Determining the Molecular Basis for the pH-dependent Interaction between the Link Module of Human TSG-6 and Hyaluronan*♦

TSG-6 is an inflammation-associated hyaluronan (HA)-binding protein that has anti-inflammatory and protective functions in arthritis and asthma as well as a critical role in mammalian ovulation. The interaction between TSG-6 and HA is pH-dependent, with a marked reduction in affinity on increasing the pH from 6.0 to 8.0. Here we have investigated the mechanism underlying this pH dependence using a combined approach of site-directed mutagenesis, NMR, isothermal titration calorimetry and microtiter plate assays. Analysis of single-site mutants of the TSG-6 Link module indicated that the loss in affinity above pH 6.0 is mediated by the change in ionization state of a histidine residue (His4) that is not within the HA-binding site. To understand this in molecular terms, the pH-dependent folding profile and the pKa values of charged residues within the Link module were determined using NMR. These data indicated that His4 makes a salt bridge to one side-chain oxygen atom of a buried aspartate residue (Asp89), whereas the other oxygen is simultaneously hydrogen-bonded to a key HA-binding residue (Tyr12). This molecular network transmits the change in ionization state of His4 to the HA-binding site, which explains the loss of affinity at high pH. In contrast, simulations of the pH affinity curves indicate that another histidine residue, His45, is largely responsible for the gain in affinity for HA between pH 3.5 and 6.0. The pH-dependent interaction of TSG-6 with HA (and other ligands) provides a means of differentially regulating the functional activity of this protein in different tissue microenvironments.

both covalent and noncovalent complexes (8,17,25). This diverse range of ligands is probably unusual for a Link module, the primary role of which in other Link module-containing proteins appears to be binding to HA but serves to underline the multifunctional role TSG-6 plays in inflammatory processes.
The best understood interaction of TSG-6 is that with HA (21, 26 -29). HA is a ubiquitous high molecular weight glycosaminoglycan (up to 10 7 Da) composed entirely of a repeating disaccharide of D-glucuronic acid (GlcUA) and N-acetyl-D-glucosamine (GlcNAc). It is found in the ECM of most vertebrate tissues and displays diverse biological functions, including roles in embryonic development, cell migration, and ovulation, as well as being implicated in many disease processes (30 -32). Although free HA is present in synovial fluid and the vitreous of the eye, in other tissues (e.g. cartilage, skin, brain) it is found mainly in complex with proteins, forming vital structural components of the ECM. The wide range of functions ascribed to HA is thought to arise mainly through interactions with HAbinding proteins that lead to the formation of protein-HA complexes with distinct architectures and functional activities (12,33,34). As such, the production of TSG-6 at sites of inflammation is expected to result in modulation of these architectures and has clear importance for ECM biology and inflammatory disease processes (3,12). In this regard, changes in the affinity of TSG-6 for HA (and other ECM molecules) by alterations in tissue microenvironments could have a crucial role in the regulation of TSG-6 function.
We have previously reported that the interaction of HA with Link_TSG6 is pH-dependent, having maximal affinity at pH 5.5-6.0 and a significant loss of function with either increasing or decreasing pH (see Fig. 1A) (22). In marked contrast, the pH dependences of both Link protein and the G1 domain of aggrecan (two constitutively expressed HA-binding proteins) for HA have no drop in affinity between pH 6.0 and 8.0 (22). Recently, the pH dependence of the HA-Link_TSG6 interaction has been questioned (35), with suggestions that the earlier observation results from an artifact of the solid-phase assays used (i.e. where either the protein or HA was immobilized on the surface of the microtiter plate). Here we present isothermal titration calorimetry data for defined oligosaccharides of HA (i.e. where both the protein and HA are in the solution phase) that confirms the original observation.
We have determined the solution structure of Link_TSG6 in the absence and presence of an HA octasaccharide (HA 8 AN ) (28) and recently have used these and other data to produce a model for the HA 8 AN -Link_TSG6 complex (see Fig. 1B) (29). 6 Therefore, it should now be possible to come to a molecular understanding of the pH dependence of HA binding. To produce a curve with a rise and then fall in activity as seen in Fig.  1A, two pH-dependent factors are required. Previously it was proposed that the gain in activity up to pH 6.0 was due to the pH-dependent folding of Link_TSG6, which was observed (by a qualitative method) to fold mostly over the range of pH 4.5 to 6.0 (22). Since Link_TSG6 showed no unfolding or gross perturbation to the structure between pH 6.0 and 7.5, it was concluded that a change in ionization state of one (or more) charged residues was likely to be responsible for the loss of activity above pH 6.0 (36). The four histidine residues (His 4 , His 29 , His 45 , His 96 ) within Link_TSG6 are reasonable candidates for this activity, as is an aspartate residue (Asp 89 ) buried within the protein (see Fig. 1C) (21,36,37). However, none of these residues is directly within the HA-binding site (as determined by site-directed mutagenesis, chemical shift mapping, and the HA-bound conformation of the protein) (6, 26 -29, 38), i.e. they are very unlikely to make direct contact with the bound HA molecule.
In this study, we have demonstrated, using isothermal titration calorimetry (ITC) that the previously reported decrease in affinity for HA between pH 6.0 and 7.5 (22) is observed in the solution phase. Analysis of single-site mutants of the TSG-6 Link module (Link_TSG6) indicated that the loss in affinity above pH 6.0 is mediated by the change in ionization state of a particular histidine residue (His 4 ). Nuclear magnetic resonance spectroscopy (NMR) experiments allowed the pH-dependent folding profile of the protein and the pK a values of all relevant residues within Link_TSG6 to be determined. Combining these functional and NMR data, we show how the change in charge state of His 4 is relayed to HA-binding residues via a network involving a salt bridge and a hydrogen bond, thus providing novel insights into the molecular basis of this important pH-dependent interaction.

