Histone H3 recognition and presentation by the WDR5 module of the MLL1 complex

Abstract

WDR5 is a core component of SET1-family complexes that achieve transcriptional activation via methylation of histone H3 on Nζ of Lys4 (H3K4). The role of WDR5 in the MLL1 complex has recently been described as specific recognition of dimethyl-K4 in the context of a histone H3 amino terminus; WDR5 is essential for vertebrate development, Hox gene activation and global H3K4 trimethylation. We report the high-resolution X-ray structures of WDR5 in the unliganded form and complexed with histone H3 peptides having unmodified and mono-, di- and trimethylated K4, which together provide the first comprehensive analysis of methylated histone recognition by the ubiquitous WD40-repeat fold. Contrary to predictions, the structures reveal that WDR5 does not read out the methylation state of K4 directly, but instead serves to present the K4 side chain for further methylation by SET1-family complexes.

The structure and dynamics of chromatin are increasingly recognized to be essential for mediating both global and local regulation of the genome^1,2. Vital to maintaining and remodeling the various chromatin states are a collection of post-translational modifications of histones that are introduced in a highly regulated fashion and cause specific functional consequences³. In particular, post-translational introduction of methyl groups onto lysine side chains of histone proteins greatly affects chromatin function in complex and often opposing ways, resulting in either transcriptional repression or activation, contingent upon the precise location and degree of methylation^4–7. For example, trimethylation of histone H3 at K9 or K27 is implicated in silencing of the underlying DNA^6,8, whereas methylation at other sites, notably H3K4, is coupled to transcriptional activation^4,7. (Throughout this manuscript, single-letter amino acid abbreviations are used to refer to histone residues, whereas three-letter abbreviations are used for WDR5 residues.) K4 is trimethylated in nucleosomes proximal to promoters and coding regions of actively transcribed genes⁹.

SET domain proteins are the major catalytic components of a number of histone methyltransferase (HMTase) complexes that effect lysine methylation^10–15. The sole yeast enzyme capable of catalyzing H3K4 methylation is the SET domain protein Set1, which was identified as the catalytic component of the Set1 complex (also called the COMPASS)^11,13,15. In higher eukaryotes, the situation is more complex: there are a number of yeast Set1 orthologs (SET1a and SET1b) and paralogs (MLLs and Ash1) collectively referred to as the SET1 family¹⁶. Among these proteins, the most extensively studied is mixed lineage leukemia-1 (MLL1), a proto-oncogene whose fusion with a number of partners causes a variety of human leukemias¹⁷. The SET1 family of transcription-activating K4 HMTase complexes (MLL1, MLL2 and SET1 in humans) share a common set of core elements, including Ash2L, two WD40-repeat proteins (WDR5 and RbBP5) and a SET domain polypeptide¹⁸. Whereas WDR5 knockdown produces a genome-wide deficiency in H3K4 trimethylation¹⁹, MLL1 knockdown does not (Y. Dou and R.G. Roeder, Rockefeller University, personal communication). WDR5 is essential to vertebrate development and hematopoiesis¹⁹, and deficiency phenotypes are analogous, at least in part, to those observed in Mll^−/− and MllΔSET mice^20,21, probably owing to a common Hox gene regulation defect. Thus, WDR5's function is not restricted to the MLL1 complex, although participation in MLL1 complexes seems to be important in organismal viability.

The molecular mechanisms of WDR5-dependent gene activation have only recently begun to be elucidated. Although it has been demonstrated that WDR5 preferentially associates with histone H3 dimethylated at K4 (H3K4me2)¹⁹, the mechanism by which this binding preference results in global K4 trimethylation remains obscure (Fig. 1a). To understand how WDR5 recognizes its substrates and to begin to describe a more precise molecular mechanism for the function of this subunit in the MLL1 complex and in a variety of other SET1-family complexes, we undertook the structural and biochemical characterization of human WDR5 and histone H3 peptides. A very recent report of WDR5 in complex with an H3K4me2 peptide has provided initial characterization of one of the five complexes reported here²². Nevertheless, only the complete set of structures and quantitative binding measurements presented herein permit insight into a more complex mechanism of K4 methyl-form discrimination by WDR5 than has previously been suggested²². Our exhaustive set of structures also suggests a previously uncharacterized role for WDR5 in MLL1 and related methylation complexes: WDR5 binding to histone H3 appears to present the K4 side chain for further methylation.

Figure 1.

The overall structure of WDR5 complexes with histone H3. (a) The set of possible methylation states of H3K4. In general, nucleosomes di- and trimethylated at K4 of H3 are associated with transcriptional activation. (b) The overall structure of WDR5 in complex with H3K4me2. WDR5 is a canonical seven-bladed β-propeller (red); the histone H3 peptide binds across the top face of the protein (green, with amino (N′) and carboxyl (C′) termini and R2 labeled). (c) Electrostatic surface of WDR5 contoured from −8 to +8 kT using ABPS⁵² shows a series of acidic patches on the upper face of WDR5 (arrows 1–4), whereas the remainder of the protein is highly positively charged. H3K4me2 peptide is shown in green; potential aromatic cage residues are situated around arrow 5. Remaining arrows indicate patches of positive electrostatic potential discussed in the text. Cutaway view of this surface along the pseudo seven-fold symmetry axis (top) demonstrates how deeply buried R2 is in the cavity of the β-propeller fold's torus.

