Mutations Targeting Intermodular Interfaces or Calcium Binding Destabilize the Thrombospondin-2 Signature Domain*

Thrombospondins (THBSs) are a family of secreted calcium-binding glycoproteins with roles in angiogenesis, cell motility, apoptosis, cytoskeletal organization, and extracellular matrix organization. The THBS-2 signature domain (three epidermal growth factor (EGF)-like modules, a wire module with 13 calcium-binding repeats, and a lectin-like module) binds 30 calcium ions and forms extensive interactions among its parts. We explored the significance of these structural elements by examining the impact of 10 different mutations known to result in pseudoachondrodysplasia or multiple epiphyseal dysplasia when found in the homologous wire and lectin-like modules of thrombospondin-5 (THBS-5). A variety of observations indicate that the mutations result in unstable THBS-5 proteins that aggregate in the endoplasmic reticulum. We introduced the mutations into homologous sites of a THBS-2 construct, for which the crystal structure is known, and determined the effects of the mutations on structure as assayed by differential scanning calorimetry and expression of the epitope for the 4B6.13 conformation-sensitive antibody. Abnormalities were found in one or more of several readouts: stability of interactions between the wire and lectin-like modules, stabilities of the EGF-like and wire modules, expression of the 4B6.13 epitope in soluble protein, and expression of the 4B6.13 epitope in substrate-adsorbed protein at different calcium concentrations. The patterns of abnormalities support the idea that the EGF-like, wire, and lectin-like modules constitute a dynamic and interactive calcium-sensitive structure in which a distortion at one site is transmitted to distal sites, leading to global changes in the protein.

There are five thrombospondins (THBSs) 3 in humans that are involved in diverse processes such as angiogenesis, cell motility, apoptosis, cytoskeletal organization, and extracellular matrix organization (1). These secreted calcium-binding glycoproteins fall into two groups (see Fig. 1A). Group A THBSs (THBS-1 and THBS-2) form trimers and are composed of an N-terminal module, an oligomerization domain, a von Willebrand Factor type C homology module, three THBS type 1 or properdin-like modules, and a signature domain containing three epidermal growth factor-like modules (EGF1, EGF2, and EGF3), a calcium-binding wire, and a lectin-like C-terminal module (see Fig. 1A). Group B THBSs (THBSs 3-5) form pentamers, lack von Willebrand factor type C and properdin modules, and have an extra EGF-like module (EGF2Ј) in the signature domain (see Fig. 1A) (1). Crystal structures of portions of THBS-1 and THBS-2 revealed that the wire module is composed of 13 calcium-binding repeats: eight C-type (23 residues) and five N-type (13 or 15 residues), with a disulfide-bonded insert in the first C-type repeat (repeat 1C) (2,3). The THBS-2 signature domain binds 30 calcium ions, one in EGF2 at the EGF1-EGF2 interface, 26 in the wire, and three in the lectin-like module. The THBS-2 signature domain denatures in two major peaks as measured by differential scanning calorimetry (DSC) (4). Each of these can be deconvoluted into two components (51.0 and 53.1°C for the low temperature peak; 78.8 and 82.7°C for the high temperature peak) (5). The four components likely represent the nonreversible melting of the wire repeat 1C-lectin and wire repeat 9C-lectin interactions followed by the reversible melting of the EGF-like and wire modules (4 -6). The melting events, therefore, are presumed to report on the interactions among the EGF2, EGF3, wire, and lectin-like modules revealed in the crystal structures of the THBS signature domain (2,3) (see Fig. 1B). These interactions involve the wire module wrapping around the lectin-like module, making contact at repeats 1C and 9C and forming a hairpin turn composed of repeats 10N-13C. The hairpin turn is in close association with EGF2 and EGF3 (3). In addition, the insert in wire repeat 1C interacts extensively with EGF3 (see Fig. 1B).
Pseudoachondroplasia (PSACH) and multiple epiphyseal dysplasia (MED/EDM1) are autosomal dominant forms of skeletal dysplasias (7). PSACH is caused exclusively by mutations in THBS-5, also called COMP (cartilage oligomeric protein; the gene name is COMP). PSACH is marked by short stature, lax joints, and early osteoarthritis. MED results in mild to moderate short stature and joint pain and is caused by mutations in collagen IX or matrilin-3, as well as in THBS-5 (7). Patients with disease-linked THBS-5 mutations, of which there are more than 100 (3), therefore suffer from a spectrum of phenotypes from severe PSACH to mild MED. Mice genetically manipulated to lack THBS-5 do not have a PSACH or MED phenotype (8), whereas mice expressing mutant THBS-5 recapitulate much of the phenotype of the same mutations in humans (7,9,10). The mouse results strongly suggest that diseases caused by mutations of THBS-5 are caused by an altered protein rather than absence of the protein. Pathological examinations of cartilage from PSACH patients and chondrocytes expressing mutant THBS-5 revealed enlarged distended endoplasmic reticula containing an accumulation of THBS-5, collagen IX, and matrillin-3 in association with chaperones (11)(12)(13)(14)(15). The accumulation is thought to lead to endoplasmic reticulum stress and increased cell death (9, 16 -18). However, because less protein is secreted to the extracellular matrix, and the resulting extracellular matrix is poorly organized, disease may also be due to an extracellular matrix less able to support the stresses and strains of normal activity (7,9,11,14,15,19,20).
