Defining the Native Disulfide Topology in the Somatomedin B Domain of Human Vitronectin*

The N-terminal 44 amino acid residues of the human plasma glycoprotein vitronectin, known as the somatomedin B (SMB) domain, mediates the interaction between vitronectin and plasminogen activator inhibitor 1 (PAI-1) in a variety of important biological processes. Despite the functional importance of the Cys-rich SMB domain, how its four disulfide bridges are arranged in the molecule remains highly controversial, as evidenced by three different disulfide connectivities reported by several laboratories. Using native chemical ligation and orthogonal protection of selected Cys residues, we chemically synthesized all three topological analogs of SMB with predefined disulfide connectivities corresponding to those previously published. In addition, we oxidatively folded a fully reduced SMB in aqueous solution, and prepared, by CNBr cleavage, the N-terminal segment of 51 amino acid residues of intact vitronectin purified from human blood. Proteolysis coupled with mass spectrometric analysis and functional characterization using a surface plasmon resonance based vitronectin-PAI-1-SMB competition assay allowed us to conclude that 1) only the Cys5–Cys21, Cys9–Cys39, Cys19–Cys32, and Cys25–Cys31 connectivity is present in native vitronectin; 2) only the native disulfide connectivity is functional; and 3) the native disulfide pairings can be readily formed during spontaneous (oxidative) folding of the SMB domain in vitro. Our results unequivocally define the native disulfide topology in the SMB domain of human vitronectin, providing biochemical as well as functional support to the structural findings on a recombinant SMB domain by Read and colleagues (Zhou, A., Huntington, J. A., Pannu, N. S., Carrell, R. W., and Read, R. J. (2003) Nat. Struct. Biol. 10, 541–544).

Vitronectin is a multidomain plasma glycoprotein involved in a variety of biological processes such as cell adhesion, cell migration, modulation of the immune system, and regulation of blood coagulation and fibrinolysis (1,2). Many regulatory functions of vitronectin result from its ability to interact with plasminogen activator inhibitor 1 (PAI-1), 2 a member of the serine protease inhibitor superfamily that inhibits both tissue-and urinary-type plasminogen activators (3)(4)(5)(6). PAI-1 plays important roles in thrombosis and fibrinolysis and has been implicated in hemostasis, angiogenesis, and tumor metastasis (7)(8)(9)(10)(11). Low abundant PAI-1 circulates in blood complexed with vitronectin (12); unliganded PAI-1 undergoes rapid "selfinactivation" into an inactive "latent" form in which the inhibitory reactive-site loop inserts as a new strand into the main ␤-sheet of the molecule (13)(14)(15). Ample evidence suggests that vitronectin regulates the activity of PAI-1 and PAI-1-mediated cellular events by stabilizing the active form of the inhibitor and delaying its conformational transition to the latent state (16 -18). Conversely, PAI-1 mediates, via binding, vitronectin-dependent cell adhesion and migration (19 -23). Thus, molecular recognition between vitronectin and PAI-1 is of great importance in biology.
Loskutoff and colleagues (24 -27) first pinpointed a high affinity PAI-1-binding site in vitronectin to the N-terminal somatomedin B (SMB) domain of the adhesive glycoprotein. The SMB domain consists of 44 amino acid residues (28 -31), including eight conserved cysteines that form four functionally indispensable intramolecular disulfide bridges (24). Kamikubo et al. (32) reported that an N-terminal fragment of VN of 97 amino acid residues, expressed in the cytoplasm of Escherichia coli and purified by immuno-affinity chromatography, showed activity in PAI-1 binding and antibody recognition similar to urea-activated vitronectin (uVN) 3 purified from human blood. CNBr cleavage of the recombinant VN fragment released the SMB domain elongated C-terminally by an extra seven-residue RGD motif (we term rVN 1-51 -1). Partial reduction and S-alkylation of rVN 1-51 -1 coupled with proteomics techniques identified a consecutively arranged, uncrossed pattern of disulfide bridges in the SMB domain, i.e. Cys 5 -Cys 9 , Cys 19 -Cys 21 , Cys 25 -Cys 31 , and Cys 32 -Cys 39 (32). As rVN 1-51 -1 was functionally indistinguishable from intact uVN, the linear disulfide topology was thought to exist in native vitronectin (32).
