Distinctive Properties of the Hyaluronan-binding Domain in the Lymphatic Endothelial Receptor Lyve-1 and Their Implications for Receptor Function*

The lymphatic endothelial hyaluronan (HA) receptor Lyve-1 is a member of the Link protein superfamily most similar to the leukocyte HA receptor CD44. However, the structure of Lyve-1 and the nature of its interaction with ligand are obscure. Here we present new evidence that Lyve-1 is functionally distinct from CD44. Using truncation mutagenesis we confirm that Lyve-1 in common with CD44 contains an extended HA-binding unit, comprising elements flanking the N and C termini of the consensus lectin-like Link module, bridged by a third conserved disulfide linkage that is critical for HA binding. In addition, we identify six essential residues Tyr-87, Ile-97, Arg-99, Asn-103, Lys-105, and Lys-108 that define a compact HA-binding surface on Lyve-1, encompassing the epitope for an adhesion-blocking monoclonal antibody 3A, in an analogous position to the HA-binding surface in CD44. The overtly electrostatic character of HA binding in Lyve-1 and its sensitivity to ionic strength (IC50 of 150 mm NaCl) contrast markedly with CD44 (IC50 > 2 m NaCl) in which HA binding is mediated by hydrogen bonding and hydrophobic interactions. In addition, unlike the extended Link module in CD44, which binds HA efficiently when expressed as a soluble monomer (Kd = 65.7 μm), that of Lyve-1 requires artificial dimerization, although the full ectodomain is active as a monomer (Kd = 35.6 μm). Finally, full-length Lyve-1 did not form stable dimers in binding-competent 293T transfectants when assessed using bioluminescent resonance energy transfer. These results reveal that elements additional to the extended Link module are required to stabilize HA binding in Lyve-1 and indicate important structural and functional differences with CD44.

features are also anticipated within the Lyve-1 hyaluronanbinding domain (HABD), which shares 51% similarity with that of CD44, including the presence of a third conserved disulfide linkage (15,23). Thus far, however, the failure to generate a high resolution Lyve-1 crystal structure has hindered further confirmation of predicted similarities, and no detailed structure function analyses have yet been reported for this protein.
The Lyve-1 molecule was originally shown to bind and internalize high molecular weight HA in studies of the recombinant receptor expressed in HEK293T cells (14). Like CD44, the receptor showed specific saturable binding to HA that could be blocked selectively by appropriate Lyve-1 mAbs and that was not diminished by competition with other glycosaminoglycans such as chondroitin or heparan sulfates (13,14). More recent studies of native endogenous Lyve-1 in primary lymphatic endothelial cells and tissue lymphatics indicate that the receptor is constitutively inactivated in vivo through terminal sialylation (24), ostensibly the same mechanism that can maintain CD44 in its low affinity state in non-activated leukocytes. However, unlike CD44, which can be activated by inflammatory agents such as tumor necrosis factor-␣ and lipopolysaccharide in cells expressing the appropriate sialidase activity (25)(26)(27), the factors required for unmasking Lyve-1 remain elusive. These various observations suggest that Lyve-1, although evidently similar to CD44, may have quite distinct properties, possibly explaining why two such related receptors might have evolved such markedly segregated patterns of expression.
Here we have carried out the first detailed structure-function analysis of the Lyve-1 HABD. We provide confirmation that the domain shares significant structural similarity with that of CD44 insofar as it requires contributions from critical N-and C-terminal flanking regions and a third conserved disulfide linkage for function. Furthermore, the predicted HA-binding patch is more compact than that of CD44 despite sharing a similar location and arrangement of critical binding residues. However, unlike CD44, in which HA binding is dominated by mixed hydrophobic and hydrogen bonding character, the Lyve-1 binding patch appears to be dominated by charged residues that support mainly electrostatic interactions. Moreover, in contrast to CD44, where soluble monomeric forms of the extended Link module bind HA with appreciable affinities (K d ϭ 10 -65 M) (28), the equivalent unit in Lyve-1 is shown to be inactive unless either dimerized as an Fc fusion protein or expressed as a full-length Lyve-1 ectodomain monomer (K d ϭ 35.6 M). Finally, we present evidence from physical studies indicating that the full-length functional receptor in transfected 293T cells is almost entirely monomeric. These characteristics indicate unexpected differences in HA binding properties between Lyve-1 and CD44 that are likely to have important consequences for biological function.
HA Preparations-High molecular weight hyaluronan derived from rooster comb (Sigma) was used either without modification or after conjugation with fluorescein or biotin using methods described previously (21,22). Size-fractionated hyaluronan (300 kDa) modified by addition of a single biotin at the reducing terminus (end-labeled hyaluronan) was kindly provided by Prof. Paul DeAngelis (Hyalose, LLC). Purified hyaluronan 4-to 22-mer oligosaccharides (21,22) were obtained from Prof. Tony Day (University of Manchester, UK).
