In Vitro and in Vivo Analyses of a Phe / Tyr Switch Controlling Product Specificity of Histone Lysine Methyltransferases

The functional significance of mono-, di-, and tri-methylation of lysine residues within histone proteins is under investigation. Evidence from several model organisms suggests that different methylated states of H3 Lys(9) (H3K9) are generated by specific histone methyltransferases (MTases) to mark distinct types of silent chromatin. Sequence alignment of all histone lysine MTases with known product specificity suggested that a key residue in the active site determines how many methyl groups they add. We examined this possibility both in vitro and in vivo and found that a Phe at the position equivalent to Phe(281) of Neurospora crassa DIM-5 or Phe(1205) of human G9a allows the enzyme to perform di and tri-methylation, whereas a Tyr at this position is restrictive, inhibiting tri-methylation and thus yielding a mono- or di-MTase. Phe to Tyr mutants of both DIM-5 and G9a restrict product specificity in vitro and in vivo without compromising overall catalysis. These mutants were employed to probe the biological significance of mono-, di-, and tri-methylation of H3K9 in both mouse embryonic stem cells and N. crassa. G9a F1205Y, when expressed in G9a (-/-) embryonic stem cells, rescued only H3K9 mono-methylation, but not di-methylation, to wild-type levels yet silenced Mage-a gene expression. When expressed in dim-5 strains, DIM-5 F281Y generated significant levels of mono- and di-H3K9 methylation (which are not observed in wild type Neurospora) as well as tri-methyl H3K9. The altered DIM-5 rescued the growth defect characteristic of dim-5 N. crassa but did not fully rescue the gross DNA hypomethylation of dim-5 strains.

By comparing the active sites of SET7/9 (a mono-MTase) and DIM-5 (a tri-MTase), we designed point mutations in both enzymes that profoundly altered their product specificities without affecting their catalytic activities (9). This led us to propose that a few key residues in the active site of HKMTs determine how many methyl groups they add, and the product specificity may be predictable from their primary sequences. We show that G9a forms mono-and di-methyl Lys 9 as initial products and then slowly adds the third methyl group, consistent with reports that G9a is a global mono-and di-MTase (11,12) but can generate trimethyl-H3K9 in some situations (26). We also demonstrate that a single point mutation, F1205Y, converts G9a to a mono-MTase. When expressed in G9a (Ϫ/Ϫ) ES cells, the F1205Y mutant rescues mono-, but not di-methyl H3K9 levels, yet represses Mage-a gene expression to wild-type levels. Similarly, the F281Y mutant of DIM-5, when expressed in a dim-5 null background, yields significant levels of monoand di-methyl H3K9 that are not observed in the wild-type Neurospora. The presence of the mono-and di-methyl marks is incompatible with the function of DIM-5 in maintaining normal DNA cytosine methylation. In contrast, the remaining trimethylation and/or the increased levels of mono-and di-methylation by the F281Y mutant is sufficient to rescue the characteristic growth defect of the dim-5 strain.

MATERIALS AND METHODS
Protein Expression and Purification-A fragment encoding the Cterminal 280 residues of human G9a was amplified from an expressed sequence tag clone BC002686 (Resgen) and subcloned between the BamHI and EcoRI sites of pGEx2T (Amersham Biosciences), yielding pXC428. BL21(DE3) Codon-Plus RIL (Stratagene) cells harboring pXC428 were grown in LB supplemented with 100 mg/liter ampicillin, 50 mg/liter chloramphenicol, and 25 M ZnSO 4 at 37°C until A 600 ϭ 0.5. The culture was shifted to 16°C and induced by the addition of 0.4 mM isopropyl ␤-D-thiogalactopyranoside for 16 h. The cells were collected by centrifugation, lysed with a French press, and applied to glutathione-Sepharose 4B (Amersham Biosciences) in a batch slurry. Thrombin cleavage of the glutathione S-transferase tag was performed on the beads, yielding the G9a catalytic domain following four residues: GSHM from the vector. Eluted protein was concentrated to 1 mg/ml, flash frozen, and stored at Ϫ80°C in 25 mM Tris, pH 8.5, 50 mM NaCl, 2 mM dithiothreitol, and 5% glycerol. DIM-5 was expressed and purified as described (27).