Patients bearing disease-associated COMP mutations are identified with relative ease. Thus, the known COMP mutations can be considered a fairly comprehensive list of residues that are important for signature domain structure and function. When mapped on the structures of THBS-1 and THBS-2 constructs, all of the THBS-5 mutations that cause PSACH or MED localize to the wire or lectin-like modules (2, 3) (see Fig. 1B). Based on such mapping, it has been proposed that disease-causing mutations impair interactions between the wire and lectinlike modules or binding of calcium to the wire, both leading to protein destabilization (2,3). Indeed, equilibrium dialysis, far UV circular dichorism spectra, sedimentation velocity, electron microscopy, and NMR studies of recombinant wire repeats of THBS-5 or full-length THBS-5 protein with mutations of calcium-binding residues suggest that the wire module is unstructured and/or disrupted and binds less calcium than wild-type protein (21)(22)(23)(24). However, Thur et al. (25) found that D361Y and delD469 mutations of calcium-binding residues in recombinant full-length THBS-5 protein do not lead to changes in far UV circular dichroism, even though these same mutations impact far UV circular dichroism, intrinsic fluorescence, and calcium-binding of constructs comprising only the wire. These findings indicate that other parts of THBS-5 compensate for mutations that are highly disruptive of the wire module in isolation.
Reasoning that the THBS-5 mutations serve as a guide on how to perturb the extensive intermodular interfaces and calcium binding found in THBS signature domains, we introduced 10 such mutations into the THBS-2 signature domain, for which the structures of the wire and lectin-like modules are known (see Figs. 1B and 2) (3). We concentrated on six PSACH or MED mutations that would be expected to disrupt interactions between the wire and lectin-like modules and compared these with four of the many mutations that target calcium binding to residues in the wire repeats (see Table 1 and Figs. 1B and 2). Because changes in protein stability and structure are implicated in the pathology of the mutations when found in THBS-5, we examined the impact of the mutations on protein stability using DSC and on protein structure using binding to 4B6.13, an anti-THBS-2 monoclonal antibody to an epitope in repeat 1C (Fig. 1B) for which the interactions among EGF3, the wire, and the lectin-like module are required for binding, as a sensitive probe of globally determined local structure (26). The 10 mutations caused a spectrum of changes in protein stability and structure that are manifested in multiple parts of the signature domain.

EXPERIMENTAL PROCEDURES
Protein Cloning-The pAcGP67.coco baculovirus transfer vector encoding a secretion signal peptide 5Ј to the cloning site followed by DNA encoding a short linker and six-histidine tag, was used to enable baculovirus-driven protein expression, secretion, and subsequent purification (27). The wild-type human and mouse proteins and the mouse protein mutated to carry the 4B6.13 epitope have been previously described (26). The signature domain of human THBS-2 comprises residues 551-1172 with ADP and ARGHHHHHH N-and C-terminal tails. The signature domain of mouse THBS-2 also comprises residues 551-1172, but with ADL and NAGHHHHHH N-terminal and C-terminal tails. The missense mutation or deletion was introduced by PCR mutagenesis into DNA encoding the signature domain THBS-2. Correct orientation and sequence of PCR-amplified DNA were verified by sequencing.
Expression and Purification of Recombinant Proteins-The wild-type and mutated THBS-2 signature domain proteins were expressed by infecting High Five insect cells in SF900II serum-free medium (Invitrogen) at 27°C with high titer virus (Ͼ10 8 plaque-forming units/ml) at a multiplicity of infection of ϳ5. Conditioned medium was collected ϳ65 h post-infection. Histidine-tagged proteins were purified at room temperature from the medium in the presence of 2 mM CaCl 2 using Ni 2ϩnitrilotriacetic acid resin (Qiagen) and fast protein liquid chromatography purification using an ion exchange HiTrap Q HP column (GE Healthcare), as described previously (3). Protein concentration was determined using absorbance at 280 nm less the absorbance at 320 nm and divided by the calculated extinction coefficient of 1.24 ml mg Ϫ1 cm Ϫ1 for all proteins (28).