Preparation of Soluble Monomeric Lyve-1 and CD44 Ectodomains-Human Lyve-1⌬232 and the murine Lyve-1 constructs mLyve⌬165 and mCD44⌬172 were constructed as ectodomain Fc fusion proteins in the vector pCDM7IgTEV containing an optimized cleavage site (ENLYFQG) at the N terminus of the Fc region for cleavage by tobacco etch virus (TEV) protease (30) (Invitrogen). After transfection of 293T cells, Fc fusion protein dimers were buffer-exchanged into 50 mM Tris-HCl, pH 8.0, 0.5 mM EDTA, 1 mM dithiothreitol and digested with one unit of enzyme per 3 g of protein for 30 min at 20°C. Cleaved Fc fragments and undigested Fc fusion protein were removed by adsorption to protein A-Sepharose (Sigma), and histidine-tagged TEV protease was removed by adsorption to nickel-nitrilotriacetic acid-agarose (Invitrogen). The integrity of cleaved monomers was assessed by SDS-PAGE and enzymelinked immunosorbent assay, using Lyve-1-or CD44-specific mAbs as appropriate.
Three-dimensional Structure Modeling-A model of human Lyve-1 was constructed using Modeler 9v6. The coordinates of human CD44 (PDB code 1UUH) were used as a template. A CLUSTAL alignment of the sequence of human Lyve-1 to human CD44 was optimized manually by introduction of gaps between hLyve-1 residues 146/147 and 166/167. The resulting alignment and the template structure were provided to the automodel algorithms of Modeler. The resulting model had a GA341 score of 0.9911. This score assumes a value from 0 to 1, with a value of 1.0 indicating a native-like fold.
Immunocytochemical Staining and Flow Cytometry-Transfectants were detached in ice-cold phosphate-buffered saline, pH 7.5, containing 5 mM EDTA, followed by gentle pipetting and stained with the human Lyve-1-specific mAbs 8C or 3A (5 g/ml) or with affinity-purified Lyve-1 polyclonal Ig prior to washing in FACS buffer (5% v/v fetal calf serum, 0.1% w/v sodium azide in phosphate-buffered saline, pH 7.5) and incubation with phycoerythrin-conjugated anti-mouse IgG (Dako, UK) or phycoerythrin-conjugated goat anti-rabbit IgG as appropriate (from Sigma). Cells were then washed (ϫ3) in FACS buffer and resuspended in FACS fix (2% v/v formalin, 0.02% w/v sodium azide in phosphate-buffered saline, pH 7.5) and analyzed on a BD Biosciences FACSCalibur TM flow cytometer with CellQuest TM software.
Hyaluronan Binding Assays-Binding of hyaluronan to Lyve-1 site-directed mutants in HEK 293T cells and to soluble Lyve-1 ectodomain truncation mutants was assessed using either a flow cytometric or a plate binding assay, respectively.
Flow Cytometric Binding Assay-Cells were double stained with fluorescein-labeled HA (Fl-HA) and the Lyve-1-specific (non HA-blocking) mAb 8C followed by phycoerythrin-conjugated anti-mouse IgG-specific antibody (24). Fluorescence was quantitated using a FACSCalibur TM flow cytometer, and data were analyzed with CellQuest TM software. To determine HA binding as a function of Lyve-1 surface density, the phycoerythrin channel was divided into six equal gates for estimation of Fl-HA mean fluorescence intensity during flow cytometric analysis essentially as described in a previous study (24). The HA binding capacity of mutants was scored by comparison with that of wild-type Lyve-1 and categorized as being either 75-100% (ϩϩϩ), 50 -75% (ϩϩ), 25-50% (ϩ), or 0 -25% (Ϫ) at the highest level of expression. Lyve-1 mutants that demonstrated levels of HA binding that were 50% or less of the wildtype value were considered significantly perturbed.
Plate Binding Assay-Truncation mutants were coated on Nunc MaxiSorp plates at 2.5 pmol/well. After blocking, biotin-ylated HA (5 g/ml) was added either alone or in combination with affinity-purified human Lyve-1 polyclonal Ig (final concentration, 50 g/ml) and incubated for 1 h at 20°C. Detection of bound HA was carried out as described previously (21,24). Coating efficiency was determined by a parallel enzyme-linked immunosorbent assay in which MaxiSorp plates were coated with either hLyve-1⌬232Fc or appropriate truncation mutants, and the amount bound was determined using the Lyve-1-specific mAb 8C (5 g/ml) followed by a horseradish peroxidaseconjugated anti-mouse IgG antibody (1/2000 v/v, Pierce). Plates were developed by adding orthophenylenediamine substrate before quenching with one volume of 2.5 M H 2 SO 4 . Absorbance values were read at 490 nm. Amounts coated on the plate were found to be equivalent for the three truncations and the hLyve-1⌬232Fc fusion protein (data not shown).
