significant levels of H3K36me3, but embryos doubly deficient in mes-4 and met-1 are completely devoid of H3K36me3 (REFS ^26,27). This suggests that MES-4 collaborates with MET-1 to perform trimethylation. As each of these studies has used different antibodies, it will be important to validate these experiments using additional, antibody-independent techniques, such as mass spectrometry. Collectively, the combination of in vitro and in vivo data suggests that NSD1 and its orthologues probably catalyse the addition of either the monomethyl or dimethyl groups onto H3K36 and so indirectly regulate the levels of trimethylation by altering the availability of monomethyl and dimethyl substrates for the trimethylating enzymes.

NSD2 as a mono- and dimethylase for H3K36

Similarly to NSD1, NSD2, the product of Wolf–Hirschhorn syndrome candidate gene 1 (WHSC1; also known as MMSET), has been reported to target H3K36, as well as H4K20, H3K4 and H3K27 (REFS ^16,28–32). Here again, these discrepancies may lie, at least in part, with the nature of the substrates used in the assays. For example, NSD2 acts as a dimethylase towards H3K36 when presented with nucleosomes, but it preferentially dimethylates H4K44 when presented with octamers¹⁶. Interestingly, when short single-stranded and double-stranded DNA molecules that are notably smaller than the length that is necessary to generate a nucleosome are added to octamers, NSD2 preferentially dimethylates H3K36, a result that has been attributed to the ability of DNA to act as an allosteric effector of NSD2 (REF. ¹⁶). This is consistent with another study, in which mass spectrometry and immunoblotting against native and recombinant nucleosomes showed that NSD2, as well as its major splicing variant, RE-IIBP, acts as a mono- and dimethylase specific for H3K36 (REF. ³⁰). By contrast, there have been reports that the NSD2 isoform RE-IIBP is an H3K27-specific HMTase²⁸ and that H3K36 trimethylation is lost in embryonic stem (ES) cells derived from NSD2-defective mice³². But the idea that NSD2 might act as a trimethylase would not be consistent with a previous study that identified SETD2 as the sole trimethylase in mammalian cells²⁵. Furthermore, H3K36me3 levels are reduced when SETD2 is depleted, despite normal levels of H3K36me2 (REFS ^33,34). Regardless, these data are all consistent with a role for NSD2 in H3K36 methylation. This is also supported by the observation that, in ~15% of patients with multiple myeloma (a haematological malignancy that accounts for 1% of all cancers³⁵), NSD2 is overexpressed as a result of a translocation with the immunoglobulin locus (t(4;14)⁺) and the global levels of H3K36me2 are increased³⁶.

Possible specificity of NSD2 for H4K20

Despite strong evidence linking NSD2 to H3K36me, it may also act as an H4K20-specific methyltransferase during the cellular response to DNA double-strand breaks (DSBs)^37,38. NSD2 was isolated in an RNA interference screen for genes involved in resistance to hydroxyurea, a DNA-replication inhibitor³⁸. Consistently, NSD2-defective cells are sensitive to DNA damage and NSD2 localizes to sites of DNA damage. Moreover, NSD2 is phosphorylated at Ser102 by the kinase ataxia telangiectasia mutated (ATM) upon damage³⁷, and this mediates the recruitment of NSD1 to sites of DNA damage, where it has been suggested to dimethylate H4K20, which is an important event in recruiting the DNA damage response regulator p53-binding protein 1 (53BP1)³⁹. Although one could envision that NSD2 phosphorylation and damaged DNA may both alter the catalytic specificity of NSD2 through allosterism, it should be noted that recombinant NSD2 methylates H4 in in vitro assays with H4 peptides³⁷. Thus, it is possible that NSD2 does have some specificity for H4 in addition to H3 even in the absence of DNA damage.

The Set2 trimethylase

In yeast, Set2 is non-essential, performs all three methylation reactions at H3K36, and couples H3K36me3 with transcriptional elongation through an interaction with Rpb1, the large subunit of RNAPII¹⁰. The human orthologue of Set2, SETD2, also interacts with RNAPII during elongation, but it is an essential protein and has been identified as a huntingtin-interacting protein. The role of SETD2 in Huntington’s disease is obscure. SETD2 is a trimethylase^25,33,34, and some reports have indicated that, in vitro, recombinant human SETD2 can add all three methyl groups to H3K36 (REFS ^32,34). By contrast, in vivo knockdown experiments targeting SETD2 revealed reduced levels of H3K36me3 only³⁴. This underscores the importance of using multiple approaches to determine enzyme specificity and also raises an important point about enzyme processivity. Does a trimethylase obligatorily require a dimethyl substrate or can it processively place all three groups on a Lys substrate? SETD2 associates with heterogeneous nuclear ribonucleoprotein L (hnRNPL), and hnRNPL knockdown analyses show decreased levels of H3K36me3 but not H3K36me1 or H3K36me2 (REF. ³⁴). Thus, the ability to act as a processive enzyme could depend on the presence and regulation of distinct cofactors, as well as on the presence of a previously methylated substrate.

