Histone lysine methyltransferases in biology and disease

Abstract

The precise temporal and spatial coordination of histone lysine methylation dynamics across the epigenome regulates virtually all DNA-templated processes. A large number of histone lysine methyltransferase (KMT) enzymes catalyze the various lysine methylation events decorating the core histone proteins. Mutations, genetic translocations and altered gene expression involving these KMTs are frequently observed in cancer, developmental disorders and other pathologies. Therapeutic compounds targeting specific KMTs are currently being tested in the clinic, although overall drug discovery in the field is relatively under-developed. Here we review the biochemical and biological activities of histone KMTs and their connections to human diseases, focusing on cancer. We also discuss the scientific and clinical challenges and opportunities in studying KMTs.

An important mechanism for regulating chromatin involves the reversible covalent modification of histones by chemical moieties such as methyl and acetyl groups. These different chemical ‘marks’ on histones are linked to discrete chromatin states and regulate the accessibility of DNA to trans-acting factors that mediate a wide variety of chromatin-templated processes such as transcription, DNA repair and DNA replication¹. Chemically, lysine methylation entails the addition of one, two or three methyl groups to the ε-nitrogen of a lysine side chain, forming mono-, di- and trimethylated derivatives (referred to here as Kme1, Kme2, and Kme3, respectively; Fig. 1a). This reaction, while only subtly changing the primary structure of the modified polypeptide, greatly increases the information encoded within the molecule, a feature highlighted by the unique activities frequently coupled to the specific extent of methylation. Methylation of lysines on histone and non-histone proteins is generated by protein lysine methyltransferases (KMTs; referred to as ‘writers’) and removed by protein lysine demethylases (KDMs; referred to as ‘erasers’) (Fig. 1a). In the human genome, there are predicted to be over 100 KMTs, and mass spectrometry–based studies suggest that more than 1,000 proteins in the human proteome harbor lysine methylation^2–4.

Fig. 1 |. Main sites of lysine methylation on mammalian histones and chromatin functions.

a, Chemical structures of methylated derivatives of lysine. Lysine residues can be monomethylated, dimethylated or trimethylated. b,c, Canonical (b) and non-canonical (c) lysine methylation marks on core nucleosomal histone H3 and H4 and their basic functions. Numbers adjacent to ‘K’ indicate the positions of the methylated lysines on histone H3 or histone H4. DNA is shown as black lines wrapped around blue histones. Key (right), chromatin-related functions associated with the methylation at left.

Lysine methylation was first described in 1959 on a bacterial flagellar protein⁵ and soon thereafter identified on histone proteins⁶. Indeed, the core histones contain numerous evolutionarily conserved lysine residues that are methylated in vivo. In humans, the canonical lysine methylation sites are found on histone H3 at lysine 4 (H3K4), lysine 9 (H3K9), lysine 27 (H3K27), lysine 36 (H3K36) and lysine 79 (H3K79), and on histone H4 at lysine 20 (H4K20). These modifications regulate an array of chromatin functions (Fig. 1b)¹. In addition to these canonical sites, there are several less well characterized sites of lysine methylation on the core histones (for example, H3K23me, H3K63me3, H45me1 and H4K12me1) (Fig. 1c)^4,7. Together, the substantial numbers of methylation sites and differentially methylated states present in histones illustrate the potential complexity that this signaling system can provide in the regulation of chromatin biology and how its deregulation can lead to disease.

In 2000, the discovery of SUV39H1, the first known histone KMT⁸, was a major breakthrough in the field that revealed a direct connection between histone methylation and a classic chromatin-mediated epigenetic phenomenon in flies known as position-effect variegation (PEV)⁸ (for a detailed review of PEV, see ref.⁹). Over the past two decades, the discovery and characterization of many additional histone KMTs has uncovered an elaborate network connecting chromatin regulation, epigenetic processes and human disease. In this context, the majority of research on lysine methylation has naturally focused on histone substrates and its role in chromatin and epigenetic regulation. One unintended consequence of this emphasis has been the emergence of biases in the initial characterization of the catalytic activities of orphan KMTs as histone-modifying enzymes. For example, the availability of reagents such as state-specific antibodies with which to study histone methylation, combined with the potential underappreciation of the limitations of these reagents, has led to the mischaracterization of some enzymes as histone KMTs² (discussed below). As the correct assignment of catalytic specificity for KMTs is crucial for understanding the role of chromatin in disease and for efforts to develop therapeutics, here we offer our perspective in classifying the reported histone KMTs as (1) bona fide histone-modifying enzymes, (2) enzymes that are referenced in the literature as histone KMTs but clearly are not, or (3) enzymes for which further work is necessary before any meaningful conclusions about catalytic activity and specificity can be drawn. Our rationale for making these distinctions, and their implications for disease etiology, are discussed below.