Differential Scanning Calorimetry-DSC experiments were performed at the University of Wisconsin Biophysics Instrumentation Facility using a Microcal VP-differential scanning calorimeter equipped with Origin 7 software. Proteins (7.9 -15.8 M) were dialyzed into 10 mM MOPS, 150 mM NaCl, 2 mM CaCl 2 , pH 7.5. The scans were conducted from 15 to 95°C at a rate of 60°C h Ϫ1 . The solutions were cooled to 15°C and then repeated to assess reversibility. As previously seen for the wildtype THBS-2 signature domain (4,5), in all proteins the low temperature peak was nonreversible, and the high temperature peak was partially or fully reversible. Appropriate buffer scans were used for reference subtraction, and the data were normalized using protein concentrations, including an estimated 2% dilution upon introduction into the DSC sample cell. The scans were done in duplicate, and the reported patterns and peak melting points were reproducible. Area-fitting analyses were performed with the DSC Origin 7 software package. Manual fit base lines were subtracted from sample scans that were corrected for reference sample and concentration. This curve was then fitted to a non-two-state model, based on the nonreversibility of the low temperature peak, with various numbers of components. The fits were refined using several iterations until the 2 value stabilized. Based on similar analyses on wild-type THBS-1 and THBS-2 signature domain proteins, which both modeled best as four components (5), we started with a fourevent model. The optimal number of components was determined as the minimum that resulted in a substantial decrease in the 2 value. For example, if two components gave a 2 value of 1 ϫ 10 5 , three components gave a 2 value of 1 ϫ 10 4 , and four components gave a 2 value of 9 ϫ 10 3 , we would consider three to be the likely number of components.
Enzyme-linked Immunosorbent Assay (ELISA)-Proteins in 2 mM calcium were tested in competition ELISA for binding to 4B6.13 as described (26) to assess whether various concentrations of soluble protein in 2 mM calcium were able to compete with surface adsorbed recombinant full-length THBS-2 for binding of the monoclonal antibody. These competition ELISA experiments were repeated, each with triplicate data points, on multiple occasions and with at least two separate preparations of protein. The data were normalized to the appropriate control and expressed as the means of the different experiments Ϯ S.E. A variation of the competition ELISA was done with a fixed concentration of soluble protein and varying calcium concentrations, with the lowest concentration maintaining the interaction of 4B6.13 with the absorbed full-length wild-type THBS-2 protein. These experiments used 1.5 g/ml full-length THBS-2 for coating and 1 M soluble protein for competition. These competition ELISA experiments each had triplicate data points. The data were normalized to the appropriate control and expressed as the means of the different data points Ϯ standard deviation. Direct ELISA experiments with monoclonal antibody 4B6.13 were conducted as described (5), plating protein at 10 g/ml and using Costar 3590 high binding polystyrene plates. Polyclonal rabbit anti-human THBS-2 antibodies (29) were used to test whether the various proteins were adsorbed to the plate at similar concentrations. These direct ELISA experiments were repeated, each with triplicate data points, on multiple occasions and with at least two separate preparations of protein. The data were normalized to the appropriate control and expressed as the means of the different experiments Ϯ S.E. To relate the amount of plated protein to density of 4B6.13 epitope in adsorbed protein in 2 mM calcium, proteins were adsorbed at 0.3, 1, 3, 10, and 30 g/ml, and binding to 4B6.13 or rabbit polyclonal antibodies (29) was determined by direct ELISA. These direct ELISA experiments each had triplicate data points. The data were normalized to the appropriate control and expressed as the means of the different data points Ϯ standard deviation. The data were plotted as percentages of control versus log concentration of competing proteins or log calcium concentration in the case of the competition ELISAs and calcium concentration or log protein concentration during adsorption in the case of the direct ELISAs.

Mutations at Intermodular Interfaces Destabilize the THBS-2
Signature Domain-Mapping the THBS-5 PSACH and MED mutations onto the THBS-2 signature domain structure reveals clusters of mutations of residues that lie at the intermodular interfaces between wire repeat 1C and one side of the lectin-like module and between wire repeat 9C and the opposite side of the lectin-like module ( Fig. 1B and Ref. 3). These residues are not predicted to bind calcium (2,3). To evaluate two related hypotheses, that the mutations disrupt the interface (3) and that the low temperature nonreversible DSC components reflect the melting of the interfaces (5), we investigated mutations that lie at either side of the two interfaces (Table 1 and Fig. 2) for effects on the DSC profile of the protein (Figs. 3 and 4). Thus, choosing from among eight described disease-associated missense mutations involving non-calcium-binding residues at the wire repeat 1C-lectin interface, we introduced four into homologous sites in the THBS-2 signature domain: L697P and S726L in wire repeat 1C, and T1013M and T1013R in the lectinlike module ( Fig. 2A). The two mutations of Thr 1013 were selected because protein trafficking studies have indicated a Ϫ45°around the x axis. This image shows only the 10 mutations used in this paper, which are labeled with arrows. In addition, also labeled are Leu 703 and His 722 , the critical residues for binding to 4B6.13, used in Figs. 5 and 6. Glycosylation sites are shown as green sticks with red oxygen atoms and blue nitrogen atoms. The image was prepared with Pymol (36). difference between these proteins, with the T1013R homologue leading to more severe trafficking defects (20). Of the seven described disease-associated missense mutations involving non-calcium-binding residues at the wire repeat 9C-lectin interface, we introduced two into homologous sites in the THBS-2 signature domain: G868E in wire repeat 9C and G1147S in the lectin-like module (Fig. 2C).