Defining further HMTase specificities for H3K36

Like other NSD family members, the HMTase NSD3 (also known as WHSC1L1) appears to be specific for H3K36 (REFS ^16,40). So far, relatively little is known about the substrate specificities of the Trithorax protein ASH1-like (ASH1L) (an orthologue of the fly protein Absent, small and homeotic discs 1 (ASH1)), SET domain and mariner transposase fusion gene-containing (SETMAR; also known as METNASE), SETD3 and SET and MYND domain-containing 2 (SMYD2). Each has been reported to methylate H3K36, particularly with respect to dimethylation, but other substrate specificities have also been described for these enzymes^12,41–43. However, the assays for SETD3 specificity were performed with peptides¹¹, and it will be important to confirm this enzyme specificity using nucleosome-based assays^16,44.

In addition to targeting H3K36, SMYD2 methylates H3K4 and non-histone substrates, such as p53 and retinoblastoma (RB)^45,46. The in vitro methylase activity of SMYD2 towards H3K4, but not towards H3K36, depends on its ability to bind heat shock protein 90α (HSP90α)⁴². Although the expression of SMYD2 appears to be largely confined to the brain and heart, its conditional deletion from mouse cardiomyocytes showed that it is dispensable for heart development and, surprisingly, its deletion had no effect on the global levels of methylated H3K36 (REF. ⁴⁷). This may suggest that the physiological target for SMYD2 is a non-histone substrate, providing an additional example of how the interpretation of knockout phenotypes for HMTases must be carefully considered.

Rigorous experiments using recombinant nucleosomes have demonstrated that ASH1L is a dimethylase that is specific for H3K36 (REFS ^41,48). However, ASH1L has also been reported to target H3K4 and to localize within the transcribed regions of active target genes⁴⁹; and in vivo levels of H3K4me3 were reduced in the absence of ASH1L. Additional studies have shown that ASH1L also targets H3K36 and H4K20 (REFS ^44,46). Thus, it remains to be seen what the full substrate repertoire of ASH1L is. Notably, similarly to NSD1, ASH1L has an autoinhibitory loop that resides between the SET and post-SET domains, which is a feature that has been proposed as a hallmark feature of H3K36-specific enzymes^41,48.

Additional factors that influence H3K36 methylation

Several mechanisms have been described for how H3K36 methylation levels are regulated at a given locus (REFS ^50–52). In addition to cofactors such as HSP90α and hnRNPL, Pro isomerization of histone H3 and additional histone modifications on both H3 and H4 have been shown to regulate the total levels of H3K36 methylation by Set2, although it is not clear how this occurs. Additional studies in yeast have shown that cyclins (encoded by bypass UAS requirement 1 (BUR1) and BUR2) are required for H3K36me3 (REF. ⁵³). Moreover, the yeast histone chaperone anti-silencing function 1 (Asf1), the RNAPII kinase C-terminal domain (CTD) kinase 1 (Ctk1) and the elongation factor Spt6 (a subunit of the FACT (facilitates chromatin transcription) complex) all regulate the levels of H3K36 trimethylation but not dimethylation^54,55. Other elongation factors, such as polymerase-associated factor 1 (Paf1), have also been shown to alter the levels of H3K36 methylation⁵⁶. Large cells 1 (Lge1), a factor associated with the ubiquitin pathway, was also identified in a genome-wide screen for candidate genes that specifically regulate H3K36 methylation⁵⁷. This study also confirmed the role of Paf1 in regulating H3K36me3: mutations in Paf1 disrupted the levels of H3K36me3, but not those of H3K36me2 (REF. ⁵⁶). Collectively, these studies suggest an emerging theme: that HMTase cofactors may alter substrate specificity on a gene-to-gene basis, providing an additional layer of regulation for their activity at distinct targets.

Roles of H3K36 methylation in gene expression

Methylation of H3K36 has been observed at multiple stages of RNA biosynthesis and is not always associated with gene activation. Rather, its ultimate output and where it fits into gene expression pathways are functions of multiple variables: where in the gene body the methylated H3K36 mark is placed, when H3K36 is methylated and what reader protein binds to this modification. Further complicating these variables is the fact that the degree of methylation can result in different biological outcomes.

H3K36 methylation and transcriptional activation

Numerous studies in multiple systems support a role for H3K36 methylation in transcriptional activation. It has been observed that, in general, there is a progressive shift from monomethylation to trimethylation of H3K36 between the promoters and the 3′ ends of genes⁵⁸. However, in zebrafish, somatic cells also show a bias for H3K36me3 in the 3′ end of actively transcribed genes, but this mark is curiously present in the 5′ promoter regions of quiescent genes that are developmentally regulated during spermatogenesis⁵⁹. The functional significance of this is unknown, and it will be important to determine whether promoter-proximal H3K36me3-containing nucleosomes can recruit repressive histone deacetylases to promote gene silencing.

