Crystals of mouse Cd44-HA complex show two conformations

We have solved three new structures of the murine Cd44 HABD (residues 25–174; mouse Cd44_25–174): the unliganded apoprotein and two HA₈-bound complexes termed type A and type B. For these studies, mouse Cd44 was used because extensive attempts to cocrystallize the human ortholog were unsuccessful.

The fold of the apoprotein mouse Cd44_25–174 is essentially as described for the human CD44 HABD¹⁶. Their superimposition shows the C traces to be extremely similar (Supplementary Fig. 1a online), except for a small kink in the 5-6 loop caused by the insertion of a valine in the mouse protein (see alignment in Supplementary Fig. 1b). The structural similarity between mouse and human CD44 HABDs is reflected in their similar binding affinities for HA₁₀ (K_d 50 M, see Supplementary Table 1 and Supplementary Fig. 2 online), as assessed by isothermal titration calorimetry (ITC).

The overall fold of the mouse Cd44_25–174–HA₈ complex is similar in type A and B crystals. However, the structures differ in the conformation of the 1-1 loop: specifically, type A crystals present a conformation that more closely resembles the apo structure, with less intimate contact between HA oligomer and polypeptide, and correspondingly less burial of surface area. Thus, the type B structure probably represents a higher-affinity ligand-bound state. The structural differences between these two crystal forms and their possible significance are described in more detail below. First, we present the features of the HA contact surface and the bound ligand.

Cd44 contacts only a short region of bound HA octamer

In the bound HA₈ in the crystal complex (illustrated as the type A structure in Fig. 1), electron density is apparent for seven of the eight sugar residues (GlcNAc at position 2 through GlcNAc at position 8, denoted GlcNAc2–GlcNAc8), indicating that the nonreducing terminal sugar (GlcUA1) does not adopt a stable conformation. Overall, the HA conformations and protein contacts are very similar for type A and B complexes, with GlcNAc2 making no obvious contact with the polypeptide and GlcUA3 forming only a tentative association with Arg155. Analysis of the change in accessible surface area of HA₈ upon complex formation suggests that the GlcUA5–GlcNAc8 region forms the core of the Cd44-HA recognition site in both costructures (Figs. 1 and 2): a large temperature factor (42.3 Ų) is found for GlcNAc2 in the crystal A structure, and much smaller ones for GlcUA5–GlcNAc8 (14.0–17.5 Ų). The relatively small sugar footprint is consistent with our earlier finding that maximal chemical shift perturbations and optimal competition of polymeric HA binding can be achieved with an octamer¹⁶. In the present study, we also compared binding of HA oligomers to mouse Cd44_25–174 by ITC and found that HA₈ and HA₁₀ bind appreciably and with similar estimated affinities (Supplementary Table 1 and Supplementary Fig. 2). The slight discrepancy between the length of HA making direct contacts to the protein in the crystal complex and that necessary for effective competition is consistent with previously described 'end effects' caused by HA dynamics¹⁷.

Figure 1: The HA-binding site in mouse Cd44.

(a) Final refined 1.25-Å-resolution 2F_o – F_c electron density for the binding site in the type A crystal complex, calculated using SIGMAA³¹ weighted map coefficients generated by REFMAC³² is contoured at 0.25 e^- Å^-3 (equivalent to the s.d. of the final map). Refined structure is shown as sticks colored by atom type (green, Cd44 carbons; cyan, HA carbons; blue, nitrogen; red, oxygen; yellow, sulfur). Individual sugar rings in the bound HA₈ oligosaccharide are numbered from the nonreducing end. (b) A ribbon diagram of mouse Cd44 (type B complex), with secondary structure identified using the DSSP algorithm³³. Pink, -helices; white, loops; green and gold, -sheets I and II, respectively; cyan, bound HA. (c) Surface representation of the HA-binding site in the type B crystal complex. The shallow HA-binding groove is shown as molecular surface. Gold, supplementary lobe formed from N- and C-terminal Link extensions; cyan, HA. Selected residues marking the boundaries of the groove are labeled. The type A crystal form shows similar features but lacks the lower platform for the HA interaction provided by reorientation of Arg45.