KMTs that catalyze canonical histone lysine methylation

In the human proteome, there are two domains with annotated lysine methyltransferase activity: the SET domain (named for three Drosophila melanogaster proteins originally recognized as containing the domain: Su(var)3–9, enhancer of zeste and trithorax) and the seven-beta-strand (7βS) domain (which is found on enzymes ranging from the histone KMT hDOT1L (Fig. 2a) to DNA methyltransferases)^1–3. In humans, there are 55 SET-domain-containing proteins. Of these, half are active KMTs (methylating histone and/or non-histone substrates), one protein (SETD3) is a histidine methyltransferase¹⁰, and the enzymatic activities of the remainder are unclear² (Fig. 2a,b and Table 1). The 7βS family is larger and more diverse than the SET family, with approximately 150–160 members in humans^3,11. Different 7βS-containing proteins methylate a wide range of substrates including lysine, arginine, other amino acid side chains, N-terminal α-amines, DNA, RNA and various metabolites³.

Fig. 2 |. Histone KMTs in the human proteome.

a, Human histone KMTs categorized by their established substrate specificity. b, Top: Examples of additional histone KMT activities. Bottom: Methylation is also detected at the non-canonical H3K18, K23, K56 and K64 sites, but the enzymes catalyzing these events are not known. c, Top two rows: generation of H3K36 trimethylation is not dependent on existing dimethylation. Bottom row: generation of H4K20me2 and H4K20me3 is dependent upon SETD8-generated H4K20me1.

Table 1 |.

Putative KMTs reported in the literature to be specific histone-modifying enzymes compared to their actual activity on histones and other substrates

Putative KMT	Histone physiological substrate	Physiological substrate(s)
SETDB221	?	?
SETMAR19,26	?/–*	?
SETD310,30	–*	Actin-H73
SETD522,24	?	?
MLL522	–	–
SMYD128	?	?
SMYD227,94	–*	p53, MAPKAP3,
SMYD320,95,96	H4K5me	MAP3K2, VEGFR1
SMYD529	H4K20me3?	?
PRDM131	H3K9me?	?
PRDM231	H3K9me?	?
PRDM397	H3K9me?	?
PRDM831	–	?
PRDM1697,98	H3K4me?, H3K9me?	?

^?, unknown;

^–, no methylation activity;

^*, no histone methylation activity on nucleosomes; histone site followed by “?”, more evidence required to determine whether the reported activity is reproducible.

The canonical histone lysine methylation marks found in humans are shown in Fig. 1b. These various modifications are generated in a context-dependent manner by a total of 24 different enzymes: 23 different SET proteins and one 7βS protein (Fig. 2a). In general, histone KMTs are highly selective: that is, the enzymes that methylate H3K36 do not methylate a different lysine if K36 is mutated. One exception is the meiotic recombination factor PRDM9, which trimethylates H3K4 in vivo in meiotic cells, but in vitro also methylates H3K9 and H3K36 (Fig. 2a,b). The physiological importance of the H3K9 and H3K36 activities of PRDM9 remains to be determined. The 7βS protein hDOT1L is the only enzyme in the human proteome that generates H3K79me, one of the few histone modifications found within the globular region of the nucleosome (Fig. 1b). In contrast, multiple enzymes mediate methylation events at H3K4, H3K9 and H3K36 (Fig. 2a). This enzymatic redundancy is used for targeting specific activities in a context-dependent manner such as differential genomic localization (such as methylation at an enhancer versus promoter region) and for the selective generation of different methylation states (such as me2 versus me3).

For example, H3K36me2 is generated by four related enzymes (NSD1, NSD2, NSD3 and ASH1L), whereas SETD2 is the only enzyme in somatic cells that synthesizes H3K36me3 (Fig. 2C). Notably, the generation of H3K36me3 by SETD2 is not dependent on the presence of H3K36me2^12,13; i.e., the initial recognition of the nucleosome as a substrate by SETD2 is far more efficient on unmethylated H3K36 than H3K36me2 (Fig. 2c). At H4K20, the monomethylated state is generated solely by SETD8, and the higher methylation states are synthesized by the KMTs SUV420-H1 and SUV420-H2 (Fig. 2c)¹⁴. However, unlike SETD2, SUV420-H1 and SUV420-H2 prefer a methylated substrate (H4K20me1) to unmethylated H4K20^14,15 (Fig. 2c). As a consequence, deletion of SETD8 leads to loss of all H4K20 methylation states even though SETD8 generates only the monomethyl species^14–16.