Multiple lines of evidence in budding yeast have showed that Set2, a non-essential enzyme that is responsible for all three forms of methylated H3K36, is coupled to transcriptional elongation. Deletions in yeast Set2 cause sensitivity to the elongation inhibitor 6-azauracil and phenocopy mutations in other known elongation factors¹⁰. However, it remains possible that this phenotype may arise from indirect effects on nucleotide metabolism. Nevertheless, Set2 associates with the hyperphosphorylated form of RNAPII and deposits the trimethyl group onto H3K36 during elongation^10,60–62 in various systems, including yeast and humans. This interaction is regulated by phosphorylated residues in the CTD of Rbp1, the large subunit of RNAPII: the Ser2-phosphorylated form of the CTD is indicative of elongation and the Ser5-phosphorylated form is characteristic of a paused polymerase at a promoter⁶³. Indeed, human SET2 proteins also appear to bind RNAPII and to target H3K36 (REFS ^34,64).

One well-established function of Set2 in yeast is the prevention of aberrant transcriptional initiation within coding sequences (FIG. 2). Set2 catalyses H3K36 methylation co-transcriptionally and recruits the reduced potassium dependency 3 small (Rpd3S) deacetylase complex here^65–67 through association of the chromodomain-containing Rpd3S subunit ESA1- associated factor 3 (Eaf3) with H3K36me3; this enforces a deacetylated chromatin state in the wake of transcribing RNAPII^65,68. Rpd3S preferentially associates with histones containing H3K36me2 and H3K36me3 but not H3K36me1 (REF. ⁵⁷). The PHD finger of the Rpd3S subunit Rco1 can also cooperate with the Eaf3 chromo-domain to promote these interactions⁶⁹. This indicates that H3K36me2 and H3K36me3 might act redundantly in the Set2–Rpd3S pathway.

Figure 2. H3K36me3-dependent prevention of aberrant transcription in yeast.

Actively transcribing RNA polymerase II (RNAPII) displaces acetylated nucleosomes. These evicted histones are reincorporated into nucleosomes and chromatin behind the polymerase. As SET domain-containing 2 (Set2) binds RNAPII, this promotes the trimethylation of Lys36 on histone H3 (H3K36me3) in the newly incorporated nucleosomes. H3K36me3 serves as a ‘mark’ that the reduced potassium dependency 3 (Rpd3) deacetylase complex binds; this complex facilitates local nucleosome deacetylation, preventing aberrant, spurious transcription in the wake of RNAPII progression through a region. CTD, carboxy-terminal domain; HAT, histone acetylase; m⁷G, 7-methylguanosine.

The role of histone Lys methylation in the maintenance of a repressive chromatin environment during transcriptional elongation is also used in humans but may be independent of acetylation⁷⁰. In this case, Lys demethylase 2 (LSD2; also known as KDM1B) demethylates methylated H3K4 in the intragenic regions of active target genes. LSD2 resides in complexes with NSD3, which acts as a mono- and dimethylase that is specific for H3K36 (REF. ¹⁶). Additionally, LSD2 binds the H3K9-specific HMTase G9A (also known as EHMT2). Consistent with this, the LSD2 complex has robust HMTase activity towards H3K9 and, to a lesser extent, H3K36 (REF. ⁷⁰). Moreover, in chromatin immuno-precipitation followed by microarray (ChIP–chip) experiments, depletion of G9A alters both H3K9 methylation levels and transcriptional regulation at LSD2 targets. In light of these observations and the fact that LSD2 resides in complexes with elongation factors such as cyclin T1 (CCNT1; also known as PTEFb) and the Ser2-phosphorylated form of RNAPII, it appears that the methylation-dependent enforcement of a repressive chromatin state through H3K9 and possibly H3K36 methylation at actively transcribed genes is evolutionarily conserved. Additionally, it seems that NSD3 can participate in transcriptional elongation⁷⁰. Interestingly, the LSD2 complex contains multiple proteins with PWWP domains, including NSD3. Originally identified in NSD2, PWWP domains also bind H3K36me3 to act as reader proteins^7,8,71. In particular, the PWWP domain of DNA methyltransferase 3A (DNMT3A) binds H3K36me3 and methylates nearby DNA, demonstrating a link between H3K36me3 and DNA methylation that could represent a new mechanism for H3K36 methylation-mediated repression of gene expression.

H3K36 methylation in dosage compensation

Coordination between methylation and acetylation also occurs during dosage compensation in D. melanogaster⁷². To compensate for having only one X chromosome, male flies use the Male-specific lethal (MSL) complex to upregulate the expression of X-linked genes by a factor of two in order to achieve the same level of transcription as females, which have two X chromosomes. This may occur through a two-step model requiring H3K36me3 and increased elongation^73–75. The MSL complex is initially recruited to X-linked genes through a GA-rich recognition site and subsequently facilitates transcriptional elongation through an enhancement of RNAPII activity. Next, the MSL3 subunit (an orthologue of yeast Eaf3) is thought to promote spreading of the upregulated state through interactions that involve the chromodomain of MSL3 and H3K36me3 (REF. ⁷³). Additional active marks, such as H4K16 acetylation, also affect X-linked gene expression, requiring the Males absent on the first (MOF) subunit of MSL. Thus, a collection of activating marks together achieves chromosome-wide gene expression⁷⁴. Although interactions between chromodomain proteins and H3K36me3 marks regulate acetylation in both yeast and D. melanogaster, they have opposite effects: transcription is repressed in yeast but upregulated in flies. This clearly illustrates that the function of a given histone modification, such as H3K36me3, is context-dependent and is much more complicated than a simple code.