Competition with HA Oligosaccharides-hLyve-1⌬232Fc was coated on Nunc MaxiSorp plates (2.5 pmol/well). After blocking, biotinylated HA (5 g/ml) was added either alone or in combination with HA oligosaccharides (4-to 22-mer) or high molecular weight HA (calculated as 22-mer equivalents) over a range of concentrations from 0 to 1000 g/ml (0 -0.66 mM) and incubated for 1 h at 20°C. Detection of bound HA was achieved by addition of horseradish peroxidase-conjugated streptavidin (Dako, 1/500 v/v). Plates were developed by adding orthophenylenediamine substrate before quenching with one volume of 2.5 M H 2 SO 4 . Absorbance values at 490 nm were measured in a spectrophotometer.
Effect of Ionic Strength on HA Binding-hLyve-1⌬232Fc and hCD44Fc were coated on Nunc MaxiSorp plates at a concentration of 2.5 pmol/well. After blocking, 250 ng of Fl-HA was added to each well in 50 mM phosphate buffer, pH 7.5, containing NaCl at concentrations ranging from 50 to 1600 mM, and plates were incubated at room temperature for 1 h prior to washing with 50 mM phosphate buffer, pH 7.5, containing the appropriate NaCl concentration (ϫ3). Bound Fl-HA was detected using an automated dual probe fluorescence plate reader (Biotek Synergy HT) with an excitation wavelength of 488 nm. The means of triplicate determinations for binding at each ionic strength value were expressed as a percentage of the maximum value.
For experiments with intact cells, HEK 293T stable transfectants were suspended in 20 mM phosphate buffer, pH 7.3, 5 mM glucose, 5 mM KCl supplemented with NaCl over the range 0 -250 mM with constant osmolarity (500 mosM) maintained by appropriate addition of sucrose. Cell suspensions were then mixed with Fl-HA (5 g/ml final) in the presence or absence of Lyve-1-blocking mAb 3A (50 g/ml) and incubated at 4°C for 30 min followed by washing and fixation prior to analysis by flow cytometry.
Determination of HA Binding Affinity by Surface Plasmon Resonance-Experiments were performed on a Biacore 3000 in HEPES-EP running buffer at a constant flow rate of 10 l/min. Prior to analysis, dimeric and monomeric Lyve-1Fc and CD44Fc fusion proteins were subjected to gel filtration using a HiLoad 26/60 Superdex 200 chromatography column (Amersham Biosciences) to remove aggregates. End-biotinylated 300-kDa HA (Hyalose) was coupled to streptavidin-coated Biacore (SA) chips. The binding analyses for dimeric hLyve-1⌬232Fc and dimeric hCD44⌬267Fc were performed using maximum protein concentrations of 50 M and 349 M, respectively. The maximum protein concentration was 195 M for hLyve-1⌬232 monomer and 1 mM for mLyve-1⌬165 monomer. Controls for nonspecific binding were performed in each case by subtracting responses from a parallel flow cell coated with biotinylated chondroitin 4-sulfate. Values for K d were obtained by non-linear curve fitting of the data to a Langmuir isotherm.
BRET Analysis-For quantitative analysis of Lyve-1 self-association at the cell surface, full-length hLYVE-1 and hCD44 (standard form) cDNAs were ligated into the bioluminescence resonance energy transfer (BRET) vectors prLuc-N3 and pGFP-N3, fusing either luciferase (Luc) or GFP with the end of the cytoplasmic tail. Additional Luc and GFP fusion constructs for CD86 (constitutive monomer) and CTLA-4 (constitutive dimer) were donated by Prof. Simon Davis (University of Oxford). Constructs were introduced into HEK 293T cells by transfection with FuGENE (Roche Applied Science). In each case, Luc:GFP ratios for the transfectants ranged from 2:1 to 1:39, whereas the total amount of transfected plasmid was constant (31). Cells were lifted 24 h post transfection and collected by centrifugation at 600 ϫ g for 3 min before re-suspension (ϳ1.5 ϫ 10 6 /ml) in minimal essential medium. For measurement of BRET, the luciferase substrate DeepBlueC (PerkinElmer) was added (10 M final concentration) to a 100-l sample of cells and light emission in the 370 -450 nm (BRET-A) and 500 -530 nm (BRET-B) spectral regions was collected over 1 s on a Fusion microplate analyzer (PerkinElmer). To determine actual GFP and Luc expression levels for each transfection ratio, separate 100-l samples of cells were excited at 425 nm, and the emission was collected at 500 -530 nm (fluorescence units) before addition of coelenterazine-h (10 M final concentration) and acquisition of the total light emission. BRET eff was calculated as BRET-B/BRET-A after background subtraction (untransfected cells) and correction for luciferase expression alone (typically 7% of LU-A). The values from the fluorescence and luminescence units measurements were converted to concentrations of GFP, and Luc and resulting ([GFP]/ [Luc]) ratios were plotted against BRET eff as described previously (31). Data points were tested for optimal fit to either a hyperbolic or a linear relationship (for [GFP]/[Luc] Ͼ 2.0) by the method of least squares analysis (see also supplemental Fig. S1) to determine behavior of receptors as dimers or monomers according to previously described criteria (31).