RESULTS

Structures of WDR5–histone H3 complexes

N-terminally truncated WDR5 was crystallized alone and in complex with a series of histone H3 peptides bearing each of the possible methylation states at K4 (Fig. 1a). The five structures of the unliganded and peptide-bound WDR5 were solved to 2.2- to 1.5-Å resolution by molecular replacement, using the C-terminal domain of Tup1 as the search model²³. All six crystal forms have two nearly identical WDR5 protomers per asymmetric unit (Supplementary Fig. 1 online), with the exception of our additional H3K4me2 crystal complex (form II), which contains only one, and therefore below we depict only one of each protomer pair. There is also extensive similarity shared by the unliganded protein and the series of four peptide-liganded structures (Supplementary Fig. 1); this is discussed in further detail below. The use of histone peptides of differing lengths yielded crystals of varying quality, with different space groups and unit cell dimensions (data not shown). Neither different packing environs for the surface-bound peptide nor differing lengths of bound peptide produced any discernable changes in the bound peptide conformation.

WDR5 has the expected canonical β-propeller fold made up of seven WD40 repeats (Fig. 1b). The β-propeller has an overall shape of a cone truncated at the apex, with a water-filled cavity running through the center along the seven-fold symmetry axis (Fig. 1b,c). The majority of the protein surface is positively charged, owing to the presence of numerous lysine and arginine residues, but a few patches of negative charge are found on the upper face of the protein in the orientation shown in Figure 1c.

For all five liganded crystal forms, omit maps calculated from high-resolution datasets revealed unambiguously interpretable electron density for the histone H3 peptides situated across the narrow end of the β-propeller cone (Fig. 1b,c). The N-terminal 6–8 amino acids of the peptide ligand could be easily modeled into the excellent density, whereas the C-terminal 1–3 residues were less ordered and therefore were not modeled (Fig. 2a). Indeed, the diffraction data were of sufficiently high quality to enable direct observation of the methyl groups on K4 in the various methylated forms, and even to distinguish the trigonal geometry about the Nε of K4 in the dimethyl form from the planar geometry of the similarly sized guanidinium group of R2 (Supplementary Fig. 2 online).

Figure 2.

Bound histone H3 peptide electron density and conformation. (a) Stereo view of the H3K4me2 peptide model in a simulated annealing omit map contoured at 1.0 σ about the peptide, calculated from the protein model and the 1.5-Å data set (form I). (b) Main chain superposition of all of the WDR5-bound peptides. Nitrogen atoms are colored in blue, oxygen atoms in red, and carbon atoms are colored differently for each peptide as indicated in key. Points of conformational divergence are localized to the K4 and Q5 side chains.

Histone H3 peptide recognition

Histone H3 peptides are bound in a nearly identical manner throughout the entire series of liganded structures, with the minor differences being localized to the K4 side chain conformation (discussed below) and to the C-terminal end of the peptide, where the interpretable electron density extended furthest in the structures having the highest resolution. Main chain superpositions of the peptides alone (Fig. 2b) closely resemble superpositions generated using the entire WDR5 complex (Supplementary Fig. 1), indicating that the overall binding mode of the peptide to WDR5 is not substantially affected by the methylation state of K4. This being the case, we will first detail the conserved features of peptide recognition and then discuss the differences.

The first three residues of the peptide N terminus (A1, R2, T3) adopt a backbone conformation best described as a truncated helix, with dihedral Φ and Ψ angles characteristic of α- and 3₁₀-helices, but interrupted at the C-terminal end so that only a single intramolecular i → i + 4 hydrogen bond is observed (that between the A1 carbonyl oxygen and the K4 amide proton, Fig. 3a). An ordered water molecule is involved in a bridging intrapeptide contact between the Nα and the Oγ-hydroxyl of T3. The A1 carbonyl also participates in a hydrogen bond to another ordered water molecule that seems to serve as the hub of a hydrogen-bonding network conserved throughout the liganded structures; this water also spans the Oγ-hydroxyl of T6 and the K4 carbonyl (Fig. 3a). This set of interactions serves to orient the main chain of K4, enforcing the projection of the side chain outward from the body of WDR5.

Figure 3.

Peptide recognition by WDR5 and conformational changes upon peptide binding. (a) The peptide is recognized by an elaborate series of direct and indirect contacts. Orientation of the peptide–WDR5 complex is the same as in the lower panel of Figure 1c. The majority of direct contacts from WDR5 are made to the N terminus and the first three residues. These residues adopt an approximately helical main chain conformation, with one hydrogen bond between the A1 and K4 backbone. Water-mediated contacts are important in recognition of the C-terminal residues of the peptide, as all waters shown (red spheres) are conserved among the peptide-bound structures. Tyr191 apparently acts as a central platform in this peptide-bound water network. (b) Phe133 and Phe263 form an aromatic sandwich about the R2 guanidinium moiety, equatorially flanked by a number of backbone carbonyl–mediated hydrogen bonds. These tight hydrogen bonds are thought to impart specificity for arginine over dimethyllysine, particularly the one from Nε of R2 to the Ser91 carbonyl. (c) Apparent coordinated movement of Phe133 and Phe149 to form the top of the aromatic sandwich recognition element when peptide is bound. The relevant apostructure side chains are depicted in gray and a representative liganded structure (H3K4me2 complex I) is in crimson. (d) Retraction of the loop bearing Lys259 causes a reorganization of the residues lining the central cavity, which permits tight R2 coordination. Coloring is as in c. This movement may be driven by a steric clash between this lysine and the incoming peptide Q5 side chain.