Read and colleagues (18) subsequently reported the crystal structure of PAI-1 complexed with a similarly obtained recombinant VN 1-51 (we term rVN 1-51 -2), which was shown to contain a crossed pattern of disulfides, i.e. Cys 5 -Cys 21 , Cys 9 -Cys 39 , Cys 19 -Cys 32 , and Cys 25 -Cys 31 . As Zhou et al. (18) did not address the difference in disulfide topology between rVN 1-51 -1 * This work was supported by National Institutes of Health Grants AI058939 and AI061482 (to W. L.). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. 1 To whom correspondence should be addressed. and rVN 1-51 -2, the question of which disulfide linkages represent the topology in native vitronectin was left unanswered. Added to the brewing controversy was a recent finding by Peterson and colleagues (33) that the first 51 amino acid residues cleaved by CNBr from natively purified vitronectin 3 (we term nVN 1-51 -3) had a third type of disulfide topology (Cys 5 -Cys 9 , Cys 19 -Cys 31 , Cys 21 -Cys 32 , and Cys 25 -Cys 39 ), which matched neither the pattern reported for rVN 1-51 -1 nor for rVN 1-51 -2. Horn et al. (33) further suggested that rVN 1-51 -1 and rVN 1-51 -2 were stable folding intermediates with non-native disulfide linkages, a view hotly contested in a latest report by Kamikubo et al. (34). While sharply pointing out the lack of any reported functional data for nVN 1-51 -3 and the possibility of disulfide scrambling during the handling of nVN 1-51 -3, Kamikubo et al. (34) showed that a fully reduced, denatured rVN 1-51 -1 could be efficiently folded in vitro into a conformationally homogeneous, thermodynamically stable, and functionally active molecule. Furthermore, none of the folding intermediates accumulated showed any biological activity as judged by PAI-1 binding and antibody recognition (34).
The conflicting reports by several laboratories have raised a series of important questions that remain to be answered. However, first and foremost, what is the native disulfide topology in vitronectin? To resolve this increasingly controversial issue that has so far clouded the understanding of PAI-1-vitronectin recognition, we chemically synthesized, by using native chemical ligation and orthogonal protection of selected Cys residues, three topologically different forms of the SMB domain, sSMB-1, sSMB-2, and sSMB-3, with predefined disulfide connectivities corresponding to those reported for rVN 1-51 -1, rVN 1-51 -2, and nVN 1-51 -3, respectively. We also oxidatively folded a fully reduced SMB domain in aqueous solution using the protocol recently published by Kamikubo et al. (34), yielding sSMB-4. In addition, we prepared, by CNBr cleavage, the SMB-containing fragments nVN 1-51 and uVN 1-51 from natively purified as well as urea-activated vitronectin of human blood origin. Biochemical and functional characterization of sSMB-1-4, nVN 1-51 , and uVN 1-51 reported here, while fully accounting for various published discrepancies, unequivocally establish the native and the only functional disulfide topology in vitronectin as Cys 5 -Cys 21 , Cys 9 -Cys 39 , Cys 19 -Cys 32 , and Cys 25 -Cys 31 , in support of the earlier findings by Read and colleagues (18).

MATERIALS AND METHODS
Peptide Synthesis, Native Chemical Ligation, and Oxidative Folding-Stepwise chemical synthesis of the 44-residue SMB domain on solid phase by t-butoxycarbonyl chemistry turned out to be extremely difficult if not impossible. We took advantage of the native chemical ligation technique pioneered by Kent and colleagues (35)(36)(37), assembled the following four full-length peptides in high purity (hereinafter referred to as peptides 1-4; C denotes Acm-protected Cys; 2 depicts the site for native chemical ligation).