Notably, SETD8 and several other KMTs that methylate histones also modify non-histone substrates². For instance, SETD7, G9A, GLP and SETD8 methylate p53 (as well as other non-histone substrates)². In this context, knockout of Setd8 in Drosophila is lethal, whereas flies harboring a substitution of H4K20A, which prevents methylation of this residue, have a substantial delay in development but are otherwise normal¹⁷. The more severe phenotype that results from the Setd8 deletion versus the H4K20A mutation argues for physiologically important roles of SETD8 outside of H4K20 methylation. Thus, for select histone KMTs, their ability to methylate non-histone substrates must be taken into account in evaluating potential inhibitory compounds as candidate therapeutics.

Considerable efforts have been made to develop small-molecule inhibitors of different histone KMTs as tool compounds and for therapeutic purposes¹⁸. At present, active clinical trials (phase 1 and 2) are focused on several inhibitors of EZH2 (the main H3K27 KMT) and one inhibitor of the essential EZH2 cofactor EED; these compounds are being evaluated for efficacy in the treatment of a wide range of adult and pediatric neoplasm types (for example, ClinicalTrials.gov identifiers , , and ). Patients enrolled in the EZH2/EED inhibitor trials have tumors that share a common molecular signature: they either are positive for EZH2 gain-of-function mutations or harbor loss-of-function mutations in other chromatin-regulatory factors that are predicted to create cellular dependency on EZH2 activity. Beyond EZH2, a clinical compound targeting hDOT1L was evaluated in a phase 1 trial that was completed in 2016 (ClinicalTrials.gov identifier ), but as of this writing a phase 2 trial has not commenced. Tool and preclinical compounds also exist for several other histone KMTs (for example, SETD8 and G9A)¹⁸, arguing that KMTs, as an enzyme class, are druggable. However, several obstacles need to be overcome in developing drugs against some of the more promising KMT targets, including the lack of structural information about the enzymes, the need to use nucleosomes as substrates for in vitro drug screening, and the still limited, although growing, understanding of the types of compounds best suited to engage KMTs.

Mistaken identity: not all KMTs methylate histones

In addition to the enzymes listed in Fig. 2a,b, the following candidate KMTs—many of which are linked to human disease—have been reported to generate at least one canonical histone methylation mark (the putative modified residue(s) is (are) provided in parentheses after each enzyme symbol): MLL5 (H3K4), SETD3 (H3K4, H3K36), SETD5 (H3K9), SETDB2 (H3K9), SMYD1 (H3K4), SMYD2 (H3K36), SMYD3 (H3K4), SMYD5 (H4K20), SETMAR (H3K36), PRDM1 (H3K9), PRDM2 (H3K9), PRDM3 (H3K9), PRDM8 (H3K9) and PRDM16 (H3K4, H3K9) (Fig. 2a). Of these enzymes, the specific canonical histone methylation activities reported for MLL5, SETD3, SETDB2, SMYD3 and SETMAR have been independently tested and not reproduced^10,19–22; the original report on MLL5 was retracted²³. Moreover, biophysical and biochemical analyses of MLL5 indicate that it is not an active enzyme²². SETD5, an important protein etiologically linked to intellectual disability disorders^24,25, is similar in structure to MLL5 and therefore is not likely to be an active enzyme²². We recently demonstrated that SETD3 is a highly selective histidine methyltransferase and that it has no detectable activity on nucleosomes¹⁰ (Fig. 2d). SETDB2, given its sequence similarity to SETDB1, is assumed to be an H3K9 methyltransferase; however, to date, no activity for SETDB2 has been rigorously identified²¹. SMYD3 is a largely cytoplasmic protein that methylates non-histone substrates such as MAP3K2 and does not methylate H3K4 on free histones or on nucleosomes²⁰. SETMAR is a DNA-repair protein that consists of a fusion between a SET domain and a DNA transposase domain^19,26. In vitro, SETMAR methylates free H3 and H2B but has no activity on nucleosomes, and its activity on free H3 does not target K36, as determined by tandem mass spectrometry¹⁹. Thus, the physiological substrate of SETMAR and its potential role in DNA repair remains to be elucidated. SMYD2 is a relatively promiscuous enzyme (as far as KMTs go) and methylates many substrates. However, it has no activity on nucleosomes and lacks specificity on free histones, in contrast to its interaction with p53, one of SMYD2’s better-characterized substrates, where it shows a highly selective activity²⁷. Zebrafish SMYD1 has activity toward histones²⁸, but methylation of histones has not been demonstrated for the human homolog. There is one report showing that SMYD5 has H4K20 trimethylation activity²⁹. However, deletion of both Suv420-H1/2 in mice eliminates H4K20me3¹⁴, leaving the status of SMYD5 as a bona fide H4K20-modifying enzyme unresolved. Finally, several conflicting reports have suggested that PRDM1, PRDM2, PRDM3, PRDM8 and PRDM16 methylate H3K9 or H3K4, but other researchers have been unable to reproduce such activities, and thus more definitive work is required to determine whether these biologically important proteins are truly active enzymes, and if they are, to identify their physiological substrates² (Fig. 2d).