NSD enzymes and transcriptional initiation

NSD1 regulation of H3K36 methylation appears to affect transcriptional initiation¹⁷. For example, NSD1 binds upstream of the promoter of BMP4 and regulates the levels of H3K36me1, H3K36me2 and H3K36me3 within the body of the gene. This seems to be required for the recruitment of RNAPII to the BMP4 promoter, an observation that links NSD1 to initiation events through RNAPII. NSD3 can also bind promoters and then influence the levels of H3K36 methylation within the body of a gene⁴⁰. For example, the extraterminal domains of bromodomain-containing 4 (BRD4) specifically recruit NSD3 to the promoter region of the gene encoding CCND1 and the decapping enzyme DCPS, and loss of either BRD4 or NSD3 reduces the levels of H3K36me3, particularly within the body of the gene. This is consistent with the NSD1-dependent regulation of BMP4, during which promoter-proximal bound NSD1 regulates H3K36me3 levels within the body of the gene¹⁷. Although NSD3 localization is biased towards the promoter of CCND1, it also can be found, albeit at reduced levels, at the 3′ end of the gene⁴⁰, suggesting that NSD3 participates in initiation and elongation events. Moreover, NSD3 levels were found to be higher in the coding regions than in the promoter of the proto-oncogene Ser/Thr kinase PIM2 (REF. ⁴⁰). Thus, NSD3 appears to reside in multiple complexes (including LSD2 and BRD4 complexes) and can localize to either promoter or internal regions to promote H3K36 methylation and thereby influence initiation and elongation processes. This is in apparent contrast to NSD1, which has been reported to localize primarily near the 5′ ends of its targets¹⁷; it will be important to confirm this using high-resolution ChIP followed by sequencing (ChIP–seq) analysis. However, previous ChIP–seq data show that NSD2 is enriched near transcription start sites³⁰.

H3K36 methylation in transcriptional repression

Fascinating clues have emerged regarding the relationship between methylation at H3K36, transcriptional repression and worm development^26,27. In worms, maternally provided MES-4 is essential for germ cell viability and is involved in X-chromosome silencing in the germline^26,27. Early primordial germ cells (PGCs) do not engage in transcription as they lack the active hyperphosphorylated CTD form of RNAPII. But, in the absence of mes-4, active RNAPII persists, indicating that MES-4 is important for the establishment and/or maintenance of transcriptional repression in late stage PGCs^26,27. Loss of MES-4 also reduces the levels of H3K36me3, independently of RNAPII function. Thus, MES-4 maintains the H3K36me3 state in germline precursors independently of transcription — an observation that has not been made in other organisms so far.

Beyond H3K36 methylation-mediated regulation of spurious initiation, H3K36 methylation-mediated repression has also been reported. Yeast Set2 represses transcription from a lacZ reporter and, in Set2-deletion mutants, the basal levels of GAL4 transcription are increased^76,77; however, in these studies, Set2 is artificially recruited to the promoter, so it will be important to confirm whether this reflects the true nature of Set2 activity. Initial studies with tethered reporters in mammalian cells have implicated NSD1 in both transcriptional activation and repression¹³, although its targets have not yet been determined. In addition, an RNA interference study has shown that NSD1 represses the expression of the homeobox regulator MEIS1 in neuronal cell lines¹⁹. This repressive event is likely to be direct, as NSD1 localizes to the promoter of MEIS1 (REF. ⁷⁸). Furthermore, NSD2 collaborates with the cardiac-specific factor NKX2-5 to repress targets such as platelet-derived growth factor-α (PDGFRα), probably through modulating H3K36 methylation levels³². Thus, H3K36 methylation appears to act as both an activating and inhibitory signal, and the overall biological readout might depend on the context of additional surrounding marks and their corresponding reader proteins.