The histone H3 N terminus appears to be specifically recognized by WDR5 through a series of hydrogen bonds from the α-amino group of A1 to the Asp107 and Ser91 side chains (Fig. 3a). The hydrophobic side chain methyl group of A1 appears to be specifically recognized through van der Waals interactions with the side chains of Tyr131 and Phe149. This methyl-recognition pocket appears to be large enough to accept a methyl group, but too small to accommodate any larger side chain without engendering a severe steric clash. The guanidine-bearing side chain of R2 extends deep into the acidic end of the WDR5 central cavity and creates multiple specific intermolecular interactions. The top and bottom faces of the R2 guanidinium moiety are framed in a sandwich-like configuration of cation-π and π-π stacking interactions by Phe133 and Phe263 (Fig. 3b). A further layer of arginine-specific recognition is contributed by interactions with the edge of R2: main chain carbonyls contributed by the N-terminal strand (A-strand) of five of the seven β-propellers directly or indirectly hydrogen-bond to every nitrogen in the R2 guanidinium functional group. The indirect hydrogen bonds are mediated by two structurally conserved water molecules (Fig. 3b). This convergence of backbone amides on a single moiety of a bound ligand is reminiscent of the oxyanion hole of serine proteases, but in such proteases, it is the opposite ends of the amide dipoles that converge on the developing anion²⁴. T3 is locked in place by a short water-mediated contact with the peptide N terminus, while the γ-methyl group projects into a hemispherical hydrophobic pocket comprised of the side chains of Ala47, Ala65 and Leu321 and the aliphatic methylene of Ser49 (Fig. 3a). This extensive array of interactions with the N-terminal end of the H3 protein provides an ample basis for specific recognition of that sequence over other N-terminal peptide sequences present in the cell.

Apparent in our structures, but absent in the recently reported WDR5–H3K4me2 complex²², are the apparently specific contacts that give rise to recognition of the backbone and side chains of Q5, T6 and R9. We attribute this additional detail to the enhanced resolution of the datasets. Though the newer structures suggest that the first three histone H3 residues are more specifically recognized than those that follow C-terminally, there are nonetheless sufficient numbers and types of contacts to residues 4–9 to impart additional binding energy and specificity. Indeed, the structure suggests that WDR5 recognizes a longer stretch of bound peptide than most other histone modification–recognition modules.

Many of the interactions between the peptide and protein consist of waters conserved throughout our liganded structure series that mediate H3 peptide-protein backbone contacts via bridging hydrogen bonds. As these waters are present in each of the peptide-bound structures and largely absent or substantially shifted in the apoprotein structure, we consider them to be an important part of peptide recognition (Fig. 3a). Notably, these water-promoted contacts are almost exclusively restricted to recognition of the peptide backbone and oxygen-bearing side chains; in particular, the backbone of peptide residues 4–7 forms a central spine of hydrated peptide. Specificity for the latter portion of this sequence is imparted by direct and indirect contacts to Q5 and T6 side chains by Tyr260 and Tyr191, respectively (Fig. 3a). By contrast, the A7 side chain is recognized by a shallow hydrophobic cleft composed of Leu234 and an aromatic face of Tyr191, analogous to the A1 side chain recognition mode (Fig. 3a). R8 forms a bidentate hydrogen bond with the backbone carbonyl of Asn214, whereas the Ne contributes a hydrogen bond to the carbonyl of Tyr191. A large patch of acidic residues is present in the vicinity of this side chain, as evident in the surface electrostatic potential renderings (Fig. 1c, arrow 3), and the acidic patch presumably forms favorable coulombic interactions with the R8 guanidinium group. Partial occupancy at this site is still apparent in the structures in which this residue is less ordered, although another conformer that positions the R8 and K9 near a different area of negatively charged surface also exists, confounding unambiguous interpretation (Fig. 1c, arrow 2).

Conformational changes upon binding

The unliganded structure is quite similar overall to the four peptide-bound species, with an average main chain r.m.s. deviation of 0.269Å that is comparable to the respective coordinate errors of each model (Supplementary Fig. 1). That stated, a few differences in WDR5 conformation between the bound and free structures are worth noting, and all of these are localized to the peptide-binding interface. In the bound structures, two aromatic residues flip down into the cavity relative to the apo form: Phe133 forms the top of the aromatic sandwich involved in the recognition of R2, and Phe149 stacks edge-to-face with Phe133 (quadrapole-to-quadrapole stacking) to stabilize the sandwich (Fig. 3c). Consistent with this important role in R2 recognition and their putative coordinated function, alanine substitutions at either position produce similar H3 peptide binding deficiencies, with mutation of Phe133 most severely attenuating binding (Fig. 4c). A loop comprising residues Lys259 to Cys261 shifts distal to the seven-fold symmetry axis, retracting Lys259 from a potential steric clash with the incoming peptide Q5 side chain and concomitantly freeing the water-filled central cavity for R2 guanidinium binding. This loop movement also shifts the carbonyl of Cys261 0.51Å closer to the Nη1 of R2, presumably strengthening this hydrogen-bonding interaction in the complex (Fig. 3d). However, this is the only portion of the R2 recognition pocket aside from Phe133 that shows appreciable movement, suggesting that this structural element is largely static. When the R2 pocket is vacated, a new side chain rotamer of Cys269 is established that forms a hydrogen bond with Tyr191's hydroxyl (Fig. 3d). The subtle withdrawal of Tyr191 from the pocket upon peptide binding permits a direct hydrogen bond to the T6 side chain in addition to stabilizing the conserved water network found in the liganded structures (Fig. 3). Mutation of this residue leads to an approximately ten-fold loss of binding energy for the H3K4me2 species (Supplementary Fig. 3 online), consistent with an important but nonessential role in peptide binding.