EXPERIMENTAL PROCEDURES
Sources of Materials-Wild-type (WT) and mutant forms of Link_TSG6 were expressed in E. coli and purified as described previously (27). In addition to the mutants H4K (6) (i.e. His 4 replaced by Lys) and H29K (38), six new mutants (H4S, H29A, H45A, H45K, H45S, H96K) were prepared and analyzed by one-dimensional NMR and mass spectrometry as before (27); all mutants had molecular masses within 2 Da of their theoretical values. Uniformly 15 N-and 15 N, 13 C-labeled WT Link_TSG6 were produced by expression in M9 minimal medium (26). Full-length TSG-6 (Q allele) was generated as detailed previously (39). Medical grade HA (Hylumed Medical; molecular mass, 0.5-1.5 MDa) was obtained from Genzyme. HA oligosaccharides of defined lengths (i.e. HA 8 AN and HA 20 AN , which comprise 4 and 10 disaccharide repeats, respectively) were purified from this high molecular weight HA following digestion with testicular hyaluronidase (40). Mono-biotinylated Link_TSG6 and biotinylated rooster comb HA were prepared as described previously (20,27).
ITC Analysis of the HA-Link_TSG6 Interaction-The interaction between Link_TSG6 (WT and folded Link_TSG6 mutants H4K, H29K, H45S, H96K) and HA oligosaccharides HA 8 AN and HA 20 AN was investigated on a Microcal VP-ITC instrument at 25°C in 5 mM MES (pH 6.0 or 7.5), using 28 injections for each measurement, as described previously (6,27,28); the sugar (HA 8 AN , 0.2 mM; HA 20 AN , 0.1 mM) was added in 5-l injections to the protein solution (0.015-0.017 and 0.012-0.017 mM for the 8-mer and 20-mer, respectively) in the 1.4-ml calorimeter cell. Single-stock solutions of HA 8 AN and HA 20 AN , for which the concentrations were determined as described previously (26), were used throughout all experiments to reduce errors. Data were fitted to a one-site model by nonlinear least squares regression after subtracting heats resulting from the addition of oligosaccharides into buffer alone. The binding constants for each interaction were determined by averaging the results from three experiments.
Microtiter Plate-based HA Binding Assays-The HA binding activities of Link_TSG6 (WT and  mutants H4K, H29K, H45S, H96K) and full-length TSG-6 were determined using the microtiter plate assay described previously (27). WT and mutant Link_TSG6 were coated overnight onto Maxisorb microtiter plates at 25 pmol/well (27), and full-length TSG-6 was coated at 8 pmol/well (39); previously we showed that these concentrations of Link_TSG6 and fulllength TSG-6 give equivalent signals in HA-binding assays (39). To vary the pH of the binding reaction, after the initial washes with standard assay buffer (SAB; 100 mM NaCl, 50 mM sodium acetate, 0.05% (v/v) Tween 20, pH 6.0), individual wells were rewashed once with buffers of specific pH (containing 100 mM NaCl, 50 mM Na-HEPES, 0.05% (v/v) Tween 20) and dried, then biotinylated HA (27) in the appropriate pH buffer was added to each well (12.5 ng/well), and the mixture was incubated at room temperature for 4 h. Wells were then rewashed in SAB, and the assay was continued as described previously (27).
General NMR Methodology-Samples for NMR analysis comprised 15 N-or 15 N, 13 C-labeled WT Link_TSG6 with or without oligosaccharide (in a 1:1 molar ratio) in 600 l of 10% (v/v) D 2 O, 0.02% (w/v) NaN 3 . All NMR experiments were performed at 25°C on spectrometers at the Oxford Centre for Molecular Sciences (University of Oxford, UK) at a proton resonance frequency of 500 MHz (unless otherwise stated) following the methods detailed previously (26,28,29 ) samples at pH values ranging from 3.0 to 7.5, in 0.25 pH unit increments. Acquisition parameters were 161.80 ms (t 1 , 15 N, 128 complex points) and t 2 128.00 ms (t 2 , 1 H, 512 complex points) with the 15 N carrier frequency at 119.0 ppm. The samples were adjusted with 0.1 M NaOH and HCl, and the pH was measured directly in FIGURE 1. The pH-dependent binding of HA to TSG-6. A, the pH-dependent change in affinity of polymeric HA binding to Link_TSG6. The curve and error bars were produced by taking the mean at each pH value of the previously reported plate assay data (22) for biotinylated HA binding to Link_TSG6-coated plates and biotinylated Link_TSG6 binding to HA-coated plates. B, stereo representation of the TSG-6 Link module showing locations of the key HA-binding residues (27) within the lowest energy HA 8 AN -bound Link_TSG6 structure (28). Aromatic (blue) and basic (red) residues are shown in their experimentally determined conformations (28), whereas the HA (gold) is derived from a model for the complex (29). The ␤1-␣1 and ␤4 -␤5 loops that rearrange upon HA binding are indicated. C, locations of the four histidine and buried aspartate (Asp 89 ) residues within Link_TSG6 (in a stereo representation). The lowest energy structure of the free protein is shown, with amino acid side-chain conformations resulting from the family of 20 structures (28).

pH-dependent Interaction between TSG-6 and HA
the NMR tube. Sample volume lost because of handling (Յ10 l/titration point) was replaced as necessary by the addition of 10% D 2 O, 0.02% NaN 3 .
The extent of folding of free and HA 8 AN -bound Link_TSG6 at a particular pH value was assessed by comparing the total volume of all the amide resonances in a 1 H, 15 N-HSQC spectrum arising from folded material at that pH value that were not occluded by resonances resulting from unfolded material at any pH value (ϳ70 resonances) with the total volume of the entire spectrum at that pH value (i.e. the combination of both folded and unfolded material). In both the free and HA 8 -bound protein, ϳ25 resonances were occluded and could not be included in the total value for the "folded material." However, assuming the intensity from these occluded peaks changes proportionally with the non-occluded peaks, the ratio of volumes from the non-occluded peaks to the total volume of the entire spectrum quantitatively reflects the fraction of folded protein present at each pH value. This method is superior to plotting the absolute intensities from folded material at each pH value, because such intensities are susceptible to large errors arising from differences in the protein concentration caused by NMR sample handling and changes in magnet shimming.
The titration curves resulting from the pH-dependent changes in chemical shift were fitted by nonlinear least squares analysis to both one-and two-site models for the pK a based on the Henderson-Hasselbach equation (29). Two-site models for the data were only accepted where the 2 value for the fit was more than 10 times better than that for the one-site model. The titrating residue responsible for the pH-induced change in chemical shift of a particular amide resonance was identified using both the determined pK a value and the location of the amide group in the structure (Protein Data Bank codes 1o7b and 1o7c for free and HA 8 AN -bound forms, respectively) (28). The chemical shifts at each pH value have been deposited in the BioMagResBank under accession codes 7221 and 7222.
