Understanding histone H3 lysine 36 methylation and its deregulation in disease

Abstract

Methylation of histone H3 lysine 36 (H3K36) plays crucial roles in the partitioning of chromatin to distinctive domains and the regulation of a wide range of biological processes. Trimethylation of H3K36 (H3K36me3) demarcates body regions of the actively transcribed genes, providing signals for modulating transcription fidelity, mRNA splicing and DNA damage repair; and di-methylation of H3K36 (H3K36me2) spreads out within large intragenic regions, regulating distribution of histone H3 lysine 27 trimethylation (H3K27me3) and possibly DNA methylation. These H3K36 methylation-mediated events are biologically crucial and controlled by different classes of proteins responsible for either ‘writing’, ‘reading’ or ‘erasing’ of H3K36 methylation marks. Deregulation of H3K36 methylation and related regulatory factors leads to pathogenesis of disease such as developmental syndrome and cancer. Additionally, recurrent mutations of H3K36 and surrounding histone residues are detected in human tumors, further highlighting the importance of H3K36 in biology and medicine. This review will elaborate on current advances in understanding H3K36 methylation and related molecular players during various chromatin-templated cellular processes, their crosstalks with other chromatin factors, as well as their deregulations in the diseased contexts.

Introduction

In eukaryotic cells, the genetic information is stored in the form of chromatin. The basic structural unit of chromatin is the nucleosomal particle that consists of approximately 146 base pairs of DNA wrapped around by a histone octamer composed of two copies of each core histone, H3, H4, and H2A and H2B [1]. Variations in amino acid compositions of the core histones H2A and H3 produce histone variants with distinctive functions. For example, the replication-independent H3 variant termed as H3.3 differs from canonical H3 isoforms, H3.1 and H3.2, by four and five amino acid residues, respectively, and H3.3 is expressed throughout the cell cycle, with its deposition occurring preferentially at active gene bodies [2]. Besides packaging of DNA, the nucleosome is subject to a myriad of dynamic and reversible histone post-translational modifications (PTMs) that occur on both the flexible protruding N-terminal tails and the core globular domains, such as acetylation, methylation, phosphorylation, ubiquitylation and ADP-ribosylation, thereby partitioning chromatin fibers into different domains. It has been increasingly appreciated that histone PTMs can either directly remodel chromatin structure or provide docking platforms for downstream ‘reader’ proteins [1, 3, 4]. Histone PTM represents a fundamental means for regulating virtually all of the chromatin-templated processes.

Histone methylation is mainly attached to the basic side chains of lysine and arginine residues. Various histone lysine methylations impart either the activating or repressive effect on gene transcription, which depends on the site, degree of methylation, genomic location, and the status of other coexisting PTMs. In general, methylations of histone H3 at lysine 4, 36 and 79 (H3K4, H3K36 and H3K79) are linked to the transcriptionally active state, whereas methylations of histone H3 at lysine 9 and 27 (H3K9, H3K27) and histone H4 at lysine 20 (H4K20) are associated with gene silencing [4, 5]. In addition to gene transcription regulation, histone lysine methylations are also found involved in numerous other cellular events including DNA damage repair, DNA replication, and mRNA splicing.

Abundance and genomic distribution of H3K36 methylation

H3K36 methylation exists in three states—mono-, di- and trimethylation (H3K36me1, H3K36me2 and H3K36me3). Mass spectrometry-based measurements of histone modifications have quantified the relative abundance for different H3K36 methylation states in cells [6–8]. Regardless of H3 isoforms (H3.1, H3.2 or H3.3), H3 proteins with either un-methylated or di-methylated H3K36 are the two most abundant forms, each accounting for ~ 20–45% of total H3 in most of the examined mouse tissues, whereas those with H3K36me3 occur in around 5% of total H3 [7]. Similar patterns were found in stem cells and human cancer cells [6, 8, 9]. Thus, methylation of H3K36 appears to be abundant [6–9]. In addition, while H3 proteins with dual H3K36me3 and H3K27me3 are very rare, those with combinatorial low-degree methylations of H3K36 (H3K36me1/2) and H3K27 (H3K27me1/2) are fairly abundant [6, 7, 9, 10], indicating an intimate crosstalk between the two histone PTMs.

Chromatin immunoprecipitation (ChIP) following by sequencing (ChIP-seq) further reveals genomic landscapes of H3K36 methylation. There is a strong correlation of H3K36me3 to gene bodies of the actively transcribed genes [11], and this feature has aided in identification of the previously unknown transcripts such as long non-coding RNAs [12]. H3K36me2, however, shows a very distinctive genomic occupancy pattern. Unlike H3K36me3, H3K36me2 displays a significant enrichment in intergenic regions [13–17] and promoter-associated H3K36me2 was also reported in various cell types [13, 15, 17–19]. Distinctive distributions of H3K36me3 and H3K36me2 are likely due to different enzymes responsible for catalyzing these two methylations, which will be covered in the next sections. While the gene body-associated H3K36me3 is widely perceived as a hallmark of active gene transcription, H3K36me2 remains relatively understudied.

Evidence supporting the functional importance for H3K36 and its methylation

Genetic manipulation of the histone genes represents a powerful tool for assessing functional relevance of an individual histone amino acid that is subject to potential PTMs. While it remains technically infeasible to mutate histone genes in human cells due to existence of a total of 64 histone gene clusters among different chromosomes, such a strategy has been used in model organisms that carry only one H3 gene or one histone gene cluster. In yeast, mutating H3K36 to a non-modifiable residue such as arginine or alanine does not perturb overt growth but leads to a transcription elongation defect as assayed with 6-azauracil [20]. Moreover, yeast cells carrying H3K36A or H3K36R mutations or lacking the H3K36-specific methylase KMT3/Set2 (see also the below section) are sensitive to nutrient stress and exhibit a shortened life span due to loss of H3K36me3 and the resultant upregulation of cryptic transcripts [21, 22]. In Neurospora crassa, a type of filamentous fungi and bread mold, an amino acid substitution of H3K36 to leucine caused various phenotypes including slow growth, poor conidiation and female sterility [23]. A similar developmental requirement for H3K36 was also demonstrated using the fruit fly where mutating H3K36 into arginine led to lethality before completion of pupal development, concurrent increase in acetylation of histone H4, as well as globally dysregulated mRNA expression from the genomic regions that are demarcated by H3K36me3 in wild-type animals [24, 25]. The importance of H3K36 is further strengthened by recent identification of so-called ‘onco’-histones targeting the H3K36 site. ‘Onco’-histone is a term used to describe cancer-associated, recurrent mutations of histones that dominantly drive the malignant development [26]. In this case, the heterozygous mutation of H3K36 to methionine (H3K36M) frequently occurs in specific human cancer types including chondroblastoma [27] and papillomavirus (HPV)-negative head and neck squamous cell carcinomas (HNSCCs) [28], and such an H3K36M mutant was shown to interfere with the global landscape of H3K36 methylation in cancer cells [14, 17, 28–30]. Together, these works support that H3K36 methylation is an evolutionarily conserved event essential for normal development in various eukaryotic organisms.