H3K36 methylation and exon definition

There has been increasing support for the idea that biased nucleosome positioning favours exonic over intronic sequences^79–81. Such a bias seems to be accompanied by distinct histone modifications, including H3K36me3. This should not be surprising, as the 147-base-pair length of a DNA fragment associated with a histone octamer correlates with the average length of an internal exon. This is likely to be more than just mathematical serendipity; it is probably evidence of interplay between chromatin and splicing. Several large-scale bioinformatics studies have analysed both the positions of nucleosomes and their modification status within the genomes of humans, C. elegans, D. melanogaster and mice^79–81. In each case, nucleosomes were enriched specifically at exonic sequences. Although the increased deposition of nucleosomes at exons guarantees a bias in histone modifications within exons relative to those within introns, it is also clear that a subset of modifications is specifically enriched here. This is particularly true for H3K36me3 but also includes methylation at H3K79, H4K20 and H2BK5 (REF. ⁸⁰). Each analysis also found that the H3K36me3 bias is more pronounced within exons further downstream of the transcription start site. This preference may reflect the propensity of RNAPII to abort transcription early in the transcription cycle, thereby reducing the number of nucleosomes that are displaced further downstream. The known association between Set2 and the RNAPII CTD may also further explain the particular increase in H3K36me3 signatures seen at downstream exons^61,82. The implication of nucleosome enrichment and the increase in H3K36me3 modifications is twofold. First, nucleosomes probably act as intrinsic pause sites for elongating RNAPII which could alter splice site recognition and, hence, change exon inclusion. Others have found that the introduction of pause sites within minigenes can increase the inclusion of alternatively spliced exons^83,84. Furthermore, expression of a ‘slow’ mutant RNAPII in D. melanogaster results in different inclusion patterns of the exons within the Ultrabithorax mRNA⁸⁵. A second possibility, which is not mutually exclusive with effects on RNAPII pausing, is that the H3K36me3 modification relays a specific signal to the splicing machinery to alter how it defines exons, leading to the specific inclusion or exclusion of particular exons.

Although the global analyses of H3K36me3 positioning in various genomes provide compelling evidence that this modification affects splicing, functional evidence beyond this has been lacking. However, an interesting connection has been made between SETD2, the reader protein MORF-related gene 15 (MRG15; which contains a chromodomain) and polypyrimidine tract-binding protein (PTB), the last of which is a known antagonist of exon definition that affects splicing of fibroblast growth factor receptor 2 (FGFR2) pre-mRNA^86,87. FGFR2 contains two mutually exclusive exons (IIIb and IIIc) that encode a region within the extracellular immunoglobulin-like domain and are each responsible for receptor binding to a unique range of FGFs⁸⁸. Exon IIIb is included in epithelial cells through the action of epithelial splicing regulatory protein (ESRP), whereas exon IIIc is included in cells of mesenchymal origin⁸⁹ (FIG. 3). Furthermore, the splicing pattern switches from exon IIIb to exon IIIc inclusion as prostate epithelial cells become androgen-independent, an important factor in metastasis. Analysis of nucleosome modifications throughout FGFR2 show that H3K36me3 is specifically enriched within exon IIIb and is restricted to mesenchymal cells, which exclude this exon. This H3K36me3 modification is recognized by MRG15, which also interacts with PTB. Thus, by recruiting PTB to its target exon, these interactions position PTB to bind to its intronic splicing silencer sites, which flank the repressed exon as they emerge from the transcribing RNAPII complex (FIG. 3). PTB repression of this exon can be alleviated by downregulating either MRG15 or SETD2. This regulatory module also exists at other alternatively spliced exons, with a bias towards exons that contain weaker PTB-binding sites⁸⁶. What remains to be seen is how two distinct cell types achieve this differential methylation of H3K36 within nucleosomes at alternatively spliced exons in order to regulate splicing.

Figure 3. H3K36me3 influences alternative splicing in a cell-type specific manner.

The fibroblast growth factor receptor 2 (FGFR2) locus that undergoes alternative splicing consists of two mutually exclusive exons, IIIb and IIIc, which are located between the constitutive exons 7 and 10. Mesenchymal stem cells favour the inclusion of exon IIIc and achieve this by repressing splicing of exon IIIb. Nucleosomes present near exon IIIb contain the SET domain-containing 2 (SETD2)-dependent trimethylated Lys36 on histone H3 (H3K36me3) ‘mark’ and its reader protein MORF-related gene 15 (MRG15). MRG15 also interacts with polypyrimidine tract-binding protein (PTB), a known repressor of exon inclusion, and this may be the mechanism by which the methylated H3K36 mark can influence splicing at this locus. In epithelial cells, FGFR2 expresses exon IIIb but excludes exon IIIc. Epithelial splicing regulatory protein (ESRP) is expressed and stimulates the inclusion of exon IIIb; reduced levels of H3K36me3 present at this exon, possibly as a result of lower SETD2 levels, allow its derepression. The role of dimethylases, such as the proteins of the nuclear receptor SET domain-containing (NSD) family, in this process has yet to be determined but these enzymes could also influence H3K36 methylation here.

This example of FGFR2 control exemplifies how the methylation status of H3K36 can affect splicing. However, this crosstalk is bidirectional: mutations in splice sites that abrogate intron removal of β-globin reporter genes cause a shift in H3K36me3 signatures towards the 3′ region of genes⁹⁰. Moreover, others have observed that H3K36me3 signatures are markedly lower in genes without introns⁹¹. In both cases, inhibition of splicing with pharmacological agents causes a rapid and global redistribution of H3K36me3 throughout the genome and reduces SETD2 recruitment to these genes^90,91. Although it is unclear how the spliceosome regulates H3K36 methylation, these studies suggest that it is a general phenomenon.