Evidence That Lyve-1 Contains an Extended HA-binding
Domain-Previously we demonstrated that an Ig Fc fusion protein comprising residues 1-232 of the hLyve-1 extracellular domain displayed similar HA binding properties to the intact membrane-anchored molecule when assayed in vitro (13,14). To date, however, no further delineation of the HA-binding unit within this domain has been reported. As illustrated by the alignment in Fig. 1A, the primary sequence of Lyve-1 (residues 54 -130) encodes a consensus Link module similar to that of CD44 comprising four key conserved cysteine residues (Cys-61, Cys-85, Cys-106, and Cys-128) together with the six ␤ strands (␤1-␤6) and two ␣ helices that adopt the characteristic lectin fold. As previously reported (13), this unit is bracketed by an additional conserved cysteine pair (Cys-36/Cys-139) equivalent to the Cys-28/Cys-129 pair that makes a critical third disulfide bridge in CD44 linking the N-and C-terminal extensions (beta strands ␤0, ␤7, ␤8, and ␤9) that are essential for correct folding and HA binding in that receptor. To assess whether similar structural elements might be encoded within the Lyve-1 HABD we analyzed the sequence flanking the Link module using the secondary structure prediction programs JPRED and JUFO (Fig. 1A). Despite low sequence homology between Lyve-1 and CD44 in these regions (particularly the C-terminal flanking region) the JUFO program identified potential elements of beta structure in areas corresponding to the ␤0/␤1 junction (residues 36 -45) the ␤7/␤8 region (residues 143-148) and the ␤8/␤9 region (residues 157-164/168 -172). To determine whether these might contribute to bona fide structural extensions of the Lyve-1 HA-binding domain we prepared a series of soluble ectodomain truncation mutants hLyve-1⌬172Fc, hLyve-1⌬141Fc, and hLyve-1⌬130Fc, designed to delete appropriate segments of the C-terminal region, including the critical cysteine, Cys-139. Comparison of the HA binding properties of these mutants using a plate binding assay with bHA revealed a graded reduction in HA binding with increasing truncation for hLyve-1⌬172Fc and hLyve-1⌬141Fc (Fig. 1B). However, the most dramatic loss of HA binding was seen upon deletion of Cys-139 in hLyve-1⌬130Fc. These results are consistent with the presence of additional elements of the Lyve-1 HABD, C-terminal to the consensus Link module, including a third essential disulfide, and constitute important evidence for an extended binding domain analogous to that of CD44.
Identification of Putative HA Contact Residues by Site-directed Mutagenesis-The recent crystallographic analysis of a murine CD44 HA 8 complex identified thirteen contact residues located within the loop regions between the ␤1-␣1, ␣2-␤3, and ␤4 -␤5 strands of the consensus Link module that together constitute a shallow HA-binding groove on one face of the CD44 HABD (22). Inspection of the Lyve-1 amino acid sequence within the equivalent regions indicates that only three of these residues (excluding conserved cysteines), specifically Tyr-87, Ile-97, and Asn-103 (equivalent to hCD44 residues Tyr-79, Ile-88, and Asn-94) are fully conserved in Lyve-1. Interestingly, some of the key contact residues of CD44 such as Arg-41, Tyr-42, and Arg-78 and the aliphatic side-chain amino acids Ile-96 and Ala-98 in the conserved ␤4/␤5 hook-like region have no obvious counterparts in Lyve-1. To assess whether corresponding non-conserved residues might form an HA-binding site in Lyve-1 we mutated 29 individual amino acids distributed mostly within the ␤1/␣1 ␣2/␤3, ␤4/␤5, and ␤5/␤6 loop regions deduced from the sequence alignment in Fig. 1 and a Lyve-1 structural model based on crystal coordinates of human CD44 (corresponding to the so-called type A structure (22); see below). In all, some 43 individual site-directed mutants carrying non-conservative or semi-conservative substitutions were generated, and each was assayed for its capacity to bind Fl-HA after expression on the surface of transfected 293T cells, with assessment for correct folding by the conformation-sensitive 8C mAb. The results as summarized in Table 1 (see supplemental Fig. S1 for quantitative data) show that non-conserva-tive mutation of eight residues in particular (R37A, Y87S/Y87F, I97T, R99A, N103D, K105D/K105Q, K108A, and K117A) led to significant perturbation of HA binding, although one mutant (K117A), which failed to bind the 8C mAb, was discounted on the basis of incorrect folding. Among these residues, Tyr-87, Ile-97, and Asn-103 show exact conservation with the HA-binding counterparts Tyr-79, Ile-88, and Asn-94 in hCD44, whereas Lys-105 and Lys-108 occupy the same positions as HAbinding residues Ile-96 and Ala-99 in hCD44. These features are discussed further in the following section.