Compared with the wild-type THBS-2 signature domain in 2 mM calcium, all the interface mutations, irrespective of whether the amino acid change was located in the wire or the lectin-like module, resulted in changes in the low temperature peak that encompasses the nonreversible events presumed to arise from melting of wire-lectin interfaces ( Table 1 and Fig. 3A). Change ranged from a 6°C decrease for the overall melting temperatures of the repeat 1C-lectin interface mutants to nearly complete loss of the peak for the repeat 9C-lectin interface mutants (Table 1 and Fig. 3, A and B).
Deconvolution of the DSC curves of the two repeat 1C mutants (L697P and S726L) fit best as four components (Table  1 and Fig. 4, B and C). The shapes and magnitudes of deconvoluted components, however, were different. L697P shows a larger decrease in the height of one of the deconvoluted nonreversible low temperature components than the other, suggesting a selective disruption of the one of the wire-lectin interfaces compared with the other. In addition, the ratio of heights of components three and four are switched relative to these components in all the other proteins, such that component four is smaller than component three, indicating an impact of the mutation on how the EGF-like and/or wire modules melt. For S726L, there was a decrease in the magnitude of both the first and second nonreversible low temperature components, but the impact on the first component was greater, again suggesting a selective interference of one of the wire-lectin interfaces compared with the other. The size and shape of the third and fourth components were not significantly altered for S726L, suggesting less of an impact on the EGF-like and wire modules melting than L697P. The two different Thr 1013 mutations resulted in similar DSC curves that each fit best as three components ( Table 1) with loss of one of the low temperature interface melts but little to no impact on the EGF-like or wire module melts (Fig. 4, D and E). G868E at the repeat 9C-lectin interface also deconvoluted best into three components, suggesting the loss of one of the low temperature wire-lectin interface melts (Table  1 and Fig. 4F). G1147S fit best as four components (Table 1), with the low temperature components being small (Fig. 4G), indicating a large impact on both of the wire-lectin interfaces. The high temperature peak of the G868E and G1147S proteins was not different from wild type (Table 1 and Fig. 3B), indicating little to no impact on the EGF-like or wire module melts.
The finding that all of the interface mutations altered one or both of the two nonreversible low temperature components is consistent with the hypothesis that these components arise from the wire-lectin interfaces. Because the components are close together in deconvolutions of wild-type protein (Fig. 4A) and because they are altered in mutant proteins, it is not possible to assign a given wire-lectin interface to a given component.  THBS-2 wire structure and the partial THBS-1 wire structure show an intricate conserved calcium-binding strategy, employing residues that are found in all human THBSs, including THBS-5 (30). Calcium is coordinated by a combination of groups, including monodentate and bidentate coordinating aspartic acid side chains, main chain carbonyl groups, and side chain carboxyl or carbonyl coordination through water (30). There are 47 missense mutations that involve calciumbinding residues and when present in THBS-5 are associated with PSACH or MED (3,31). To examine the impact of THBS-5 mutations targeting calcium-coordinating residues, we introduced four mutations of aspartate residues into the THBS-2 signature domain (Table 1 and Fig. 1B): D738V in wire repeat 2N (Fig. 2B), D789Y in wire repeat 5N (Fig. 2B), delD897 deleting an aspartate in wire repeat 10N (Fig.  2D), and D901G in wire repeat 10N (Fig. 2D).