Additional 1 H, 15 N-HSQC spectra with wider spectral widths and different carrier frequencies in the 15 N dimension were recorded on the 1 mM 15 N-Link_TSG6ϩHA 8 AN sample at pH 3.75 in order to examine the amide resonances that become visible because of the reduction in their chemical exchange rate with low pH, i.e. Arg N and Lys N . A 1 H, 13 C-NOESY-HSQC spectrum was recorded (600 MHz) at pH 7.5 on a 2.8 mM 15 N, 13 C-Link_TSG6 sample, with parameters identical to the one recorded previously at pH 6.0 (see Ref. 28)). 1 H, 15 N-HSQC spectra were recorded with the same parameters used in the pH titration described above on a 0.2 mM 15 N-Link_TSG6 ϩ 0.1 mM HA 20 AN sample (i.e. a 2:1 protein:HA ratio) at pH 6.0. 1 H, 15 N-HMQC experiments, adjusted to allow the detection of histidine side-chain nitrogen nuclei (i.e. N ␦1 , N ⑀2 ), were recorded on 2 mM free and HA 8 AN -bound 15 N-Link_TSG6 samples at pH 6.0. The 15 N carrier frequency was set to 210 ppm and the delay, generating 1 H and 15 N antiphase magnetization, was 22 ms (41). Acquisition times were 115.00 ms (t 1 , 15 N, corresponding to a sweep width of 100 pm, 256 complex points) and 128.00 ms (t 2 , 1 H, 512 complex points), achieving folding of backbone amide resonances into the (unoccupied) center of the spectrum.

Structure Refinements to Include Salt Bridges and Side-chain
Hydrogen Bonds-All potential salt bridges and side-chain hydrogen bonds were investigated individually and collectively to see if they were compatible with the solution structures determined previously (Protein Data Bank codes 1o7b and 1o7c (28)). Each member of the family of 20 structures was refined with a single cycle of simulated annealing followed by extensive energy minimization (28), using all of the original restraints, with the inclusion of new restraints for the salt bridge(s)/hydrogen bond(s) under investigation. Each hydrogen bond was modeled as two distance restraints, i.e. N-O at 2.4 Å, H-O at 1.7 Å (42). Salt bridges involving arginine were included as two hydrogen bonds (examining both possible bonding arrangements), whereas those involving lysine were included as one hydrogen bond. Hydrogen bonds and salt bridges were considered to be consistent with the NMR structure if their inclusion did not result in an increase in energy or violations of previously satisfied structural restraints. Molecular models have been displayed using MOLMOL (hugin.ethz.ch/wuthrich/software/ molmol/).

Production and Characterization of Link_TSG6 Mutants-
Six new mutant forms of Link_TSG6 (i.e. H4S, H29A, H45A, H45K, H45S and H96K) were generated to investigate the role of histidine residues in the pH-dependent interaction of HA with Link_TSG6. Only the H45S and H96K mutants were demonstrated by NMR to have wild-type folds, also found to be the case previously for H4K (6) and H29K (38). The other four Link_TGS6 mutants made here (i.e. H4S, H29A, H45A, and H45K) all had perturbed folds (based on the criteria described in (Ref. 27)) and thus were not analyzed further. The D89A mutant described previously also has a perturbed fold (27), which is not surprising given that it is completely buried within the protein (21,28).
ITC Analysis of the Interaction of Wild-type and Mutant Link_TSG6 with HA 8 AN -The affinity of Link_TSG6 for HA 8 AN (the shortest HA AN oligosaccharide that binds with maximal affinity (28)) was determined by ITC at pH 6.0 and 7.5 (see Table 1 and Fig. 2). In the case of wild-type protein, the binding constant at pH 7.5 is less than half that determined at pH 6.0. This indicates that the interaction of Link_TSG6 with HA 8 AN is indeed pH-dependent in the solution phase and follows the same trend observed previously in the microtiter plate assay (22).
Both of the H29K and H96K mutants exhibit HA 8 AN binding properties very similar to the wild-type protein with essentially equivalent binding constants at pH 6.0 (97 and 90% of wild type, respectively) and with reductions in their affinities at pH 7.5 of ϳ50% (Table 1). On the basis of these data, it is concluded that neither His 29 nor His 96 is responsible for the pH-dependent interaction of Link_TSG6 with HA 8 AN . This is not particularly surprising, because both residues are distant to the HA-binding site (see Fig. 1, B and C).
From Table 1 it can be seen that the H45S mutant has a decreased affinity for HA 8 AN at pH 6.0 (48% of the wild-type value). Previously we had found that mutation of residues directly involved in mediating the interaction with HA leads to pH-dependent Interaction between TSG-6 and HA APRIL 27, 2007 • VOLUME 282 • NUMBER 17 JOURNAL OF BIOLOGICAL CHEMISTRY 12979 a 0.1-0.01-fold change in the binding constant (i.e. 10 and 1% of the WT value, respectively (6,(27)(28)(29)38)). Therefore, it seems unlikely that His 45 participates directly in the interaction with HA, but rather it is assumed that this drop in affinity on mutation to serine is because of a small but significant long-range perturbation to the binding site. In this regard, His 45 is positioned adjacent to the ␤4-␤5 loop ( Fig. 1C) that has been shown to undergo a conformational change on binding (28). At pH 7.5 the binding constant for the interaction of H45S with HA 8 AN is about 0.5-fold the value at pH 6.0 (i.e. a similar reduction in affinity to that seen for wild-type Link_TSG6). This indicates that His 45 does not mediate the pH-dependent loss of affinity with HA above pH 6.0.
The affinity of H4K for HA 8 AN is somewhat reduced relative to wild-type protein at pH 6.0 (68%), indicating that this mutation may also cause a long-range disruption to the binding site. Crucially, however, the binding constant for this mutant is essentially unaffected by increasing the pH to 7.5 ( Table 1). The finding that the pH-dependent reduction in HA binding activity is abolished by mutating His 4 to lysine, while being unaffected in the H29K, H45S, and H96K mutants (relative to wild-type), indicates that His 4 may be responsible for the pH dependence.
