An Activating Mutation Reveals a Second Binding Mode of the Integrin α2 I Domain to the GFOGER Motif in Collagens

The GFOGER motif in collagens (O denotes hydroxyproline) represents a high-affinity binding site for all collagen-binding integrins. Other GxOGER motifs require integrin activation for maximal binding. The E318W mutant of the integrin α2β1 I domain displays a relaxed collagen specificity, typical of an active state. E318W binds more strongly than the wild-type α2 I domain to GMOGER, and forms a 2:1 complex with a homotrimeric, collagen-like, GFOGER peptide. Crystal structure analysis of this complex reveals two E318W I domains, A and B, bound to a single triple helix. The E318W I domains are virtually identical to the collagen-bound wild-type I domain, suggesting that the E318W mutation activates the I domain by destabilising the unligated conformation. E318W I domain A interacts with two collagen chains similarly to wild-type I domain (high-affinity mode). E318W I domain B makes favourable interactions with only one collagen chain (low-affinity mode). This observation suggests that single GxOGER motifs in the heterotrimeric collagens V and IX may support binding of activated integrins.


Introduction
The integrins are heterodimeric adhesion receptors, with 24 permitted pairings selected from 18 a and 8 b subunits [1]. Integrin a2b1 is the best-researched of the four collagen-binding integrins, which share with the closely-related leukocyte b2 integrins the presence of an autonomously-folding VWF A, or inserted (I), domain that contains the primary ligand-binding site [2]. The I domain, which adopts the Rossman-fold [3], is located between blades 2 and 3 of the a subunit b-propeller and believed to interact at its base with the b subunit I-like domain, a similar structure that serves as the primary ligand-binding site in integrins without an a subunit I domain. In all integrins, a series of loops at the distal surface of these I domains bind a Mg 2+ ion that is crucially involved in the binding of ligand [4]. This arrangement is known as the metal ion-dependent adhesion site (MIDAS).
The integrins are bi-directional signalling receptors [5]: ligand binding to the extracellular region of integrins signals to the interior of the cell, and integrin affinity for ligand can be increased by stimulation of the cell through other pathways. Upon activation, the gross conformation of the extracellular region of integrins changes from a bent to an upright posture (reviewed in [5,6]), allowing unimpeded access to the ligand-binding head of the receptor. Since integrin ligands are generally macromolecular structures, this is an important aspect of the activation process, but its basis remains incompletely understood.
The I domain conformation is also mobile [7]. The I domain consists of a parallel b-sheet surrounded by a-helices. The elucidation of ligated and free forms of the a2 I domain revealed that, upon ligation, helix 7 travels downwards by ,10 Å , away from the MIDAS [8,9]. This provides a bidirectional conduit for transmission of information. In the high-affinity form of the integrin, a glutamic acid in the linker segment following helix 7 is believed to interact with the b subunit MIDAS [10,11]. A second mobile feature of the I domain is the C-helix, a short a-helix close to the MIDAS that prevents ligation of the resting integrin. The Chelix is stabilised by a salt bridge between an arginine at its Nterminus and a glutamic acid at the top of helix 7 (R288 and E318 in the a2 I domain). Upon ligation (or activation) the salt bridge is broken by downwards motion of helix 7, and the C-helix is remodelled to an additional turn in helix 6 [9].
Conformational change in the a2 I domain was revealed by its co-crystallisation with a collagen-like, triple-helical, peptide of sequence [GPO] 2 GFOGER[GPO] 3 , O denoting hydroxyproline [9]. In this co-crystal, the glutamate carboxylate anion of the GER triplet co-ordinates the bound Mg 2+ ion directly, leading to a rearrangement of the ion's octahedral co-ordination shell. Additional contacts with the a2 I domain are made by phenylalanine and arginine residues of the GFOGER peptide. GFOGER constitutes a high-affinity motif for all the collagenbinding integrins reported to date [12][13][14]. However, cellular activation is reported to be necessary for full binding to related, lower-affinity motifs. The interaction of platelets through a2b1 with short triple-helical peptides containing GMOGER, for example, required activation with ADP before full platelet binding was achieved [15]. Similarly, the activatory monoclonal antibody TS2/16 increased platelet binding to longer triple-helical Toolkit peptides containing GLOGER, GMOGER and other sub-optimal motifs [16]. Platelet adhesion to a GFOGER-containing peptide was not significantly affected in either study, showing that the phenylalanine sidechains of triple-helical GFOGER confer highaffinity binding.
Various strategies have been adopted to produce constitutively active or inactive forms of the a2 I domain. Inactivation has been achieved by mutating the Mg 2+ ligand T221 to alanine, disrupting the MIDAS [16,17], and by introducing a disulphide bridge between helices 1 and 7 such that helix 7 is locked into the lowaffinity (''closed'') conformation (Hamaia et al., in preparation). Activation has been achieved by disulphide-locking helix 7 into the high-affinity (''open'') conformation that is seen in the GFOGER-I domain co-crystal structure [15]. An alternative strategy was used by Tuckwell, who replaced E318 with tryptophan, resulting in a destabilised C-helix and a MIDAS that was more open to ligation [18]. A recent crystal structure of the analogous E317A mutant of the a1 I domain was reported by the group of Heino [19]. In their unligated structure, the C-helix was unwound, but helix 7 remained in the low-affinity conformation.
Here we report a functional and structural characterisation of the a2 I domain E318W mutant. Unlike the wild-type I domain, the E318W mutant binds the GFOGER peptide with 2:1 stoichiometry. Because the three chains of a collagen triple helix are not topologically equivalent, two distinct binding modes are observed crystallographically, one resembling that of the wild-type I domain and the other sacrificing one of the two favourable contacts with the phenylalanine residues of the GFOGER motifs. The structure suggests how activated integrin a2b1 might bind to heterotrimeric collagens harbouring a high-affinity motif in only one of the three chains of the triple helix.