Full size image (96 KB)

Figure 2: Atomic interactions of HA.

(a,b) The core of the HA-binding site (a) and the full binding groove in the type B crystal complex (b) are shown as sticks, colored as in Figure 1a. Dotted lines denote hydrogen bonds (identified as contacts between polar atoms closer than 3.4 Å). Numbers indicate individual sugar rings, as in Figure 1a.

Full size image (82 KB)

HA binds in shallow groove dominated by hydrogen bonds

The GlcUA5–GlcNAc8 residues of HA lie in a shallow groove on the surface of the Cd44 HABD, toward the top of the molecule as viewed in Figure 1. Thirteen residues of the Link module (Arg45, Tyr46, Cys81, Arg82, Tyr83, Ile92, Asn98, Ile100, Cys101, Ala102, Ala103, His105 and Tyr109) make prominent contacts with the oligosaccharide (Fig. 2), and there is a notably large contribution from aliphatic side chains to HA binding (including Ile92, Ile100, Ala102 and Ala103, which are all located in the 4-5 loop). The remainder of the binding groove is lined by aromatic and basic residues (Arg45, Tyr46, Arg82 and Tyr83) brought together from the 1-1 and 2-3 loops, together with Tyr109 from the 5 strand.

The Cd44 protein shows substantial specificity for recognition of HA over other glycosaminoglycans¹⁶. The structure of the mouse Cd44_25–174–HA₈ complex suggests that this specificity arises from molecular recognition of a number of characteristics peculiar to a repeating polymer of GlcUA and GlcNAc. Central to this is recognition of the carboxylate group of GlcUA7 and the N-acetyl group of GlcNAc6 (Fig. 2). The O6A and O6B carboxylate oxygens of GlcUA7 form hydrogen bonds with the main chain nitrogens of Ala102 and Ala103, respectively. Further recognition of this carboxylate is achieved through a direct hydrogen bond from O6B to the hydroxyl oxygen of Tyr83 (Fig. 2a).

The N-acetyl group of GlcNAc6 is also recognized through the formation of multiple interactions. The methyl group sits in a hydrophobic pocket lined by the phenyl ring of Tyr83, the side chain of Ile92 and the disulphide bond between Cys81 and Cys101, and the carbonyl oxygen forms a direct hydrogen bond with the hydroxyl oxygen of Tyr46 and a water-mediated hydrogen bond with the guanidino group of Arg45 (see Fig. 2a,b). The amide nitrogen donates a hydrogen bond to the carbonyl oxygen of Ile100. Main chain interactions of Ile100 with GlcNAc6, and of Ile100, Ala102 and Ala103 with GlcUA7, are facilitated by the unusual hook-like loop conformation of this part of Cd44. Finally, the more hydrophobic surface of the glycosidic ring of GlcNAc6 finds a matching extended hydrophobic surface in the side chain of Ile100 (Fig. 2a).

Further contributions to specific recognition come from sugar moieties that flank this key disaccharide pair. In particular, O6 of GlcNAc8 forms a hydrogen bond with the hydroxyl oxygen of Tyr109, and the O2 and O3 hydroxyl groups of GlcUA5 form hydrogen bonds with the guanidino groups of Arg45 and Arg82, respectively. GlcUA5, and to a lesser extent GlcNAc4, also form extensive hydrophobic and van der Waals interactions with the solvent-exposed side chain of Ile100 and the disulphide bond between Cys81 and Cys101 (Fig. 2a,b and Supplementary Video 1 online).

Notably, the interactions described above may be facilitated by a conformational change in the HA molecule upon binding to Cd44. When the lowest-energy conformation of a free HA oligomer obtained from fiber diffraction analysis (PDB 3HYA)¹⁸ is superimposed upon the bound conformation of the HA octamer in the Cd44 crystal complex, a marked kink is seen in the latter, between GlcUA5 and GlcNAc6 (Supplementary Fig. 3 online), that allows more extensive interactions between the HA and the protein.

