Recognition of Histone H3 Lysine-4 Methylation by the Double Tudor Domain of JMJD2A

See allHide authors and affiliations

Science  05 May 2006:
Vol. 312, Issue 5774, pp. 748-751
DOI: 10.1126/science.1125162


Biological responses to histone methylation critically depend on the faithful readout and transduction of the methyl-lysine signal by “effector” proteins, yet our understanding of methyl-lysine recognition has so far been limited to the study of histone binding by chromodomain and WD40-repeat proteins. The double tudor domain of JMJD2A, a Jmjc domain–containing histone demethylase, binds methylated histone H3-K4 and H4-K20. We found that the double tudor domain has an interdigitated structure, and the unusual fold is required for its ability to bind methylated histone tails. The cocrystal structure of the JMJD2A double tudor domain with a trimethylated H3-K4 peptide reveals that the trimethyl-K4 is bound in a cage of three aromatic residues, two of which are from the tudor-2 motif, whereas the binding specificity is determined by side-chain interactions involving amino acids from the tudor-1 motif. Our study provides mechanistic insights into recognition of methylated histone tails by tudor domains and reveals the structural intricacy of methyl-lysine recognition by two closely spaced effector domains.

Posttranslational modifications of histones have a profound influence on fundamental biological processes such as transcriptional regulation and epigenetic inheritance (1, 2). However, the mechanisms by which covalent histone modifications are recognized and deciphered remain poorly understood. Currently, both the recognition of acetylated histones by the bromo domain (3, 4) and the binding of methylated histone tails by the chromo domain and a WD40-repeat protein, WDR5, are the only modified histone/protein interactions understood in any detail (510). The chromodomain is a member of the tudor domain “royal family,” which includes the tudor, plant Agenet, chromo, PWWP, and MBT domains. Many of the proteins containing these domains are associated with chromatin (11). Using a protein-array approach, we recently showed that in addition to chromo domains, certain tudor and MBT domains specifically bind to methylated histone tails (12). Here we focus on the tudor domains of JMJD2A, which is a member of a conserved family of Jmjc domain–containing proteins (Fig. 1, A and B) belonging to the Jmjc domain histone demethylase superfamily (13). JMJD2A has been reported to interact with histone deacetylase complexes and Rb and to function as a transcription repressor (14, 15).

Fig. 1.

JMJD2A structure. (A) A schematic drawing of the domain structure of JMJD2A. (B) Alignment of the double tudor domain sequences of three human JMJD2 homologs, one from zebrafish (Ensemble ID: DARP00000024692) and another from frog (Ensemble ID: XETP00000001152), and human 53BP1. Secondary-structure elements (orange and green for the first and second tudor motifs, respectively), their nomenclature, and the amino acid numbering of JMJD2A are shown above the sequences. The secondary-structure elements of the 53BP1 double tudor domain are delineated below the sequences. JMJD2A residues subjected to mutational studies are indicated by stars. (C) The structure of the JMJD2A double tudor domain (ribbon representation) in complex with a trimethylated H3K4 peptide (ball-and-stick model). Regions spanning the first and second tudor motifs are colored orange and green, respectively. The dotted line indicates a segment of seven disordered residues. (D) The structure of the double tudor domain of 53BP1 shown in a ribbon representation. Secondary-structure elements in (C) and (D) are colored as in (B).

Previous structural studies of the tudor domain of survival motor neuron (SMN) revealed that it contains a binding site for a symmetrically dimethylated arginine (16). The double tudor domain of 53BP1, a p53 binding protein, has been implicated in binding methylated Lys79 of histone H3 (17), methylated Lys20 of histone H4 (12), as well as DNA and arginine-glycine–rich sequences (18). However, the structural basis for ligand binding by tudor domains is not understood. JMJD2A contains two tandem tudor domains, and the double tudor domain is known to bind methylated histone H3-K4 or H4-K20 (12). To gain insights into the structural basis for binding methylated histone tails, we have solved the crystal structure of the double tudor domain of JMJD2A both in the presence and absence of a trimethylated H3-K4 peptide (H3K4Me3). A detailed description of structure determination and data statistics can be found in the supporting online material.

Surprisingly, the 2.1Å structure revealed that the two tudor domains interdigitate and form a bilobal, saddle-shaped structure with each lobe resembling the canonical tudor domain structure (Fig. 1C). We term these tudorlike lobes hybrid tudor domain 1 and 2 (HTD-1 and HTD-2). The methyl-H3-K4 peptide is bound in a cleft of HTD-2 (Fig. 1C). The interdigitated folding of the double tudor domain of JMJD2A differs appreciably from the structure of the homologous double tudor domain of 53BP1, which comprises two independently folded tudor domains (Fig. 1D) (17, 18). The hybrid-tudor domains are formed by the exchange of the third and fourth β strands with respect to the canonical tudor domain. The swapped β3 forms a long, contiguous strand with the nonswapped β2, which we call β2β3, and the two long strands, β2β3 and β2'β3' (an apostrophe indicates secondary-structure elements of the second tudor motif, as denoted in Fig. 1B), bridge the two compact HTDs (Fig. 1, B and C).