Results
Binding of Integrin a2 I E318W to Immobilised Triplehelical Peptides We previously used a solid-phase binding assay to investigate binding of a2 I E318W to the Collagen Toolkit III, which represents the entire triple helix of collagen III as a set of overlapping peptides [16]. In the present study, we have extended this analysis to the Collagen Toolkit II [20]. Wild-type a2 I domain bound a limited set of peptides from Toolkit II: II-7 and II-8 containing the GLOGER motif, and II-28 containing GFOGER (Fig. 1A). The E318W mutant bound the same three peptides with similar amplitude, and additionally bound (in order of decreasing amplitude) II-23, II-31, II-44, II-24, II-55, II-56, II-20, II-21 and II-18 (Fig. 1B). These emerging positive peptides include the motifs GMOGER, GQRGER and GAOGER, all reported as intermediate or weakly-binding ligands for integrin a2b1 [15].
We also investigated the binding of a2 I E318W to a larger panel of triple-helical Gxx'GEx'' peptides than previously used [15,16] (Fig. 2A). We observed a modest, but significant (p,0.001; 2-way ANOVA) increase in the binding of a2 I E318W across the board, compared with wild-type a2 I domain. In particular, binding of GMOGER stood out as doubling in amplitude (p,0.001; 1-way ANOVA), whilst amongst lower-affinity peptides, GLKGEN, GQRGER and GASGER showed several-fold increases in binding activity, albeit to low absolute amplitudes (p,0.001 for each; 1-way ANOVA). Both wild-type and E318W a2 I domain bound well to GFOGER, GROGER, GLOGEN and GLOGER, all present in collagens II and III, to GLOGEA and GFOGEK, occurring in other collagens, and to GFPGER, engineered into a bacterial collagen-like protein [21]. The relative increase in prominence of the lower-affinity Toolkit peptides in these solid-phase binding experiments may be interpreted as a relaxation in the ligand specificity of a2 I E318W, resembling that of integrin a2b1 in activated platelets [15,16]. In binding experiments in which E318W and wild-type a2 I domains were applied in increasing dose to GFOGER, GMOGER and GAOGER, E318W showed a modest increase in affinity for GFOGER compared with wild-type and a substantially greater increase in affinity for GMOGER and GAOGER (Fig. 2B). The wild-type I domain bound these peptides to a greater extent than the negative control GPP-10 (not shown) only at the highest concentration tested.
Binding of a2 I E318W to triple-helical peptides in solution.
As a prelude to crystallisation trials, we investigated the interactions of the wild-type and mutant a2 I domain with short collagen-like peptides ([GPO] 2 GxOGER[GPO] 3 ,6 = F, M, A) in solution using analytical size exclusion chromatography (SEC). Because the collagen-like peptides do not absorb at 280 nm, they can be added in molar excess without obscuring the elution of the protein peak. The wild-type a2 I domain formed a complex with the GFOGER peptide, as evidenced by a 0.6 ml reduction in elution volume (Fig. 3). The apparent mass of the complex derived from the calibration of the SEC column is 31 kDa, in perfect agreement with the calculated mass of the previously crystallised 1:1 complex [9]. Addition of the GFOGER peptide to a2 I E318W caused a much greater reduction in elution volume (1.4 ml), suggestive of a larger complex containing two mutant I domains. The asymmetric peak profile suggests that some dissociation and re-association was occurring during the SEC run, which is further supported by the lower-than-expected apparent mass of the complex (44 kDa from calibration, 56.1 kDa calculated). The GMOGER peptide formed a stable 1:1 complex with a2 I E318W, but not with wild-type a2 I, and the GAOGER peptide did not interact with either I domain (not shown). We interpret the collective data as a2 I E318W having gained affinity for GMOGER and a second, less favourable, binding mode for GFOGER.