Figure 4.

Peptide binding by WDR5. (a) Peptide injections over immobilized WDR5-biotin. Reference and blank traces were subtracted and traces from each functionalized flow cell (three traces per concentration) were averaged for ease of viewing. Injection concentrations of peptides (residues 1–14): H3 unmodified and H3K4me3, 0.5, 0.75, 1.0, 2.0, 4.0, 6.0, 10.0 and 12.0 μM; H3K4me1, 1.0, 2.0, 4.0, 6.0, 10.0 and 12.0 μM; H3K4me2, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.75, 0.8, 0.9, 1.0, 2.0, 4.0 and 5.0 μM. The association/injection is from 0 to 200 s and the dissociation/buffer flow is from 200 to 460 s. The more gradual association and dissociation of H3K4me2 indicates both a slower on and slower off rate than those of the other peptides. Similar experiments with H3_1–9-derived peptide had slightly weaker binding but recapitulated the methyl-form binding trend shown. (b) Equilibrium K_d analysis. Equilibrium K_ds were determined by fitting response saturation fractions calculated by the Langmuir binding isotherm (see Methods). H3 unmodified, K_d 3.3 ± 0.2 μM; H3K4me1, K_d = 8.7 ± 0.3 μM; H3K4me2, K_d = 1.02 ± 0.05 μM; H3K4me, K_d = 7.8 ± 0.2 μM. (c) Single-protein injections of wild-type WDR5 and mutants, all at 750 nM, over immobilized H3_1–20-biotin peptides. Reference and blank traces were subtracted. Consistent with the importance of interactions with the N terminus of the H3 peptide, mutation of Asp107 to alanine caused the greatest decrease in peptide binding among all the mutant proteins examined. Mutation of Tyr131 to alanine leads to slightly enhanced binding of the H3 peptide, suggesting that even a methyl group is slightly larger than optimal for the A1-binding pocket. By contrast, the F149A mutant showed greatly decreased H3 peptide binding, perhaps owing to Phe149's additional role in stabilizing the R2-binding pocket.

Although most protein modules that bind the H3 N terminus impart a fairly extended peptide conformation proximal to K4, with Ψ-Φ angles within the β-strand region of Ramachandran space^25–28, H3 peptides bound to WDR5 are quite deformed. Notably, the peptide conformation is similar to those found in complexes with SET domains^29,30. However, a shallow binding cleft in the MLL1 SET domain would be required to accommodate the composite surface of a WDR5-presented peptide without steric infractions upon complex formation.

Measurement of peptide binding

To understand the energetic contribution made by various contacts between WDR5 and the H3 peptide and the dependence of binding on the methylation state at K4, we performed detailed binding analyses using surface plasmon resonance, formatted using chip preparations having either WDR5 or the H3 peptide immobilized. With immobilized full-length protein, equilibrium analysis and kinetic-parameter fitting of the association and dissociation rates furnished similar K_d values for the series of peptides binding to the wild-type protein, validating the measurements (Fig. 4a,b). The assays with immobilized peptides were performed with wild-type and mutant proteins to assess the roles of residues deemed crucial for peptide binding on the basis of the structure (Fig. 4c). In these experiments, exact numerical analysis is confounded by moderate nonspecific adhesion of WDR5 to the dextran-streptavidin matrix; nevertheless, qualitative comparison of affinities can be made on the basis of the data.

Our binding studies show that the initially reported specificity of WDR5 for H3K4me2 is not as pronounced as that measured by peptide pull-down assays¹⁹. However, the general trend of binding is recapitulated: H3K4me2 binds WDR5 most strongly, with a measured K_d of 1.02 ± 0.05 μM, whereas the analogous monomethylated (H3K4me1) and trimethylated (H3K4me3) peptides bind more weakly (K_d = 8.7 ± 0.3 μM and 7.8 ± 0.2 μM, respectively). However, substantial binding to unmodified H3 peptide is detected, with a K_d of 3.3 ± 0.2 μM. The minor disparity from the previously reported trend¹⁹ is not surprising, as peptide pull-down experiments are qualitative measurements dominated by the k_off term (due to resin-washing steps), and therefore they may amplify small differences in apparent affinity. Overall, the affinities measured here are similar: the difference between the most tightly and the most weakly binding peptide is only about eight-fold, indicative of very modest binding-affinity differences, on the order of 1.2 kcal mol⁻¹.

Differences in the kinetics of binding suggest that recognition of the K4 methylation state and decomposition of the bound complex may proceed through some transient intermediate before achieving the final bound form observed in the crystal structures. The rates of association and dissociation for H3K3me2 binding to WDR5 are discernibly unique: relative to the other peptides, H3K3me2 has a much slower k_on, manifested as a slower approach to equilibrium, and a slower k_off, apparent in the sluggish decay of bound peptide signal in the dissociation phase (Fig. 4a).