Signaling of H3K36 methylation via ‘writer’, ‘reader’ and ‘eraser’

Using a variety of model organisms and cell systems, prior studies have identified the molecular players responsible for deposition, recognition or removal of H3K36 methylation, which are often termed as ‘writer’, ‘reader’ and ‘eraser’.

‘Writer’

KMT3/Set2 is the sole H3K36 methyltransferase in yeast that is responsible for production of all three states of H3K36 methylation, whereas there is a labor division and a more complicated landscape of H3K36 methylation in mammalian cells, with at least six different H3K36 ‘writers’ or methyltransferases (Table 1). Specifically, the enzymatic activities of the NSD family proteins (KMT3B/NSD1, KMT3G/NSD2/MMSET/WHSC1, KMT3F/NSD3/WHSC1L1), KMT2H/ASH1L and Metnase/SETMAR are restricted to lower degree methylation of H3K36me2/1 [31–34]; only KMT3A/SETD2, the mammalian homologue of yeast Set2 can further produce H3K36me3, and disrupting SETD2 results in global loss of H3K36me3 but not H3K36me2/1 in mammalian cells [14, 35–37]. Studies also show that NSD1 or NSD2 acts as one of the main ‘writers’ for H3K36me2 in embryonic stem cells (ESCs) [16] and cancer cell lines [13, 15], probably due to its relatively high expression in these cell types. As mentioned above, the epigenomic landscapes of H3K36me3 and H3K36me2 in mammalian cells are quite different, most likely resultant from differential targeting of their corresponding ‘writers’. Thus, H3K36me3 and H3K36me2/1 appear to have distinctive biological functions. Technically, knockout or knockdown of SETD2 [14, 35–37] and NSD1-3 [13, 15] provides a genetic approach for manipulating cellular levels of H3K36me3 and H3K36me2, respectively.

Table 1.

List of H3K36 methylation ‘writers’, ‘readers’ and ‘erasers’

	Domain	Function	References
Writer
KMT3/SETD2	SET	Trimethylation of H3K36	[31, 33]
KMT3B/NSD1	SET	Mono- and di-methylation of H3K36	[31, 34]
KMT3G/NSD2	SET	Mono- and di-methylation of H3K36	[34]
KMT3F/NSD3	SET	Mono- and di-methylation of H3K36	[31, 34]
KMT2H/ASH1L	SET	Mono- and di-methylation of H3K36	[31]
Metnase/SETMAR	SET	Di-methylation of H3K36 during DSB repair	[85]
Reader
MRG15	Chromo	Reads H3K36me3 and recruits downstream effector proteins such as KDM5B	[66, 67]
PHF1; PHF19	Tudor	Reads H3K36me3 and directs PRC2 complex to active genes	[48, 72, 73]
DNMT3B	PWWP	Reads H3K36me3 and guides DNMT3B to catalyze gene-body DNA methylation	[40, 68, 69, 82]
DNMT3A	PWWP	Reads H3K36me2/3 in vitro	[40, 81, 82]
MSH6	PWWP	Reads H3K36me3 and recruits the mismatch sensor hMutSa during DNA mismatch repair	[40, 83]
LEDGF/PSIP1	PWWP	Reads H3K36me3 and recruits CtIP to facilitate HR repair of DSB; also reads H3K36me3 and recruits splicing machinery; also reads H3K36me2 and recruits MLL to oncogene promoters in leukemia	[19, 86, 88, 91]
ZMYND11/BS69	PWWP	H3.3K36me3-specific reader	[40, 50, 92]
Eraser
KDM2A/2B	JmjC	Demethylates H3K36me2	[51, 52]
KDM4A/4B/4C	JmjN;JmjC	Demethylates H3K9me3 and H3K36me3	[53, 55]

‘Reader’

Following its deposition, H3K36 methylation can serve as docking sites for recruiting the so-called ‘reader’ or effector proteins. Members of ‘Royal family’ protein domains, which include Tudor, PWWP, and chromodomain [38–40], have been documented to engage and ‘read’ different degrees of H3K36 methylation (Table 1; Fig. 1). These ‘reading’ domains establish a ‘cage’-like structure involving a cluster of aromatic residues for engagement of the side chain of the methylated H3K36 (Fig. 1), a phenomenon also seen with ‘reader’ domains specific for other histone methylation as elaborated extensively in previous reviews [39–43]. As the site of H3K36 is in close proximity to DNA in the nucleosomal context, the H3K36 methylation-‘reading’ Royal domains including PWWP [40, 44] and Tudor [45] can synergistically bind H3K36 methylation and DNA. Here, binding to DNA is mediated through a set of electrostatic interactions between the phosphate backbone of DNA and a highly positively charged patch on the surface of Royal domain, therefore, robustly increasing the binding affinity to modified nucleosome without displaying DNA sequence specificity [40, 44, 45].

Fig. 1.