To obtain sSMB-1, sSMB-2, and sSMB-3, Acm-protected peptides 1-3 were first oxidized at 0.25 mg/ml in phosphatebuffered saline buffer, pH 7.4, by 20% (v/v) Me 2 SO. An overnight oxidation of each peptide at room temperature yielded three chromatographically distinct species. After disulfide mapping, three desired folding intermediates with Cys 19 -Cys 21 /Cys 32 -Cys 39 (for sSMB-1), Cys 5 -Cys 21 /Cys 25 -Cys 31 (for sSMB-2), and Cys 5 -Cys 9 /Cys 21 -Cys 32 (for sSMB-3) were selected for Acm deprotection/disulfide formation. Specifically, the two-disulfide-bridged peptides at 0.5 mg/ml in an acidic solution containing 0.1 M citric acid, 0.2 M HCl, and 20% methanol were treated by 1 mM iodine for 15 min, each resulting in three fully oxidized and topologically different SMB domains, from which sSMB-1, sSMB-2, or sSMB-3 was eventually decoded. Spontaneous folding of the fully unprotected peptide 4 in aqueous solution under redox control of reduced and oxidized glutathione was carried out essentially as described for rVN 1-51 -1 (34), giving rise to a predominant folding species termed sSMB-4. All four synthetic SMB domains were purified to homogeneity on RP-HPLC and verified by ESI-MS. Quantification of SMB domains was carried out by UV absorbance measurements at 280 nm using molar extinction coefficients calculated according to a published algorithm (41).
Dissection of Two-disulfide-bridged Oxidation Intermediates-We strategically selected Cys residues to be protected by Acm in the SMB sequence so that discerning the pattern of the first two disulfides after Me 2 SO oxidation could be achieved by onestep proteolysis coupled with liquid chromatography-MS analysis. Bovine chymotrypsin and trypsin and Staphyloccocus aureus Glu-specific V8 protease were purchased from Worthington Biochemical Co. Enzymatic digestion was carried out for 1 h at 37°C in 50 mM Tris, 20 mM CaCl 2 , 0.005% Triton X-100,  39 , and was readily identified by digestion with chymotrypsin that cleaved (Tyr 27 -Tyr 28 or Tyr 28 -Gln 29 ) between Cys 21 and Cys 32 . The desired oxidation intermediates derived from peptides 2 and 3 both contained two linear and uncrossed disulfides, Cys 5 -Cys 21 / Cys 25 -Cys 31 , and Cys 5 -Cys 9 /Cys 21 -Cys 32 , respectively. For the peptide 2 derivative, Glu-specific V8 protease made a clean split (Glu 23 -Leu 24 ) between Cys 21 and Cys 25 , whereas trypsin cleaved (Lys 17 -Lys 18 or Lys 18 -Cys 19 ) between Cys 9 and Cys 21 of the peptide 3 derivative. Using this highly simplified approach, all undesired Me 2 SO-oxidized intermediates with crossed patterns of disulfide bonding were readily excluded.
Dissection of Fully Oxidized sSMB-1, sSMB-2, and sSMB-3-To confirm the linear and uncrossed disulfide pattern in sSMB-1, the peptide was treated with chymotrypsin/Glu-specific V8 protease for 1 h at 37°C, generating, as judged by ESI-MS, the following four major fragments: [SC 5  Isolation of VN 1-51 from Intact Vitronectin-Natively purified and urea-activated vitronectins of human blood source were purchased from Molecular Innovations, Inc. The SMBcontaining VN 1-51 was released from vitronectin (0.5 mg/ml in 2.5% trifluoroacetic acid) by CNBr (20 mg/ml). After an overnight cleavage at room temperature, VN 1-51 was purified by analytical C18 RP-HPLC. Two products resulted, the major component containing a C-terminal homoserine lactone and the minor species ending with homoserine. At basic pH, spontaneous hydrolysis quickly converts homoserine lactone to homoserine.
Proteolytic Fingerprinting of sSMB-2, sSMB-4, nVN  , and uVN 1-51 -50 g of sSMB-2 or sSMB-4 was dissolved at 1 mg/ml in 50 mM Tris/HCl buffer containing 20 mM CaCl 2 and 0.005% Triton X-100, pH 8.3, to which 1 g of each enzyme was added in one of the following three binary combinations: chymotrypsin/trypsin, chymotrypsin/V8 protease, and trypsin/V8 protease. After an 18-h incubation at 37°C, the cleavage reaction was terminated by addition of 50 l of 10% acetic acid, followed by liquid chromatography-MS analyses. Due to limitations in quantities, much less nVN 1-51 or uVN 1-51 was used in otherwise identical experiments.