H3K36 methylation in DNA replication, recombination and repair

As replication origins initiate at distinct times in S phase, the decision to fire at any particular origin depends on a number of factors, including the transcriptional status of nearby genes. Origins of replication are bound by the origin recognition complex (ORC), which recruits various factors, including CDC6, minichromosome maintenance (MCM) proteins and CDC45, before loading of the DNA polymerase. Studies in budding yeast have revealed a role for H3K36 methylation in regulating replication origin firing, as deletions in SET2 cause a delay in Cdc45 loading at origins⁹². H3K36 methylation has also been linked to DNA replication checkpoint control in fission yeast⁹³. Furthermore, yeast mutants in the FACT elongation complex are sensitive to the replication inhibitor hydroxyurea, and this sensitivity can be suppressed by mutations in SET2 (REF. ⁹⁴). How H3K36 methylation influences these S phase events is unknown, but it could keep chromatin in an active state, perhaps through recruiting specific reader proteins that can influence DNA replication control.

Histone methylation is also critical for the maintenance of genomic stability, including the biological response to DSBs^39,95. For example, H3K36me2 has been implicated in the repair response to DSBs through the non-homologous end-joining (NHEJ) pathway⁹⁶. Here, the methylase SETMAR, a SET domain-containing enzyme previously reported to affect NHEJ as well as DNA replication, was shown to directly mediate H3K36me2 formation near sites of DSBs⁹⁶ (FIG. 4). SETMAR-dependent formation of H3K36me2 was shown to enhance the rate of association of the DNA repair factors KU70 and Nijmegen breakage syndrome 1 (NBS1; also known as nibrin) near DSBs⁹⁶. However, it is not yet clear how methylated H3K36 acts at the sites of DSBs, particularly with respect to which proteins may respond to the presence of this mark, or what relationship this mark might have with other known histone modifications that are important for DSB repair. Most notably, this may include crosstalk with dimethylation of H4K20, an event that recruits the DNA damage response regulator 53BP1 to sites of damage³⁹. It is also possible that methylated H3K36 in this context mediates its effects by antagonizing transcriptional elongation, as RNAPII is displaced in response to DNA damage⁹⁷.

Figure 4. Model for H3K36 methylation at sites of DNA damage.

Upon the generation of DNA damage, SET domain and mariner transposase fusion gene- containing (SETMAR) dimethylates Lys36 on histone H3 (H3K36) near sites of DNA double-strand breaks (DSBs), possibly by recruiting currently undefined reader proteins that facilitate the binding of KU70 and the MRE11 RAD50 NBS1 (MRN) complex to facilitate DNA repair. How methylation at H3K36 is coordinated with other chromatin modifications that are known to participate in DNA repair (such as, p53-binding protein 1 (53BP1) binding to dimethylated H4K20 (H4K20me2)) is unknown. NSD2, nuclear receptor SET domain-containing 2.

H3K36 methylation: links with disease

In higher eukaryotes, defects in the genes that maintain the levels of H3K36 methylation cause developmental defects and disease^32,33,98 (TABLE 2). Defects in SETD2 are causal for sporadic clear renal cell carcinoma, a disease that is marked by the loss of the short arm of chromosome 3 (REF. ⁹⁹). SETD2 has also been hypothesized to be a tumour suppressor in breast cancer¹⁰⁰. Defects in each member of the NSD family have been implicated in multiple diseases and cancer types^{31,78,101–106}. For example, haploinsufficiency in NSD1 is causal for Sotos overgrowth syndrome⁹⁸. NSD1 has also been implicated in breast, lung and prostate cancer, acute myeloid leukaemia (AML) and refractory anaemia^{19,31,78,101–103,107–112}. NSD1-defective mice display an embryoniclethal phenotype: the embryos initiate mesoderm formation but fail to complete gastrulation, apparently as a result of apoptosis¹⁴. NSD2-defective mice die shortly after birth and display features that are consistent with Wolf–Hirschhorn syndrome³², including facial abnormalities and cardiac defects. Thus, NSD1-knockout mice and NSD2-knockout mice have different phenotypes, despite the fact that both proteins catalyse the formation of H3K36me2. No knockout-mouse phenotype has yet been reported for NSD3.

Table 2.

H3K36 methyltransferases and their roles in diseases

Gene	Diseases	Molecular defects	Phenotypes	Refs
NSD1	Sotos syndrome	Haploinsufficiency, point mutations, deletions, translocations	Macrocephaly, hypertolerism, cognitive and/or motor skill deficiencies	98
NSD1	Myelodysplastic syndrome	Translocations	Anaemia, cytopenia	111
NSD1	Cancers: AML, prostate, neuroblastoma, breast	Overexpression, gene silencing, translocations	Numerous tumour types	19,78,101, 110,112
NSD2	Wolf–Hirschhorn syndrome	Deletions	Learning difficulties, microcephaly, heart defects	32
NSD2	Multiple myeloma	t(4;14)+ translocation	Renal failure, anaemia, bone lesions	30,115
NSD3	Breast cancer	Gene amplification at 8p11	Solid tumours	107
NSD3	AML	Translocations	Leukaemic cells in bone marrow	105
NSD3	Myelodysplastic syndrome	Translocations	Anaemia, cytopenia	106
SETD2	Renal cell carcinoma	Deletions, missense mutations	Haematuria, flank pain	99

The table lists the known diseases resulting from alterations in the genes encoding enzymes that are specific for Lys36 on histone H3 (H3K36). AML, acute myeloid leukaemia; NSD, nuclear receptor SET domain-containing; SETD2, SET domain-containing 2.