Structures of the H3K36 methylation-specific ‘reader’ domains, as exemplified by a the chromodomain of MRG15 (PDB: 2F5K), b the PWWP domain of DNMT3B (PDB: 5CIU) and c the Tudor domain of PHF19 (PDB: 4BD3). The H3K36me3-engaging residues are labeled in each panel, with the histone peptide shown in gold

Studies also reveal distinctive mechanisms by which H3K36 methylation-specific ‘readers’ elicit their respective biological consequences. First, binding to H3K36 methylation by certain ‘readers’ can alter physical properties of nucleosomes, which are again due to proximity of H3K36 to DNA in the nucleosomal context. For instance, it was shown that engagement of H3K36me3 by the Tudor domain of PHF1 enhances accessibility of DNA to regulatory proteins within the H3K36me3-containing nucleosomes [45, 46]. Such a role of PHF1 may be involved in modulation of the methyltransferase activity of Polycomb repressive complex 2 (PRC2), to which PHF1 interacts, or PHF1-mediated DNA damage repair occurs [47, 48]. Additionally, as a more commonly appreciated mode-of-action, H3K36 methylation-specific ‘readers’ can recruit various interacting factors, which carry activities for regulating the chromatin-templated processes. In particular, a list of PWWP domain-containing proteins bind H3K36me3/2 (Table 1) and are known to be involved in gene transcription, chromatin modification and remodeling, mRNA splicing and DNA damage repair, thus linking H3K36me3/2 to modulation of these essential cellular events. Interestingly, NSD2, an H3K36me2 ‘writer’, also harbors a PWWP domain that can bind H3K36me2 and such ‘reading’ of the catalytic products by ‘writer’ suggests a mechanism for genomic spreading and propagation of H3K36me2 [49], a PTM that indeed often displays a pattern of large intergenic domains as shown by ChIP-seq [14, 16]. Furthermore, the H3K36me3-binding PWWP motif of ZMYND11/BS69 is flanked by a bromodomain and an embedded zinc finger, which create a second ‘pocket’ for engaging the H3.3 isoform-unique residue serine 31 (H3.3S31), thus providing a delicate strategy for ‘reading’ H3K36me3 in an H3 isoform-specific manner [50]. Together, recruitment of ‘readers’ represents an important mechanism through which H3K36 methylation elicits its functional consequences. Manipulating the interaction between H3K36 methylation and its specific ‘readers’ provides a means for experimentally dissecting the biology related to this histone PTM.

‘Eraser’

H3K36 methylation is reversible and can be actively removed by certain members of the JmjC signature domain-containing histone lysine demethylases or ‘erasers’ (Table 1), which include the KDM2/JHDM1 and KDM4/JHDM3/JMJD2 family proteins [51–55]. Functional and structural analyses uncovered that the conserved JmjC domain coordinates two cofactors, Fe(II) and α-ketoglutarate, to mediate the hydroxylation and ultimate demethylation of methyl groups. The KDM2/JHDM1 group of ‘erasers’ specifically demethylates H3K36 methylation with a preference for H3K36me2; in contrast, the KDM4/JMJD2/JHDM3 group enzymes exhibit dual-substrate specificity towards H3K9me3 and H3K36me3 [51–55].

Biological processes regulated by H3K36 methylation and associated factors

H3K36me3-mediated silencing of spurious intragenic transcripts

H3K36me3 is a hallmark of active transcriptional elongation. In yeast, Set2 associates with the serine 2 phosphorylated, and to a lesser extent, the serine 5 phosphorylated C-terminal domain (CTD) of RNA polymerase II (RNAPII) through its Set2–Rpb1-interacting (SRI) domain, thereby traveling across the gene body together with RNAPII to lay down H3K36me3 during elongation [20, 56–58] (Fig. 2a). Such transcription-associated H3K36me3 recruits the Rpd3S histone deacetylase complex to the actively transcribed regions, inducing local histone deacetylation to suppress transcription initiation from cryptic alternative start sites within the gene body [59, 60] (Fig. 2a). Mechanistically, recognition of the gene-body-associated H3K36me3 relies on the combined action of the chromodomain of Eaf3, a subunit of Rpd3S, and the plant homeodomain (PHD) of the Rco1 subunit [61–63]. Similarly, SETD2 in mammalian cells interacts with the elongation-competent form of RNAPII to carry out H3K36me3 at the gene bodies of actively transcribed genes [64, 65] (Fig. 2b). Although silencing of SETD2 also leads to aberrant intragenic transcription initiation, no change in histone acetylation levels happens at the affected genes [64], indicating a regulatory mechanism distinctive from what was observed in yeast.

Fig. 2.

Suppression of cryptic intragenic transcripts by H3K36me3. Both the yeast Set2 and the mammalian SETD2 bind the elongating RNAPII and decorate the gene body regions with H3K36me3. In yeast (a), the H3K36me3 recruits the Rpd3S histone deacetylase complex to deacetylate the transcribed body regions to suppress spurious intragenic transcripts. Unlike yeast, the mammalian cells (b) employ different H3K36me3 ‘readers’, a de novo methyltransferase DNMT3B and the MRG15–KDM5B complex, to methylate DNA and remove H3K4me3, respectively, thus leading to inactivation of intragenic promoters; as well, SETD2 recruits the FACT complex to facilitate nucleosome structure recovery after RNAPII passage, thus preventing RNAPII entry into the intragenic promoters

Indeed, recent studies have uncovered multiple SETD2- and H3K36me3-directed signaling pathways for ensuring fidelity of gene transcription. First, similar to its yeast orthologue Eaf3, MRG15 also ‘reads’ H3K36me3 via a conserved chromodomain [66] (Figs. 1a, 2b; Table 1). Here, besides a mammalian Rpd3S-like complex [67], MRG15 also interacts with KDM5B/JARID1B, a histone ‘eraser’ specific to H3K4me3/2. H3K4me3 is a histone PTM correlated with activation of gene promoters. ChIP-seq in mouse ESCs revealed a predominant targeting of KDM5B to intragenic regions in the MRG15-dependent fashion, and depletion of either MRG15 or KDM5B resulted in the increased levels of intragenic H3K4me3 and concurrent cryptic intragenic transcription [67]. This work demonstrated an important requirement of a pathway involving H3K36me3–MRG15–KDM5B for ‘erasing’ H3K4me3 and restricting the usage of alternative intragenic promoters within the gene body, thus safeguarding transcription fidelity during the wake of RNAPII.

Also, recent works found that the SETD2-catalyzed H3K36me3 is recognized by the PWWP domain of DNMT3B (Figs. 1b, 2b; Table 1), which then catalyzes the de novo DNA methylation at the gene body of the actively transcribed gene [68, 69]. In mouse ESCs, SETD2 knockdown resulted in a drastic reduction of H3K36me3, a concomitant loss of DNMT3B binding to gene bodies, and activation of cryptic intragenic transcription initiation. Importantly, loss of Dnmt3b also resulted in an increase of spurious transcripts at the regions affected by SETD2 knockdown and such an abnormal increase in spurious transcripts was suppressed by restoration of a wild-type Dnmt3b but not its enzymatic-dead mutant or a PWWP mutant defective in ‘reading’ H3K36me3. Clearly, there exists an additional signaling pathway in mammalian cells involving SETD2–H3K36me3–DNMT3B, which operates to maintain high degree of DNA methylation at the gene body to protect from spurious RNAPII entry and to ensure transcript fidelity [68, 69].