Surface Plasmon Resonance Spectroscopy-The ability of synthetic SMB domains and VN 1-51 to interact with PAI-1 (purchased from Oxford Biomedical Research) in solution was quantified on a Biacore 3000 surface plasmon resonance instrument according to a competition assay protocol developed by Loskutoff and colleagues (32,34,42). Briefly, 1100 resonance units of urea-activated vitronectin were immobilized (in 10 mM acetate buffer, pH 4.0) to a CM5 sensor chip using the 1-ethyl-3-(dimethylaminopropyl)carbodiimide hydrochloride/N-hydroxysulfosuccinimide coupling chemistry and procedures recommended by the manufacturer. Kinetic analysis of the binding to vitronectin of PAI-1, either alone or in the presence of sSMB, was carried out at 25°C in HBS-EP buffer (10 mM HEPES, 150 mM NaCl, 3 mM EDTA, 0.005% surfactant P20, pH 7.4). For competition, 25 nM PAI-1 was incubated at room temperature for 15 min with varying concentrations of synthetic SMB and injected at a flow rate of 20 l/min for 5 min, followed by 5-min dissociation. The concentration of free PAI-1 in solution (not complexed with sSMB) was deduced, based on the initial rate (slope) of VN association, from a calibration curve established by resonance unit measurements of different concentrations of PAI-1 injected alone. Non-linear regression analysis was performed using GraphPad Prism 4 to give rise to IC 50 values, concentrations of SMB at which 50% of PAI-1 was sequestered in SMB-PAI-1 complexes, thus unavailable for VN binding.

RESULTS
The SMB domain with eight Cys residues possesses 105 unique disulfide connectivities. We selectively protected four Cys residues in SMB with the orthogonal protecting agent Acm and oxidatively folded the resultant peptide containing four free cysteines, yielding three two-disulfide-bridged intermediates separable on reversed phase HPLC. After the disulfide linkages were discerned for all three species by mass mapping of peptide fragments generated by proteolysis, the folding intermediate with desired disulfide bridges was selected for Acm deprotection and simultaneous formation of two remaining disulfide bonds, also in three unique combinations. Mass mapping aided by enzymatic digestion was performed again to definitively establish disulfide connectivities in three fully oxidized products. The strategy for the synthesis of sSMB-1, sSMB-2, and sSMB-3 with predefined disulfide connectivities corresponding to the ones previously reported for rVN 1-51 -1, rVN 1-51 -2, and nVN 1-51 -3, respectively, is illustrated in Fig. 1.
For comparison, a fully reduced and unprotected synthetic SMB was oxidatively folded in aqueous solution according to the published protocol (34), yielding conformationally homogeneous and thermodynamically stable sSMB-4. In addition, nVN 1-51 and uVN  , which contain the SMB domain C-terminally connected to a seven-residue RGD motif (RGDVFTM), were prepared by CNBr cleavage, at the Met 51 -Pro 52 peptide bond, of natively purified and urea-treated vitronectin from human blood, respectively. All synthetic SMB domains as well as nVN 1-51 and uVN 1-51 were purified by RP-HPLC to homogeneity, and their molecular masses ascertained by ESI-MS. Shown in Fig. 2 are sSMB-1, sSMB-2, sSMB-3, and sSMB-4 analyzed by C18 RP-HPLC and ESI-MS. The molecular masses of synthetic sSMB domains were found to be 5003.2 Ϯ 0.3 Da, in agreement with the expected value of 5003.5 Da calculated on the basis of average isotopic compositions of fully oxidized SMB.
High resolution analytical RP-HPLC is a powerful tool for chromatographically differentiating peptides with different disulfide connectivities (43). As shown in Fig. 2, the three topological analogs sSMB-1, sSMB-2 and sSMB-3, at least 1.5 min apart from each other, were fully separated on analytical C18 RP-HPLC. One immediate anomaly was that sSMB-4 had different retention from sSMB-1 (1.6 min apart), suggesting that sSMB-4 and sSMB-1 were topologically different. This finding was surprising because, based on the recent report by Kamikubo et al. (34), we had expected sSMB-4 to be identical to sSMB-1. Proteolytic digestion of sSMB-1 and sSMB-4 by Glu-specific V8 protease confirmed their difference in disulfide bonding pattern (data not shown). Notably, sSMB-4 and sSMB-2 showed identical retention on RP-HPLC (Fig. 2), and, in fact, co-eluted when injected as a mixture. We therefore hypoth-   FEBRUARY 23, 2007 • VOLUME 282 • NUMBER 8 esized that folding of fully reduced SMB under redox control in aqueous solution had produced, in contrast to the published assumption (34), a non-linear disulfide-bonding pattern identical to the one reported by Zhou et al. (18) for the crystal structure of rVN 1-51 -2 complexed with PAI-1.