Compared to the corresponding surface area of HTD-1, the surface of HTD-2 surrounding a cluster of aromatic residues has a more negatively charged electrostatic potential, consistent with its role in binding the positively charged H3K4Me3 peptide (Fig. 2A). Three aromatic residues, Trp967 and Tyr973 of the tudor-2 motif and Phe932 of the tudor-1 motif, form an open cage that accommodates the binding of trimethylated Lys4 (Fig. 2B). The binding buries a total accessible surface area of 1039Å2. In addition, a negatively charged residue, Asp934, lines up at the edge of the aromatic cage (Fig. 2C). This arrangement of aromatic residues and the negatively charged amino acid is reminiscent of the H3K9Me3 binding site of the HP1 chromo domain (Fig. 2B) (5, 6). The aromatic cage is required for binding methylated lysines: Changing Tyr973 to an alanine (Y973A) or Trp967 to a histidine (W967H) abolishes the binding of H3K4Me3, whereas the wild-type double tudor domain binds the peptide with an apparent equilibrium dissociation constant (Kd) of 10.4 μM. The binding constants were determined using surface plasmon resonance (SPR) (Fig. 3, A and B).

Fig. 2.

JMJD2A double tudor domain–H3K4Me3 peptide interactions. (A) The double tudor domain as seen in a surface representation with electrostatic potential distribution colored red for negatively charged, white for neutral, and blue for positively charged areas. The peptide is shown as a stick model superimposed with the surrounding 2FoFc electron density map (displayed at 1.2σ contour level). (B) A stereo view of the methyl-H3K4 (colored cyan) bindingaromaticcagesuperimposedwiththe methyllysine binding aromatic cage of the chromo domain of HP1 (gray). Tudor-1 and Tudor-2 residues are colored orange and green, respectively. (C) A detailed view of JMJD2A-H3K4Me3 interactions. The peptide and the HTD-2 residues involved are shown in a stick model. Dashed lines indicate hydrogen bonds. The same coloring scheme as in Fig. 1 is used.

Fig. 3.

Binding of the JMJD2A double tudor domain to histone H3 peptides. (A) Representative binding curves for the wild-type double tudor domain and a biotinylated H3K4Me3 peptide (Btn-H3K4Me3) as measured using SPR. Curves represent measurements with an increasing concentration of the protein used. From bottom to top: 1 μM(red), 2 μM(orange),5 μM (yellow), 10 μM(green), 20 μM(blue), and 40 μM (magenta). The vertical and horizontal axes show the Biacore response unit (RU) and the time scale (seconds). (B) Tabulation of pairwise protein-peptide interactions measured. The wild-type or mutant JMJD2A double tudor domains used are shown in the first column, and the peptide used is indicated in the second column. Btn-H3K4: an H3K4 peptide with an unmethylated Lys4; Btn-R2A: an H3K4Me3 peptide with an alanine in place of Arg2. The derived equilibrium dissociation constants are shown in the third column, where n/d denotes bindings too weak to be reliably quantitated. (C) Representative binding curves of the wild-type double tudor domain with the Btn-R2A peptide. (D) Alignment of amino acid sequences encompassing known methylated lysine residues of histone H3 and H4. The position of methyl-lysines is designated position 0. Both the 0 and –2 positions are highlighted in red.

A number of side chain-side chain interactions between HTD-2 and the H3K4Me3 peptide contribute to their binding specificity. In contrast to the binding of methylated histone peptides by chromo domains (59), there are only two β sheet–like main-chain hydrogen bonds between the H3K4Me3 peptide and HTD-2, which involve the main-chain atoms of Gln5 of the histone peptide and Phe937 of HTD-2. In the present structure, Arg2 of the histone H3 peptide interacts with Asp945 of HTD-2 through ionic and hydrogen bonds (Fig. 2C). Thr3 of the peptide forms a hydrogen bond with Asn940, and the peptide Gln5 interacts with Phe937 through van der Waals contacts. In addition, the terminal amino group of the peptide interacts with HTD-2 through two weak hydrogen bonds (both ∼3.4Å), one with the hydroxyl group of Tyr942 and another with the carbonyl group of Ala941. It should be noted that all of the HTD-2 residues involved, except the ones in the methyl-lysine binding aromatic cage, belong to the tudor-1 motif (Fig. 2C), which explains the observation that tudor-2 alone cannot bind H3K4Me3 (Fig. 3B).

Because Arg2 of the H3K4Me3 peptide plays a prominent role in the interaction with the double tudor domain of JMJD2A, we tested the binding effect of mutations of Arg2 and its interaction partner, Asp945 of HTD-2, by changing Arg2 to an alanine (R2A) and changing Asp945 to an alanine (D945A) or an arginine (D945R). The wild-type double tudor domain binds the R2A H3K4Me3 peptide with a Kd of 127.9 μM, which is ∼1/12th the level of binding to the wild-type H3K4Me3 peptide (Fig. 3). The D945A mutant binds to the wild-type H3K4Me3 peptide with a Kd of 118.4 μM, a level comparable to that between the wild-type double tudor domain and the R2A H3K4Me3 peptide, and the D945R mutant no longer binds the peptide. In the vicinity of Lys9, Lys27, and Lys36 of histone H3, either an alanine or a glycine occupies the position corresponding to that of Arg2 (Fig. 3D). These small residues cannot efficiently interact with Asp945 of JMJD2A, which is likely to account for the preferential binding of the JMJD2A double tudor domain to H3K4Me3 over other histone H3 methyl-lysine marks.