Modeling of an HA-binding Surface in Lyve-1-To visualize the HA-binding site in Lyve-1 we prepared a three-dimensional structural model based on the crystal coordinates of human CD44 (PDB code 1UUH). This represents the CD44 "type A" crystal structure (22), which differs subtly from the "type B" structure insofar as the critical HA contact residue Arg-45 within the ␤1/␣1 loop is oriented away from the ligand and more likely represents the "off" or   22)). B, space-filled representation of the Lyve-1 HABD model with key residues comprising the predicted ligand-binding surface defined by site-directed mutagenesis (see Table 1) colored pink. The position of the Cys-85 to Cys-106 structural disulfide that may also contribute to HA binding is colored yellow. An HA octasaccharide docked onto the putative binding site is shown with individual atoms colored turquoise, red, or blue (backbone carbon, oxygen, and nitrogen, respectively). The region corresponding to the Link module extension in CD44 (supplementary lobe) is colored blue. C, space-filled representation of the HABD from human CD44 with contact residues colored pink and a bound HA oligosaccharide in green. D, space-filled representation of the Lyve-1 HABD model with residues identified as part of the epitopes for the Lyve-1-specific mAbs 3A and 8C (see Table 1) colored magenta and orange, respectively. ϩϩϩ ϩ ϩϩϩ ϩϩϩ a Expression of mutated Lyve-1 on the surface of transfected 293T cells. Mean fluorescence was measured after staining samples with an affinity-purified polyclonal antibody specific for human Lyve-1 and compared with the value for wild-type transfectants, where ϩϩϩ ϭ 75-100% (or above), ϩϩ ϭ 50 -75%, ϩ ϭ 25-50%, and Ϫ ϭ 0 -25% of the mean fluorescence value of the wild type. b Fl-HA binding for mutated full-length human Lyve-1-transfected 293T cells compared to binding to wild type. Scores used are as described in footnote a. c Reactivity of mutated Lyve-1-transfected 293T cells with the human Lyve-1-specific monoclonal antibodies 8C and 3A compared to binding to the wild type. Symbols used are as described in footnote a. All mutagenesis experiments were repeated three times. Representative quantitative data are provided in supplemental Fig. S1. lower affinity state of the receptor. The Lyve-1 sequence can be threaded onto the CD44 structure without causing significant structural perturbations, with the exception of some minor variation in the loop region between ␤5 and ␤6 ( Fig. 2A). Mapping of the seven candidate HA-contact residues onto a space-filling model of the Lyve-1 structure (Fig. 2B) reveals that all except Arg-37 come together to form a compact binding patch that encompasses the same characteristic hook-like region (centered on Lys-105) that is a prominent feature of the ligandbinding groove in CD44 (Fig. 2C) as identified recently by x-ray crystallography and NMR analysis of the mCD44/HA8 complex (22). The disruption of HA binding in the R37A mutant is not readily explained, but may signify a long range effect as reflected by the subtle change in mAb 8C reactivity ( Table 1). The near identical location of the ligand interaction surface in the two receptors is highlighted by docking a model of an HA8mer in its CD44-bound conformation into the putative binding groove of Lyve-1 (Fig. 2, B and C).
To further confirm the authenticity of the HA-binding patch, we probed each of the 43 site-directed mutants for binding to the Lyve-1 mAb 3A (supplemental Fig. S1), a reagent that was recently identified as a potent inhibitor of HA binding to both soluble and membrane-attached Lyve-1 (29). As summarized by the data in Table 1, non-conservative mutation of either Tyr-87, Arg-99, Trp-116, or Arg-122 drastically reduced mAb 3A binding but did not disrupt binding to mAb 8C. Plotting of these residues onto the model of Lyve-1 shows they lie within or in close proximity to the predicted HA-binding patch, thus providing further confirmation of its authenticity (Fig. 2D).