Why have these enzymes been potentially mischaracterized? As mentioned above, in some cases, interpretation of data relying solely on state-specific antibodies can be misleading, particularly in the absence of tandem mass spectrometry studies as an independent approach to confirm the specific methylation event. In addition, although technically challenging, the use of nucleosomes as substrates in addition to histone peptides can provide important information about whether a putative histone methylation activity is likely to be physiologically relevant. Finally, some studies have relied on mass spectrometry data in which the mass shifts attributed to methylation reactions are the wrong molecular weight for methylation, raising doubt about the studies’ conclusions (for example, ref.³⁰). Taken together, these complications emphasize that it is important that those undertaking any research, including drug-development efforts, focused on these potentially mischaracterized KMTs—many of which have clear links to disease (for example, PRDM1 and PRDM2 are important tumor suppressors³¹, PRDM16 is a key regulator of adipogenesis³¹ and SETDB2 is a regulator of fibrotic diseases²¹)—consider starting with a rigorous and unbiased analysis of their enzymatic activities.

H3K36-specific lysine methyltransferases in cancer

The established link between histone lysine methylation dynamics, gene expression regulation and oncogenic programming provides a paradigm for the way that pathological alterations of histone KMTs can promote the development and progression of diverse cancers (Fig. 3). The findings in this field are vast, and many excellent and comprehensive reviews on the topic are available for the interested reader^1,32,33. Here we focus on the pathological roles of the enzymes that either dimethylate or trimethylate H3K36 (see Fig. 2c) as model histone KMTs and discuss examples of crosstalk between H3K27 and H3K36 methylation in epigenetic-mediated oncogenic programming.

Fig. 3 |. Spectrum of cancers associated with H3K36 methyltransferases.

The potential tumor-suppressive functions listed are informed by the identification of recurrent deletions, frameshifts, or truncating or damaging missense mutations, and by biological studies, including mouse models. Potential oncogenic functions are informed by overexpression, focal amplifications, gain of function or identification of a fusion oncogene, and by biological studies, including mouse models. CNS, central nervous system; HSTL, hepatosplenic T cell lymphoma; EATL-II, enteropathy-associated T cell lymphoma, type II; LSCC, lung squamous-cell carcinoma.

The state of methylation at H3K36 defines distinct biological outcomes, and mutations in the H3K36 KMTs are linked to a variety of developmental disorders and cancer (Fig. 3 and Table 2). SETD2, which synthesizes H3K36me3 in humans, regulates DNA methylation, RNA processing, DNA repair and tumor suppression^13,34–37. In contrast, it is less clear what specific molecular functions are associated with H3K36me2, although this modification has been linked to DNA methylation, gene activation and cellular transformation^12,33,38,39. There are four enzymes that generate H3K36me2: NSD1, NSD2, NSD3 and ASH1L (Fig. 2c). The enzyme(s) responsible for generating H3K36me1 is (are) unknown, and a cellular function for H3K36me1 is not clear at present, although it is likely that the mark itself is synthesized through the combined actions of KMTs and KDMs.

Table 2 |.

Histone KMTs from Fig. 2 in human disease, observed murine phenotypes, and putative roles in cancer