Further, it is noteworthy that SETD2 per se was also documented to attenuate spurious intragenic transcription through recruiting the FACT chromatin-remodeling complex [64]. Specifically, FACT is recruited in a SETD2-dependent manner to the gene body of actively transcribed genes, where it not only facilitates the RNAPII passage by removing H2A–H2B dimers during the elongation phase but also helps to re-establish nucleosomes after RNAPII has passed through (Fig. 2b). Such activities of the FACT remodeling complex also contribute to prevention of cryptic intragenic promoter activation [64].

Together, to ensure fidelity of gene transcription in the wake of RNAPII, cells employ multiple mechanisms involving Set2/SETD2 and the catalyzed H3K36me3 to suppress spurious transcriptional initiation within the gene bodies.

Shaping chromatin landscapes via crosstalk between H3K36 methylation and PRC2

H3K36 methylation can also mediate its biological outputs through its antagonistic effect on other biologically important histone PTMs. Initially, the in vitro histone methyltransferase assays showed that the H3K27me3-catalyzing activity of PRC2 is impaired in cis by the presence of H3K36me2 or H3K36me3 [70]. A similar observation was seen for H3K4me3 [70]. These in vitro studies indicated a mechanism by which the transcriptionally active chromatin marks counteract the actions of PRC2. Such an antagonism between H3K36me3/2 and PRC2 was substantiated by the in vivo results obtained from various tissue types that include ESCs [16], adipose tissue [71] and cancer cell lines [6, 14, 37]. These findings collectively support that restricting genomic distribution of the PRC2-catalyzed H3K27me3 represents one of the main mechanisms underpinning the biological functions of H3K36me3/2. However, unlike H3K27me3, H3K27me2 often co-exists with H3K36me2/1 on the same H3 tails in cells [6, 7, 9, 10], indicating a more complicated interplay of the two PTMs. In a recent mass spectrometry-based analysis of PRC2-interacting proteins carried out in mouse ESCs, Streubel et al. uncovered Nsd1 [16]. Consistently, ChIP-seq found co-localization of the Nsd1-mediated H3K36me2 with the PRC-catalyzed H3K27me2, and Nsd1 depletion led to genome-wide loss of H3K36me2 and concurrent expansion of H3K27me3 [16] (Fig. 3). Furthermore, in ESCs, PRC2 also associates with the H3K36me3/2 ‘erasers’ [72, 73], as well as the polycomb-like proteins (PHF1 and PH19), which can ‘read’ H3K36me3/2 thereby assisting with PRC2’s intrusion and spreading into the H3K36me3-marked active genes (Figs. 1c, 3; Table 1) [48, 72, 73]. Apparently, biological outcomes of these above complexes oppose each other, thus enabling conversion between different chromatin states, either active, repressed or poised. It remains to be determined how these antagonizing complexes reach a fine balance during dynamic modulation of H3K36me3/2 and H3K27me3/2.

Fig. 3.

Crosstalk of H3K36 methylation with H3K27 methylation and DNA methylation. In the intergenic regions, the NSD1-catalyzed H3K36me2 prevents spreading of the PRC2-catalyzed H3K27me3, and loss of NSD1 causes genome-wide loss of H3K36me2 and DNA methylation, as well as concomitant gain of H3K27me3. Intrusion and spreading of the PRC2-catalyzed H3K27me3 into the active gene bodies are assisted by PHF1 or PHF19, which mediates ‘reading’ of H3K36me3. Furthermore, CpG island elements recruit KDM2A to remove H3K36me2

Shaping epigenomic landscapes via crosstalk between H3K36 methylation and DNA methyltransferases

Growing evidence has linked H3K36 methylation to DNA methylation. First, NSD1 abnormality or mutation is a hallmark for Sotos syndrome, a genetic disorder characterized by facial dysmorphism, macrocephaly, learning disability, and childhood overgrowth [74], and DNA methylome analysis of blood samples from Sotos syndrome patients carrying NSD1 abnormalities has revealed loss of DNA methylation at 7,038 CpG sites, which account for 99.3% of all the differentially methylated CpG sites [75]. Importantly, these hypomethylated genomic sites were found enriched in genes related to neural and cellular development pathways [75]. Remarkably, a comprehensive analysis of genomic datasets from a large set of human tumor samples also found a strong correlation between NSD1 mutation and DNA hypo-methylation [76]. As well, using unsupervised hierarchical clustering of DNA methylation dataset obtained from HPV-negative HNSCCs, Papillon-Cavanagh et al. uncovered a previously unappreciated HNSCC subtype that is featured with either H3K36M substitution or NSD1 mutations, with the most pathogenic missense mutations of NSD1 clustering around its catalytic SET domain [28]. Interestingly, reminiscent to the NSD1-mutated Sotos syndrome, this HNSCC subtype is characterized by global DNA hypo-methylation, particularly at enhancers and intergenic regions rather than promoters [28]. The fact that H3K36me2 reduction is a shared epigenomic alteration in both NSD1-mutated Sotos syndrome and HNSCC patients supports a potential role of H3K36me2 in regulating DNA methylation (Fig. 3). The exact underlying mechanisms remain to be defined. Another evidence highlighting the interplay between H3K36me2 and DNA methylation is from an observation that CpG islands (CGIs) are normally refractory to DNA methylation and also are uniquely devoid of H3K36me2 [77]. Here, un-methylated CGIs serve as the initial recruitment and nucleation sites for the H3K36-specific ‘eraser’ KDM2A through its zinc finger-CxxC domain [77] (Fig. 3). Similarly, KDM2B specifically binds to non-methylated CGIs genome-wide via its zinc finger-CxxC domain. In contrast to KDM2A, KDM2B is preferentially enriched at Polycomb-repressed CGIs where it recruits PRC1 complexes to suppress lineage-specific genes in mouse ESCs [78, 79]. More interestingly, a recent work further demonstrated that H3K36me2 marks are tightly bound by a cysteine-rich domain of Ten-eleven translocation 2 (TET2), a DNA dioxygenase and demethylase, implying that H3K36me2 may also act to fine-tune the dynamic processes of DNA methylation and demethylation [80].