DSC analysis revealed significant destabilization of each mutated protein compared with wild type. D738V in repeat 2N resulted in a minor change in the first overall peak, a large change in the position of the low temperature peak (shift of Ϫ6.5°C), and a shoulder on the high temperature peak (Table 1 and Fig.  3C). This curve fit best as four components (Table 1 and Fig. 4H). The component sizes and shapes suggest a greater disruption of one of the wire-lectin interfaces than the other and a large shift in either the high temperature EGF-like or wire module melt. The D789Y mutation in repeat 5N resulted in moderate decreases in temperatures of both peaks (Table 1 and Fig. 3C). This curve fit best as four components (Table 1 and Fig. 4I). The shapes of the components mimic the wildtype protein and suggest a similar impact on all of the elements of the signature domain. Both delD897 and D901G resulted in striking loss of the low temperature peak but only slight changes to the high temperature peak (Table 1 and Fig. 3C). These curves both fit best as three components (Table 1 and Fig. 4, J and K), suggesting the loss of one of the low  Disruption of Protein Stability Is Specific for Disease-associated Mutations-As a control for the changes found in the DSC profile, we studied the signature domain of wild-type mouse THBS-2, which has the same spacing of residues as the human protein and changes in residues at 37 sites (of which 17 are conservative), and mouse THBS-2 containing N703L and K722H mutations in repeat 1C. The wild-type and mutated mouse proteins had been produced to map the 4B6.13 epitope (26). Wild-type mouse THBS-2 signature domain had a DSC with major peak temperatures at 52.2 and 79.9°C and fitted components at 51.0, 53.2, 69.6, and 79.7°C (Table 1 and Figs. 3D and 4L). The greatest difference between the DSC profiles of wild-type human (Fig. 4A) and mouse THBS-2 signature domain is in the position of the presumptive EGF-like module melt, which shifts to a lower temperature and results in a shoulder on the high temperature peak. Differences were not observed between the DSC profiles of the wild-type and mutated mouse proteins (major peaks at 52.5 and 81.5°C) ( Table 1 and Figs. 3D and 4M). These experiments, therefore, indicate that changes in the DSC profiles of mutant human THBS-2 signature domain proteins report significant structural changes that might be expected to cause disease.
Mutations Alter the Conformation of the THBS-2 Signature Domain in Solution-We investigated the impact of the interface mutations on the ability of the soluble protein to express the epitope of the conformation-sensitive monoclonal antibody 4B6.13 in wire repeat 1C (Fig. 5). Wire repeat 1C residues Leu 703 and His 722 (Fig. 1B) are necessary for full expression of the 4B6.13 epitope, which is conformation-determined and requires the presence of the EGF3, wire, and lectin-like modules in a calcium-replete form (26). Thus, 4B6.13 acts as a reporter of wire repeat 1C structure, the calcium-bound status of THBS-2, and interactions within the EGF3-wire-lectin complex. As previously described (26), in 2 mM calcium, wild-type signature domain competes with recombinant full-length THBS-2 for binding to 4B6.13, inhibiting 50% of 4B6.13 binding to full-length recombinant THBS-2 at a concentration of 0.1 M protein in competition ELISAs (Table 1 and Fig. 5A). The interface mutations located in wire repeat 1C, L697P and S726L, caused a 30-fold increase in concentration of the soluble mutant protein over wild-type protein that was required for competition of 4B6.13 binding to adsorbed full-length THBS-2 (Table 1 and Fig. 5A). Both of the Thr 1013 mutant proteins (in the lectin-like module at the interface with wire repeat 1C) competed poorly in the competition ELISA, with a greater difference for the T1013R than the T1013M protein (Table 1 and Fig. 5A). The G868E mutation of wire repeat 9C bound 4B6.13 like wild-type protein in the competition ELISA (Table 1 and Fig. 5A). However, the complementary mutation in the lectinlike module, G1147S, caused greatly altered binding to 4B6.13, resulting in a Ͼ30-fold increase in soluble mutant protein over wild-type protein that was required for competition of 4B6.13 binding to adsorbed full-length THBS-2 (Table 1 and Fig. 5A). Of the calcium-binding mutants, both the D738V and D789Y mutated proteins bound 4B6.13 normally in 2 mM calcium, as demonstrated by competition ELISA (Table 1 and Fig. 5B). However, the wire repeat 10N mutations, delD897 and D901G, did not compete for 4B6.13 binding at any concentration tested in competition ELISAs (Table 1 and Fig. 5B).
Mutations Alter the Impact of Calcium on the Conformation of the THBS-2 Signature Domain in Solution-4B6.13 binds to wild-type THBS-2 signature domain protein only when THBS-2 is in a calcium-replete form (26). To examine whether the mutations alter the sensitivity of the THBS-2 signature domain to calcium, we tested proteins in solution, using competition ELISAs (Table 1 and Fig. 5C). These assays are limited by the requirement that 4B6.13 binds to the surface-adsorbed ligand and were therefore conducted in calcium concentrations that allow binding of 4B6.13 to full-length THBS-2. The seven mutants with little to no binding to 4B6.13 in 2 mM calcium (Fig.  5, A and B) did not gain the epitope in lower calcium (Table 1). More detailed experiments focused on the mutants that showed normal binding in 2 mM calcium: D738V, D789Y, and G868E ( Table 1). The wild-type signature domain competed for 4B6.13 at all calcium concentrations tested (down to 160 M). The D738V protein showed a requirement for increased calcium, with an EC 50 of ϳ200 M calcium. The D789Y protein showed a slight shift toward an increased calcium requirement, with a 83% competition at 160 M calcium versus 97% competition by the wild-type protein at 160 M calcium. Lastly, the G868E protein showed a requirement for greatly increased calcium concentrations, with an EC 50 of ϳ400 M calcium.