Mutagenesis-Mutagenesis was performed with the QuikChange site-directed mutagenesis kit (Stratagene). The human G9a amino acid, corresponding to 1205 of the G9a mouse long isoform (10), in pXC428 was converted from Phe (TTT) to Tyr (TAT), yielding pXC444. The construct was fully sequenced to confirm the desired mutation and the lack of unintended mutations. Generation of DIM-5 mutants Y178V and F281Y has been described (9).
Mass Spectrometry Analysis-Methylation reactions for G9a were performed in 25 mM Tris, pH 8.5, 5 mM dithiothreitol, 250 M Sadenosyl-L-methionine, 10 M peptide (H3 amino acids 1-15, 1-24, or 21-34), and 0.1 mg/ml enzyme at room temperature. For DIM-5, 20 M of peptide and 0.02 mg/ml of the enzyme were used. The reactions were stopped at various times by the addition of trifluoroacetic acid to 0.5%. MALDI-TOF mass spectrometry was performed on an Applied Biosystem Voyager System 4258 (Emory University Chemistry Department) with ␣-cyano-4-hydroxycinnamic acid as matrix. The final spectra are the averages of 200 shots/position at 10 different positions.
Analysis of DNA Methylation-Genomic DNA was isolated from liquid cultures grown for 2 days at 32°C and analyzed for DNA methylation as described (30). The DNA (ϳ1 g) was digested with DpnII or Sau3AI and analyzed by Southern hybridization. Probes for methylated DNA regions 1d21, 8A6, and 8F10 used were PCR products generated with the primer pairs described above.
G9a ES Cells-Undifferentiated ES cells were maintained in 10% fetal calf serum and leukemia-inducing factor (500 units/ml)-containing medium. G9a long isoform (G9a-L) cDNA carrying the F1205Y mutation was generated by conventional double PCR mutagenesis and subcloned into the expression vector, pCAGGS. The G9a-L F1205Y expression vector was introduced into ES cells using Lipofectamine 2000 reagent (Invitrogen) according to the manual.

G9a Is a Fast Mono/di-MTase and a Slow Tri-MTase-We
have previously used MALDI-TOF mass spectrometry to monitor the kinetic progression of methylation reactions and determined that DIM-5 is a tri-MTase and SET7/9 is a mono-MTase (8,9) (Fig. 1A). Here we report a similar investigation of the product specificity of the G9a catalytic domain by monitoring the methylation of a synthetic histone H3 peptide, residues H3 1-24 ( Fig. 1B) or H3 1-15 (Fig. 2, left panels), by mass spectrometry. G9a produced both mono-and di-methylated peptides early in the reaction. Di-methylated H3K9 was the dominant form after about 1 min on both the 1-24 and 1-15 peptides, shortly after a burst of mono-methyl product coincident with the rapid disappearance of unmethylated peptides. Notably, tri-methylated H3K9 was first detected only after the di-methyl product was maximal and both un-modified and mono-methylated peptides had disappeared. The conversion of di-to tri-methyl lysine was much slower, not becoming complete until after 30 min in both cases. These observations suggest that mono and di-methyl H3K9 are the initial products of G9a. This is in sharp contrast to DIM-5, which shows a direct progression to tri-methyl lysine without significant accumulation of di-methyl lysine (8,9) (Fig. 1A) and SET7/9, which produces only mono-methyl lysine plus a trace of di-methyl lysine after overnight incubation (9). These biochemical findings are consistent with in vivo observations that G9a is a global mono-and di-HKMT; specifically, G9a (Ϫ/Ϫ) ES cells lose global euchromatic mono and di-methyl H3K9 immunostaining (11,12). Interestingly, our results suggest G9a could generate tri-methyl H3K9 at regions where it has high local concentrations. In fact, G9a tethered to a synthetic mini-locus as a Gal4DBD fusion generates tri-methyl lysine in vivo (26). Whether G9a makes trimethyl-H3K9 under normal circumstances in vivo remains to be determined.