Each NSD family member behaves as an oncogene in multiple cancers^{35,78,105,112}. Translocations in NSD1 or NSD3 lead to the development of AML^105,112. Moreover, NSD1, an androgen receptor co-regulator¹¹³, has been identified as a candidate gene capable of discriminating between malignant and non-malignant prostate tissue¹⁰¹. In addition, silencing of NSD1 has also been shown to cause sensitivity to the oestrogen receptor antagonist tamoxifen¹¹⁴, but how these two observations are linked to the ability of NSD1 to bind and regulate either the androgen or oestrogen receptors is unknown. Overexpression of NSD2 has also been implicated in multiple myeloma and additional cancers, such as neuroblastoma^102,115, and overexpression of NSD3 via gene amplification occurs in about 15% of breast cancers¹⁰⁷ and correlates with poor prognosis. Therefore, NSD proteins are instrumental in the development of cancer, but their mechanisms of action in this context are only beginning to be elucidated. One possible mechanism underlying NSD-mediated disease is that these proteins enforce H3K36 methylation patterns to change gene expression. This is discussed below in the context of well-known NSD-dependent cancers.

Through overexpression or their incorporation into fusion proteins via translocations, NSD proteins act as potent oncoproteins^30,78. For example, fusion proteins between NSD1 and the nucleoporin NUP98 are well established as being causal for a subset of AML¹¹², and they regulate homeobox (Hox) gene expression in a mouse model of AML⁷⁸. In the case of AML, NSD1–NUP98 fusions enforce an H3K36 methylation signal that contributes to the inappropriate activation of Hox genes during development. These increased H3K36 methylation levels seem to be antagonistic to the repressive methylated H3K27 marks that normally silence Hox gene expression, leading to reduced H3K27 methylation levels and gene activation. As a consequence, unscheduled cell proliferation and the generation of AML are observed in the mouse model⁷⁸.

Overexpression of NSD2 via a translocation between chromosome 4 and chromosome 14 alters the global profile of H3K36 methylation in KMS11 multiple myeloma cells^30,36. Moreover, the catalytic activity of NSD2 is required for tumorigenesis in a mouse xenograft model³⁰. And loss of NSD2 function (through either short hairpin RNA-mediated depletion or homologous recombination) globally decreases H3K36me2 and H3K36me3 levels but has no effect on H4K20 or H3K4 methylation³⁶. It is, however, accompanied by a concomitant decrease in the global levels of H3K27me2 and H3K27me3, and this downregulation results in the expression of target genes that would normally be quiescent. This suggests that the overexpression of NSD2 alters transcriptional programming by promoting the formation of H3K36me2 and thereby changing the chromatin structure. Indeed, ChIP–seq analysis has precisely mapped the distribution of H3K36me2 in KMS11 multiple myeloma cells overexpressing NSD2 (REF. ³⁰). In cells that have one normal allele of NSD2, the H3K36me2 signal revealed by ChIP–seq was preferentially enriched in intragenic regions, which is consistent with previous ChIP–chip results in flies²³. However, overexpression of NSD2 disperses an increased H3K36me2 signal throughout the genome, including intergenic regions. Cells expressing one normal allele of NSD2 have a modest H3K36me2 signal in promoter regions, followed by a peak at transcriptional start sites, and then the signal decays downstream to the 3′ end³⁰, and a high signal corresponds with higher transcriptional activation. By contrast, cells overexpressing NSD2 have little variance in H3K36me2 signal intensity across the average gene and the correlation between signal levels and transcription is lost. Instead, as some genes are much more sensitive to H3K36me2 levels than others, this aberrant enrichment of the H3K36me2 signal appears to trigger the expression of quiescent oncogenes, including transforming growth factor alpha (TGFA), MET, p21 activated kinase 1 (PAK1) and RRAS2 (related RAS viral (r-ras) oncogene homologue)³⁰.

H3K36 and H3K27 methylation may be mutually exclusive modifications⁴⁴. Nucleosomes pre-methylated at H3K27 are largely refractory to enzymatic catalysis at H3K36 and vice versa. As overexpression of NSD3 is causal for the generation of breast cancer through 8p11–12 amplifications^107,109, one exciting possibility is that global gene expression profiles are drastically altered through a switch from H3K27 to H3K36 methylation. When considering this and the fact that antagonism between H3K36 and H3K27 methylation has also been observed in worms²², one might anticipate that this could be a general mechanism for the oncogenic behaviour observed for enzymes that promote H3K36 or H3K27 methylation (FIG. 5). But NSD proteins are not solely restricted to acting as oncoproteins; they can also act as tumour suppressors, as a lack of NSD1 expression is observed in neuroblastomas¹⁹.