Discrimination of histone H3 methylation states

Among the complex structures in our series, it is difficult to discern any specific interactions arising as a consequence of differential methylation at K4 that appear to be important for peptide discrimination. This is surprising, given that all other methylated lysine–binding modules that have been studied use an aromatic cage, presumably as a consequence of convergent evolution^25–28. The K4 side chain adopts two general conformers divergent at χ₃ (the torsion angle between Cγ and Cδ), in a manner apparently dependent on the methylation status of K4 (Fig. 5). Perhaps this is a consequence of crystal packing, although the entire methylated series was crystallized under the same conditions, in the same space group, with nearly identical unit cells (Fig. 5). Furthermore, we obtained structures of several other H3K4me2-bound WDR5 complexes crystallized in different space groups that were not appreciably different from the corresponding complex with the 9-residue peptide (complex II and data not shown). Moreover, the conformation of K4me2 in the recently reported structures of this complex²² appears to be identical to that reported here.

Figure 5.

Differences in K4 conformation in the different methylation states. (a–d) The crystal-packing interface relevant to K4 conformation is depicted for the H3K4me3 complex (C2 space group, a), H3K4me2 complex I (C2, b), H3K4me1 complex (C2, c) and unmodified H3 complex (P2₁, d). Peptides are colored as in Figure 2b; the principal WDR5 protomer is red; and the symmetry-related protomer at the peptide interface is gray. Note the rotation about χ3 moving from the tri- and dimethylated species to the monomethylated and unmodified species. In H3K4me2 complex I, the distances between the ζ-methyl carbons and the Glu322 carboxylate Oδ1 are 3.27 Å and 3.37 Å for the closest methyl group in each of the two complexes per asymmetric unit, whereas these distances are 3.83 Å and 3.87 Å for the more distant methyl group. For comparison, the previously reported shorter distances for these measurements were 3.15 Å and 3.42 Ų².

The recent report on WDR5 in complex with H3K4me2 peptides suggested unconventional hydrogen bonds between the ζ-methyl groups and Glu322 as the source of discrimination among the methylation states of K4 (ref. 22). The distances measured in this 1.9-Å structure between the methyl carbons and the Glu322 carboxylate oxygen fell slightly within the sum of atomic van der Waals radii, and these groups were therefore assumed to be repulsive in the absence of the putative hydrogen bonds. In our 1.5-Å H3K4me2 structure, these distances are longer than those previously measured (Fig. 5b). Furthermore, the interceding proton between the postulated donor and acceptor atoms would be 2.49–2.57Å away from the Oδ2, a distance longer than the average H···O hydrogen bond distance³¹. According to a calculation with the AMBER force field (ignoring the attractive force of a putative CH···O hydrogen bond), there is still an attractive potential resulting from a NH·m=.·O hydrogen bond and salt bridge that outweighs van der Waals repulsion at the distances in our structure³². Thus, even in this intimate distance regime, there is still modest attraction, suggesting that unconventional hydrogen-bonding is not required to explain the observed distances. Regardless, the K4 side chain of H3K4me2 forms more extensive intracomplex interactions than that of any other peptide, generating a hydrogen-bonding network between the ammonium-proximal waters and Glu322 (Fig. 5b).

In the H3K4me1-bound WDR5 structure, the K4 conformer in the crystal has a different χ₃ angle from that found in the other two methylated peptide structures, although there is ~ 15% occupancy of the conformer observed for K4me2 and K4me3. Consequently, there is no substantial interaction between the K4me1 side chain and Glu322 in this structure, despite predictions that one unconventional hydrogen bond should exist (Fig. 5c)²². Rather, there are contacts from the K4 side chain to Glu151 and a backbone carbonyl of a symmetry-related protomer in the crystal (Fig. 5c). The unmodified peptide is found in a nearly identical conformation in a different space group, although this packing arrangement produces a similar set of contacts (Fig. 5d).

With H3K4me3, the Glu322 side chain retracts from what would be a very repulsive Cζ-Oδ1 distance of 2.76 Å between this side chain and the third methyl group. This conformation of the Glu322 side chain places the carboxylate so that the shortest distance between Oδ2 and the two H3K4 ζ-methyl groups is 3.75 Å (Fig. 5a), longer than the putative range of CH···O hydrogen bonds ³¹. Thus, the interactions predicted by ref. 22 appear not to exist in the H3K4me3 structure. Rather, the distances observed are consistent with a solvent-exposed salt bridge.

DISCUSSION

The role of WDR5 in SET1-family complexes

A paradox is apparent in the previous WDR5 literature: WDR5 binds preferentially to histone H3 dimethylated at K4, yet somehow affects global trimethylation at the very same residue¹⁹. This paradox is resolved, at least in part, by the present data coupled with the report of Dou et al.³³ The most noteworthy result of structural analysis is that K4 of the peptide appears to be presented for recognition by another component of the MLL1 complex, and the binding data confirm that dimethylation at this position is modestly recognized. This structure-derived hypothesis is further validated biochemically in the paper of Dou et al.³³, which shows that WDR5 mutations that attenuate H3-binding activity, devised by inspection of the present structures, decrease the HMTase activity of the MLL1 core complex when incorporated in place of the native WDR5.