Structure of the a2 I E318W-GFOGER Complex
To reveal the details of the presumed 2:1 complex, we determined the structure of the a2 I E318W-GFOGER complex using crystals diffracting anisotropically to 3.04 Å resolution ( Table 1). The unbiased electron density of the collagen peptide clearly indicated the register of all three chains (Fig. 4A) and the final residual difference density map also showed no evidence of register disorder (not shown). Model building and refinement was carried out with close reference to the high-resolution structure of the wild-type a2 I-GFOGER complex [9] and converged at a free R-factor of 0.298 for a model with excellent stereochemistry (Table 1).
Consistent with the SEC data, the asymmetric unit of the crystals contains one GFOGER peptide interacting with two molecules of a2 I E318W, termed A and B (Fig. 5). The collagenlike peptide is straight and not bent as in the wild-type a2 I-GFOGER complex [9]. The glutamic acid residues of the trailing and leading collagen chains (defined as in [22]) interact, respectively, with the MIDAS Mg 2+ ions of molecules A and B.
The two a2 I E318W molecules in the complex are very similar to each other (r.m.s. deviation of 0.49 Å for 177 Ca atoms), as well as to the ligated conformation of the wild-type a2 I domain [9] (r.m.s. deviation of 0.58-0.59 Å ). In molecule A, the C-terminal helix 7 has clear electron density for two helical turns, including the solvent-exposed side chain of W318. In molecule B, helix 7 has patchy electron density and W318 is disordered. The crystal lattice is formed from two types of 2-fold symmetric contacts, one involving the collagen peptide and a2 I domain residues 286-287 (highlighted in Fig. 4B) and the other involving only the a2 I domains at residues 228-233.
The interaction of a2 I E318W molecule A with the GFOGER peptide is essentially the same as described for wild-type a2 I [9]. The trailing collagen chain provides the crucial glutamic acid that coordinates the Mg 2+ ion of the MIDAS; the same chain also provides the phenylalanine that makes van der Waals contacts with N154 and Q215, and the arginine that interacts electrostatically with D219 (Fig. 5B, C). The middle collagen chain provides the phenylalanine that makes van der Waals contacts with Y157 and L286, the hydroxyproline that donates a hydrogen bond to the peptide carbonyl oxygen of residue 154, and the arginine that stacks against H258 and is close to E256.
Molecule B also interacts with two of the three collagen chains, but because the three chains in a triple helix are not topologically equivalent, some of the interactions made by molecule A are not possible. The leading chain interacts with molecule B in the same way as the trailing chain with molecule A (Fig. 5B, D). In contrast, new interactions are formed by the second, trailing chain: a proline takes the place of the phenylalanine near Y157 and the arginine interacting with H258 is replaced by a hydroxyproline. These changes are expected to result in a weaker interaction, consistent with a smaller buried collagen surface area for molecule B (470 Å 2 ) compared with molecule A (530 Å 2 ).
We considered whether it is theoretically possible to achieve identical binding modes in a 2:1 complex. We modelled a complex in which I domain A is bound to the leading and middle chains, and I domain B is bound to the middle and trailing chains (the leading-middle and middle-trailing relationships are topologically equivalent). In contrast to the crystal structure, this modelled arrangement leads to clashes between the two I domains in a region that is key for allosteric regulation (residues 285-291) [9]. Thus, the crystal structure of the a2 I E318W-GFOGER complex represents the only sterically reasonable arrangement that can account for the observation of a 2:1 complex in solution (Fig. 3).