Mutations Alter the Impact of Calcium or Binding to Plastic on Surface-adsorbed THBS-2 Signature Domain-As previously reported (26) and described above for full-length THBS-2, 4B6.13 binds to adsorbed wild-type THBS-2 signature domain protein in direct ELISA transitioning from 0 to 100% binding within a range of 140 -200 M calcium and with a midpoint of 150 M (Table 1 and Fig. 6, A and B). The interface mutations located in wire repeat 1C, L697P and S726L, caused a loss or reduction of 4B6.13 binding in direct ELISA (Table 1 and Fig. 6A). The two different mutations of Thr 1013 bound 4B6.13 as well as wild-type protein did in direct ELISA (Table 1 and Fig. 6A), which contrasts to the lack of binding by soluble mutant protein in competition ELISAs (described above). Protein with the G868E mutation in wire repeat 9C bound 4B6.13 like wild-type protein in the direct ELISAs (Table 1 and Fig. 6, A  and B). However, the complementary mutation, G1147S, resulted in low binding to 4B6.13 in direct ELISA (Table 1 and Fig. 6A). Of the calcium-binding mutants, the D738V and D789Y proteins required increased calcium, 600 and 250 M, respectively, for 4B6.13 binding in direct ELISA assays (Table 1 and Fig. 6B). Adsorbed protein bearing the wire repeat 10N mutations, delD897 and D901G, which did not compete for 4B6.13 binding in solution (described above), bound 4B6.13 like adsorbed wild-type protein in direct ELISA assays (Table 1 and Fig. 6B). The discrepancy between the competition and direct ELISAs seen for the S726L, G868E, T1013M, T1013R, delD897, and D901G mutants suggested that the 4B6.13 epitope becomes available upon adsorption of these mutants to polystyrene surfaces, i.e. adsorption to the plastic results in a protein conformation in which the 4B6.13 epitope is expressed. This possibility was explored by relating epitope recognition in 2 mM calcium to density of adsorbed mutant protein. When adsorbed to the plastic in increasing concentrations, more S726L, G868E, T1013R, D901G, or delD897 protein than wild-type protein was required for saturation of the binding to 4B6.13 (Table 1 and Fig. 6C). This result indicates that four mutant proteins can gain the 4B6.13 epitope upon adsorption to the plastic less efficiently than wild-type protein. The T1013M protein, which showed a smaller difference between the two ELISAs than the T1013R mutant, required only a small increase in protein concentrations compared with the other mutants (Table 1 and Fig.  6C). The D789Y mutant proteins, which did not show any disagreement between the direct and competition ELISA in 2 mM calcium, showed only small increases in adsorbed protein con-  centrations required to bind 4B6.13 compared with wild type. (Table 1 and Fig. 6C).

DISCUSSION
Based on THBS-5 mutations that cause PSACH or MED, we introduced 10 mutations into the signature domain of THBS-2. Homology considerations and the crystal structure of the THBS-2 signature domain predict that the chosen residues are involved in either an interface between the wire and lectin-like modules or in the coordination of calcium ions (Fig. 2). Eight different patterns of altered read-outs were detected in the following parameters: 1) change in the low temperature components in DSC, 2) change in the high temperature components in DSC, 3) binding of 4B6.13 in solution, and 4) binding of 4B6.13 when adsorbed to plastic ( Table 1). The finding that characteristic patterns of alterations were found for the two mutations of Thr 1013 and for the two repeat 10N aspartate mutations indicates that patterns are specific for specific residues and regions of the signature domain.
The similarities among the DSC profiles of wild-type human THBS-2 signature domain, wild-type mouse THBS-2 signature domain, and the mutated mouse THBS-2 signature domain changed to express the epitope for the 4B6.13 mouse anti-human THBS-2 antibody are further evidence that the diseaseassociated mutations disrupt structure specifically. Only two of the 37 sites that are different between mouse (Ser 882 and Thr 955 ) and human THBS-2 (Ala 882 and Ser 955 ) signature domain proteins are sites of disease-causing mutations in THBS-5 (S454R and T527A) (32,33), and at these two sites the mouse THBS-2 signature domain has the same amino acids as the wild-type human THBS-5 signature domain. We have mapped the 37 differences between mouse and human THBS-2 signature domain proteins on the crystal structure of the human THBS-2 protein and found only one pair of residues (Ser 955 -Ile 941 ) that are within 4 Å of one another, as is regularly the case when the protein in one species contains a residue that causes disease when present in the homologous protein of another species and is accompanied by an apparently compensating change of a nearby residue (34). Further analyses will be required to learn what changes in signature domains are compatible with normal structure and function.