The significance of the modest differences in H3 peptide binding to isolated WDR5 remains unclear; however, in the context of histone methylation complexes such as MLL1, WDR5 clearly associates with nucleosomes bearing di- and trimethylation at H3K4 (ref. 19). This strong preferential association is the result of either additional binding affinity imparted by some other complex module or processive turnover by the SET domain acting on the presented K4. There is precedent for SET domains having preferences for the methylation state of their lysine substrates manifested as K_m effects³⁴ and k_cat effects^29,35 such that further discrimination of the methylation form could be accomplished by the SET domain in its binding and turnover. The recombinant MLL1 SET domain (like several other isolated SET proteins³⁶) has very poor HMTase activity, particularly in performing the di-to-trimethylation reaction¹². Yet the addition of WDR5 does not enhance the HMTase activity of recombinant MLL (data not shown). Thus, other complex components must be important both in modulating catalysis and in amplifying the modest methyl-form discrimination in WDR5's intrinsic binding activity.

Global H3K4 trimethylation is lost upon WDR5 knockdown in animals¹⁹, yet the MLL1 complex may only be present at a fraction of actively transcribed promoters³⁷. Thus, it is likely that WDR5 has a ‘peptide presentation’ role in most if not all HMT complexes involved in H3K4 trimethylation.

Methyl-form discrimination by WDR5

Although we cannot rule out the possibility that unconventional CH···O hydrogen bonds have some role in determining methyl-form specificity²², this explanation does not satisfactorily explain the observed trend of binding affinity and is inconsistent with our mutagenesis studies. If noncanonical hydrogen bonds (from the methyl groups of the more highly methylated forms of K4 to Glu322) were the key determinant of the observed binding specificity, a loss of methyl-form preference would be expected upon mutation of this residue. Although the E322A mutant binds all forms of the peptide more weakly than wild-type WDR5, identical preferences for the different methylated states are retained (Fig. 4c). Moreover, according to the above hypothesis, more highly methylated forms of K4 should bind more tightly than the unmodified form²², so that affinity decreases monotonically with methylation. Our binding data do not show such a simple trend.

In the absence of a convincing structural basis for the established preference for dimethylation at K4, this binding preference may be attributed to a modest kinetic discrimination for H3K4me2, manifested as a slower k_off relative to other methylated forms of the peptide (Fig. 5a). This modest preference may be imparted by a transient binding intermediate that collapses to the thermodynamic minima under saturating peptide concentrations on a crystallographic time-scale, so that the crystallized structures do not show this transient intermediate state. Perhaps the H3K4me2 preference results from binding of the methylated lysine by an aromatic cage–like structure, analogous to recognition motifs observed in all other methyllysine-binding modules^25–28. It is tempting to speculate that the aromatic residues Tyr131, Phe133 and Phe149 near the final resting place of the K4 side chain have this role (Fig. 1c, arrow 5), but this is not clear from the mutagenesis, given the additional functions of these residues in R2 recognition. The presence of negative surface potential patches proximal to the observed peptide-binding groove also could represent intermediate association sites for this polycationic peptide (Fig. 1c, arrows 1–4). Furthermore, the positions of waters in liganded structures relative to the apo form indicates that appreciable solvent rearrangement must occur between these two limiting states. Perhaps differential water rearrangement or differences in the entropy of freeing waters contribute to the overall modest preference for H3K4me2. Any of these factors could produce a ~ 1.2 kcal mol⁻¹ difference in affinity without apparent structural basis.

All known SET domains perform methyltransferase chemistry within a deep cavity^29,30. Thus, WDR5 would not be expected to interrogate the methylation state of the K4 side chain and simultaneously present this same moiety for SET-mediated methylation. Regardless of how the modest kinetic selection is actually achieved, it enables the dual functions of H3K4 recognition and presentation by WDR5 to occur sequentially. Moreover, the somewhat promiscuous binding with regard to K4 methylation also permits, albeit to a lesser extent, the presentation of less highly methylated and unmodified forms of H3 for methylation. Thus, we envision the WDR5 presentation of H3K4 and its subsequent methylation to be sensitive to, yet not wholly dependent upon, preexisting K4 methylation. It is noteworthy that a slight reduction in monomethylation of K4 accompanies the abrogation of global K4 trimethylation upon WDR5 knockdown¹⁹, consistent with the measured peptide affinity trend.

WD40-repeat proteins in histone modification

WD40-repeat proteins that are coupled with SET domain proteins may represent a common mechanism for substrate selection and presentation preserved among HMTase complexes that methylate other histone positions, even those that promote the lysine methylation events that cause DNA silencing. EZH2, the archetypal SET domain protein involved in silencing, does not show HMTase activity outside of a complex that also contains two WD40-repeat proteins, EED and RbBP4 (also called RbAp48), the latter of which has been shown to bind histones^38,39.

There is good reason to anticipate that peptide recognition and presentation by WD40-repeat proteins is a more general phenomenon extending to other post-translational modification complexes. There are a number of WD-repeat domain proteins that physically and genetically interact with histone-modification catalytic proteins. Examples of such WD40-repeat proteins include Groucho proteins, which bind the histone H3 N-terminal tail in Rpd3–H3–HDAC complexes⁴⁰; RbBP4 and RbBP7, which are required for the human Hat1 acetyltransferase complex activity and bind the complex's histone H4–tail substrate³⁹; and the HIRA histone H3 chaperone, which associates with HDAC-1 in several complexes⁴¹. From an evolutionary standpoint, there is some logic and efficiency to the notion of delegating specific substrate recognition and catalysis to separate components: recognition modules could evolve without the constraint of maintaining catalytic power and combinatorially associate with alternative catalytic subunits. Although it is intriguing to speculate that these WD40-repeat proteins may serve similar histone peptide–recognition and substrate-presentation roles in their complexes, attribution of these functions awaits further study.