Analysis of HA-Protein Interactions by Microtiter Plate Assays-The change in HA binding activity with pH for fulllength TSG-6 and each of the folded histidine mutants of Link_TSG6 was investigated using a microtiter plate-based assay; this assay determines the binding of polymeric HA (biotinylated) to protein immobilized on the plate. From Fig. 3 it can be seen that the decrease in HA binding with pH is also a property of full-length TSG-6. As the profile for the full-length protein is very similar to that of the isolated Link module (wildtype Link_TSG6), it can be concluded that neither the CUB module nor N-and C-terminal extensions are involved in mediating the pH dependence. The Link_TSG6 mutants H29K, H45S, and H96K also gave a similar binding profile, indicating that none of these histidine residues are involved in mediating the pH-dependent loss of affinity with HA, as was concluded above from the ITC data with HA 8 AN . The H4K mutant binds very weakly in this plate assay, making it impossible to determine whether it has a pH-dependent decrease in affinity for HA (Fig. 3). This lack of HA binding is particularly surprising given that the NMR and ITC analyses (described above) demonstrated that the protein is fully folded and binds HA 8 AN . In fact, as assessed by ITC, H4K has a higher   20 AN to WT and H4K Link_TSG6 at pH 7.5. All data were fit with one-site models (smooth line) (27). The derived binding constants and stoichiometries are presented in Table 1. . The pH dependences of HA binding to full-length TSG-6 and WT and mutant forms of Link_TSG6. The interaction was assessed by a microtiter plate assay in which the binding of biotinylated HA to wells coated with each protein was measured colorimetrically (27). Values are plotted as the mean absorbance (n ϭ 3) at 405 nm after a 10-min development Ϯ S.E. affinity for HA 8 AN (68% of WT) than H45S (47% of WT), which binds well in the plate assay (Table 1). One possibility is that this anomaly could arise from the H4K mutant coating onto the plate differently from the other mutants. However, this seems unlikely because the loss of HA binding function on immobilization is also seen for the "H4K" mutation when this is made in the context of the full-length TSG-6 protein expressed in Drosophila S2 cells. 7 Furthermore, of the 15 Link_TSG6 mutants that have been assessed to date by both the microtiter plate assay and ITC (6,27,38), including those described here, the H4K mutant is the only one in which we have found conflicting results from these two methods. The major difference between these experimental techniques is the length of the HA utilized, i.e. polymeric HA (ϳ1000 kDa) in the plate assay versus an HA octasaccharide (ϳ1.5 kDa) in the ITC. An HA octasaccharide only has a binding site for a single protein molecule (26,28), whereas in the polymeric HA used there are likely to be more than 500 binding sites for Link_TSG6 on an individual HA chain. In this latter case it is conceivable that interactions between neighboring protein molecules may take place, leading to an increased affinity for HA and the observed difference between these techniques.
Investigation of Potential Protein-Protein Interactions-The binding of wild-type and H4K Link_TSG6 to HA 20 AN was investigated using ITC at pH 6.0 and 7.5 (Table 1); preliminary ITC experiments with HA 18 AN and HA 20 AN demonstrated that the 20-mer is the minimum length of HA n AN that can accommodate two Link_TSG6 molecules with optimal affinity (data not shown). As can be seen from Fig. 2, the ITC experiments performed with HA 20 AN gave titration curves with a single phase, which can arise from both noncooperative and positively cooperative systems. The absence of a second phase to the titration curve indicates that there is unlikely to be negative cooperativity present at either pH with either wild-type Link_TSG6 or the H4K mutant. These single-phase titrations with HA 20 AN can be fitted using a standard one-site model (i.e. the most appropriate model for two identical binding sites) allowing a comparison of experimentally derived values for the wild-type and H4K proteins fitted in an identical manner.
The fitting of the ITC data gave rise to stoichiometries of close to 0.5 for all of the experiments conducted (Table 1), consistent with two protein molecules binding to the same HA chain. As seen previously for HA 8 AN (described above), the binding constant derived for the interaction of HA 20 AN with wild-type protein is also smaller at pH 7.5 compared with that at pH 6.0; in the experiments with the 20-mer there is a larger reduction in affinity (ϳ0.14-fold) compared with that with HA 8 AN (ϳ0.38-fold). This provides further evidence for the pH dependence of the interaction between TSG-6 and HA (in the solution phase) but also indicates that this pH dependence may become more pronounced as the HA increases in size. This finding is consistent with the hypothesis that protein-protein interactions may occur in HA chains containing two or more protein-binding sites.
Importantly, the H4K mutant shows no such reduction in binding affinity for HA 20 AN at pH 7.5 compared with pH 6.0; in fact, there is a small apparent increase in affinity (ϳ1.4-fold) at higher pH. This provides further evidence that His 4 is responsible for mediating the pH dependence of Link_TSG6-HA binding and suggests that it may be required for stabilizing protein-protein interactions between neighboring Link_TSG6 molecules on a common HA chain. This amino acid has been implicated previously in mediating TSG-6 self-association (38). NMR spectroscopy was used in an attempt to determine whether protein-protein interactions do occur when Link_TSG6 binds to an HA 20-mer, i.e. by mapping of chemical shift perturbations within the 1 H, 15 N-HSQC spectra of 15 N-Link_TSG6 (0.2 mM) in the absence and presence of HA 20 AN (at a 2:1 molar ratio). However, upon the addition of HA 20 AN to the protein, extensive precipitation occurred. This was surprising, because precipitation with HA 8 AN (or HA 10 AN ) does not occur in NMR samples with protein concentrations more than 10 times higher than used with HA 20 AN (26,28), nor did it occur in the ITC experiments with HA 20 AN described here. Nevertheless, this precipitation (although unfortunate) perhaps constitutes further evidence that protein-protein interactions may occur in the context of longer lengths of HA (i.e. ՆHA 20 ).