As pointed out earlier, the double tudor domain can also bind methylated H4-K20 (12). A trimethylated H4-K20 peptide binds the double tudor domain of JMJD2A with a Kd of ∼25 μM, and our study shows that the same binding pocket is involved in binding H3K4Me3 and methylated H4-K20, because the W967H and Y973A HTD-2 aromatic cage mutants also lost the ability to bind methylated H4-K20. In a H4-K20 peptide, the residue corresponding to Arg2 is a histidine (Fig. 3D), which cannot interact with Asp945 as does Arg2 of histone H3 but should still be able to bind favorably in the area formed by Leu941, Tyr942, and Asp945 (Fig. 2C). Furthermore, the HTD-2 region surrounding the methyl-lysine binding site is quite negatively charged (Fig. 2A), which provides a favorable environment for the binding of three positively charged arginine residues near H4-K20 (Fig. 3D).

The two HTDs share ∼23 and ∼44% sequence identity and similarity, respectively, and the main chains of the two domains can be superimposed with a root mean square (RMS) deviation of 1.05Å. In particular, a HTD-1 aromatic cage differs from that of HTD-2 only by one residue, a histidine (His909) instead of a tryptophan (Trp967). Superimposition of the two HTD domains reveals that His909 occupies a position that occludes the binding of a methyl-lysine (Fig. 4A). To determine whether His909 is solely responsible for the inability of HTD-1 to bind H3K4Me3, we changed the histidine to a tryptophan while at the same time changing Trp967 of HTD-2 to a histidine. As described earlier, the latter mutation alone precludes HTD-2 from binding H3K4Me3. The W967H/H909W double mutant does not bind H3K4Me3 (Fig. 3B). Thus, residues outside the aromatic cage are also important for the functional differences between HTD-1 and HTD-2. It is possible that HTD-1 could be involved in binding yet unidentified ligands.

Fig. 4.

Tudor domain ligand binding sites. (A) A stereo view of superimposed HTD-2 and HTD-1 aromatic clusters. The latter is formed by His909 and Tyr915 of the tudor-1 motif (orange) and Phe990 of tudor-2 (magenta). His909 clashes with the superimposed methyl-H3K4. (B) Conformational differences of the β1-β2 loop among HTD-1 (green) and HTD-2 (orange) of JMJD2A, SMN tudor (magenta), and the first tudor domain of 53BP1 (blue). Two of the aromatic residues implicated in binding methylated lysine, or arginine for SMN tudor, reside in this loop.

There is little conformational change between the H3K4Me3 bound and free double tudor domain structures of JMJD2A, showing that the interdigitated structure is not due to H3K4Me3 binding and that conformational dynamics do not play a major role in regulating histone binding (fig. S1). HTDs and canonical tudor domains have similar secondary structural arrangements: HTD-1 and HTD-2 can be superimposed with the tudor domain of SMN (16) with RMS deviations of 1.26 and 0.87Å, respectively. The locations of the aromatic cages in both HTDs also generally coincide with the sites in SMN and 53BP1 implicated in binding dimethylated arginine and methylated histone H3-K79, respectively (16, 17). A comparison of the tudor domain structures reveals that the loop connecting β1 and β2, in which two of the three aromatic residues reside, has the most notable conformational variation among different tudor domains (Fig. 4B), perhaps reflecting the structural flexibility needed for accommodating different binding partners.

Because of the involvement in transcription regulation, it is tempting to speculate that the double tudor domain is responsible for directing JMJD2A to chromatin regions enriched with H3-K4 or H4-K20 methylation. However, alternative scenarios are also possible, and the functional relation between methyl-histone binding by the double tudor domain and histone demethylation by JMJD2A in a physiological context remains to be established. Nevertheless, our results demonstrate that the hybrid-tudor domain structure of JMJD2A is required for the formation of a functional methyl-histone H3 binding module. The unusual fold requires two tudor domain motifs in tandem, which is only present in a subset of tudor domain–containing proteins including JMJD2 family members and 53BP1. It will be extremely interesting to understand the principle underlying the distinct folding of the double tudor domains of JMJD2A and 53BP1 despite their sequence similarity. Many chromatin-associated proteins have closely spaced tandem repeats of effector domains implicated in histone binding, including the bromo, chromo, MBT, Agenet, and tudor domains. Our discovery reveals the potential for forming novel histone binding modules from the familiar effector domains.

Supporting Online Material

Materials and Methods

Fig. S1

Table S1


References and Notes

View Abstract

Stay Connected to Science

Navigate This Article