METHODS

WDR5 protein and histone peptide preparation

WDR5 was PCR-amplified from MGC clone 1025 and inserted into pMCGS7 by ligation-independent cloning to furnish tobacco etch virus protease–cleavable N-terminal His₆ fusions corresponding to amino acid residues 2–335 (full length) and 23–335 (ΔN-WDR5). For the surface plasmon resonance experiments, a sequence-specific BirA biotinylation sequence (SGLNDIFEAQKIEWH) was fused N-terminally to the full protein (b-WDR5) for endogenous biotinylation at the lysine indicated (bold). Mutagenesis to make S91K, D107A, Y131A, F133A, F149A, Y191F, F219A, E322A and combinations thereof was performed with the QuikChange kit (Stratagene). All proteins were expressed in the Rosetta 2(DE3)PlysS expression host (Novagen) and purified to homogeneity on a series of nickel–nitrilotriacetic acid (Qiagen), Heparin Hitrap, then Superdex 75 chromatography columns, with intervening tobacco etch virus protease cleavage of the His₆ tag. An additional monomeric avidin (Pierce) chromatographic step was used to isolate biotinylated b-WDR5 before gel-filtration chromatography in GF buffer (50 mM Tris-HCl, 500 mM NaCl and 10 mM β-mercaptoethanol (βME) (pH 7.4)). Histone H3 peptides (1–9, 1–14 and 1–20-Y-biotin) were prepared by the conventional Fmoc-SPSS method (with protected amino acids purchased from EMD and Bachem), purified to >95% homogeneity by reverse-phase HPLC, then characterized and quantified by LC-ESI mass spectrometry and amino acid analysis. The protein and peptide preparation for H3K4me2 complex II and the unmodified peptide complex is detailed in Supplementary Methods online.

Crystallization and data collection

ΔN-WDR5 crystals (250 μm × 200 μm × 50 μm) were grown by hanging drop vapor equilibration in Nextal plates as follows: 1 μl of 10–15 mg μl⁻¹ protein solution (10 mM Tris-HCl (pH 7.4), 50 mM NaCl and 10 mM bME) were mixed with 1 μl of well solution composed of 50 mM HEPES (pH 7.5), 100 mM potassium formate and 10%–20% (w/v) polyethylene glycol 3,350 and equilibrated at room temperature overnight against 1 ml of well solution. Upon equilibration, cat whisker streak seeding from smaller crystals grown in higher PEG concentrations consistently provided diffraction-quality crystals. The complexes of WDR5 with various H3 peptide methylation forms were grown similarly, by premixing 3 molar equivalents of peptide in distilled water with 1 ΔN-WDR5 equivalent just before hanging drop setup. Crystals of the H3K4me2 complex II and the unmodified peptide complex were grown under different conditions, detailed in Supplementary Methods. Truncation of the first 23 amino acids of WDR5 in the recombinant protein was requisite for effective crystallization; this region of the protein is highly divergent among higher eukaryotes and is predicted to be unstructured¹⁹, and it therefore is anticipated to be dispensable for function. All data was collected at 100 K at the locations described in Table 1.

Table 1.

Data collection and refinement statistics

	Apo-WDR5a	Unmodified H31–15–WDR5 complexb	H31–9K4me1–WDR5 complexc	H31–9K4me2–WDR5 complex Ic	H31–9K4me2–WDR5 complex IId	H31–9K4me3–WDR5 complexd
Data collection
Space group	C2	P2 1	C2	C2	C2	C2
Cell dimensions
a, b, c (Å)	134.6, 46.6, 112.4	64.9, 47.0, 101.8	133.5, 46.5, 112.1	134.7, 46.1, 112.7	100.3, 46.8, 64.8	134.3, 46.4, 111.4
α, β, γ (°)	90.0, 117.2, 90.0	90.0, 107.0, 90.0	90.0, 116.9, 90.0	90.0, 117.1, 90.0	90.0, 107.1, 90.0	90.0, 116.7, 90.0
Resolution (Å)	50.0–1.79 (1.86–1.79)e,f	20.0–2.25 (2.33–2.25)e,f	50.0–1.89 (1.97–1.89)e,f	50.00–1.50 (1.55–1.50)e,f	20.0–2.10 (2.17–2.10)e,f	50.00–1.88 (1.95–1.88)e,f
R merge	11.9 (16.9)	15.3 (56.5)	7.1 (25.5)	14.9 (46.4)	7.5 (23.6)	9.1 (27.1)
I / σI	10.9 (7.5)	15.8 (3.07)	16.2 (3.8)	9.2 (2.9)	27.2 (5.1)	19.8 (6.6)
Completeness (%)	99.6 (99.8)	100.0 (100.0)	98.9 (92.9)	99.9 (99.6)	98.5 (97.0)	99.4 (95.1)
Redundancy	3.9 (3.5)	6.7 (6.5)	3.2 (2.5)	4.3 (3.5)	6.8 (6.0)	6.9 (5.4)
Refinement
Resolution (Å)	1.78	2.25	1.89	1.50	2.10	1.88
No. unique reflections	57,091	28,238	48,816	98,522	16,940	50,232
Rwork / Rfree	17.0 / 19.2	19.0 / 22.6	14.6 / 19.0	17.9 / 19.8	19.2 / 22.3	15.2 / 18.5
No. atoms
Protein	4,723	4,711	4,735	4,721	2,369	4,705
Peptide	–	84	141	134	44	114
Water	581	297	737	632	192	498
B-factors
Protein	15.3	23.0	16.1	15.8	29.6	16.2
Ligand/ion	–	25.4	24.1	17.6	27.2	21.7
Water	30.7	32.7	32.8	32.4	39.9	33.9
R.m.s. deviations
Bond lengths (Å)	0.012	0.006	0.011	0.009	0.006	0.013
Bond angles (°)	1.38	1.53	1.26	1.29	1.59	1.37

^aData collected at NSLS X-29.

