Nucleosomes, the main components of chromatin, consist of histones, and histone proteins have positively charged amino-terminal tails that are exposed on the outside of nucleosomes. These tails are subject to several post-translational covalent modifications, including acetylation, phosphorylation, ubiquitination, sumoylation and methylation (reviewed in 1). Methylation been found on a range of lysine residues in various histones: K4 (using the single-letter amino-acid code for lysine), K9, K27, K36 and K79 in histone H3, K20 in histone H4, K59 in the globular domain of histone H4 2 and K26 of histone H1B 3. Several proteins responsible for the methylation of specific residues have been characterized, and all but one of these contains a SET domain; they make up the SET-domain protein methyltransferase family (Table 1). The exception to the rule is the DOT1 family, members of which methylate K79 in the globular region of histone H3 and which are structurally not related to SET-domain proteins 456. Recent work suggests that SET-domain-containing proteins methylate a few proteins in addition to histones (see later); they should therefore be named protein lysine methyltransferases rather than histone lysine methyltransferases. The function of SET-domain proteins is to transfer a methyl group from S-adenosyl-L-methionine (AdoMet) to the amino group of a lysine residue on the histone or other protein, leaving a methylated lysine residue and the cofactor byproduct S-adenosyl-L-homocysteine (AdoHcy). Methylation of specific histone lysine residues serves as a post-translational epigenetic modification that controls the expression of genes by serving as 'markers' for the recruitment of particular complexes that direct the organization of chromatin.

The SET domain (Figure 1) was first recognized as a conserved sequence in three Drosophila melanogaster proteins: a modifier of position-effect variegation, Suppressor of variegation 3-9 (Su(var)3-9) 7, the Polycomb-group chromatin regulator Enhancer of zeste (E(z)) 8, and the trithorax-group chromatin regulator trithorax (Trx) 9. The domain, which is approximately 130 amino acids long, was characterized in 1998 10 and SET-domain proteins have now been found in all eukaryotic organisms studied. There are currently 157 entries for human SET-domain proteins in the SMART database 11 and 93 entries in the Pfam database 12, although both databases contain duplicate entries. Seven main families of SET-domain proteins are known - the SUV39, SET1, SET2, EZ, RIZ, SMYD, and SUV4-20 families - as well as a few orphan members such as SET7/9 and SET8 (also called PR-SET7; see Table 2 for a list of the members of each family in humans and their properties). Proteins within each family have similar sequence motifs surrounding the SET domain, and they often also share a higher level of similarity in the SET domain.

The SUV39 family has been characterized in most detail. Members of this family - human SUV39H1, murine Suv39h2, and Schizosaccharomyces pombe Cryptic loci regulator 4 (CLR4) - were the first SET-domain protein lysine methyltransferases to be characterized, following the discovery of sequence homology between their SET domains 13. These proteins, with other members of the family such as D. melanogaster Su(var)3-9, specifically methylate lysine 9 of histone H3 (H3 K9) 13. Human SUV39H1 and its closely related paralog, SUV39H2, are 55% identical at the amino-acid level. The structures of the genes encoding the two proteins are shown in Figure 2a,b: both have six exons, and they have identical intron-exon junctions. It appears that SUV39H1 and SUV39H2 resulted from a recent gene-duplication event, as only mammals have two copies. The SUV39-family proteins of the frog Xenopus laevis (GenBank accession number AAH70805) and the zebrafish Danio rerio (AAH76417) are more closely related to human SUV39H1 than to SUV39H2; they share 75% and 63% amino-acid identity with SUV39H1, respectively. The zebrafish gene also shares all intron-exon junctions with human, and the frog gene shares at least four of the junctions. The D. melanogaster Su(var)3-9 protein is only 30% identical to human SUV39H1, and their genes share no intron-exon junctions. The S. pombe member of the SUV39 family, CLR4, is only 27% identical to the human protein and its gene contains no introns.

The members of the SUV39 family discussed above are involved in both euchromatin and heterochromatin, but another member of the same family, G9a, is the predominant histone H3 K9 methyltransferase in mammalian euchromatin 14. There are two isoforms of G9a in the mouse: the short form (GenBank accession number NP_671493) corresponds to human G9a and the long form (NP_665829), which lacks intron one, has additional Arg-Gly repeats at the amino terminus. No human expressed sequence tag (EST) corresponding to the long form of G9a has yet been isolated, although the sequence is present in the genome. Similar to the situation with SUV39H1, G9a also has a closely related paralog in mammals, Gga-like-protein-1 (GLP1). The human G9a gene has 28 exons and is about 17.3 kilobases (kb) long (Figure 2c). GLP1 is 45% identical to G9a and most of the divergence is in the amino-terminal third of the protein. The GLP1 gene has 25 exons - it lacks homologs of the first three introns of G9a- and the 20 exons from the 3' end have identical junctions to those found in G9a. The GLP gene is quite large, 120 kb in human and 92 kb in mouse, with introns as long as 16 kb (Figure 2d). No obvious orthologs of G9a or GLP can be found in the worm, frog or yeast genomes; in the D. melanogaster genome there is one gene (CAB65850) encoding a protein that is distantly related to human G9a (20% identity) or GLP (18% identity) in the carboxy-terminal half of the protein. The chicken genome also encodes one protein (CAH65313) that shares 75% identity with human GLP. Interestingly, both a frog (Xenopus tropicalis) and three species of fish (D. rerio, Tetraodon nigroviridis, and Takifugu rubripes) have both G9a and GLP in their genomes, although most have not yet been annotated as such. The zebrafish GLP ortholog (CAE49087) is 45% identical to human GLP, and the gene shares all but three of its 23 intron-exon junctions with human GLP1.