Having established with ITC and plate assay experiments that His 4 is likely to be involved in mediating the pH dependence of binding of Link_TSG6 to HA, we required a molecular explanation of this phenomenon. Therefore, NMR was used to determine how the extent of folding changes with pH (which we proposed previously as responsible for the gain in affinity up to pH 6.0 (22)) and to determine the pK a of all titratable side chains within Link_TSG6. The locations of intramolecular salt bridges and hydrogen bonds were investigated in order to fully characterize all pH-dependent changes in Link_TSG6.
pH-dependent Folding of Link_TSG6-A method for quantitatively measuring the extent of folding with pH was developed (see "Experimental Procedures"). 1 H-15 N HSQC spectra were recorded at different pH values from 3.0 to 7.5, and the combined volume of all amide resonances arising from folded material that were not occluded by unfolded material at any pH value (ϳ70 resonances) was measured. The ratio of this volume to the total volume of the entire spectrum quantitatively reflects the fraction of folded protein present at each pH value, providing a relative change in extent of folding with pH. These data were scaled to 100% at pH 7.5 (at which the protein is assumed to be 7 W. Tongsoongnoen and A. J. Day, unpublished observation.

pH-dependent Interaction between TSG-6 and HA
completely folded). At no point in the titration of either the free or bound protein were partially folded species observed (e.g. the presence of amide resonances from ␣-helices but not ␤-sheets).
The folding curve for the free protein (Fig. 4) is consistent with the previous one-dimensional NMR studies (22). At pH 3.75, only ϳ20% of the protein is folded, and this value increases steadily to ϳ85% at pH 5.5. Above pH 5.5, there is a shallow increase in the percentage of folded material present up to the final data point (pH 7.5). This curve cannot be fitted to a onesite model (i.e. it does not result from a single pK a ), indicating that it arises from a combination of multiple pH-dependent factors. The folding curve for the HA 8 AN -bound protein is quite different. At pH 3.75, ϳ60% of the protein is still folded (i.e. ϳ3 times that of the free protein), and this value increases steadily up to pH 7.5. Below pH 3.75, the protein rapidly unfolds, until at pH 3.0 there is no folded protein detectable, i.e. 60% of the protein unfolds over 0.75 pH units, whereas in the free protein, a similar percentage reduction occurs over a pH range twice as wide (pH 3.75-5.25). It is therefore apparent that the binding of HA 8 AN to Link_TSG6 stabilizes the protein fold considerably, allowing it to stay folded at much lower pH values than the free protein; this stabilization was also observed in deuterium exchange experiments at pH 6.75 (see supplemental Fig. S1) that were used previously to infer the presence of hydrogen bonds during the structure determination (28). The sudden unfolding observed below pH 3.75 for the HA 8 AN -bound protein is likely because of the protonation of carboxylate groups within the bound HA 8 AN molecule (intrinsic pK a value 2.9 (43)), which would abolish intermolecular salt bridges (28,29,36) and thereby destabilize the complex.
Determination of pK a Values in Free and HA 8

AN -bound Link_TSG6-
The pH-dependent change of amide proton and nitrogen chemical shifts (determined from 1 H, 15 N-HSQC spectra) were fitted to either one-or two-site models (see Fig. 5) to determine the pK a value(s) of the titrating group(s) influencing them. Since there is little folded protein present below pH 4.0 in the free Link_TSG6, it was not possible to obtain accurate pK a values of Asp and Glu side chains (intrinsic values of ϳ4.0 and ϳ4.5, respectively). In the HA 8 AN -bound protein, however, there is a significant amount of folded material present at pH 3.25 (Fig. 4), and thus the pK a values of these acidic residues could be determined accurately. It was therefore possible to obtain the pK a value of every group titratable between pH 3.0 and 7.5 within the HA 8 AN -bound protein ( Table 2). Fig. 5 shows the locations of "reporter" amide groups for each histidine residue within the structure of the HA 8 AN -bound protein. Despite the seemingly far-ranging sensitivity of some groups to ionization of particular histidine residues (e.g. Gly 55 reports His 29 , despite being ϳ10 Å away), there is no evidence for either perturbations to the Link_TSG6 fold (i.e. resonances characteristic of folded protein are unaffected by changes in pH) or alterations in the side-chain conformations of the His residues themselves ( 1 H, 13 C-NOESY-HSQC spectra recorded at pH values of 6.0 and 7.5 showed identical NOE patterns at the C ␣ and C ␤ positions).
As seen from Table 2, the pK a values of His 4 , His 29 , and His 96 , as well as the N-terminal primary amine, are within the usual range expected for these functionalities (i.e. histidine ϳ6.0 -6.5, N-terminal amine ϳ7.5) in both the free and HA 8 AN -bound protein, although the value for His 4 in the free protein is somewhat lower than typical (pK a 5.8). The pK a of His 45 is also sup-

pH-dependent Interaction between TSG-6 and HA
pressed in both cases (pK a 5.4 and 5.7, respectively), which is probably because of the proximity of the positive charge on Lys 63 that is generally within ϳ5 Å of the His 45 side chain in the Link module structures (28), i.e. disfavoring protonation of this residue. Modified 1 H-15 N-HMQC spectra, which were recorded at pH 6.0, provided an independent evaluation of the ionization state of each histidine side chain (41). In agreement with the determined pK a values (Table 2), these spectra showed that His 4 , His 29 , and His 96 are charged in both free and HA 8 ANbound protein at pH 6.0, whereas His 45 is clearly deprotonated (the data for the HA 8 AN -bound protein are shown in supplemental Fig. S2).
The pK a values of His 29 and His 96 are not significantly altered when Link_TSG6 binds to HA 8 AN (Table 2), which is not surprising because they are on the opposite face of the protein from the HA-binding surface (Fig. 1, B and C). In contrast, the pK a values of His 4 and His 45 both change on binding (by ϩ0.5 and ϩ0.4 units, respectively), consistent with the observed sidechain chemical shift perturbations (His 4 , C ␤ Ϫ0.30 ppm (negative values indicate upfield shifts), C ␦2 0.66 ppm, C ⑀1 Ϫ0.63 ppm; His 45 , C ␤ 0.23 ppm, C ⑀1 1.00 ppm, H ⑀1 0.53 ppm) and small changes in average conformation (His 4 , 1 ϩ60°; His 45 , 2 Ϫ50°) seen between the free and HA 8 AN -bound spectra/structures (28). Several residues within the ␤4 -␤5 loop, the chemical shifts of which are affected by deprotonation of His 45 in the free protein (Fig. 5), are no longer affected in the complex (i.e. Lys 63 , Gly 65 , Asn 67 , Cys 68 , Phe 70 , Gly 74 , Ile 76 ; see Table 2), indicating that the effects of His 45 are less widespread in the HA 8 AN -bound state. Precise pK a values were also measured for all carboxylate groups (except Glu 86 , which was somewhat less accurately determined) within the HA 8 AN -bound protein ( Table 2). These were used in the identification of intramolecular salt bridges and transient hydrogen bonds as described below. Intramolecular Salt Bridges and Hydrogen Bonds in HA 8 ANbound Link_TSG6-Analysis of the HA 8 AN -bound Link_TSG6 structure (28) showed that eight pairs of oppositely charged residues are potentially close enough to each other to form intramolecular salt bridges (His 4 7Asp 89 , Arg 5 7Glu 26 , Lys 20 7Glu 24 , Lys 34 7Glu 37 , Arg 40 7Glu 37 , Lys 41 7Glu 37 , Arg 56 7Asp 77 , and Arg 81 7Glu 86 ). Three criteria must be satisfied to demonstrate that a salt bridge is present in solution: 1) the favorable charge interaction results in a depression of the carboxylate pK a below its intrinsic value; 2) the proximal basic residue should display side-chain chemical shift titrations with the same pK a as its acidic partner; and 3) inclusion of the salt bridge in restrained energy minimization calculations should not result in any violations of the original restraints used to define the solution structure.