Considering that G9a has also been reported to have activity on H3K27 (10), we also tested G9a product specificity on a H3 peptide containing Lys 27 (Fig. 1C). We found that G9a has very poor activity on this lysine, some 100-fold worse than on peptides of similar length that include Lys 9 . Nevertheless, G9a did produce mono-and di-methyl H3K27 simultaneously in the early stages of the reaction and very slowly generated trimethyl Lys 27 (ϳ24 h). This low activity is consistent with the finding that methyl-H3K27 levels are unaffected in G9a (Ϫ/Ϫ) ES cells (11) and that G9a is a strong H3K9 MTase but has low or nondetectable H3K27 MTase activity, even when assayed as a member of a complex isolated by co-immunoprecipitation (13,15). We conclude that G9a has poor activity on H3K27 in vitro, and it is unclear whether it has any significant effect on H3K27 methylation in vivo.
Replacement of Phe 1205 with Tyr Converts G9a to a Mono-MTase-Comparison of the crystal structures for DIM-5 and SET7/9 led us to generate mutants (F281Y in DIM-5 and Y305F in SET7/9) with altered product specificity (9). To test the hypothesis that the corresponding residue Phe 1205 in G9a (Table I) is the major determinant of product specificity, we replaced this residue with Tyr (F1205Y). The mutation did not significantly affect the catalytic activity of G9a, because the rate of loss of unmethylated peptide was roughly equal to that of wild-type (WT) enzyme (Fig. 2D, right panel). However, the reaction by F1205Y stalled at the mono-methyl stage (compare 1.5 and 10 min; Fig. 2, A and B), supporting our hypothesis. A trace amount of di-methylated product was observed only after  1-24). B, G9a on histone H3 peptide (amino acids 1-24). C, G9a on histone H3 peptide (amino acids 21-34). The average relative amounts of each peptide across the time course of the reaction as a percentage of the sum intensity of all of the peaks from an averaged (Ͼ200 shots) MALDI-TOF spectrum is reported .  FIG. 2. Generation of mono-methyl H3K9 (H3 amino acids 1-15) by G9a F1205Y variant without impaired catalytic ability. A-C, representative mass spectra (from a single reaction) at various time points for WT G9a (left panels) and F1205Y (right panels). Unmodified, mono-, di-, and tri-methyl peptides are labeled. D, the relative amount of each peptide species over the full time courses of the reactions, expressed as a percentage of the sum of intensity of all related peaks. a.i., absolute intensity. prolonged incubation (45 min; Fig. 2C), which indicated that di-methylation by the altered enzyme was actually slower than the tri-methylation by the WT G9a. We conclude that the F1205Y mutation changed the product specificity of G9a from a fast mono/di-MTase with a slow tri-MTase activity to a predominantly mono-MTase without affecting overall catalytic activity, analogous to the F281Y mutation for DIM-5 (9).
Monomethyl G9a Silences Mage-a Gene Expression-G9a (Ϫ/Ϫ) ES cells lack global mono-and di-methyl H3K9 and show aberrant expression of Mage-a genes (11,12,31). A G9a transgene with the F1205Y mutation was expressed as a full-length G9a (long isoform) transgene in G9a (Ϫ/Ϫ) mouse ES cells. When expressed at WT levels ( Fig. 3A), the F1205Y mutant rescued H3K9 mono-but not di-methylation to WT levels ( Fig.  3B), consistent with our biochemical data. Interestingly, Northern hybridization results showed that this mono-methylation of Lys 9 was sufficient to repress Mage-a (Fig. 3C). This result may reflect an unknown repressive lysine-binding module that recognizes mono-and/or di-H3K9, or it may be that mono-and di-H3K9 can both block H3K9 acetylation and neighboring marks required for transcription of the Mage genes.