A more detailed view of the interaction site also emphasizes the highly conserved spatial arrangement of key HA-binding residues within this Lyve-1 groove, exemplified particularly by Tyr-87, Ile-97, and the Cys-85 to Cys-106 disulfide whose counterparts in CD44 constitute a hydrophobic "pocket" for the methyl group of GlcNAc ( Fig. 3; see also Ref. 22). This suggests that the bound sugar lies in a similar orientation in both receptors. Interestingly, the total number of potential binding residues identified in Lyve-1 (seven) is considerably smaller than the 13 or more identified in CD44 by either site-directed mutagenesis or crystallography and the binding site is significantly more compact (see Fig. 2). Moreover, the character of the putative binding residues in Lyve-1 appears to differ from those of CD44. Most notably, of the six semi-contiguous HA-binding residues identified in Lyve-1, four have charged or polar side chains (Arg-99, Asn-103, Lys-105, and Lys-108), whereas those of CD44 are predominantly aliphatic (Ile-88, Ile-96, Cys-97, Ala-98, and Ala-99) or aromatic (Tyr-42, Tyr-79, and Tyr-105).
Clearly these differences predict major dissimilarities in HA binding properties between Lyve-1 and CD44. Distinctive Features of the Lyve-1 HA-binding Interaction-To assess whether the chemical nature of the HA-protein interaction in Lyve-1 was different to that of CD44, we determined the effect of ionic strength using the bHA plate binding assay. As shown in Fig. 4, binding of bHA to CD44⌬267Fc was largely independent of salt concentration up to and beyond 2 M NaCl. In contrast, however, binding to Lyve-1Fc was fully disrupted at NaCl concentrations as low as 250 mM. Importantly, this sensitivity was not due to an effect of ionic strength on the Fc partner, as similar findings were obtained in analyses with Lyve-1⌬232 monomers  from which this Ig domain had been removed (data not shown). These results are consistent with a binding interaction that is mediated predominantly by charge interactions, in contrast to CD44, which binds HA primarily through H-bonds and hydrophobic interactions (22).
To further characterize the Lyve-1 HA interaction and the unit length of sugar polymer required for optimal binding, we measured the competition for high molecular weight bHA binding by unlabeled HA oligosaccharides of defined length using immobilized Lyve-1 ectodomain Fc fusion protein (Lyve-1⌬232). The results (Fig. 5) show the minimal effective saccharide unit was an 8-mer (IC 50 0.55 mM). Furthermore, there was a sharp increase in potency as oligosaccharide length was increased from HA16 (IC 50 0.25 mM) to HA22 (IC 50 0.05 mM) that may indicate cooperativity. Again these properties of Lyve-1 are distinct from CD44 in which the minimal sugar unit is a 6-mer (21), an unexpected finding in light of the foregoing evidence that the area of the binding patch in Lyve-1 is smaller than that of CD44.
Analysis of Lyve-1 HA Binding Affinity by Surface Plasmon Resonance-The relatively compact dimensions of the HAbinding surface in Lyve-1 and the marked sensitivity of the binding interaction to ionic strength are both suggestive of a low affinity receptor. To investigate this possibility more directly we estimated the apparent HA binding affinities of Lyve-1 and CD44 ectodomain Fc fusion proteins by means of surface plasmon resonance, using high molecular weight HA immobilized on the sensor chip and Lyve-1/CD44 as the analyte. Equilibrium binding analyses (Fig. 6A) revealed that both receptors have broadly similar apparent affinity for HA (K dapp hLyve-1⌬232Fc, 11.4 M; K dapp hCD44⌬267Fc, 24 M). Because Fc fusion proteins constitute covalent dimers linked by an interchain disulfide within the Ig hinge region, the measured binding affinities do not represent true monomeric binding constants. To generate appropriate monomers, we first transferred the hLyve-1⌬232Fc ectodomain fusion construct into the vector pCDM7IgTEV, which contains an engineered TEV protease cleavage motif within the Fc hinge region. As shown in Fig. 6 (A  and B), the HA binding affinity was reduced to only a small extent (K d hLyve-1⌬232 monomer, 35.6 M; K dapp hLyve-1⌬232 dimer, 11.4 M). Likewise, a CD44 monomer comprising the extended Link module (mCD44⌬172) showed only a modest reduction in binding affinity compared with the dimeric receptor (K d mCD44⌬172 monomer, 65.7 M; K dapp hCD44⌬267 dimer, 23.8 M), consistent with the known capacity of this extended Link domain to form an independent folding unit. Curiously, however, an equivalent monomer comprising the extended Link module of Lyve-1 (mLyve-1⌬165Fc) exhibited little or no HA binding in the Biacore (data not shown). Possible explanations for this loss of binding include aggregation or partial unfolding of the monomerized ectodomains. Indeed, we have observed that similar hLyve-1Fc fusion constructs (e.g. Lyve-1⌬165) yielded poor recoveries on size-exclusion chromatography following monomerization. Nevertheless, the particular mLyve-1⌬165Fc monomers analyzed in this study eluted normally, retained normal reactivity with the conformation-sensitive mAbs C1/8 and B1/10 (29) in enzyme-linked immunosorbent assay, and yielded one-dimensional NMR spectra consistent with a folded structure (data not shown). Hence, we conclude that the HA-binding domain of Lyve-1 differs from that of CD44 in requiring additional elements from the C-terminal membrane proximal ectodomain for functional stability.