	Human disease or syndrome	Knockout mouse phenotype(s)	Putative roles in cancer
H3K4
SETD7		−/−: no identifiable phenotype99
PRDM9		−/−: infertile; involved in speciation100	Overexpressed in specific cancers
SETD1A	Schizophrenia101	−/−: lethal
		Conditional: impaired B-cell development102	Oncogenic/tumor suppressive
SETD1B	12q24.31 microdeletion103	−/−: lethal, hematopoietic defects104	Oncogenic/tumor suppressive
MLL1–MLL4	MLL1: Wiedmann-Steiner105	−/−: lethal106–109	Mll fusions oncogenic
		Conditional: pleiotropic, e.g., impaired hematopoiesis in MLL1111
	MLL2: dystonia110		Wild-type Mll
	MLL3: Kleefstra-2112		Tumor suppressive
	MLL4: Kabuki113
H3K9
G9A		−/−: lethal114	Oncogenic
		Conditional: impaired learning and memory115
GLP	Kleefstra-1116	−/−: lethal117	Oncogenic
		Conditional: impaired learning, adipogeneis115,118
SUV39H1, SUV39H2		−/−: no identifiable phenotype
		dKO: increased tumor risk, impaired fertility114
SETDB1		−/−: lethal; defects in neural development119	Oncogenic
		Conditional: impaired spermiogenesis, oogenesis120
H3K27
EZH1		−/−: no identifiable phenotype121
		Conditional: impaired hematopoeisis122
EZH2	Weaver89	−/−: lethal123	Oncogenic/tumor suppressive
		Conditional: broadly impaired development
H3K36
NSD1	Sotos88, Beckwith-Wiedmann125	Conditional: lethal124	Oncogenic/tumor suppressive
NSD2	Wolf-Hirschhorn126–128	−/−: lethality shortly after birth	Oncogenic
		−/+: WHS-like defects129
		Conditional: impaired B-Cell development130
NSD3			Oncogenic
ASH1L	Intellectual disability131	−/−: lethal132	Oncogenic
		Conditional: impaired hematopoesis133
SETD2	Luscan-Lumish134,135	−/−: lethal, vascular defects136	Tumor suppressive
		Conditional: impaired osteogenesis137,
		myogenesis138, germ cell development62,139, hematopoesis140
H3K79
DOT1L		−/−: lethal: impaired cardiac development141	Oncogenic
		Conditional: impaired hematopoesis142,143
H4K20
SETD8		−/−: embryonic lethal144
SUV420H1	Intellectual disability145	−/−: lethal shortly after birth; short stature14
SUV420H2		−/−: no apparent defects
H4K12
KMT9			Oncogenic

dKO, double knockout; WHS, Wolf-Hirschhorn syndrome.

The initial evidence for a potential tumor-suppressive role of SETD2, and by proxy H3K36me3, came from sequencing studies of renal-cell carcinoma (RCC). These studies found recurrent biallelic loss of SETD2, a classic hallmark of known tumor-suppressor genes^40,41. Subsequent sequencing studies have identified recurrent SETD2 mutations across a broad spectrum of human malignancies, including lung adenocarcinoma (LUAC)⁴², multiple types of leukemia and other hematological malignancies^43–48, central nervous system tumors⁴⁹, bladder cancer⁵⁰ and gastrointestinal tumors⁵¹. The remarkably high frequency of SETD2 mutations across a wide variety of cancer types is reminiscent of other classical tumor suppressors and suggests a broad, general role in prevention of cancer.

Further sequencing efforts in RCC found that inactivating mutations in SETD2 occurred in a subclonal fraction of the tumor, arguing against a role for SETD2 loss in tumor initiation. However, the same studies identified individual tumors with distinct inactivating mutations in SETD2⁵², demonstrating parallel evolution toward loss of SETD2. These results suggest that SETD2 loss plays a role in cancer progression. Indeed, biallelic loss of SETD2 in patients with RCC is unfortunately associated with significantly lower survival rates⁵³. Furthermore, although SETD2 is lost in only a fraction of all RCCs (11.3%), its loss is far more prevalent in more aggressive subtypes of RCC (63%)⁵⁴. The role of SETD2 in cancer progression extends to high-risk gastrointestinal stromal tumors⁵¹. Finally, patients with acute lymphocytic leukemia (ALL) who unfortunately relapsed after chemotherapy frequently acquire loss-of-function mutations in SETD2⁵⁵.

More recent analyses of multiple cancer subtypes have provided additional evidence that loss of SETD2 drives tumor progression. In comprehensive studies of acute myeloid leukemia (AML), ALL and LUAC, SETD2 was mutated at a much higher frequency in tumors driven by fusion oncogenes⁵⁶. Specifically, mutations in SETD2 were detected in 22.6% of patients with MLL-rearranged leukemia but in only 4.6% of patients without the oncogenic fusion. Similarly, in LUAC, SETD2 mutations were found in 18% of patients with cancers driven by fusion oncogenes, compared with 9% of those lacking oncogenic fusions. Other studies have observed co-occurrence of SETD2 loss and silencing of CDKN2A⁴², as well as Ras-activating mutations⁵⁵. This co-occurrence suggests that the requirement for SETD2 in tumor suppression may be enhanced in specific contexts.