Using these criteria we have been able to confidently identify intramolecular salt bridges between Arg 5 7Glu 26 , Arg 40 7Glu 37 , and Arg 56 7Asp 77 (see Fig. 6) and also to conclude that a salt bridge does not form between Arg 81 and Glu 86 . It also is very likely that a salt bridge is formed between Lys 20 7Glu 24 . A detailed description for these identifications (i.e. based on pK a values determined for Asp and Glu residues ( Table 2), 1 H, 15 N-HSQC experiments (supplemental Fig. S3), and structure calculations (see Fig. 6)) is provided on-line as supplemental information.
Of particular interest and meriting description here is Asp 89 , which has a pK a of 4.2 ( Table 2) that is very similar to its intrinsic value (4.0). This is surprising, because being buried within the protein (i.e. zero solvent accessibility), it would have been expected to be considerably higher (e.g. a buried Asp residue has a pK a of 7.5 in thioredoxin (44)). The best explanation for this unexpectedly low pK a is that His 4 makes a salt bridge to Asp 89 (see Figs. 6 and 7), favorably neutralizing the buried negative charge and thereby reducing the pK a value. This salt bridge is also consistent with the NMR structures of the free and HA 8 AN -bound protein (28), where the hydrogen bond from the carboxylate of Asp 89 is formed with the H ⑀2 proton of His 4 as indicated by the NOE-defined side-chain conformation.
Criteria similar to those used for the assignment of salt bridges can be used to predict the presence of transient intramolecular hydrogen bonds (see the supplemental information). The H N of Tyr 12 is a good candidate for such a hydrogen bond given the proximity of this moiety to the side chain of Asp 89 (within 2 Å; see Figs. 6 and 7). Its apparent pK a is identical to that of the Asp 89 carboxylate (pK a 4.2; see Table  2), and its inclusion in structure calculations does not perturb or violate any other restraint (including the proposed salt bridge between His 4 and Asp 89 ). On titration of Asp 89 (with pH), the H N of Tyr 12 has a large upfield shift of the amide proton of at least 0.30 ppm (the precise value is hard to determine because the secondary titration of His 4 also perturbs the chemical shift; see Fig. 5), which is characteristic of a hydrogen bond to a carboxylate group (see supplemental information). Therefore, it seems probable that this hydrogen bond is present in solution. Thus, there are likely to be two hydrogen bonds to the carboxylate oxygens of Asp 89 :  Fig. 7). As discussed in detail below, this network of hydrogen bonds, linking His 4 to Tyr 12 (via Asp 89 ) is likely to form the molecular basis for the pH-dependent decrease of affinity of TSG-6 for HA above pH 6.0. The deprotonation of His 4 with increasing pH would lead to loss of the salt bridge to Asp 89 and the consequent perturbation of its hydrogen bond to Tyr 12 , a residue that has been shown previously to have a crucial role in HA binding (6,27) by making hydrogen bond contact with the sugar via its hydroxyl hydrogen (28).

DISCUSSION
A combined approach of NMR, ITC, and microtiter plate binding assays has been used to investigate the molecular basis of the pH-dependent interaction between HA and TSG-6. In solid-phase microtiter plate assays, the interaction of polymeric HA with both full-length TSG-6 and the isolated Link module (Link_TSG6) is pH-dependent, showing a marked reduction in binding on increasing the pH from 6.0 to 8.0. Consistent with this finding, ITC experiments (i.e. carried out in the solution state) showed that the affinities of Link_TSG6 for HA 8 AN and HA 20 AN are higher at pH 6.0 than at pH 7.4 (ϳ2.5-and 7-fold, respectively). These data reaffirmed our previous findings on the pH dependence of the Link_TSG6-HA interaction (22) but appear to be inconsistent with the results of a recent study (35). However, it should be noted that Wisniewski et al. (35) examined the binding of TSG-6 to immobilized HA in the presence of 0.5 M NaCl (i.e. a salt concentration much higher than physiological). Here we have shown that the pH-dependent component of the TSG-6-HA interaction relies upon the change in ionization state of His 4 , and therefore it is to be expected that the pH dependence will not be observed in a salt concentration as high as 0.5 M NaCl because this would severely weaken all ionic interactions. We also note that a study using confocal-fluorescent recovery after photobleaching (FRAP), has also demonstrated that the pH dependence of the Link_TSG6 interaction with polymeric HA occurs in solution. 8 As noted in the Introduction, at least two pH-dependent factors are required in order to produce a bellshaped binding curve of the type seen in Fig. 1A. Here we have quantified how the extent of Link_TSG6 folding changes with pH, determined the pK a values of aspartate, glutamate, and histidine side chains, and identified intramolecular salt bridges and hydrogen bonds to carboxylate groups within the protein. This has allowed us to identify the components that could potentially contribute to the pH dependence of the Link_TSG6/HA interaction (i.e. the folding of the protein and the pK a values of His 4 , His 45 , and Asp 89 ); these are shown in Fig. 8.