F281Y DIM-5 Partially Rescues Phenotypes of dim-5 Mutant-A lack of DIM-5 HKMT leads to gross hypomethylation of H3K9, which is a required mark for DNA methylation in Neurospora (7,8). In addition, dim-5 strains are characterized by growth defects and poor fertility. Neurospora strains that lack all DNA methylation because of a null mutation in the DNA MTase gene dim-2 do not show these defects, suggesting that DIM-5, or its product trimethyl-H3K9, has functions beyond DNA methylation. To determine whether DNA methylation and other functions of DIM-5 activity rely on tri-methylation of H3K9, we introduced the F281Y substitution, which, in vitro, dramatically changed the product specificity of the enzyme from tri-to mono-H3K9 (with slight residual di-methyl activity) into a Neurospora strain bearing a nonsense mutation in dim-5. We then tested the transformed strains for mono-, di-, and tri-methylation and scored for physiological effects. As controls, we also transformed the dim-5 host strain with the WT dim-5 gene as well as with a second mutated gene (Y178V), resulting in a catalytically inactive variant (ϳ0.5% WT activity) (9). All of the constructs were targeted to the his-3 locus. The ectopic WT and F281Y mutant expressed at similar levels as wild-type DIM-5, as determined by Western blot (Fig. 4A), but the Y178V mutant was not detected, presumably because of instability of the mutant protein.   4. DIM-5 F281Y variant partially rescues defects of dim-5 mutant. A, DIM-5 expression is similar from both the ectopic dim-5 WT allele and the dim-5 HT1 transformants carrying dim-5 WT or dim-5 F281Y alleles at his-3. The dim-5 Y178V allele, however, is either not expressed or the resulting mutant protein is unstable. B, expression of DIM-5 F281Y variant in N. crassa produces mono-, di-, and tri-methylated H3K9, whereas the WT DIM-5 produces exclusively trimethylated H3K9. Western blot analysis of N. crassa histone H3 using antibodies against mono-, di-, or tri-methylated H3K9. Coomassie Blue staining of a replica gel confirmed that the amount of histone H3 was approximately even between the WT, dim-5 Y178V , and dim-5 F281Y protein samples. C, chromatin immunoprecipitation experiments with chromatin from dim-5 HT1 transformants carrying dim-5 Y178V (inactive), dim-5 WT , or dim-5 F281Y alleles at his-3 were carried out using the indicated antibodies. Total DNA and immunoprecipitated DNA were subject to duplex PCR to amplify pairs of unmethylated (hH4 and pcn) and naturally methylated (1d21, 8A6, and 8F10; underlined) DNA regions of N. crassa. Note that the F281Y mutant, but not the inactive and wild-type alleles, show significant mono-and di-methyl-H3K9 signals in the regions that are subject to DNA methylation. The specificity mutant (F281Y) also shows a tri-methyl H3K9 signal, although it is somewhat weaker than in the ChIP with the wild-type allele (most obvious in the 1d21 region). In one experiment (right panel), regions were probed for dimethyl-H3K4, which is underrepresented at heterochromatic regions. D, DIM-5 F281Y variant poorly complements DNA methylation defect in dim-5 HT1 strain. Genomic DNA samples from two independently isolated dim-5 HT1 transformants (1 and 2) carrying dim-5 WT , dim-5 F281Y , or dim-5 Y178V alleles at his-3 were digested with DpnII (D) or Sau3AI (S) and analyzed by gel electrophoresis and Southern hybridization using probes for the indicated regions. A photograph of the ethidium bromide-stained gel (total) shows partial complementation on a global level. Note that Sau3AI and DpnII are isoschizomers, but only Sau3AI is sensitive to DNA cytosine methylation.