Thus, the recognition of HA by Cd44 appears to be driven by hydrogen bonds and the burial of an extended surface area, rather than by the sugar-aromatic stacking interactions expected from the results of site-directed mutagenesis¹⁹, which had suggested key roles for aromatic residues Tyr42, Tyr79 and Tyr105 in human CD44. Although their murine equivalents (Tyr46, Tyr83 and Tyr109) contact the HA oligomer (Fig. 2), none of these residues form stacking interactions, instead making hydrogen bond contacts. In addition, the aromatic ring of Tyr109 appears to stabilize the conformation of the hook region of Cd44 through an interaction with the side chain of Ala102 (Fig. 2a).

HA binds almost exclusively within the Link module

As indicated above, the structure of the mouse Cd44_25–174–HA₈ complex reveals that all atoms in close contact (<4 Å) with the bound octamer come from residues in the Link module. However, previous mutagenesis of human CD44 (refs. 19,20) has suggested that a tract of basic residues (Arg150, Arg154, Lys158 and Arg162) in the C-terminal extension also contributes to HA binding. To address this inconsistency, we reassessed the effect of mutating Arg150, Arg154, Lys158 or Arg162 to alanine on HA binding to human CD44 (residues 20–178; human CD44_20–178) using NMR (Supplementary Fig. 4 online). Accordingly, ¹H-¹⁵N HSQC spectra were acquired for each human CD44_20–178 mutant in the presence of varying concentrations of HA oligomers, as described previously for the wild-type protein¹⁶. Apart from perturbations attributable directly to the mutation, the spectra of R150A, R154A, K158A and R162A were essentially the same as that of the wild-type hCD44_20–178 in the presence of either HA₆ or HA₁₀, suggesting that these mutations affect neither the strength nor the molecular details of HA binding to CD44.

We also assessed binding of these C-terminal extension mutants to high–molecular weight HA by surface plasmon resonance (SPR). Wild-type human CD44_20–178 yielded a K_d of 20 M, in good agreement with the value of 27 M reported previously²¹ and the K_d of 13 M determined by ITC for the interaction of mouse Cd44_23–174 with HA₂₆ (see Supplementary Table 1). The R150A mutant had an approximately two-fold lower affinity (K_d = 41 M; Supplementary Fig. 5a online), whereas R154A, K158A and R162A all bound more tightly (K_d values of 11, 8 and 7 M, respectively). This indicates that these residues are not involved to any appreciable extent in mediating HA binding, which is consistent with our structural studies.

Finally, we considered the possibility that a role for residues outside the Link module might be apparent only in native glycosylated Cd44 rather than in the bacterially expressed HABD. To test this, we introduced equivalent mutations into full-length mouse Cd44 expressed on the surface of human 293T cells and measured their capacity to bind fluorescein-tagged HA (Fl-HA) by flow cytometry. None of the mutants (R155A, R159A, K163A or R167A) showed markedly less Fl-HA binding than the wild-type protein when measured at either high or low Cd44 surface densities (Supplementary Fig. 5b); as seen for their human counterparts, R155A showed a small (about two-fold) decrease in activity, whereas R167A showed enhanced binding. These data conflict with a previous study²⁰, which indicated a functional role for these C-terminal residues in human CD44. However, in that study, mutations of Arg150, Arg154, Lys158 and Arg162 had marked effects on HA binding only in compound mutants. Hence, we conclude that the Link extension does not contribute directly to the ligand-interaction site.

Orientation of a key binding residue differs between crystals

The structure of the mouse Cd44 apoprotein (and the type A complex, with which it is essentially identical) differs appreciably from that of the type B complex with respect to the main chain conformation around residue Gly44. This results in a side chain rearrangement in the type B complex that brings residue Arg45 (equivalent to Arg41 in human CD44) into contact with HA, displaces some of the contacts made by Arg82 and imposes a subtle shift in Glu41 (Fig. 3a–c). In addition, the type B structure shows further small but widespread structural differences from the mouse Cd44_25–174 apoprotein (and the type A crystal complex), particularly in the C-terminal portion of the molecule. These structural differences may be additional consequences of the conformational rearrangement involving residue Arg45, the major effect of which is to increase the surface area of interaction between protein and sugar, particularly on the lower lip of the binding groove (Fig. 3b,c).