Native Disulfide Topology in the SMB Domain of Human Vitronectin
Two powerful lines of evidence support this hypothesis. First, sSMB-2 and sSMB-4 were structurally identical. We incubated sSMB-2 and sSMB-4 with three different binary combinations of chymotrypsin, trypsin, and Glu-specific V8 protease and generated identical proteolytic "fingerprints" for the two SMB domains (Fig. 3). Results from mass spectroscopic analyses of all peptide fragments generated were consistent with enzyme specificities and the disulfide connectivity, Cys 5 -Cys 21 , Cys 9 -Cys 39 , Cys 19 -Cys 32 , and Cys 25 -Cys 31 (Table 1). Second, sSMB-2 and sSMB-4 were functionally indistinguishable with respect to PAI-1 binding. We quantified the ability of sSMB-2 and sSMB-4 to interact with PAI-1 by using a previously established competition assay in which urea-activated vitronectin was immobilized on an optical sensor chip for kinectic analysis (34,42). As shown in Fig. 4, both sSMB-2 and sSMB-4 strongly bound to PAI-1 in solution and inhibited the binding of PAI-1 to immobilized uVN in an identical fashion. Non-linear regression analysis yielded identical IC 50 values of 6.6 and 6.5 nM for sSMB-2 and sSMB-4, respectively.
In sharp contrast, the disulfide connectivity deduced by Peterson and co-workers (33) was hugely deleterious functionally. As shown in Fig. 4, sSMB-3 showed a drastically reduced binding affinity for PAI-1 compared with sSMB-2/sSMB-4, with an IC 50 value of 39.5 M, at least 6000-fold higher than those of sSMB-2/sSMS-4. The linear and uncrossed pattern of

. Identical proteolytic fingerprints for sSMB-2 and sSMB-4 generated by chymotrypsin/trypsin (left column), chymotrypsin/V8 protease (center column), and trypsin/V8 protease (right column).
The chromatographic conditions were the same as described in the legend of Fig. 2. All the peaks numbered were identified by simulated proteolytic digestion and mass spectrometric analysis, and the peptide fragments generated were consistent with the disulfide connectivity Cys 5 -Cys 21 , Cys 9 -Cys 39 , Cys 19 -Cys 32 , and Cys 25 -Cys 31 (Table 1).

identified by simulated proteolysis and mass mapping
The experimentally determined masses (in daltons) are within experimental error of the calculated values shown in parentheses. NA, not applicable. disulfides reported by Loskutoff and colleagues (32) fared a little better with an IC50 value of 10.8 M, still 1600-fold higher than those of sSMB-2/sSMB-4. Thus, functionally, it is unlikely that the disulfide-bonding pattern in sSMB-1 and sSMB-3 represents the native topology in vitronectin.

Peak # Chymotrypsin/Trypsin
The central question remains: what is the native disulfide topology in vitronectin? To tackle this controversial issue, we biochemically and functionally characterized nVN 1-51 and uVN 1-51 prepared by CNBr cleavage of nVN and uVN. As shown in Fig. 5, two different forms of VN 1-51 , one ending with homoserine (minor peak) and the other with homoserine lactone (major product), were generated, whose experimentally determined molecular masses of 5780.2 Ϯ 0.4 and 5762.4 Ϯ 0.1 Da were in agreement with the calculated values of 5780.4 and 5762.4 Da, respectively. We focused on the properties of the major cleavage product ending with homoserine lacton because of its quantity (homoserine lacton was rapidly converted to homoserine at basic pH where most characterizations were performed).