Additionally, as discussed in the above section, the SETD2-mediated H3K36me3 tethers DNMT3B to gene bodies of the active genes via a PWWP ‘reader’ domain (Fig. 1b); consequently, DNA methylation is induced at high levels to prevent cryptic intragenic transcription (Fig. 2b). Notably, various in vitro studies also indicated that the PWWP domain of DNMT3A behaves similarly as that of DNMT3B, with a preferential binding to H3K36me2/3 [81, 82]. However, as revealed by ChIP-seq studies, the DNMT3A’s genomic binding does not follow the transcriptional elongation activity and its binding pattern is not altered following cellular loss of H3K36me3, suggesting a functional difference between DNMT3A and DNMT3B as for their interplay to H3K36 methylation [68, 69].

Overall, accumulating evidence pinpoints the intimate, yet rather complicated crosstalks among H3K36 methylation (either H3K36me3 or H3K36me2), DNMT3B/3A, PRC2 and H3K27me2/3. Further investigation is warranted to delineate the exact regulatory details under different cellular and chromatin contexts.

H3K36 methylation acting to recruit DNA damage repair factors

The DNA mismatch repair (MMR) pathway safeguards the integrity of cellular genomes by correcting replication- and damage-associated base–base mismatches and small insertion–deletion loops. Cells lacking SETD2 displayed an impaired chromatin occupancy of the MMR machinery, resulting in the elevated frequency of microsatellite instability and spontaneous genetic mutation [83]. Mechanistically, H3K36me3 serves as the docking site for recruiting hMutSa (hMSH2–hMSH6), a key mismatch sensor, through an H3K36me3-‘reading’ PWWP domain of the hMSH6 subunit (Fig. 4; Table 1) [83]. The recruited hMSH2–hMSH6 complex then surveys the naked DNA for lesion repair during both DNA replication and gene transcription. Strikingly, the abundance of H3K36me3 fluctuates in response to cell cycle changes, with it being highly enriched in the late G1/early S, the phase when DNA mismatches usually occur, and being largely removed in late S phase and maintained at lower levels throughout the G2/M phase [83]. In support of a critical involvement of H3K36me3 in MMR, the cancer-related mutation of histone H3 Gly34 (H3G34), a histone residue in the vicinity of H3K36, to a bulky side chain residue was shown to suppress MMR and to facilitate genome instability by sterically hindering the interaction of H3K36 to both SETD2 and hMutSα [84].

Fig. 4.

H3K36 methylation-mediated promotion of DNA damage repair via both DNA mismatch repair (MMR) and double-strand break (DSB) pathways. Specifically, the SETD2-catalyzed H3K36me3 is recognized by the mismatch sensor hMutSa, thereby promoting MMR during active gene transcription. In response to DSB, Metnase is activated to catalyze H3K36me2 near the DSB site, which in turn recruits the Ku70–NBS1 complex for damage repair via NHEJ. Unlike H3K36me2, the SETD2-catalyzed H3K36me3 also favors HR repair of DSB by orchestrating assembly of an H3K36me3 ‘reader’ complex, LEDGF–CtlP

In addition to MMR, H3K36 methylation is also implicated in the repair of double-strand DNA breaks (DSBs) by non-homologous end joining (NHEJ) and homologous recombination (HR) pathways. Here, Fnu et al. showed that H3K36me2 is markedly induced by Metnase (Table 1), a methylase possessing a SET domain and a transposase nuclease domain, at regions adjacent to the DSB site [85]; such an induced H3K36me2 then enhances recruitment of the NHEJ pathway proteins such as Ku70 and NBS1 for the error-prone DSB repair [85] (Fig. 4). In contrast, during the error-free HR repair of DSB, the constitutive anchoring of LEDGF/PSIP1 to the H3K36me3-marked chromatin through its PWWP ‘reader’ domain can facilitate timely recruitment of CtIP, a DNA damage response factor, to the break sites, thus facilitating the resection events associated with HR [86] (Fig. 4; Table 1). A more recent study also showed that, akin to γ-H2AX, temporal changes of global H3K36me3 levels are induced post-DSB in a SETD2-dependent manner [87]; such an elevation of H3K36me3 peaks 30 min post-exposure to the DSB-inducing agents and drops back to basal levels after 4 h [87]. Moreover, the increased H3K36me3 demarcates the immediate surrounding regions around DSBs and recruits LEDGF, which in turn brings in the histone acetyltransferase KAT5/TIP60 to catalyze acetylation of histone H4 Lys 16 (H4K16ac). H4K16ac and TIP60 are known to mitigate the interaction between the Tudor domain of 53BP1 and H4K20me2, and promote recruitment of BRCA1, thus favoring HR for DSB repair [39, 88]. It is worth noting that H3K36me2 and H3K36me3 appear to exert opposite effect on the choice for DSB repair, either NHEJ or HR. The molecular determinants underlying activation of these distinctive pathways upon DNA damage remain to be defined.

Regulation of mRNA splicing by gene body-associated H3K36me3

Lines of evidence have highlighted a bidirectional communication between H3K36me3 and co-transcriptional splicing of pre-mRNAs. Such a connection was initially implied by the observation that H3K36me3 is progressively built up in the transcribed gene bodies, with preferential enrichment in exons relative to introns within the same gene [89]. Interestingly, exons included for alternative splicing tend to have higher levels of h3K36me3 than those excluded, and the intron-containing genes are consistently marked by higher H3K36me3 levels than those intronless genes, irrespective of nucleosome occupancy, gene size and transcriptional activity [89]. Subsequent studies then revealed that H3K36me3 plays a crucial role in the recruitment of various splicing-related factors, thus influencing the outcomes of mRNA splicing. For instance, the chromodomain-containing H3K36me3 ‘reader’ protein MRG15 recruits the polypyrimidine tract-binding protein (PTB) to regulate alternative splicing choice [90]; as well, the PWWP domain-containing H3K36me3 ‘reader’ LEDGF contributes to the splicing regulation by acting as an adaptor to couple H3K36me3 with the splicing machinery assembly [91] (Fig. 5; Table 1). Likewise, ZMYND11, an H3.3K36me3-specific ‘reader’, mainly functions to regulate intron retention by antagonizing elongation factor Tu GTP-binding domain containing 2 (EFTUD2), a component of U5 snRNP [92] (Fig. 5; Table 1). Remarkably, splicing can also affect H3K36me3 levels. Here, inhibition of splicing was reported to impair the SETD2 recruitment and reduce the H3K36me3 levels without altering the transcriptional activity, which was demonstrated by the unaffected RNAPII occupancy and transcription elongation rate [89].