Figure 3: Conformational states of Cd44 involved in HA recognition.

(a) Stereo ribbon diagram showing comparison of the structure of mouse Cd44 apoprotein (gold) with the superimposed structure of the type B mouse Cd44–HA₈ complex (green). Bound sugar is shown as sticks, as are both conformations of the amino acid side chains that undergo the most substantial conformational changes (colored as in Fig. 1, but with polypeptide carbons green or gold in the respective structures). (b,c) Comparison of the shapes of the molecular surfaces of crystal complex A (possibly representing a low-affinity state; carbons colored gold) and crystal complex B (possibly representing a higher-affinity state; carbons colored green), respectively. Movement of the loop around Arg45 results in a deeper cleft being involved in HA recognition in the type B crystal complex. (d) R.m.s. differences in main chain conformation that accompany sugar binding. Surface color gradient represents movement: white, residues unaffected by HA binding; red, residues that undergo small movement (0.25 Å); yellow, residues that undergo large movement (0.5 Å). (e) Amino acid residues, equivalent to those of wild-type human CD44, indicated by NMR to experience appreciable changes in chemical environment upon HA binding. Yellow, changes of NH, HN and C atoms (all three together); red, changes of only one of these nuclei; magenta, chemical shift in the resonance of the C' atom; gray, HN and NH atom resonances are not assigned in the bound complex, generally indicating an appreciable change in chemical environment.

Full size image (97 KB)

To better understand the relationship between the conformation of the HABD and its HA-binding status, we investigated the CD44-HA interactions using NMR. First, we reanalyzed earlier data showing that the binding of HA to human CD44_20–178 induces widespread chemical shift perturbations¹⁶. It is now apparent that some of these map to residues that make direct contact with HA. Notably, the majority of the remaining shift changes, which are remote from the interaction site, can be accounted for by a conformational change similar to that distinguishing the type A and type B crystal structures (see Fig. 3d,e). The most likely interpretation of this finding is that HA binding induces a transition between a type A– and a type B–like conformation. A movie showing a morph between the A and B complex structures illustrates the HA-induced transition (Supplementary Video 2 online).

It should be noted that this transition is distinct from the HA-induced changes reported recently in another study²², which inferred extensive reordering of the 0, 8 and 9 strands within the C-terminal extension upon binding HA₆. Such extensive changes are incompatible with the subtle ligand-induced perturbations of the 1/1 loop described here. To further investigate this phenomenon, we assessed whether the same HA-induced perturbations could be detected in an R41A mutant of human CD44_20–178. HSQC spectra for R41A collected at various protein/hexasaccharide ratios showed that the R41A–HA₆ complex, unlike the wild-type protein and C-terminal extension mutants described above, is in fast exchange on the NMR timescale (data not shown), consistent with a substantial reduction in binding affinity (estimated K_d = 1.2 mM; see Supplementary Fig. 6a online). Compared with the wild-type human CD44_20–178 protein¹⁶, relatively few amides in the R41A mutant are affected by HA₆ binding and the magnitudes of the perturbations are much smaller (Supplementary Figure 6b). When these appreciably perturbed amides in the hCD44_20–178 R41A mutant are mapped onto the mCd44_23–174–HA₈ type B crystal structure (Supplementary Fig. 6c), it can be seen that they generally correspond to residues close to the bound HA and, for the most part, do not include residues perturbed in the Cd44 structure as a consequence of the HA-induced conformational change (compare Supplementary Fig. 6c,d with Fig. 3d,e). These data indicate that crystal form B is likely to correspond to the ligand-bound form of Cd44 and further support the notion that HA binding itself might promote transition of the receptor to this state.