Both nVN 1-51 and uVN 1-51 had identical retention on analytical RP-HPLC (Fig. 5), and, when injected as a mixture, coeluted as a single peak, suggesting that nVN 1-51 and uVN 1-51 shared the same disulfide topology. This was further verified by identical proteolytic fingerprints generated from combination Note that the nominal concentration of PAI-1 (25 nM) was calculated based on the quantity provided by the manufacturer, and the actual value was likely lower due to the strong tendency of PAI-1 to self-inactivate. The initial slopes of the association curves are linearly correlated to PAI-1 concentrations, thus used in the competition assay as a measurement of free (unbound) PAI-1 available for VN binding. As the concentration of SMB increases, more PAI-1 is sequestered in SMB-PAI-1 complexes in solution, resulting in less PAI-1 binding to the chip surface. Non-linear regression analysis of the data yielded IC 50 values of 10.8 M, 6.6 nM, 39.5 M, and 6.5 nM for sSMB-1, sSMB-2, sSMB-3, and sSMB-4, respectively (bottom right panel).

FIGURE 5. CNBr cleavage of 100 g of natively purified (thin line) or urea-treated vitronection (thick line) from human blood (left panel) and proteolytic fingerprints for VN 1-51 generated by chymotrypsin/trypsin (center panel) and trypsin/V8 protease (right panel).
The cleavage reaction and enzymatic digestion was monitored on analytical C18 RP-HPLC using the same conditions as described in the legend of Fig. 2. The numbered peaks were identified by simulated proteolytic digestion and mass spectrometric analysis, and the results were consistent with the disulfide connectivity Cys 5 -Cys 21 , Cys 9 -Cys 39 , Cys 19 -Cys 32 , and Cys 25 -Cys 31 ( Table 2). Note that VN 1-51 yielded from urea-treated VN was significantly less than that from natively purified vitronectin, because the quantity of uVN was grossly overestimated by the manufacturer. FEBRUARY 23, 2007 • VOLUME 282 • NUMBER 8 cleavages of the two SMB domain-containing peptides by chymotrypsin/trypsin and by trypsin/Glu-specific V8 protease (data not shown). Importantly, the proteolytic fingerprints of nVN 1-51 /uVN 1-51 matched those of sSMB-2/sSMB-4, and the peptide fragments generated were consistent with the disulfide connectivity Cys 5 -Cys 21 , Cys 9 -Cys 39 , Cys 19 -Cys 32 , and Cys 25 -Cys 31 , as judged by mass spectroscopic analyses (Table  2). Furthermore, results obtained from surface plasmon resonance-based competition assays of nVN 1-51 and uVN 1-51 using PAI-1 and immobilized vitronectin fully corroborated the structural data. As shown in Fig. 6, nVN 1-51 and uVN 1-51 bound PAI-1 equally well, with almost identical IC 50 values of 4.0 and 3.6 nM, respectively, slightly lower than the values previously determined for sSMB-2 and sSMB-4. The small difference in binding affinity for PAI-1 between sSMB-2/sSMB-4 and nVN 1-51 /uVN 1-51 may result from the extra seven-residue RGD motif at the C terminus of VN 1-51 making favorable contacts with PAI-1. Although disordered and invisible in the crystal structure of rVN 1-51 -2 complexed with PAI-1, the RGD motif is indeed positioned close to PAI-1 in the complex (18).

Native Disulfide Topology in the SMB Domain of Human Vitronectin
Taken together, our data unequivocally demonstrate that the native disulfide topology in the SMB domain of vitronectin is the same as that in sSMB2, sSMB4, and rVN 1-51 -2, i.e. Cys 5 -Cys 21 , Cys 9 -Cys 39 , Cys 19 -Cys 32 , and Cys 25 -Cys 31 . Furthermore, this crossed pattern of disulfide bonding, first identified by Read and colleagues structurally (18), represents the only functional topology in the adhesive glycoprotein. Finally, as was demonstrated by Loskutoff and co-workers (34), the correctly folded and fully active SMB domain can be readily obtained via oxidative folding of a fully reduced peptide in aqueous solution.

DISCUSSION
The studies in this report clarified a hotly debated issue with respect to the native disulfide topology in the SMB domain of human vitronectin, paving the way for a better understanding of the molecular recognition between vitronectin and PAI-1 in a variety of important biological processes. Our work, while providing direct biochemical and functional support to the structural findings by Read and colleagues, sharply contrasts the reports of Kamikubo et al. (32) and Horn et al. (33). Naturally, the biggest remaining question is how did the Loskutoff and the Peterson laboratories arrive at two different, non-native and inactive disulfide topologies for SMB when both groups clearly had correctly folded and fully functional molecules, rVN 1-51 -and nVN 1-51 -3, in the first place?