Western blots showed that the dim-5 strain transformed with the WT gene showed strong tri-methylation of H3K9, but no noticeable mono-or di-methylation of H3K9, as expected (Fig. 4B). The catalytically impaired enzyme also lacked monoor di-methylated forms of H3K9 but showed a background of tri-methyl H3K9 (as does the dim-5 null strain (8)), possibly because of an unknown HKMT that is not preferentially targeted to regions that are subject to DNA methylation.
Interestingly, results of Western blot and ChIP experiments showed that the F281Y mutant produced mono-, di-, and trimethyl H3K9 (Fig. 4, B and C). Although the anti-trimethyl antibody has been well characterized (8), we carried out an additional peptide competition experiment that demonstrated that the signal detected in the F281Y transformant is truly tri-methyl H3K9 (data not shown). Although the overall levels of trimethyl-H3K9 seem similar in the F281Y mutant and WT based on the semi-quantitative Western analysis, repeated duplex ChIP experiments, in which a region of the genome that is normally subject to DNA methylation (1d21, 8A6, or 8F10) was compared with a region not subject to DNA methylation (pcn or hH4), revealed modest reductions of tri-methyl Lys 9 , especially in the 1d21 and 8F10 regions (Fig. 4C). That the F281Y mutant produced substantial levels of tri-methyl H3K9, in addition to mono-and di-methyl H3K9, may be due to strong localization at heterochromatic loci, allowing its very slow residual di-and tri-MTase activity, detected in vitro (9), to act over a prolonged residence time. Another possibility is that an unknown HK-MT(s) can convert the mono-or di-H3K9 products generated by the mutant enzyme to tri-methyl H3K9. We have observed that WT DIM-5, for instance, has a much faster rate on modified (mono-or di-methyl peptides) than unmodified peptides (8). Nevertheless we cannot exclude the possibility that the F281Y mutant behaves principally as a tri-MTase in vivo, in contrast to the situation in vitro. Regardless of these possibilities, the presence of mono-and di-H3K9, which are not observed at significant levels in wild-type Neurospora, is clearly attributable to the reduced product specificity of F281Y DIM-5. The ChIP experiments showed that the mono-and di-methyl forms of H3K9 were specifically targeted to regions to be methylated, as is the case for DIM-5-dependent H3K9 tri-methylation in WT cells (Fig. 4C).
The Y178V mutant completely failed to complement the DNA methylation defect of dim-5 strains, as expected. More interestingly, F281Y poorly restored DNA methylation compared with ectopically expressed WT DIM-5 (Fig. 4D). We conclude that either the presence of mono-and di-methyl K9 or the slight reduction of tri-methyl H3K9 significantly compromised signals for DNA methylation. Curiously, although the F281Y mutant poorly restored DNA methylation, it rescued the dim-5 growth defect to the same degree as WT DIM-5 (Fig. 5). The catalytically impaired mutant (Y178V) did not complement the growth defect but because this mutant failed to produce stable protein (Fig. 4A), we could not determine whether the full rescue of the growth defect is due to some function of DIM-5 not dependent on its methyltransferase activity. DISCUSSION The multiplicity of methylation levels at a given residue is a newly recognized elaboration of the histone code. Each methylation mark (mono-, di-, or tri-) could serve to establish or mask unique binding sites of chromodomain proteins or associated proteins, thereby greatly expanding the complexity of the histone code. Variable degrees of methylation could also serve to tune the output of the "code" by modulating the affinity of lysine binding by chromodomains. The extent to which mono-, di-, and tri-methylation of a single lysine generates a unique mark or act to govern the signal strength of a single mark remains to be seen. The ability to predict the product of each HKMT will be useful in creating engineered mutants to test the biological impact of the variable methylation states of histone lysines. In this study, we found evidence that distinct processes may differ in their reading of a methyl mark. Specifically, in Neurospora, introduction of lower methylation states (monoand di-methyl) at H3K9 appeared to preclude full complementation of the DNA methylation defect of dim-5 but did not hamper full complementation of the growth and fertility defects. The effect of mono-and/or di-methyl H3K9 on DNA methylation may reflect a requirement for a high local concentration of tri-methyl H3K9 and perhaps both H3 molecules in the nucleosome must be tri-methylated. This is consistent with the finding that the ectopic expression of histone H3 with K9R or K9L mutations in a wild-type background has the dominant effect of bringing about gross DNA hypomethylation (7). Additional work will be required to determine whether the striking difference in complementation reflects differences in the molecules that read the marks. It is conceivable, for example, that HP1, which is known to bind both di-and tri-methylated forms of H3K9 in other organisms (32,33) and has been demonstrated to be essential for DNA methylation in Neurospora (34), forms distinct complexes that are differentially sensitive to the methylation state.