In vivo screens for tumor suppressors have identified SETD2 as a top candidate in multiple cancer models, including ALL⁵⁷, hepatocellular carcinoma⁵⁸ and gastrointestinal cancer⁵⁹. Moreover, recent work using multiplexed in vivo CRISPR-based genome editing to knock out numerous known and putative tumor-suppressor genes in a Kras-driven mouse model of LUAC demonstrated that Setd2 depletion dramatically increased tumor size⁶⁰. Indeed, loss of Setd2 resulted in the largest tumors observed in the study, surpassed only by tumors harboring p53 inactivation. These results are consistent with earlier work demonstrating a role for SETD2 in suppressing Kras-driven LUAC in mouse models⁶¹. Collectively, these studies provide compelling experimental evidence for the tumor-suppressor function of SETD2, in accordance with the numerous SETD2 mutations identified in human tumors (Fig. 3). However, further work is needed to determine the mechanism(s) of tumor suppression by SETD2, the relationship to H3K36me3 catalysis and the effects of genetic context.

At the molecular level, the ability of SETD2 to regulate several fundamental biological processes is directly linked to selective recognition of H3K36me3 by methyl-lysine reader domains. Baubec et al. demonstrated that crosstalk between DNA methylation and histone methylation is mediated by recognition of H3K36me3 by the DNMT3B PWWP domain³⁷. This interaction facilitates the targeting of DNMT3B to the bodies of transcribed genes, which are enriched for H3K36me3. This leads to focal de novo DNA methylation at these genomic regions³⁷, which may influence the expression of nearby genes. Indeed, SETD2 loss in mouse oocytes causes defects in DNA methylation, genomic imprinting and development⁶². SETD2-mediated regulation of transcription and pre-mRNA splicing is also mediated by an H3K36me3-selective reader domain^35,36. The tandem bromo-PWWP domains on the nuclear factor BS69 (also known as ZMYND11) recognizes K36me3, but only in the context of nucleosomes containing the histone variant H3.3³⁶. The binding of BS69/ZMYND11 to H3.3K36me3-enriched chromatin recruits BS69/ZMYND11 and its associated proteins, which include RNA splicing and transcription factors^35,36. Connections between DNA-repair mechanisms and SETD2 are also mediated by reader domains. For example, recognition of H3K36me3 by the PWWP-domain-containing factor MSH6 facilitates the association of the mismatch-recognition complex to facilitate DNA repair⁶³. Collectively, the discovery of selective H3K36me3-reader domains has provided crucial insight into the molecular mechanisms of action by which SETD2 regulates biology and how these functions may influence oncogenesis. For example, altered gene expression and/or compromised DNA repair in a SETD2-deficient setting could promote cellular transformation. Finally, two interesting studies recently reported that SETD2 directly methylates tubulin and STAT1^64,65. The strong substrate preference of SETD2 for intact nucleosomes suggests that it recognizes a specific three-dimensional topology during catalysis. Therefore, the molecular basis for the recognition of such disparate substrates (nucleosomal H3K36, tubulin and STAT1) and the relative contributions of H3K36me3 versus the non-histone substrates in tumor suppression are important questions to address in the future.

Whereas H3K36me3 is generated exclusively by SETD2, the biosynthesis of H3K36me2 is more complex. There are four related enzymes, NSD1, NSD2, NSD3 and ASH1L, that can generate H3K36me2 on nucleosomes in vitro (Fig. 2c)⁶⁶. In most cell types, including various cancer cell lines, NSD2 is responsible for generating the bulk of H3K36me2¹². In specific cellular contexts NSD1 replaces NSD2 as the enzyme required to generate global H3K36me2⁶⁷. ASH1L does not globally regulate cellular H3K36me2 levels; instead, its activity is localized to specific genes⁶⁸. The physiological role of NSD3 is not clear.

In contrast to SETD2 and its role as a tumor suppressor, all four of the H3K36me2-specific KMTs are thought to promote oncogenesis (Fig. 3). One clear example is the role of NSD2 in the pathogenesis of multiple myeloma (MM)⁶⁹. MM is an incurable blood malignancy that effects hundreds of thousands of people throughout the world^70,71. Among patients with MM, 15–20% carry a t(4;14) translocation, which places the transcription of NSD2 under the control of a strong IgH intronic enhancer and leads to aberrant, massive upregulation of NSD2 that is thought to drive cancer development^72–74. Consistent with this, NSD2 expression in MM cells drives xenograft tumor formation and tumor invasion in mice in a manner that depends on the catalytic activity of NSD2³⁹. Beyond MM, NSD2 overexpression is broadly found in diverse cancers³⁹ and drives metastatic progression in prostate cancer⁷⁵. Consistent with this expression profile, NSD2 depletion in multiple cancer cell lines results in decreased cellular proliferation^12,76.