Reagents.

HA oligosaccharides (HA₆, HA₈, HA₁₀ and HA₂₆) were produced as described²⁷. Biotinylated HA (bHA) was prepared from Genzyme medical-grade HA (1.2 MDa) as described²⁸. The monoclonal antibody IM7 was obtained from Pharmingen. See Acknowledments for sources of murine Cd44 monoclonal antibodies.

Expression of mouse and human Cd44 HABDs.

For bacterial expression of the mouse Cd44 HABD, a segment of the complementary DNA encoding residues 25–174 was amplified using the forward primer 5'-GGAATTCtcatgaATCAGATCGATTTGAATGTAACCTGCCGC-3' and the reverse primer 5'-GCggatccTCAATCGATGTCTTCTTGGTGTGTTCTATAC-3', each containing a BamHI site (lowercase) for ligation into the expression vector pET19b (Novagen). The polypeptide mouse Cd44_25–174 is equivalent to residues 3–152 of the mature protein, the first two residues, His23-Gln24, having been replaced by Met23-Asn24. N-terminal sequencing indicated that the translation-initiating methionine was retained when the protein was expressed in Escherichia coli. The HABD was expressed and purified to homogeneity as described for human CD44_20–178 (ref. 29). The generation and expression of human CD44 HABD site-directed mutants in E. coli and of full-length mouse Cd44 in 293T fibroblasts are described in Supplementary Methods online, along with the analysis of HA binding using surface plasmon resonance, isothermal titration calorimetry and flow cytometry.

Crystallization of mouse Cd44 HABD and HABD–HA complexes, data collection and analysis.

For the preparation of mouse Cd44_25–174 apoprotein crystals, droplets comprising 200 nl mouse Cd44_25–174 (at a concentration of 0.5 mM) mixed with 200 nl well solution were dispensed as sitting drops and underwent vapor diffusion with a well solution of 30% (w/v) PEG monomethylether 5,000 and 200 mM (NH₄)₂SO₄ in 100 mM MES buffer (pH 6.5). Crystals comprising composites of plate-like sheets grew over the period of 1 week. Cocrystals of the mouse Cd44_25–174–HA₈ complex were prepared after addition of HA₈ (2 mM final concentration) to the protein solution followed by mixing 1:1 with well solutions containing 25% (w/v) PEG 3,350 and 100 mM NaCl in 100 mM HEPES buffered at either pH 7.0 or pH 8.0. Cocrystals grew over a period of 2–8 weeks at room temperature. Crystal parameters are shown in Table 1; full details of analyses and data refinement are included in Supplementary Methods.

Table 1: Data collection and refinement statistics

Full table

NMR spectroscopy.

Samples for NMR were prepared from lyophilized human CD44_20–178 (wild-type, R41A, R150A and R154A at 0.3 mM; K158A and R162A at 0.15 mM) in 10% (v/v) D₂O (containing 0.02% (w/v) sodium azide) and adjusted to pH 6.5. ¹H-¹⁵N HSQC spectra were acquired at 500 MHz in the absence or presence of HA₆ or HA₁₀ at oligosaccharide/protein stoichiometries of 0:1, 1:1, 2:1 and 4:1, as described for the wild-type HABD¹⁶. The R41A mutant was titrated only with HA₆, at 0:1, 1:1, 2:1, 4:1, 6:1, 10:1 and 18:1 ratios. Here, the chemical shift of the HN nuclei of residues 80, 94, 97 and 110, and the NH nuclei of residues 97 and 110, were measured at each ligand concentration and normalized to a maximal value of 1, and the average was determined. These data were fit in MicroCal Origin by a sum of least-squares iterative improvement as described³⁰, using the equation.

Accession codes.

Protein Data Bank: Coordinates have been deposited with accession codes 2JCP (apo mouse Cd44), 2JCQ (type A crystal of mouse Cd44–HA₈ complex) and 2JCR (type B crystal of mouse Cd44–HA₈ complex).