As previously reported, Phe 281 of DIM-5 is critical in determining its product specificity (9). A Tyr at this position generates a mono/di-MTase. Sequence alignment including all HK-FIG. 5. DIM-5 F281Y variant fully restores normal growth in dim-5 HT1 strain. Apical growth rates of six independent his-3-targeted dim-5 HT1 transformants carrying dim-5 WT , dim-5 Y178V , or dim-5 F281Y alleles at his-3 were measured at 32°C using "race tubes" containing sucrose Vogel's agar medium supplemented with leucine and pantothenic acid.
MTs with known product specificity suggests that this rule may be generalized (Table I). Arabidopsis KYP and SUVH6 both have a tyrosine and they are both mono/di-MTases (25). From a structural perspective, it appears the tyrosine hydroxyl can block substrate lysines with methyl group(s) attached from rotating into a position where they can be further methylated.
In this work, we demonstrated that G9a, which has a Phe at this position, is a fast mono/di-MTase and slow tri-MTase. Interestingly, substituting the Phe of G9a with a Tyr also resulted in a mono-MTase, analogous to DIM-5. This mutant with "switched" product specificity rescued only mono-methyl H3K9 yet represses Mage-a gene expression. It is unknown whether di-methyl H3K9 is critical for other G9a-dependent processes (e.g. maintenance of DNA methylation at imprinting centers, PRDI-BF1 silencing in terminal B-cell differentiation, and neuron-restrictive silencing factor/REST-mediated silencing of neuronal genes).
At face value, S. cerevisiae SET1 protein appears to conflict with our Phe/Tyr switch hypothesis. It has a tyrosine at the position comparable with Phe 281 of DIM-5, and yet results of ChIP experiments suggest that it can both di-and tri-methylate H3K4 (21,22). It should be noted, however, that the published studies did not reveal whether SET1 also produces mono-methyl-Lys 4 , presumably because the mono-specific antibody was not yet available. Moreover, the product specificity of SET1 has not been determined in vitro because the enzyme appears to be only active as part of a large complex (COMPASS or SET1c), and an active recombinant protein complex has not been obtained (35,36). It is possible that SET1 is intrinsically a mono/di-MTase but capable of di/tri-methylation when present in the large complex. Alternatively, SET1 may in fact make mono-methyl-K4 in vivo, but other HKMTs that can only utilize mono-methylated H3K4 as substrate carry the methylation to di-or tri-states. ESET, a di-HKMT, is stimulated and converted into a tri-MTase by binding mAM, a murine ATFaassociating modulator, both in vivo and in vitro (37). Curiously, this enzyme (SETDB1 in Table I) has a tryptophan at the key position, unlike other SUV SET domains. It was shown recently that ALL1/MLL1, another member of the SET1 family, is an H3K4 MTase and is required for assembly of a large complex on the promoter of target genes to activate their expression (38,39). More importantly to this study, recombinant ALL1 fragments of C-terminal 224 residues (39) or 377 residues (38) have robust HKMT activity, but no activity on a di-methylated peptide, consistent with it being a mono/di-MTase. Our limited study of SET domain HKMTs suggests that a Phe/Tyr switch controls their product specificity and that alteration of this product specificity can have important biological consequences. It appears the histone code may show rigidity in some cases and surprising plasticity in others.