In addition to the t(4;14) translocation and general overexpression of NSD2, a recurrent heterozygous gain-of-function NSD2 (E1099K) variant is found in ~10% of cases of childhood ALL with a precursor-B phenotype^77,78. ALL is the most common cancer diagnosed in children, representing more than a quarter of all pediatric neoplasms⁷⁹. E1099 is found within the catalytic SET domain of NSD2, and the E1099K substitution confers a roughly 1.5-fold increase in NSD2 catalytic efficiency through a mechanism that is presently unknown. Expression of NSD2-E1099K in cells leads to elevated H3K36me2 levels, which causes a decrease in H3K27me3 levels due to the direct inhibition of EZH2 by H3K36me2^70,71,80. In this context, the ability of NSD2-E1099K to drive pediatric ALL is postulated to be mediated in part via depletion of H3K27 methylation, which in turn leads to defects in epigenetic gene silencing and oncogenic reprograming (Fig. 4).

Fig. 4 |. Model for crosstalk between methylation at H3K27 and H3K36 in oncogenic programming.

Deregulation of the dynamic interplay between methylation at H3K27 and that at H3K36 leads to pathological transcriptional activation or repression and thereby promotes oncogenic reprogramming.

Notably, beyond pediatric ALL, the NSD2 E1099K mutation is found in other neoplasms, including several types of solid tumors, such as LUAC, colon cancer and thyroid tumors^71,73,81. Together, the many links between NSD2 alterations and different cancers indicate that the NSD2–H3K36me2 axis has a broad role in promoting tumorigenesis. However, it remains unclear whether H3K36me2 has direct effect on chromatin and gene regulation beyond the suppression of H3K27me3. That said, depletion of NSD2 and H3K36me2 in HT1080 cancer cells impairs cell proliferation, and this phenotype is independent of H3K27me3, because it is not rescued by EZH2 inhibition⁸². Moreover, the PWWP domain on NSD2 itself preferentially binds to H3K36me2, and it is postulated to be important for the propagation of NSD2-mediated H3K36me2 domains⁸³. This might indicate that other yet-to-be-discovered H3K36me2-specific reader domains exist that link this modification to cancer pathways.

Like NSD2, the proteins NSD1, NSD3 and ASH1L are linked to oncogenesis. In AML, the t(5;11) fusion of NUP98, a member of the nuclear pore complex, to NSD1 (NUP98–NSD1), is found in about 5% of AML cases and is associated with poor prognosis³⁹. In mouse adaptive-transfer experiments, Wang et al. showed that bone marrow progenitor cells ectopically expressing NUP98–NSD1 rapidly developed AML⁸⁴. Mechanistically, this transformation activity is mediated by the activation of HOX genes, important developmental genes that are frequently dysregulated in cancer, via H3K36 methylation and antagonism of EHZ2-mediated repression⁸⁴. Beyond the NUP98–NSD1 fusion, the role of NSD1 in cancer is complex. Nonsense mutations in NSD1 are observed in ~10% of distinct populations of patients with squamous cell carcinoma of the head and neck (SCCHN)⁶⁷ and with lower frequency in several other cancers⁸⁵. This suggests that loss of functional NSD1 promotes oncogenesis. On the other hand, patients with SCCHN who harbor NSD1 mutations have a favorable outcome and show a better response to chemotherapy. Thus, the role of NSD1 in cancer might be dependent on the tissue and etiological context as well as the mutational landscape of the disease.

The NSD3 gene is commonly amplified in breast cancer, lung squamous cell carcinoma and squamous cell carcinoma of the head and neck⁸¹. In addition, NSD3 is involved in rare translocations in patients suffering from acute myeloid leukemia (AML), with the fusion including the NSD3 SET domain⁸⁶. Furthermore, a rare fusion lacking the SET domain but including the BET-interaction domain of NSD3 is found in midline carcinoma^84,87. Despite the links between NSD3 and cancer, the physiological role of the catalytic activity of NSD3 and its relationship to tumorigenesis remains unclear.

Like NSD3, the gene ASH1L is amplified in various cancers including breast, uterine and pancreas⁸¹. Moreover, the H3K36 dimethylation activity of ASH1L promotes MLL-dependent leukemogenesis in both mouse models and human MLL-rearranged leukemic cells through the regulation of transcription at key leukemia-associated gene⁶⁸. Taken together, these data indicate that whereas the SETD2–H3K36me3 axis plays a role in suppressing tumorigenesis, the dimethyl state at H3K36 is generally associated with promoting oncogenesis.