G9a and SUV39H1 both belong to the same family of SET-domain proteins and both have pre-SET and post-SET domains surrounding the SET domain, but they do not share any intron-exon junctions, even though a number of these junctions occur within the highly conserved SET domain. The other two SUV39 family proteins, ESET (also called SETDB1) and CLLL8 (SETDB2) also have significant similarities in their genomic structures with each other but not with G9a or SUV39H1 (data not shown). Several proteins in other SET families are also found in closely related pairs: EZH1 and EZH2 (members of the EZ family), MLL1 (also called HRX) and MLL2 (HRX2, both members of the SET1 family), SET1 and SET1L (SET1 family), NSD2 (WHSC1) and NSD3 (WHSC1L1; SET2 family), and SUV4-20H1 and SUV4-20H2 (SUV4-20 family).

The structures of SET-domain proteins that are currently known include: the crystal structures of two members of the SUV39 family, Neurospora crassa Decrease in DNA methylation 5 (DIM-5) 1516 and S. pombe CLR4 17; four structures of human SET7/9 in various configurations 18192021; a nuclear magnetic resonance (NMR) structure of a viral protein that contains only the SET domain 22; and a structure of the non-histone protein methyltransferase Rubisco LSMT, an unclassified member of the superfamily 2324. These structures revealed that the SET domain forms a novel β fold not seen in any other previously characterized AdoMet-dependent methyltransferases (reviewed in 25). The fold has a series of curved β strands forming several small sheets, packed together with pre-SET (or N-SET) and post-SET (or C-SET) domains or regions (Figure 3). The pre-SET domain of SUV39-family proteins (see Table 2) contains nine invariant cysteine residues that are grouped into two segments of five and four cysteines separated by various numbers of amino acids (CXCX₅CX₄CXC-X_N-CX₃CXCX₃C, where N is 46 in DIM-5 and 28 in CLR4). The nine cysteines of the pre-SET domain of DIM-5 coordinate three zinc ions to form an equilateral triangular cluster, Zn₃Cys₉ (Figure 3a). The SET domain, which may have evolved through the duplication of a three-stranded unit 26, is folded in all the solved structures into several small β sheets surrounding a knot-like structure by threading of the carboxyl terminus through an opening of a short loop formed by a preceding stretch of the sequence (Figure 3). This remarkable 'pseudoknot' fold brings together the two most-conserved sequence motifs of the SET domain (RFINHXCXPN and ELXFDY; see Figure 1) to form an active site in a location immediately next to the pocket where the methyl donor binds and to the peptide-binding cleft.

The post-SET region of DIM-5 contains three conserved cysteine residues, arranged CXCX₄C, that are essential for its histone lysine methyltransferase activity 15. The structure of DIM-5 in a ternary complex with an H3 K9 peptide and AdoHcy 16 reveals that, as expected from their arrangement, these three post-SET-domain cysteines coordinate a zinc ion tetrahedrally together with cysteine 244 of the SET-domain signature motif RFINHXCXPN in the pseudoknot near the active site (Figure 3a). Consequently, a narrow channel is formed to accommodate the side chain of the target lysine. Three ternary structures - SET7/9 in complex with a peptide containing histone H3 K4 21, DIM-5 in complex with a histone H3 K9 peptide 16, and Rubisco LSMT in complex with a free lysine 24 - reveal that the target lysine is inserted into a narrow channel so that the target nitrogen would be in close proximity to the methyl donor AdoMet at the opposite end of the channel.

Close examination of the region carboxy-terminal to the SET domain in many proteins, including members of the SUV39, SET1, and SET2 families, suggests that the post-SET-domain metal center observed in DIM-5 is universal among all those members of the superfamily that have the cysteine-rich post-SET domain. For almost all SET-domain proteins, there appears to be an absolute correlation between the presence of the post-SET domain and a cysteine corresponding to Cys244 of DIM-5 near the active site. Comparison of DIM-5 with SET7/9 1921 and the Rubisco LSMT 2324, two SET-domain proteins that do not have Cys-rich pre-SET and post-SET domains, reveals a remarkable example of convergent evolution. In particular, as in DIM-5, these two enzymes rely on residues carboxy-terminal to the SET domain for the formation of lysine channel, but they do so by packing of an α helix, rather than a metal center, onto the active site.