^bData collected at Memorial Sloan-Kettering Cancer Center on a RIGAKU R-AXIS IV⁺⁺.

^cData collected at Advanced Light Source 5.0.2.

^dData collected at APS 24ID-C.

^eHighest-resolution shell is shown in parentheses.

^fSingle-crystal data set.

Structure solution and refinement

Structures were solved using MOLREP⁴² and BEAST⁴³ from the CCP4 suite⁴⁴ and a loop-truncated polyalanine sequence with backbone coordinates derived from the C-terminal tup1 structure as a probe. Initial building and refinement was done in ARP/wARP⁴⁵. Further rounds of simulated annealing, conjugate gradient energy minimization and B-factor refinement, carried out in CNS⁴⁶ and interspersed with manual rebuilding in Coot⁴⁷, furnished nearly complete models. The final stages of refinement were done in Refmac⁴⁸ using TLS refinement⁴⁹, and PRODRG⁵⁰ generated parameter files for the methylated lysine species. Ramachandran statistics are as follows (residues in most favored regions, residues in additional allowed regions, residues in generously allowed regions, residues in disallowed regions): apo-WDR5, 89.8%, 10.1%, 0.0%, 0.2%; unmodified H3_1–15–WDR5 complex, 86.2%, 13.4%, 0.4%, 0.0%; H3K4me1_1–9–WDR5 complex, 88.9%, 11.1%, 0.0%, 0.0%; H3K4me2_1–9–WDR5 complex I, 90.0%, 10.0%, 0.0%, 0.0%; H3K4me2_1–9–WDR5 complex II, 89.8%, 10.1%, 0.0%, 0.4%; H3K4me3_1–9–WDR5 complex, 89.0%, 11.0%, 0.0%, 0.0%⁵¹. All structure figures were generated in PyMOL (http://pymol.sourceforge.net) with the APBS plugin⁵² for surface electrostatics. Superpositions were done using LSQKAB in the CCP4 suite⁴⁴.

Binding studies

All experiments were performed using a Biacore 3000 sensor (Biacore AB). The change in response level is proportional to the change in mass at the surface, as the interacting partner (the analyte) binds during its injection over the surface with bound ligand.

Either b-WDR5 or biotinylated peptide were immobilized on a Sensor Chip SA (Biacore AB) after the surface had been conditioned with three injections of 50 mM NaOH and 1 M NaCl for 1 min at 10 μl min⁻¹. Peptides were immobilized to relatively high levels to see the peptide response over the high nonspecific background: H3K4me1, 497 resonance/response units (RU); H3K4me2, 701 RU; H3K4me3, 675 RU. Site-specifically biotinylated WDR5 was immobilized to 4,000-5,000 RU in flow cells 2, 3 and 4, giving three traces per injection of analyte, with flow cell 1 used as the blank for subtraction. Any free avidin sites were blocked with an injection of 1 mM biotin (in 50 mM phosphate buffer (pH 8.0)) for 1 min at 10 μl min⁻¹. HBS-EP (10 mM HEPES (pH 7.4), 150 mM NaCl, 3 mM EDTA, 0.005% (v/v) surfactant P20) plus 5 mM βME was used as the instrument running buffer. Peptides and proteins were diluted directly into this buffer from a stock in water or GF buffer by at least 500 and 85 times, respectively. The runs were started with six start-up cycles, where HBS-EP plus βME was injected instead of sample, followed by sample-injection cycles, interspersed every three injections with another buffer injection. The peptide analyte injection cycles consisted of a 250-μl sample injection (75 μl min⁻¹), 180 s of buffer flow (dissociation phase), a 40-μl regeneration injection (1 or 1.5 M NaCl in 70%–80% (v/v) HBS-EP plus βME) and a 30-s buffer injection. The unbiotinylated protein analyte injection cycles consisted of a 200-μl sample injection (75 μl min⁻¹), 180 s of buffer flow (dissociation phase), a 75-μl regeneration injection (50 mM NaOH and 1 M NaCl) and a 10-μl regeneration injection (0.25% (w/v) SDS, 25 mM NaOH, 0.5 M NaCl) needed to remove the nonspecifically but tightly bound protein from the SA-dextran surface. The flow cell temperature was 25 °C for all experiments.

Scrubber 2.0 (Center for Biomolecular Interaction Analysis, University of Utah) was used for subtraction of reference and blank data, to align traces and to calculate saturation fractions for use in the equilibrium K_d analysis. KaleidaGraph 3.0 (Synergy Software) was used to fit the equilibrium K_ds, using a modified Langmuir binding isotherm: y = m1 + (1 – m1)(x/(x + K_d)), where y is the fraction of saturation, x is the concentration and m1 allows for curve fitting when the plots do not trend toward zero. Biacore 3000 BiaEvaluation software (Biacore AB) was used for kinetic fits.