Discussion
Our new binding data support the view that the E318W mutant represents an activated form of the a2 I domain that binds collagenous ligands in a largely conventional manner [15,16,18,23]. In the solid-phase binding assay with Toolkit II, the E318W mutant displayed stronger discrimination of positive peptides than the wild-type a2 I domain and bound to several new peptides (Fig. 1). All the E318W-selective Toolkit peptides contain a GEx' triplet, consistent with the pivotal role of the glutamic acid in co-ordinating the Mg 2+ ion of the MIDAS. Binding of a2 I E318W to shorter high-affinity peptides was similar to wild-type, probably because the assay was saturated. With lower affinity peptides, notably GMOGER, a substantial increase in binding of a2 I E318W was observed (Fig. 2). GFPGER, reported as able to bind integrin a2b1 but not a1b1 in recombinant human collagen [24], bound wild-type and E318W a2 I domain equally well despite the absence of the prolyl hydroxyl group. Notably, the a1b1-specific motif GVOGEA [12], either as a short peptide or within II-27, was not recognised by either form of a2 I domain.
In our recent co-crystallisation experiments with triple-helical peptides, we found analytical SEC to be a useful method to monitor protein-collagen peptide interactions in solution [22,25,26]. Using SEC, we detected a stable 1:1 complex of a short GMOGER peptide with a2 I E318W, but not with wild-type a2 I, again confirming that the E318W mutant is more active. Wild-type a2 I formed a 1:1 complex with a short GFOGER peptide, as expected [9]. Surprisingly, however, a2 I E318W appeared to form a 2:1 complex with the GFOGER peptide (Fig. 3).
The crystal structure of the a2 I E318W-GFOGER complex indeed revealed two a2 I domains, A and B, bound to the same triple helix (Fig. 5). A similar arrangement was recently described  for the complex of the collagen chaperone Hsp47 with a triplehelical peptide [27]. I domain A of the a2 I E318W-complex binds to the trailing and middle chains of the GFOGER peptide, whilst the wild-type I domain bound to the middle and leading chains [9]. The integrin-collagen interactions are identical in both structures, however, because the trailing-middle and middleleading combinations are topologically equivalent. We thus assign the interaction with I domain A as the high-affinity mode in the a2 I E318W-GFOGER complex. I domain B binds to the leading and trailing chains, and this interaction is assigned as the lowaffinity mode. The collagen chain that contributes the MIDASbinding glutamic acid makes identical interactions in the high-and low-affinity binding modes (Fig. 5). In contrast, the second (trailing) collagen chain in the low-affinity mode recapitulates only a single contact of the high-affinity binding mode (via the hydroxyproline of the preceding GPO triplet) and is lacking the interactions of the phenylalanine with Y157/L286 and the arginine with E256/H258. The peptide binding data suggest that the interaction with Y157/L286 contributes most to the binding affinity: the identity of residue6in the Gxx'GEx'' motif is highly correlated with affinity (F .L < R .M ..A), whereas substitution of arginine in position x'' has only a small effect on binding (Fig. 2). Consistent with this interpretation, mutation of Y157 in integrin a2b1 has a stronger effect on collagen binding than mutation of either E256 or H258 [28]. Whether this 2:1 I domain-collagen interaction could be reproduced exactly in nature is open to question, since GFOGER occurs in the fibrillar collagens within different primary sequence settings than the flanking GPO triplets in our peptide. However, an adjacent GxO triplet, like that contributing to the trailing chain interaction revealed with I domain B, precedes several occurrences of GFOGER in other collagens, notably GFOGFOGER in a4(IV) and GPOGFOGER in a5(IV). The single-helix structure of collagen IV might permit simultaneous interaction with two I domains. Such speculation, although interesting, does not readily lend itself to experiment.
The low-affinity binding mode in the a2 I E318W-GFOGER complex suggests that a single GxOGER motif may support a2b1 binding, especially after activation of the cell in question, so long as the other collagen chain contacting the I domain does not cause steric hindrance. The markedly enhanced recognition of GMO-  GER-containing peptides by a2 I E318W suggests that this motif could also contribute to integrin-mediated cell adhesion. GMO-GER occurs naturally in collagen a1(I), as a heterotrimer with GLOGER in the a2(I) chain, and in collagens a1(II), a1(III), a2(V) and a3(IX). In collagens V and IX, GMOGER is present without a corresponding motif in the other constituent a-chains. Recent advances in producing register-specific collagen heterotrimers [29,30] may permit this hypothesis to be tested directly. Moreover, such a mode of binding might also operate in the fibrillar collagens where a-chains are sub-optimally exposed on the surface of a collagen fibre.