Although either dimethylation or trimethylation at H3K36 antagonizes EZH2 and prevents H3K27 methylation, the distribution patterns of H3K36me2 and H3K36me3 across the genome are distinct. H3K36me2 is far more abundant than H3K36me3 and is both present in intergenic regions and enriched proximal to the transcriptional start sites of actively transcribed genes¹². In contrast, H3K36me3 is found largely within the bodies of actively transcribed genes, reaching its highest levels at the 3’ end. Overall, it is likely that their different chromatin distributions coupled with state-specific reader domains explain why H3K36me2 and H3K36me3 have divergent roles in cancer. Furthermore, because of the wider distribution of H3K36me2 across the genome and its overall greater abundance, H3K36me2 will naturally have a larger impact in antagonizing EZH2 than does SETD2-catalyzed H3K36me3. It appears contradictory that although EZH2 itself has oncogenic properties, counteracting its activity promotes H3K36me2-driven cancers. This, however, highlights an important concept in epigenetic mis-regulation: either gains or losses of histone methylation marks in a cell-context-dependent manner can select for gene expression programs that provide a fitness advantage or prevent differentiation, locking cells into a proliferative state that exists only transiently in normal development (Fig. 4).

Histone KMTs and developmental disorders

Haploinsufficiency of histone KMTs manifest in numerous developmental disorders (Table 2). Notably, there is striking similarity in the developmental phenotypes of patients with deficiencies in the main H3K27 methyltransferase complex and those with deficiencies in H3K36 methyltransferases (Table 2). Both Sotos syndrome and Weaver syndrome, largely characterized by mutations in NSD1 and EZH2, respectively, present with overgrowth and intellectual disability^88,89. Although these conditions are categorized under different names, a subset of patients with Weaver syndrome possess NSD1 mutations rather than EZH2 mutations⁹⁰. Furthermore, patients diagnosed with Sotos syndrome but lacking mutations in NSD1 have been found to have mutations in SETD2 and DNMT3A⁹¹. The remarkable phenotypic convergence observed in these patients may reflect underlying molecular relationships among the methylation of H3K27, H3K36 and DNA.

Non-canonical histone methylation sites in disease

Beyond the canonical sites, many other methylation events on histones (for example, H3K14me3, H3K56me1, H3K64me3, H4K12me1 and several others) have been identified by various methods, including mass spectrometry^4,7 (Fig. 1c). Interestingly, H3K14me3 is not normally found in human chromatin but is generated by the bacterial effector protein RomA in cells infected with Legionella pneumophila⁹². This mark is hypothesized to repress the expression of host genes encoding components of the innate immune system, which helps promote intracellular replication of the mycobacterium. Thus, a RomA inhibitor could function as an antibiotic to selectively treat legionella pneumonia⁹².

Recently, KMT9, a heterodimeric complex consisting of two 7βS enzymes (C21orf127, also known as HEMK2, N6AMT1 or PrmC; and TRMT112), was shown to monomethylate H4K12 in vitro on nucleosomes, and depletion of KMT9 in prostate cancer cells results in decreased endogenous H4K12me1 levels⁹³. Beyond DOT1L, KMT9 represents the only other 7βS enzyme known to date to have histone lysine methylation activity. In cells, H4K12me1 modification localizes to gene promoters, and depletion of this mark by knockdown of KMT9 reduces the expression of genes marked with H4K12me1, suggestive of a role for the KMT9–H4K12me1 axis in transcription initiation⁹³. Notably, the levels of KMT9 and H4K12me1 are specifically elevated in malignant prostate cancer. Furthermore, depletion of KMT9 impairs cell proliferation and xenograft tumor growth of androgen-independent prostate cancer, but not the growth of several other cell types. Interestingly, the heterodimeric KMT9 complex also functions as a protein gluta-mine methyltransferase, but Metzger et al. have identified KMT9 mutants that separate the two enzymatic functions to demonstrate H4K12me1 synthesis as the relevant activity in prostate cancer⁹³. This study suggests that targeting of a non-canonical histone mark, H4K12me1—through inhibition of KMT9—may offer a new strategy for the treatment of lethal castration-resistant prostate cancer.

Outlook

Over the last several decades, fueled by discoveries based on the integration of diverse methods, the scientific community has developed an understanding of the fundamental role of histone lysine methylation in the regulation of chromatin biology and of how this complex signaling system affects human disease. Drugs targeting EZH2, the main H3K27 KMT, are being tested as precision medicines that will hopefully soon be available in clinical settings to help patients. Over the next several years, we anticipate that new ways to chemically or biologically modulate other histone KMTs, such as NSD2, will be realized and may offer therapeutic benefit in the treatment of cancers and other human pathologies.