Given that the ITC analysis described here clearly indicates that His 4 is responsible for the loss of HA binding affinity above pH 6.0 and that it has a pK a value of 5.7 (in the free protein), it is evident that the loss of a positive charge on His 4 is the crucial determinant in the pH-dependent loss of activity. As noted above (see "Results" and Fig. 7), His 4 forms a salt bridge with Asp 89 , which is itself bonded to the backbone amide of Tyr 12 , a residue that we have shown previously to be critical for HA binding and is believed to form a hydrogen bond with HA via its hydroxyl hydrogen (28); the Y12F mutant has a much lower (1%) binding constant compared with the wild type (6). Thus, it is not surprising that the change in charge state of His 4 (with increasing pH) and the loss of the His 4 -Asp 89 salt bridge (resulting in an "unshielded" negative charge of Asp 89 buried within the protein) should lead to a local structural perturbation of the ␤1-␣1 loop, which is a key part of the HA-binding site and contains the functionally important residues Lys 11 and  codes 1o7b and 1o7c). The refined structures that include these new restraints are available upon request.

pH-dependent Interaction between TSG-6 and HA
Tyr 12 . It is likely, therefore, that the pK a of Asp 89 is also an important contributor to the pH dependence, and because its pK a is below the maximum of the curve (at ϳpH 5.7), it may contribute to the gain in affinity (i.e. between pH 3.5 and ϳ5.7). Unfortunately, the pK a value for Asp 89 could not be determined in the free protein, although it is likely to be less than that in the HA 8 AN -bound Link_TSG6 (i.e. Ͻ4.2) because the pK a of His 4 (its salt-bridging partner) is raised by 0.5 units on binding to HA 8 AN . However, combining the pK a value for Asp 89 determined for the HA 8 -bound protein with the pK a for His 4 (see above) gives rise to the pH dependence profile shown in Fig. 8B, i.e. this prediction, generated by simply combining the His 4 and Asp 89 components from Fig. 8A, is based on the hypothesis that an intact salt bridge between His 4 and Asp 89 is solely responsible for mediating the observed pH dependence. Although the overall bell shape of the curve correlates well with that seen in microtiter plate binding assays, the point of maximal affinity (pH 5.0) is considerably lower than that observed experimentally (pH ϳ5.7), indicating that these factors alone are probably insufficient to account for the pH dependence of HA binding. The maximal point of the curve occurs at the mean of the pK a values used for His 4 and Asp 89 , so a lower value for Asp 89 (as expected in the free state) would move the maximum to a lower pH (i.e. even further from that observed). When the extent of Link_TSG6 folding with pH is included in the model, this increases the maximal point of the curve (see Fig. 8B), and it is therefore concluded that the pH-dependent folding of Link_TSG6 may be a contributor to the observed gain of affinity up to pH 6 but that another titrating group is likely to be involved.
To produce a pH dependence profile with a maximum at pH ϳ5.7, another titratable residue is required such that when its pK a is combined with that of His 4 , it gives rise to a mean value of ϳ5.7. Based on our determination of pK a values in Link_TSG6 (Table 2), it is clear that the only possible residue is His 45 (pK a of 5.4), where an uncharged state of this amino acid is required for HA binding. Fig. 8C shows the predicted pH dependence of HA binding to Link_TSG6 if His 45 is involved in addition to the His 4 7Asp 89 salt bridge described above. This curve agrees well with the experimental data, and therefore it seems likely that His 45 makes a significant contribution to the gain in affinity of TSG-6 for HA up to ϳpH 5.7; the inclusion of the folding component does not cause any improvement to the model curve. This suggestion is not unreasonable, because His 45 is adjacent to the HA-binding site (see Figs. 1, B and C, and 7A) and experiences chemical shift perturbations (28), a subtle reorientation of its average side-chain conformation (28), and a change in pK a when Link_TSG6 binds to HA 8 AN . In this regard, the loss of a positive charge on His 45 with increasing pH could, for example, serve to stabilize the ␤4 -␤5 loop because of the loss of charge repulsion with Lys 63 . Thus, the full bell-shaped pH dependence curve of HA binding to Link_TSG6 appears to result from a combination of the changes in ionization state of two histidine residues (i.e. His 45 as well as His 4 ).
As described under "Results," it seems likely that proteinprotein interactions occur between TSG-6 Link modules when the protein associates with HA chains containing multiple binding sites and that such interactions might involve His 4 . We have found previously that Link_TSG6 (and the full-length protein) can form stable complexes with polymeric HA that have enhanced binding to CD44 on leukocyte cell lines compared with free HA (12,38); these complexes, which may work through the induction of receptor clustering, are presumed to correspond to cross-linked hyaluronan fibers, which are formed when there is sufficient TSG-6 to saturate all possible protein-binding sites on the HA and when the HA is above a threshold concentration. Interestingly, the H4K mutant of Link_TSG6 was found to have substantially reduced enhancement activity, as also was the case for mutants that have an impaired HA binding function (e.g. Y12F). It was concluded on the basis of these experiments that His 4 has a role in Link_TSG6 self-association (38), which is clearly consistent with the data described here.
Other ligand binding activities of TSG-6 have also been found to be sensitive to pH. For example, the interactions of the G1 domain of aggrecan, TSP1, and heparin with the TSG-6 Link module have very similar pH dependences to that described here for HA (i.e. with maximal binding seen at ϳpH 6.0 and a reduction in binding activity as the pH is increased (8,22,24)). 9 Although the location of the aggrecan-binding site on Link_TSG6 has not been investigated, the interaction can be competed by HA, suggesting that it may overlap with the HAbinding site (22). Interestingly, although the pH dependences of the heparin and HA interactions appear very similar, it is clear that the binding sites for these glycosaminoglycans on the Link module are distinct (8). In the case of TSP1, it is likely that it is interacting at a site outside of both the HA-and heparin-binding surfaces, because Link_TSG6 mutants with reduced HA or heparin binding activities do not have impaired binding to TSP-1 (24). The similarities in the ligand binding properties of TSG-6 for the ligands described above suggest that the mechanism determined here for the pH-dependent interaction of HA with TSG-6 (i.e. stabilization of the ␤4 -␤5 loop and perturbation of the ␤1-␣1 loop with increasing pH) may also underlie the pH sensitivities of other Link_TSG6-ligand interactions. In this regard, we have reported previously that there is likely to be allosteric cross-talk between the heparin-and HA-binding sites (8).