Does Native Lyve-1 Require Self-association for HA Binding?-To explore the possibility of Lyve-1 self-association, we assayed full-length Lyve-1-transfected 293T cells for BRET, a sensitive technique that measures the proximity-dependent transfer of fluorescence between individual luciferase-tagged and GFP-tagged receptor pairs at the cell surface. For the purpose of comparison we carried out a parallel assay for self-association of CD44, a receptor that has been reported to undergo clustering at the cell surface (32,33). As shown in Fig. 7, plots of BRET eff (GFP emission/Luciferase emission) for increasing ratios of Lyve-1 GFP/Lyve-1 luciferase and CD44 GFP/CD44 luciferase yielded broadly similar curves with the expected approach to a plateau at saturating levels of luciferase donor. Importantly, however, a root mean square analysis of residuals (supplemental Fig. S2) indicated a better fit of the Lyve-1 data to a linear rather than a hyperbolic relationship, implying no significant dimer formation. Moreover, BRET eff was not altered by addition of high molecular weight HA (not shown). In contrast, the CD44 data showed a closer approximation to a hyperbolic relationship, indicating the possibility of dimerization, even in the absence of ligand. Overall these data suggest that Lyve-1 undergoes neither constitutive nor ligand-induced dimerization in the plasma membrane.

DISCUSSION
The analyses presented here uncover key functional differences that distinguish Lyve-1, the lymphatic endothelial HA receptor from CD44, the major HA receptor on leukocytes and mesenchymal cells. Until now these two receptors were assumed to be mutually redundant, both being members of the Link protein superfamily with single HA-binding Link modules located toward the N terminus of a highly N-and O-glycosylated extracellular domain. In terms of function, CD44 mediates leukocyte adhesion, as triggered by inflammatory mediators that exert dual control over synthesis of HA at the luminal surface of blood vessels, and the binding status of CD44, switching the receptor from an "off" state in resting leukocytes to an "on" state in activated leukocytes to promote recruitment from the circulation during inflammation (34 -36). One of the mechanisms behind this switch in CD44 is the enzymatic cleavage of terminal sialic acid residues from N-glycans (25)(26)(27)37) that block ligation of HA by steric hindrance or other effects on the binding site (21). In the case of Lyve-1, however, the physiological function is still open to conjecture, although evidence suggests the receptor could in principle mediate leukocyte-endothelial adhesion through binding HA presented by CD44 on the surface of trafficking leukocytes (13,15,38). In keeping with this notion, there is good evidence to suggest that Lyve-1 is also maintained in an off state in resting cells (i.e. lymphatic endothelium) through a consonant mechanism of sialylation, although unlike CD44, the identity of the factors responsible for activation are not yet known (24). Clearly a better knowledge of the structure and function of Lyve-1 is needed to distinguish its true physiological role from that of CD44.