Fig. 5.

H3K36me3 modulates alternative splicing of mRNA via its ‘readers’ and recruited splicing machineries. Recognition of H3K36me3 by MRG15 recruits PTB, leading to enhanced splicing of alternative exons, whereas binding of the LEDGF–SRSF1 complex to H3K36me3 facilitates inclusion of alternative exons. Further, ZMYND11 specifically ‘reads’ H3.3K36me3 and opposes the intron-splicing activity of EFTUD2, thus resulting in intron retention

Human pathologies associated with deregulation of H3K36 methylation and related pathways

Because H3K36 methylation is involved in numerous processes critical for cellular functions, its deregulation often results in pathogenesis of various human diseases.

‘Mis-writing’ of H3K36 methylation in disease

NSD1

The t(5;11)(q35;p15.5) chromosomal translocations were reported in ~ 15% of pediatric acute myeloid leukemia (AML) patients and a smaller subset of adult AML cases, resulting in the aberrant chimeric fusion between NSD1 and a nucleoporin-encoding gene, Nup98 [93–97]. Such a NUP98–NSD1 chimeric oncoprotein carries the H3K36 methyltransferase domain of NSD1 and, using murine models, prior studies have shown that NUP98–NSD1 is sufficient in driving AML development, either alone or together with the FLT3 kinase mutation [98, 99]. Studies with the murine AML cells carrying NUP98–NSD1 further demonstrate that it enforces high expression of multiple AML proto-oncogenes such as HOX gene clusters and MEIS1 [98], which are known to arrest hematopoietic differentiation and promote progenitor/stem cell expansion [100–103]. Mechanistically, NUP98–NSD1 directly binds these downstream targets, mediating H3K36 methylation and concurrent histone acetylation to antagonize spreading of the PRC2-catalyzed H3K27me3, a gene-repressive histone PTM known to be associated with terminal differentiation of myeloid cells [98, 104]. Consistently, transcriptome profiling of samples from human AML patients carrying the NUP98–NSD1 translocation also revealed a characteristic HOX-gene expression pattern [93, 95]. In the clinic, NUP98–NSD1 is significantly associated with a poorer prognosis [93, 95], demanding new treatment strategies.

As mentioned above, germ-line haploinsufficiency of NSD1 is a hallmark mutation of Sotos syndrome [74, 105]. Additionally, inactivating mutations of NSD1 were recently found in approximately 10% of HNSCC patients featured by widespread DNA hypo-methylation [28]. A similar NSD1-mutated and DNA-hypomethylated subtype was reported in patients with lung squamous cell carcinoma [106]. Interestingly, HNSCC patients with NSD1 mutations generally display the pronounced survival advantage [107]. Recently, Bui et al. have shown that disruption of NSD1 in the HNSCC cell lines confers a favorable chemotherapeutic sensitivity with a markedly decreased response to cisplatin, presumably due to global DNA hypo-methylation seen in these cells [108]. Moreover, NSD1 inactivation in HNSCC also confers an ‘immune cold’ phenotype, featured by low levels of filtration of anti-tumor immune cells such as the M1 tumor-associated macrophages and CD8⁺ T cells, as well as low expression of the PD-1 immune checkpoint receptor and its ligands [106]. It has been postulated that NSD1 loss acts as a tumor cell-intrinsic strategy for the production of an ‘immune cold’ phenotype, which potentially affects the efficacy of immunotherapy-based agents [106]. In agreement to its role as a tumor suppressor, NSD1 was found to be epigenetically silenced through promoter DNA hyper-methylation in neuroblastoma, glioma [109] and clear cell renal cell carcinoma (ccRCC) [110], and restoration of the NSD1 expression in these tumor cells compromised their growth [109]. Thus, depending on tissue types, both oncogenic and tumor suppressive roles are reported for NSD1.

NSD2

In about 10–20% of multiple myeloma (MM), a malignancy of antibody-producing plasma cells, NSD2 is amplified and overexpressed due to recurrent t(4;14) chromosomal translocation [111]. Additionally, gain-of-function mutation of NSD2 was reported in a subset of pediatric acute lymphoblastic leukemia [112, 113]. Various studies have shown that the altered NSD2 acts as an oncoprotein, enforcing the gene-expression programs that are crucial for MM tumorigenesis [13, 15, 113–116]. Mechanistically, the pathogenic alterations of NSD2 enhanced genome-wide distribution of H3K36me2 while restricting the PRC2-dependent H3K27me3, both of which are involved in the selection for a gene-expression profile that favors oncogenic transformation of plasma cells [6, 15, 113, 116]. Thus, such the overexpressed and hyper-activated NSD2 enzymes represent an attractive therapeutic target in MM. However, despite recent effort and progress, it remains as a challenge to develop the small-molecule inhibitors of NSD2 that are selective, potent and bioactive. For example, in a recent screening with full-length NSD2 proteins and nucleosome substrates, Coussens et al. identified two out of 16,251 compounds, namely DA3003-1 and PF-03882845, that displayed strong binding affinity to the NSD2’s SET domain in vitro and potent H3K36me2-suppressing effect in U2OS cells [117]; however, multiple targets beyond NSD2 appear to be responsible for the cytotoxic mechanisms [117].

NSD3

Genomic amplification and alterations of NSD3 occur in multiple cancer types, implicating its cancer-promoting role [118–122]. For instance, the fusion between the NUP98 and NSD3 genes was detected in patients with AML or myelodysplastic syndrome [123, 124]. Presumably, the produced NUP98–NSD3 is able to promote hematopoietic transformation in the same fashion as what was already shown for NUP98–NSD1, due to a structural similarity between the two [98]. Recently, Turner-Ivey et al. used a transgenic mouse model to enforce the NSD3 overexpression in the mammary epithelium, and found that a subset of female NSD3-transgenic mice developed frank mammary tumors featured with hyperplasia, ductal dysplasia and carcinoma in situ at the age of 40–50 weeks [120]. It remains unclear, however, as for whether such a cancer-promoting role of NSD3 is dependent on its H3K36 methyltransferase activity.