The heparin-mediated augmentation of the potentiation of I␣I anti-plasmin activity by Link_TSG6 is also highly pH-dependent, where there is no significant activity at ՆpH 7 and increasing augmentation as the pH is lowered from 6.75 to 6.0 (8). This is likely to be because of the pH dependence of the interaction between heparin and TSG-6, as the potentiation of I␣I anti-plasmin activity by TSG-6, which occurs via an interaction between the bikunin chain of I␣I and TSG-6 Link mod-   Table 2) are plotted against pH. B and C, two different models of the pH dependence of Link_TSG6 binding to HA, based on various combinations of the components shown in A compared with the experimentally determined bell-shaped affinity curve (smooth line) obtained from microtiter plate assays (22). Each model was determined by taking the product, at each pH value, of the experimentally measured fraction of folded material present and the fractional ionization state of each residue being considered as a component (as determined from the measured pK a value); the resultant curve was normalized to 100% at its maximum value. B, predicted HA binding affinity profile if the pH dependence of binding activity is determined by the residues in the salt bridge between His 4 (pK a 5.8) and Asp 89 (pK a 4.2, estimated from the HA 8 AN -bound protein) with (circles) or without (diamonds) inclusion of the pH-dependent protein folding. C, predicted affinity profile if the pH dependence of binding activity also involves His 45 in addition to the salt bridge between His 4 and Asp 89 , with (circles) or without (diamonds) inclusion of the pH-dependent protein folding. ule (at a site overlapping the HA-binding surface), is similar at pH 6.0 and 7.4 (6,8). We have also found that the stable complexes formed between TSG-6 and HA (when these are incubated together at high concentrations), which have enhanced CD44 binding activities, can be formed at both pH 6.0 and 7.0 (38). In this case, it seems likely that although the initial interaction of TSG-6 with HA will be adversely affected at the higher pH (by the mechanism described here), the formation of stable complexes (by the coalescence of protein-saturated HA chains) will drive the equilibrium toward the protein-bound state, even at suboptimal pH values. To date we have found one functional activity of TSG-6 that is negatively modulated by decreasing pH. The formation of covalent complexes between TSG-6 and the heavy chains (HC) of I␣I (i.e. TSG-6⅐HC1 and TSG-6⅐HC2), which act as intermediates in the transfer of HCs onto HA, is less efficient at pH 6.0 compared with pH 6.5-8.0 (17). Thus, the functions of TSG-6 exhibit differential sensitivities to pH.
As we have noted previously (22), the presence of pH gradients and increasing acidity within tissues (as occurs in inflammation) could control the activity of TSG-6, and this is likely to be of biological significance. For example, in cartilage, the lactic acid released by chondrocytes, in combination with the Gibbs-Donnan equilibrium, could lead to pH values in the deep regions of the tissue as much as 1 pH unit lower than the equilibrating synovial fluid (45), where the pH of synovial fluid can be as low as pH 6.8 in inflammatory arthritis (46). Thus, in inflamed cartilage, where TSG-6 is expressed during rheumatoid arthritis and osteoarthritis (47,48), it is conceivable that TSG-6 may reach close to its maximal HA binding activity in certain regions of the tissue. However, in most locations where TSG-6 is produced (e.g. in the ovarian follicle prior to ovulation (49,50) and in the lungs of asthmatics and smokers (4)), although there is still likely to be lowered pH due to acidosis associated with inflammation, this reduction generally will be smaller than that envisioned for cartilage. Importantly, the gradient of the pH dependence curve means that TSG-6 is exquisitely sensitive to changes in pH, and relatively small decreases in pH can lead to a significant increase in TSG-6 activity. Moreover, the different sensitivities of the various TSG-6-ligand interactions to pH (described above) provides a mechanism for the differential regulation of TSG-6 function by the pH of the particular microenvironment.
The residues identified here as being important in mediating the pH dependence of HA binding in human TSG-6 (i.e. His 4 , His 45 , and Asp 89 ) are conserved across the 10 other species shown in Fig. 9. Although there is a high degree of sequence identity within the TSG-6 Link module (79 -99% compared with the human protein), this nevertheless suggests that the control of HA binding by pH is likely to be a conserved feature of the TSG-6 protein from these species. In this regard, His 29 and His 96 , which were shown here not to be involved in the pH dependence, are only present in 5 of 11 and 8 of 11 species, respectively. Additionally, the HA-binding residues are totally conserved across all species (except for the presence of an arginine instead of lysine at position 11 in four of the species), whereas those shown to mediate heparin binding in human TSG-6 are less highly conserved in frog and fish. All of the intramolecular salt bridges except for that between Lys 20 and Glu 24 are invariantly conserved, indicating that they may have an important role in stabilizing the Link module fold. Consistent with this, mutants R56A and D77A, which would each disrupt an intramolecular salt bridge, have perturbed folds (8,27). Two of the five residues shown to be involved in intramolecular salt bridges are also likely to be important in heparin binding (i.e. Lys 20 and Arg 56 (8)). It is possible that these salt bridges serve to orientate these side chains into a more amenable conformation for capture of heparin.
In summary, we have confirmed that the HA binding activity of TSG-6 can be differentially regulated by pH, thus allowing the exquisite control of TSG-6 function by pH gradients and different microenvironments in vivo. We have identified the molecular basis underlying this pH dependence, thus providing new insights into the interrelationship between function and structure for this important inflammation-associated protein.  (52)) is compared with those from chimpanzee (Chimp; XP_001136392), horse (AAY16441), bovine (CAH74219), dog (XP_533354), rabbit (P98065 (53)), mouse (NP_033424 (54)), chicken (chick; ABB81887), frog Xenopus laevis (Xenopus; AAH88707), zebra fish (ZebraF; XP_695561), and teleost fish Tetraodon nigroviridis (Teleost; Q4S9S8). The two residues of each intramolecular salt bridge are colored as in Fig. 6. This includes His 4 and Asp 89 , which mediate the pH dependence and are depicted in green; His 45 is shown in dark green. The key HA-and heparin-binding residues are indicated by an asterisk (*) and a caret (∧), respectively (8,27,28); additional residues predicted to be involved in heparin binding are shown by lowercase letters. Residues not conserved across the species (compared with human) are colored orange; these all locate to the protein surface. The position of the secondary structure elements relative to the primary sequence is given. APRIL