The lysine ε-amino groups of histones can be mono-, di-, or tri-methylated and, depending on the specific residue(s) modified, this methylation is associated with the formation of repressive heterochromatin, with transcriptional activation and elongation by RNA polymerase II or with the transcriptional silencing of euchromatic genes (Table 1). In heterochromatin, the H3 Kg methyl mark is recognized by heterochromatin protein 1 (HP1) in mammals and its homolog Swi6 in S. pombe; these proteins bind to the modified residue via their chromodomains, a domain shared by many regulators of chromatin structure 2728, resulting in the formation of transcriptionally silent heterochromatin. Methylation of H3 K9 and H3 K27 is also associated with transcriptional silencing in euchromatin 29. In contrast, di- and tri-methylation of Lys4 on the same histone (H3 K4) is associated with active transcription 30.

The Saccharomyces cerevisiae SET1 and SET2 complexes are involved in transcriptional elongation as part of the RNA polymerase II holoenzyme 3132. Tri-methylation of H3 K4 by SET1 is associated with regions of each gene that are transcribed early, in contrast to the SET2-mediated methylation of H3 K36, which is associated with downstream regions that are transcribed in the later stages of transcriptional elongation. The mammalian nuclear-receptor-binding SET-domain-containing protein (NSD1, a member of the SET2 family) has been found to play a crucial role in post-implantation development, methylating H3 K36 and H4 K20 33. ESET (also called SETDB1), which predominantly methylates H3 K9 in transcriptionally silent euchromatin, is also required for peri-implantation development 34. ESET has been reported to bind the co-repressor KAP-1, which acts as a molecular scaffold, targeting ESET and HP1 to euchromatic genes silenced by KRAB-domain zinc-finger proteins 35.

Human SET7/9 mono-methylates H3 K4 1621, whereas S. cerevisiae SET1 di- or tri-methylates the same residue 36. The strict lysine specificity of these enzymes is in distinct contrast to Drosophila ASH1 (a member of the SET1 family), mammalian G9a, human EZH1 and EZH2 and mouse NSD1, enzymes that can methylate two or more different lysine residues (Table 1). In some cases, the functions of SET-domain enzymes are not confined to histone methylation. For instance, human SET7/9 has recently been reported to methylate Lys189 in the general transcription factor TAF10, resulting in an increased affinity for RNA polymerase II and transcriptional activation of certain TAF10-dependent genes 37. SET7/9 has also been reported to methylate p53, increasing the stability of this short lived tumor-suppressor protein 38. These observations suggest that we should not narrowly define the SET-domain proteins as histone lysine methyltransferases but instead call them protein lysine methyltransferases.

DIM-5 tri-methylates H3 K9 16, and this marks chromatin regions for DNA methylation 39. Other members of the SUV39 family - KRYPTONITE of Arabidopsis thaliana 4041, Suv39h1 of mouse 42, and mammalian G9a 43 - have been implicated in DNA methylation. In contrast to the tri-methylation of H3 K9 by which DIM-5 marks regions for DNA methylation 39, the critical mark for DNA methylation by KRYPTONITE is di-methylation of H3 K9 40.

As the function of SET-domain proteins become clearer, it is apparent that they can also be perturbed in disease. The recent recognition of the role played by the SET-domain protein SMYD3 in the proliferation of colorectal and hepatocellular carcinomas 44 may pave the way for the development of specific inhibitors of SMYD3 activity in cancer treatment. SMYD3 expression is upregulated in these cancers, and its histone H3 K4 methyltransferase activity activates oncogenes and other genes associated with the cell cycle. MLL1, the human homolog of Drosophila Trx and a member of the SET1 family, is often implicated in leukemia as a result of aberrant Hox gene activation mediated by histone H3 K4 methylation 45. Moreover, EZH2 is involved in metastatic prostate and breast cancer 4647, suggesting that identification and targeted inhibition of SET-domain proteins involved in cancer may be useful for the treatment of patients in the future. The recent identification and characterization of two new SET-domain methyltransferases, SUV4-20H1 and SUV4-20H2, which function in H4 K20 tri-methylation 48, suggests that more SET-domain methyltransferases await characterization. In addition, the previously uncharacterized S. pombe SET9 protein has recently been shown to be able to methylate H4 K20 49. This modification does not seem to play a role in controlling gene expression or heterochromatin formation, but rather appears to be responsible for the recruitment of the checkpoint protein Crb2 to sites of DNA damage, unveiling yet another role for SET-domain proteins. The recent identification of the first histone demethylase, LSD1, which is conserved from S. pombe to humans, reveals that regulation of histone methylation is even more dynamic than was thought 50. More recently, evidence has been provided that a cytosolic EZH2-associated methyltransferase complex regulates actin polymerization in various cell types, suggesting that SET-domain proteins may have many different roles in the cell 51.