ASH1L

ASH1L represents another H3K36 methyltransferase that can counteract PRC2- and H3K27me3-mediated gene silencing [19, 125, 126]. And ASH1L was recently shown to be essential for tumorigenicity of the MLL/KMT2-rearranged leukemia [19]. Here, the ASH1L-catalyzed H3K36me2 is recognized by LEDGF, which subsequently recruits the MLL proteins onto the leukemia-related target genes, such as HOXA9 and MEIS1, to enforce their high expression [19]. In addition, this pathway can be suppressed by KDM2A, an H3K36me2 demethylase, and Polycomb group silencing factors [19]. These observations show the importance of well-balanced antagonism between H3K36 and H3K27 methylation for the prevention of malignant development.

SETD2

Loss-of-function mutation or inactivation of SETD2, the non-redundant H3K36me3-inducing methyltransferase, is found to be frequent in a wide range of human cancers, including ccRCC [127], acute leukemia [128, 129], high-grade glioma [130] and enteropathy-associated T cell lymphoma type II [131]. Clearly, SETD2 has generally been proposed to function as a tumor suppressor, a notion recently supported by various SETD2 loss-of-function models [128, 132–134]; evidence also exists, however, showing that SETD2 can be essential for tumor cell viability [135]. In various cancer models, SETD2 inactivation has been shown to perturb a range of critical processes, including gene expression [136], DNA repair [137], genomic instability [138], drug resistance [139], alternative splicing [140, 141] and metabolism [142, 143]. It should be noted that the non-histone substrates of SETD2 such as alpha-tubulin also show perturbed methylation in SETD2-mutated ccRCC cells, contributing to tumor-associated phenotypes [138, 144]. Due to a complicated picture of this topic, readers shall also be referred to comprehensive reviews published recently [33, 145, 146].

‘Onco’-histone: mutation of H3K36 in cancer

Heterozygous mutation of the H3.3 K36M (H3.3K36M) is a frequent event in chondroblastoma, suggesting a dominant-negative or neomorphic nature of the mutation [27]. Also, recurrent mutations of K36M at histone H3.1, H3.2 or H3.3 were detected in HPV-negative HNSCCs [28]. Incorporation of the H3.3K36M ‘onco’-histones into chromatin reduced the genome-wide H3K36me2 and H3K36me3 levels, which is likely due to direct inhibitory effect of H3.3K36M on NSD2 and SETD2, respectively [14, 17]. Fang et al. reported that, in the chondrocytes with the engineered H3.3K36M, the H3K36me2 peaks were drastically depleted from the intergenic regions and became relatively enriched at promoters, gene bodies and transcription end sites, despite the overall reduced levels [17]. Notably, there is a significant correlation between gene expression changes and reduction of H3K36me2 and H3K36me3 at gene bodies, although the underlying mechanism is unclear [17]. Likewise, using the mesenchymal progenitor cells, Lu et al. observed a similar dominant sequestration and repressive effect of H3.3K36M on NSD2 and SETD2, causing global decrease of H3K36me2/3 and arrest of cell differentiation [14]. Additionally, there was concurrent gain of H3K27me3 at intergenic regions, which further induced re-distribution and/or dilution of the H3K27me3-‘reading’ PRC1 complexes, thus providing an explanation for the observed de-repression of polycomb target genes associated with mesenchymal differentiation [14].

Aberrant readout of H3K36 methylation in disease

DNMT3A/3B

Mutations targeting the PWWP domain of DNMT3B were identified from patients with ICF (immunodeficiency, centromeric instability, and facial anomalies) syndrome, a genetic immune disorder featured by defective lymphocytes with pericentromeric DNA hypo-methylation. Such a pathogenic PWWP mutation of DNMT3B (S270P substitution) abrogated the ability of DNMT3B to ‘recognize’ H3K36me3, which is critical for installing proper DNA methylation at the pericentric heterochromatin [82, 147]. Importance of appropriate ‘decoding’ of H3K36 methylation in the diseased setting is further underscored by the recent identification of a heterozygous W330R mutation that ‘hit’ the PWWP domain of DNMT3A in patients of microcephalic dwarfism [148]. Such a W330R mutation disrupted DNMT3A’s binding to H3K36me2/3. However, unlike what was seen with the H3K36me3-binding-defective PWWP mutant of DNMT3B [68, 69], genome-wide profiling of DNA methylation in dwarfism patient-derived cells carrying the PWWP(W330R)-mutated DNMT3A revealed the significantly enhanced DNA methylation at the polycomb-targeted genomic regions, which are enriched with development-related genes and normally show very low degree of DNA methylation [148]. Mechanistically, it was proposed that disrupting the PWWP-mediated binding of H3K36me2/3 might re-shuffle DNMT3A from the H3K36me2/3-covered regions to polycomb-targeted genomic regions, which in return interferes with the PRC2 functions. As a result, hyper-methylation of polycomb targets, as well as concurrent increase in transcriptional activity and gene-active histone PTMs, was seen in cells from dwarfism patients relative to normal cells [148].

ZMYND11

ZMYND11, an H3.3K36me3 ‘reader’, was proposed as a putative tumor suppressor, showing downregulation in various human cancers [149, 150]. Wen et al. showed that, in breast cancer cells, ZMYND11 acts as a transcription co-repressor and blocks the transition of paused RNAPII to the elongation-competent form, and that ZMYND11 represses expression of oncogenes in a mechanism that relies on its capability as an H3.3K36me3-specific ‘reader’ [50]. In addition, chromosomal translocation and missense mutation of ZMYND11 were reported in patients with AML [151–153] and the 10p15.3 microdeletion syndrome [154–156], respectively, but the relationship between these alterations and H3K36me3 is unclear.