Mutation of residues involved in the wire 1C-lectin interface, L697P, S726L, T1013M, and T1013R ( Fig. 2A), led to a loss or change of one or both of the interface melts. Wire 1C mutations L697P and S726L, but not the complementary lectin-like module Thr 1013 mutations, led to additional changes in the melts of the EGF-like and wire modules. These effects were most apparent in the L697P mutant. Surface-adsorbed L697P protein did not express the 4B6.13 epitope, and L697P caused a 30-fold increase in the soluble mutant protein over wild-type protein that was required for competition of 4B6.13 binding to adsorbed full-length THBS-2. Soluble S726L and Thr 1013 mutant proteins also were unable to bind 4B6.13 in competitive ELISAs except at high protein concentrations. However, both of these proteins expressed the 4B6.13 epitope when adsorbed to plastic.
Of the 27 PSACH or MED missense mutations located between repeats 1C and 9C, i.e. between the interfaces with the lectin-like module, we chose two to introduce into the THBS-2 signature domain (Fig. 2B). D738V, located at position 10 of wire repeat 2N, replaces an aspartate that coordinates one calcium ion through its main chain and another through its side chain via water. D789Y, located at position two of wire repeat 5N, replaces an aspartate that uses a side chain oxygen to coordinate one calcium ion (Fig. 2B). D738V impacted both of the interface melts modestly and caused large changes in the stability of the EGF-like and wire modules, which were impacted so much that one of the high temperature components moved enough to form a discernible shoulder on the second overall peak. D789Y destabilized the interfaces and also caused moderate decreases in the stability of the EGF-like and wire modules. Both soluble D738V and D789Y proteins bound 4B6.13 normally in 2 mM calcium. However, both of the proteins showed a decreased sensitivity to calcium, as soluble and surface-adsorbed protein, as shown by a requirement for increased calcium concentrations to gain the 4B6.13 epitope relative to wild-type protein. In each case, the D738V protein exhibited a more dramatic impact. Surface-adsorbed D738V mutant required 4-fold higher calcium than wild type in the direct ELISA, whereas the D789Y mutant required 2-fold higher calcium than wild type. This difference between the D738V and D789Y mutants most likely is because the 4B6.13 epitope is in wire repeat 1C, adjacent to repeat 2N harboring D738V, but removed from repeat 5N, harboring D789Y. Similarly, a THBS-5 wire construct harboring D361Y (homologous to D789Y) in repeat 5N requires ϳ4-fold more calcium than wild type to induce quenching of a tryptophan reporter in repeat 4C, which is immediately adjacent (25). Studies of a reporter tryptophan in repeat 1C of THBS-1 titrated with calcium ions in progressively longer constructs ending with wire repeats 2N, 3C, or 4C showed that calcium binding becomes more cooperative with addition of wire repeats and that the altered kinetics of calcium binding caused by the constructs harboring the Ser 700 polymorphism in wire repeat 1C are only normalized when four repeats are present (35). Taken together, these results indicate that binding of calcium to a given repeat is modulated by calcium binding to at least four repeats in each direction, with closer repeats having the most effect. This conclusion raises the question of whether the D738V or D789Y mutant proteins at higher calcium concentrations bind calcium ions despite the loss of a calcium binding residue. A THBS-2 signature domain construct with a serine instead of the calciumcoordinating asparagine at position 10 of wire repeat 1C (N702S) binds the full complement of calcium ions (5). By extrapolation, one would predict that all of the calcium ions are present in the D738V or D789Y protein when a suitably high calcium concentration is present. However, unlike the Ser 702 protein, which was successfully crystallized, we were unable to obtain useable crystals of the D738V or D789Y protein 4 and thus cannot test this speculation directly.
At the wire 9C-lectin interface (Fig. 2C), the G868E and G1147S proteins had major changes in the melting of the wirelectin interfaces and little to no changes in the melting of the EGF-like or wire modules. A 30-fold increase of soluble G1147S over wild-type protein was required for competition of 4B6.13 binding to adsorbed full-length THBS-2, and surface-adsorbed G1147S bound 4B6.13 poorly in direct ELISA. The G868E protein did not show a disruption of binding to 4B6.13 in the 2 mM calcium competition ELISA or direct ELISAs done at various calcium concentrations. However, a disruption of binding to 4B6.13 was revealed in low calcium competition ELISAs, resulting in a requirement for increased calcium concentrations for binding. This apparent disagreement between the direct and competition ELISAs using a calcium gradient can be partially explained by the observed altered binding of G868E to the ELISA plastic.