Expression Plasmids
The GST-tagged recombinant human a2 I domain-encoding plasmids (pGEX-2T-a2 I and pGEX-2T-E318W-a2 I) for binding studies were a generous gift from Dr D. Tuckwell (F2G Ltd, Manchester, UK). For crystallisation and SEC, the His-tagged E318W I domain ORF was cloned into the bacterial expression vector pDEST-N110 (a kind gift of Dr. M. Dyson) as described [31].

Protein Production
The wild-type and E318W a2 I domains were produced in the same way for binding studies; recombinant proteins were expressed as described [12]. The procedure used to produce His-tagged E318W for crystallisation includes some modifications, and is detailed as follows. Origami TM E. coli cells (Novagen) were transformed with the respective expression plasmid and a 50 ml overnight culture used to inoculate 1 litre of Luria Broth medium containing 50 mg/ml ampicillin (Melford). After growing the culture for 2 h at 37uC, expression was induced with 0.5 mM isopropyl b-D-thiogalactopyranoside (Melford), and the culture grown overnight at 25uC. The cells were harvested by centrifugation for 20 min at 4500 g and the pellet was gently resuspended in 25 ml of ice-cold TES buffer (30 mM Tris-HCl pH 8.0, 1 mM EDTA, 20% sucrose) containing 1 tablet of EDTA-free protease inhibitor cocktail (Roche Applied Science), 50 mg of lysozyme (Sigma-Aldrich) and 25 mg of protamine sulphate (Sigma-Aldrich). The suspension was left on ice for 10 minutes and centrifuged for 10 min at 6000 g (4uC). The supernatant was collected and the pellet gently resuspended in 25 ml of ice-cold 5 mM MgSO 4 containing 1 tablet of EDTA-free protease inhibitor cocktail, 50 mg of lysozyme and 25 mg of protamine sulphate. After 10 min on ice the suspension was centrifuged for 10 min at 6000g (4uC) and the pellet discarded. The pooled supernatants were centrifuged for 20 min at 16000g (4uC) and filtered through a 0.22 mm filter. Imidazole (Merck) was added to a final concentration of 50 mM and the solution loaded onto a 5 ml HisTrap FF column (GE Healthcare) equilibrated in PBS (phosphate buffered saline) using an Ä KTA Purifier system (GE Healthcare). The proteins were eluted with PBS containing 150 mM imidazole, concentrated using Vivaspin centrifugal devices (Sartorius) and further purified by SEC using a Superdex 75 10/300 GL column (GE Healthcare) and a running buffer consisting of 100 mM Tris-HCl pH 7.5, 150 mM NaCl and 2 mM MgCl 2 (SEC buffer). The final protein yields were ,10 mg/litre of culture.

Peptide Synthesis
The general format of the peptides for solid-phase binding assay is GPC(GPP) 5 2[insert]-(GPP) 5 GPC. For Toolkit peptides, the insert contains 27 amino acids of primary collagen sequence, each subsequent peptide advancing 18 residues along the collagen triple-helical domain creating a 92residue overlap between adjacent peptides in the set [16,20]); shorter peptides contain a [Gxx'Gxx'] insert, with sequences as defined in Fig. 2, whilst the negative control, GPP10, lacks any insert. For crystallisation trials, peptides were made conforming to a shorter template, (GPO) 2 2[insert]-(GPO) 3 , the GPO flanking sequence providing greater thermal stability. Peptides, as C-terminal amides, were synthesised on TentaGel R-Ram resin using an Applied Biosystems Pioneer peptide synthesiser as described previously [16]. Shorter peptides were made using the same Fmoc chemistry in a CEM Liberty microwave synthesiser. In either case, fractions containing homogeneous product were identified by analytical HPLC on an ACEphenyl300 (5 mm) column, characterized by MALDI-TOF mass spectrometry, pooled and freeze-dried. Triplehelical conformation was confirmed by polarimetry.