Figure 5. NSD proteins can act as oncoproteins.

Through overexpression and/or translocation events that result in the fusion of nuclear receptor SET domain-containing (NSD) proteins with other proteins, such as the nucleoporin NUP98, NSD proteins can be aberrantly recruited to target loci in various tissues. As a consequence, global levels of dimethylated Lys36 on histone H3 (H3K36me2) increase and are sufficient to activate inappropriate transcription, which contributes to cancer development. In some cases, the increased levels of H3K36me2 are expected to inversely correlate with trimethylated H3K27 (H3K27me3; not shown), altering the balance between competitive activating and repressive ‘marks’. As a consequence, multiple gene sets that are sensitive to the levels of H3K36me2 are turned on, a causal event in oncogenesis. AML, acute myeloid leukaemia; AR, androgen receptor; RNAPII, RNA polymerase II.

Conclusions

Although widely perceived as an activating modification, H3K36 methylation also functions in transcriptional repression as well as in processes such as DNA repair. Thus, H3K36 methylation function extends beyond transcription and is likely to be relevant in a wide range of DNA-based processes. So far, much effort has gone into understanding the nature of the enzymes and their substrate specificities, but little is known about how the enzymes that mediate H3K36 methylation are localized and regulated beyond the well-characterized model of yeast Set2 and its relationship with RNAPII. Given that NSD1 binds various nuclear receptors, it is clear that this H3K36-specific methyltransferase, and probably others, operates in distinct tissues and perhaps in separate complexes. Similarly, NSD2 interacts with different transcription factors in mouse ES cells versus heart cells from embryonic day 12.5 (REF ³²). Thus, elucidating the distinct complexes that these enzymes form in different cell types and at different stages of development will be important. As mammalian cells possess multiple, non-redundant enzymes that methylate H3K36 to varying degrees, the roles of this modification are expected to be more complex than in yeast. Despite this, it is apparent that a role for methylated H3K36 in coordinating crosstalk between acetylation and methylation is conserved between flies and higher eukaryotes¹¹⁶. This crosstalk results in a given H3K36 modification that is interpreted by the reader protein in the context of neighbouring histone modifications, potentially changing its meaning. Undoubtedly, this ‘combinatorial histone language’ will be the subject of future studies of methylated H3K36 that will expand its scope of importance to the chromatin landscape.

Glossary

Dosage compensation

The mechanism by which expression levels from sex chromosomes are adjusted. In mammalian systems, one copy of the X chromosome is silenced in the female. By contrast, in Drosophila, genes on the male X chromosome are expressed at twofold levels

SET domain

(Suppressor of variegation 3–9, Enhancer of Zeste and Trithorax domain). A catalytic domain that uses S-adenosylmethionine to transfer methyl groups to substrates

PWWP domain

(Pro-Trp-Trp-Pro domain). A chromatin-interacting domain that has recently been shown to bind trimethylated Lys36 on histone H3

Plant homeodomain fingers

(PHD fingers). Motifs that bind zinc and have been shown to bind methylated residues

Allosteric

Pertaining to the process by which a binding event at one site influences the activity of an enzyme or protein at a second, distant site. This may lead to activation or inhibition by cooperation between ligands, when a ligand bound at one site affects the affinity of another site for its ligand by inducing transitions between distinct conformational states

Heterogeneous nuclear ribonucleoprotein L

(hnRNPL). A member of a highly abundant family of RNA-binding proteins known to associate with newly synthesized pre-mRNA

Pro isomerization

Peptide bonds to Pro residues can exist in either a cis or a trans state, and the transitions can be catalysed by prolyl isomerase enzymes

Anti-silencing function 1

(Asf1). A histone chaperone protein that associates with newly synthesized histones and is involved in chromatin synthesis

Reader protein

A protein that recognizes and binds post-translational modifications on histones

Chromodomain

A domain of ~50 residues that has been shown to bind methylated residues

Polypyrimidine tract-binding protein

(PTB). A protein that has been implicated as an antagonist of exon definition, the action of which results in the repression of exon inclusion

Fibroblast growth factor receptor 2

(FGFR2). A membrane-bound receptor that undergoes alternative splicing and is subject to regulation by methylation of Lys36 on histone H3

Epithelial splicing regulatory protein

(ESRP). An alternative splicing factor that is enriched in epithelial tissues and is responsible for enforcing specific exon inclusion

Non-homologous end joining

(NHEJ). The main pathway that is used primarily in the G1 phase of the cell cycle to repair chromosomal DNA double-strand breaks in somatic cells. In contrast to homologous-recombination repair, NHEJ is error-prone because it leads to the joining of heterologous ends