Mutations of calcium-binding residues of repeat 10N, delD897 and D901G (Fig. 2D), resulted in severe disruption of the wire-lectin interfaces and little to no changes in the melts of the EGF-like or wire modules. Each of these mutations involves an aspartate that uses side chain oxygen atoms to coordinate two calcium ions (Fig. 2D). Also, Asp 897 binds an additional calcium ion using the main chain carbonyl oxygen and is in the position of Asp 469 of THBS-5, the deletion of which is the most common PSACH mutation (31). These mutated proteins did not compete for 4B6.13 binding in competition ELISA but bound 4B6.13 when adsorbed to plastic. Both the delD897 and D901G mutations are located in the repeat 10N-13C hairpin turn (Fig. 2D), where many of PSACH and MED missense mutations, 25 at present, have been localized (Fig. 1C) (2,3). The hairpin turn is stabilized by a hydrophobic core composed of conserved residues and has been hypothesized to be critically important to the overall fold and stability of the THBS wire (2). Like the insert in wire repeat 1C, which has extensive contacts with EGF3, the hairpin turn has interactions with itself, EGF2, EGF3, and the lectin-like module (Fig. 2D). Therefore, the hairpin turn may play a particularly important role in sensing and transmitting conformational changes throughout the signature domain. Antibody studies suggest a model in which sequences in wire repeats 10N/11C and 12N/13C unravel upon removal of calcium and thus mediate the extension of THBSs that occurs under low calcium conditions (26). Such a role is compatible with the finding that delD897 and D901G lead to changes in the stabilities of the wire-lectin interfaces, even though not present at those interfaces. Significant structural impacts of delD469 (homologous to delD897) in THBS-5 have been demonstrated by far UV circular dichroism and NMR of wire constructs (21,24) and sedimentation velocity and electron microscopy of fulllength protein (23).
The combination of DSC and 4B6.13 read-outs facilitates examination of how the mutations impact the stability and structure of the overall signature domain and its specific parts. The impact of the mutations is clearly not limited to the local environment but rather is propagated to distant sites. The present findings indicate that this propagation is felt at least five repeats away along the length of the wire module and also across the lectin-like module even in the presence of 2 mM calcium, with the interfaces of the lectin-like module and wire repeats 1C and 9C serving as contact points. Both of the Thr 1013 mutations (in the lectin-like module at the interface with wire repeat 1C) cause structural changes 13 Å away to alter binding of 4B6.13 to repeat 1C. More strikingly, the G1147S and G868E mutations (of the lectin-like module and wire repeat 9C, both at the interface) result in structural changes 35 Å away to alter 4B6.13 binding. The D789Y amino acid change in wire repeat 5N is located 20 Å from either the wire 1C-lectin or wire 9C-lectin interface and yet causes large changes in the thermal stabilities of these interfaces. Thus, the changes seen in the D789Y DSC profile reflect changes in distant structural features, indicating a propagation of the conformation deformation through the wire module. Lastly, both of the mutations involving the direct coordination to two or three calcium ions, delD897 and D901G, located in the wire 10N-13C hairpin turn and not at the interfaces, impact the wire-lectin interfaces, as evidenced by a loss of one of the low temperature melting components. These structural changes are transmitted 45 Å to disrupt 4B6.13 binding to wire repeat 1C. Thus, the wire and wire-lectin interfaces provide a mechanism whereby local changes in protein structure are propagated to distant sites in the protein, causing long range conformational changes and presumably leading to changes in function.
Studies of the S726L, G868E, T1013R, delD897, and D901G mutants revealed an unexpected and interesting phenomenon: more expression of the 4B6.13 epitope by mutant proteins adsorbed to polystyrene ELISA plates than in solution. This finding indicates that the signature domains of the mutant proteins in 2 mM calcium are deformable. Further, it raises the question of whether ligands of the wild-type signature domain induce a conformation that favors expression of the 4B6.13 epitope. For instance, do ligands that preferentially bind in low calcium induce the epitope?
The results from the mutated THBS-2 signature domain proteins can be extended to the pathological mechanism of PSACH and MED, but with important reservations. Although the signature domains of THBS-2 and THBS-5 seem very similar, there are some notable differences (30). THBS-5 contains an extra EGF-like module, EGF2Ј (Fig. 1A), which is similar but not identical to EGF2 of THBS-2 yet would be predicted to replace EGF2 in the interaction with the wire 10N-13C hairpin turn. It is unclear how this module would change the structure of the signature domain and, of particular interest, the structure of the wire 10N-13C hairpin turn. Other differences when comparing THBS-5 to THBS-2 include the lack of the three-residue NAT glycosylation recognition sequence found in the wire 1C insert, an unpaired cysteine in the lectin-like module, and a C-terminal tail with 13 extra amino acids. Nevertheless, our THBS-2 results are compatible with proposed pathogenic mechanisms causing PSACH and MED (31). We found evidence for overall destabilization of the proteins, as demonstrated by decreases in melting temperatures observed in all of the mutants. Indeed, some temperatures approached the physiological range (midpoints as low as 43.2°C). We show that mutations that localize to sites of wire-lectin interfaces are important for protein stability and structure. Interestingly, the T1013R mutation caused more disruption of 4B6.13 binding than T1013M, correlating with the differences in trafficking of THBS-5 proteins bearing the T585R or T585M mutations (20). In THBS-2, the wirelectin interfaces appear to be acting as sensors and transmitters of structural changes within the signature domain. Future stud-ies on these mutations within the THBS-5 signature domain will be needed to see whether the communication network operates in a similar fashion in THBS-5.