Toolkit and Shorter Peptide Binding Experiments
I domain adhesion was determined colorimetrically using a solid-phase binding assay, modified slightly from our previous studies [15,16,32] and specified below. Wells were coated using 1 mg peptide in 100 ml 0.01 M acetic acid for 1 h at 22uC on Immulon-2 HB 96-well plates (Thermo Life Sciences, Basingstoke, UK), and blocked for 1 h with 200 ml of TBS containing 50 mg/ ml bovine serum albumin). Wells were washed four times with 200 ml of the adhesion buffer (TBS with 1 mg/ml bovine serum albumin) before adding 100 ml of adhesion buffer containing 10 mg/ml of recombinant GST I domains in the presence of either 2 mM MgCl 2 or EDTA for 1 h at room temperature. Wells were washed five times with 200 ml of adhesion buffer containing MgCl 2 , before adding 100 ml of adhesion buffer containing Amersham anti-GST-HRP conjugate (RPN1236; GE Healthcare) at 1:10,000 dilution for 1 h at room temperature. After washing, colour was developed using an ImmunoPure TMB Substrate Kit (Pierce) according to the manufacturer's instructions. Replacing Mg 2+ with EDTA abolished binding (not shown). Data for short peptide binding were compared using 2-way ANOVA to establish the overall effect of the E318W mutation, and using repeated measures 1-way ANOVA for specific comparisons (Prism 5.0 for Mac, GraphPad).

Analytical Size Exclusion Chromatography
The lyophilised triple-helical peptides (GFOGER, GMOGER, GAOGER) were dissolved in 500 ml of 1 mg/ml stock solutions of either wild-type or E318W a2 I domains. The amount of peptide was chosen so as to yield a greater than 2-fold molar excess of peptide. After incubation for 30 min at 4uC, the solutions were injected onto a Superdex 200 10/300 GL column and run at 4uC at a flow rate of 0.5 ml/min in SEC buffer.

Crystallisation
The a2 I E318W-GFOGER complex was formed by dissolving 3.1 mg of lyophilised GFOGER peptide in 2 ml of a 3.5 mg/ml solution of a2 I E318W in SEC buffer (,2:1 peptide:protein ratio). After incubation for 30 min at 4uC, the solution was subjected to SEC as in the analytical experiments. The fractions corresponding to the a2 I E318W-GFOGER complex were pooled, concentrated to 9 mg/ml and stored at 4uC overnight. A substantial precipitate was observed on the following day and pelleted by centrifugation. The clear supernatant was stable at 5.3 mg/ml and screened for crystallisation using a Mosquito nanolitre robot (TTP LabTech). Crystals grew after 2 days with 0.2 M MgCl 2 , 0.1 M bis-Tris pH 5.5, 25% PEG 3350 as precipitant. Crystals were flash-frozen in liquid nitrogen directly from the crystallisation drop.

Data Collection and Structure Determination
Diffraction data were collected from a single a2 I E318W-GFOGER crystal at 100K on beamline I04 of the Diamond Light Source (Oxfordshire, UK) using a wavelength of 0.9795 Å and a CCD detector (ADSC). The diffraction data were integrated with XDS [33]. The diffraction limit was anisotropic, extending to,3.0 Å in the c* direction but limited to ,3.5 Å in the a*b* plane. The integrated data were scaled and merged with AIMLESS, and converted to structure factor amplitudes with TRUNCATE [34]. The structure was solved by molecular replacement with PHASER [35] using as search model the wild-type a2 I-GFOGER complex [9]. The model was rebuilt with O [36], refined with CNS [37] and PHENIX [38], and validated with MOLPROBITY [39]. OMIT maps were used to verify the register of the three collagen chains (Fig. 4A) and the few changes with respect to the high-resolution structure of the a2 I-GFOGER complex [9]. The structure was refined with tight geometry restraints and restrained atomic B-factors. This treatment (which is physically more realistic than grouped B-factors) resulted in the lowest R free and the smallest difference between R work and R free . Mg 2+ -ligand distances were restrained to 2.07 Å . Difference density indicated the presence of a Mg 2+ -bound water molecule in each I domain; due to the low resolution of the diffraction data, water molecules were not included in the model, however. Crystallographic statistics are summarised in Table 1. The figures were made with PyMOL (www.pymol.org). Surface areas were calculated with AREA-IMOL [34]. The coordinates of the a2 I E318W-GFOGER complex have been deposited in the Protein Data bank (entry code 4bj3).