‘Mis-erasing’ of H3K36 methylation in disease

KDM2A/2B

KDM2A/2B was shown to have cancer-promoting roles in cancer. KDM2A is significantly amplified in a subset of non-small cell lung cancer (NSCLC), correlating with worse clinical prognosis [157]. Here, KDM2A suppressed the expression of dual-specificity phosphatase 3 (DUSP3) through ‘erasing’ the promoter-associated H3K36me2, an event that then led to ERK1/2 hyper-activation and the enhanced NSCLC cell proliferation and invasion in vivo [157]. Likewise, KDM2B is highly expressed in human AML and is critical for AML initiation and maintenance [158]. Here, the AML-causing leukemia stem cells utilize KDM2B to ‘erase’ H3K36me2 at the tumor suppressor locus p15^Ink4b, ensuring its silencing [158]. Furthermore, a previous work demonstrated that the FGF2-triggered transcriptional upregulation of KDM2B in bladder cancer operates in concert with EZH2/KMT6 to repress the EZH2-targeting miR-101, which not only results in EZH2 overexpression but also promotes robust cell proliferation, migration and angiogenesis [159]. Also, in a model of pancreatic ductal adenocarcinoma, the synergistic effect of ectopic expression of KDM2B and Kras^G12D is sufficient to drive tumor development via modulating both developmental and metabolic genes [160]. Owing to these works, chemical inhibitors of KDM2A/2B are proposed to be useful for cancer intervention. Recently, compound 9, a cyclopropyl-containing hydroxamate derivative designed based on the crystal structure of KDM7B, was shown to robustly and selectively repress KDM2A, KDM7A, and KDM7B over other KDMs, with IC50 values of 6.8, 0.2, and 1.2 μM, respectively [161]; treatment of cancer cells with compound 9 resulted in arrest of cell cycle progression [161].

KDM4A/B

KDM4A/KDM4B are lysine demethylases with dual specificity for both H3K9me2/3 and H3K36me2/3. In various human cancers, they are found overexpressed due to expression mis-regulation and/or genomic amplification [162, 163]. Prior studies of KDM4A and KDM4B in cancer have largely been focused on their role as the H3K9 demethylase. Recently, Mishra et al. examined a cancer-related phenomenon termed transient site-specific copy gain (TSSG) and reported a correlation of TSSG formation to KDM4B recruitment and H3K36me3 removal [164]. This report supported a role for H3K36me3 ‘mis-erasing’ during oncogenesis [164]. Like KDM2A/2B, the cancer-promoting property of the overexpressed KDM4A/4B makes them valuable targets for anti-cancer agent development. Compound 4 (NSC636819), a selective KDM4A/KDM4B inhibitor designed based on the crystal structure of the KDM4B–pyridine 2,4-dicarboxylic acid–H3K9me3 ternary complex, was identified to block the demethylating activities of KDM4A/4B towards H3K9me3 and H3K36me3 in vitro; however, only the H3K9me3 level was elevated in prostate cancer cells treated with Compound 4 [165].

Conclusions

In summary, H3K36 methylation is a crucial, versatile histone PTM involved in not only the regulation of epigenomic landscapes in cells but also a wide range of biological processes. In a cellular and chromatin context-dependent fashion, H3K36 methylation can recruit different ‘reader’ proteins, which carry out their respective functions for mediating chromatin remodeling, gene transcription regulation, DNA damage repair, or mRNA splicing. Such an important role for H3K36 methylation is reflected by human diseases induced by mis-regulation of various proteins that are responsible for deposition, functional ‘readout’ or removal of H3K36 methylation.

Despite recent advances, questions and challenges remained to be addressed. For example, besides methylation-modulating activities, the regulatory roles of certain H3K36-associated enzymes may rely on their scaffolding functions. An extreme example is a short isoform of NSD3 (NSD3S), which lacks the SET domain and acts mainly as an adaptor protein for bringing together BRD4 and CHD8 to sustain the growth of AML cells [166, 167]. As well, NSD3S was shown to interact with MYC [168]. Association of NSD3 and BRD4 appears to be essential for malignant growth of NUT midline carcinoma (NMC) that is featured with aberrant chromosomal translocation of BRD4–NUT or NSD3–NUT [121]. Also, the bindings of various ‘readers’ to H3K36me3 and H3K36me2 appear to be promiscuous, with a generally low affinity seen for either ligand, as exemplified by the PWWP domain of NSD2, DNMT3A and DNMT3B [19, 49, 81, 82]. Yet, in cells, DNMT3A and DNMT3B display differential patterns of targeting to the H3K36me3 versus H3K36me2-covered genomic regions [68, 69]. How could the promiscuous, low-affinity binding mediate the wanted efficient and selective events in cells? A possibility is that the ‘reader’ modules may simultaneously bind both DNA sequences and H3K36me3/2 as shown for the Tudor domain of PHF1 [45, 46]; alternatively, they may act in concert with other motifs within the same protein or protein complexes to promote a multivalent engagement of histone tails, thus enhancing both selectivity and affinity [40, 42, 44, 45, 50]. Last, unlike potent, selective chemical inhibitors developed for other epigenetic factors such as BRD4 and EZH2 [169–172], those against the H3K36-related regulators (such as the NSD family proteins as discussed above) are currently lacking. Future efforts shall also be directed towards the development of small molecules that specifically target H3K36 methylation-related ‘writers’, ‘erasors’ and/or readers, which could provide attractive therapeutic means for the genetically defined cancer types, for example, those carrying gain-of-function mutations of NSD1 or NSD2.

Abbreviations

H3K36

Histone H3 lysine 36

H3K36me1/2/3

Mono/di/trimethylation of H3K36

H3K27me3

Trimethylation of histone H3 lysine 27

PTMs

Post-translational modifications

ChIP-seq

Chromatin immunoprecipitation (ChIP) following by sequencing

ESCs

Embryonic stem cells

PWWP

Pro-Trp-Trp-Pro

PRC2

Polycomb repressive complex 2

RNAPII

RNA polymerase II

CTD

C-terminal domain

SRI

Set2–Rpb1-interacting domain

PHD

Plant homeodomain

TET2

Ten-eleven translocation 2

MMR

Mismatch repair

DSB

Double-strand DNA break

NHEJ

Non-homologous end joining

HR

Homologous recombination repair

HPV

Human papillomavirus

HNSCCs

Head and neck squamous cell carcinomas

AML

Acute myeloid leukemia

ccRCC

Clear cell renal cell carcinoma

MM

Multiple myeloma

NSCLC

Non-small cell lung cancer

NMC

NUT midline carcinoma