These studies focused primarily on delineating the HA-binding domain in Lyve-1 and the character of the ligand-binding interaction. Specifically, truncation mutagenesis confirmed the critical importance for HA binding of a third disulfide linkage bracketing the consensus Link module and secondary structure elements within the immediate C-terminal region, features that are diagnostic of the extended HA-binding domain in CD44. Indeed, equivalents for all four additional beta strands of this extension, ␤0, ␤7, ␤8, and ␤9, were anticipated by secondary structure predictions and by molecular modeling of the Lyve-1 HABD using coordinates from the recently derived CD44 crystal structure. The close resemblance with CD44 was underscored by the conserved location and orientation of the putative HA-binding site, which was shown to occupy a complementary position below the hook-like ␤4/␤5 loop region of the consensus Link fold in the Lyve-1 structural model. Notably, the binding site in Lyve-1 also contained the characteristic hydrophobic "pocket" that tethers the GlcNAc methyl group of HA in CD44 (22). Aside from these topographical features, however, the nature of the ligand-binding interaction was shown to differ markedly from that of the CD44 HABD. Specifically, the size of the HA-binding surface was considerably smaller than that of CD44. In addition, most of the residues (4/6) identified by site-directed mutagenesis were either charged or polar (Arg-99, Asn-103, Lys-105, and Lys-108), and the interaction was completely disrupted by conditions of ionic strength Ͼ200 mM. This stands in marked contrast to CD44 in which the interaction with HA is dominated by shape and hydrogen bonding complementarity and is mediated by hydrogen bonds and van der Waals forces that are unaffected by ionic strength (22). Interestingly, the mode of HA binding to Lyve-1 ectodomains resembles that of the selectins, which bind the prototype fucosylated tetrasaccharide sialyl Lewis-X through recognition of the negatively charged carboxylate group of NeuNAc by primarily lysine and arginine residues (39). Indeed the sensitivity of the Lyve-1 HA interaction to ionic strength (I 50 150 mM) is almost identical to that of L-, E-, and P-selectins with sLeX (39,40). However, preliminary experiments with Lyve-1 transfectants indicate this sensitivity to ionic strength is not apparent in intact cells (see supplemental Fig. S3). It may be that the HAbinding domain in the native receptor is shielded from the bulk solvent due to the nature of its local environment. Further experiments will be required to evaluate this possibility.
In addition to the smaller HA-binding surface and different character of HA binding in Lyve-1, we obtained evidence during Biacore analyses that Lyve-1 receptor function is influenced FIGURE 6. Equilibrium binding analysis of Lyve-1 and CD44 to immobilized HA using surface plasmon resonance. A, plots of the maximum HA binding responses with varying concentrations of dimeric hLyve-1⌬232 ectodomain Fc fusion protein (upper) and dimeric hCD44⌬267 ectodomain Fc fusion protein (lower) with inset Scatchard plots. Overlays of the sensorgrams obtained over a range of fusion protein (analyte) concentrations for each dimeric receptor ectodomain fusion protein are shown to the left. Lyve-1⌬232Fc shows incomplete dissociation. B, binding data for monomeric hLyve-1⌬232 (upper) and monomeric mCD44⌬172 (lower) generated by controlled cleavage of dimers with TEV proteinase. Each experiment was repeated a minimum of two times. by elements even beyond the extended Link module. Whereas soluble CD44 constructs that contained only the extended Link module (mCD44⌬172) were fully functional as monomers, those containing the equivalent unit in Lyve-1 (mLyve-1⌬165) were functional only as dimers. Importantly, the mLyve-1⌬165 monomers appeared to be folded correctly as assessed by conformation-selective mAb binding. Moreover, HA binding to similar dimers has been previously documented as both specific and saturable as well as open to efficient competition by free HA in preference to other glycosaminoglycans such as chondroitin and heparan sulfates and selective obstruction by Lyve-1-blocking mAbs (13,14,29). As we did not observe the same functional requirement for dimerization in constructs encompassing the full-length Lyve-1 ectodomain (hLyve-1⌬232), our results therefore point to the involvement of additional regions downstream of the extended Link module in stabilizing the monomeric receptor, forming a larger regulatory unit in Lyve-1 than previously envisaged. Notably, the C-terminal membrane proximal region of the Lyve-1 ectodomain is highly O-glycosylated and carries a single unpaired cysteine residue (13,14). Elements within this section may well interact with or otherwise stabilize the extended Link module. Experiments to define the nature of these elements and to delineate the boundaries of the stable monomer HA-binding unit using limited proteolysis are currently under way in our laboratory.
Finally, we explored the possibility that Lyve-1 might engage in receptor self-association or clustering at the cell surface, analogous to that observed for CD44 (see, for example, Refs. 32 and 33). However, physical measurement of self-association using BRET analyses showed no convincing evidence that the functional receptor on the surface of transfected 293T cells forms either constitutive or ligand-induced Lyve-1 homodimers. It remains a possibility that multivalent interactions between Lyve-1 and large HA polymers may induce longer range clustering of the receptor than that detectable by BRET (Յ10 nm). It is tempting to speculate that such clustering by appropriate conformations of HA in the form of HA-protein complexes such as HA cables (Refs. 41-43; reviewed in Ref. 44) or serum I␣I heavy chain-HA adducts including SHAP (serum hyaluronan-associated protein) (45,46) might have the capacity to activate Lyve-1 just as they activate CD44 in vivo. Further studies will be required to evaluate these possibilities.