We strongly suspect that the partial reduction/S-alkylation procedures used by both teams to decode disulfide bonding in SMB caused unbeknownst disulfide-thiol exchanges under the experimental conditions used. The problem appears to be 2-fold. First, thiol-disulfide exchanges, known to occur much more rapidly at basic pH, cannot be eliminated entirely at pH 4.6 and pH 6.5, where limited disulfide reduction of rVN 1-51 -1 and nVN 1-51 -3 by tris(2-carboxyethyl)phosphine was conducted by Kamikubo et al. (32) and Horn et al., respectively (33). Second, we found that S-alkylation of free Cys residues by N-ethylmaleimide at pH 4.6 was an incomplete reaction itself and that singly and doubly derivatized SMB peptides co-eluted chromatographically. Consequently, contamination by partially alkylated peptides with free thiols can give rise to ambiguous disulfide assignments using protein sequencing. To push the alkylation reaction to completion, Horn et al. (33) maintained pH at 6.5 for hours after tris(2-carboxyethyl)phosphine reduction, which likely intensified disulfide shuffling. Using sSMB-2 with the native disulfide topology, we replicated the partial reduction/S-alkylation procedures used by Kamikubo et al. (32) and were indeed able to identify the presence of the disulfide bond between the two N-terminal cysteine residues (Cys 5 -Cys 9 ) (data not shown). Notably, Cys 5 -Cys 9 is the only common disulfide bond shared by the two otherwise different  topologies reported by Kamikubo et al. (32) and Horn et al. (33), suggesting an inherent propensity of disulfide shuffling to form the Cys 5 -Cys 9 bond in the molecule.
Interestingly, rVN 1-51 -1, which presumably contained the native disulfide topology, was characterized by NMR spectroscopy, showing a compact solution structure stabilized by packing in the core of four linearly arranged disulfides (42). A charge-hydrophobic patch encompassing residues from the third disulfide loop (DELC 25 SYYQSC 31 ), previously mapped to be the PAI-1-binding site by mutagenesis (24,44), clustered on the surface of the molecule to form the putative binding interface (42). The overall solution structure of rVN 1-51 -1 was highly similar to the crystal structure of rVN 1-51 -2 except for the disulfide bonding pattern. Importantly, it was noted in the same study that several different disulfide arrangements, including the native topology, had comparable stabilization energies in conformational energy calculations and could all satisfy the same NMR constraints, thus raising serious doubts about the structural certainty of the linear and uncrossed pattern of disulfide bonding reported for rVN 1-51 -1 based on obviously ambiguous NMR data.
The solution structure of nVN 1-51 -3 was also studied by NMR spectroscopy (45). As was the case with rVN 1-51 -1, direct observation of the disulfide bridges was not possible based on nuclear Overhauser effect cross-peaks involving Cys residues. Consequently, the two disulfide bonds biochemically identified by Horn et al. (33) for nVN 1-51 -3, i.e. Cys 5 -Cys 9 and Cys 25 -Cys 39 , were used as constraints in simulated annealing calculations, yielded a low-resolution model significantly different from the structures of rVN 1-51 -1 and rVN 1-51 -2 (18,42). The structural studies of nVN 1-51 -3 by NMR spectroscopy, in retrospect, appeared seriously flawed due to the use of the incorrect disulfide constraints.
Two important lessons can be learned from the studies of the SMB domain of human vitronectin. First, limited tris(2-carboxyethyl)phosphine reduction/S-alkylation, while a widely used biochemical procedure for decoding disulfide bonding pattern in peptides and proteins, can cause, under certain conditions, thiol-disulfide shuffling not easily detected by conventional analytical techniques, perhaps a far more serious problem than commonly led to believe by the existing literature. To completely eliminate the possibility of disulfide scrambling, we avoided using any reduction step in dissecting the disulfide connectivities for synthetic SMB domains. Second, since disulfide bonds are often difficult to observe directly from NMR data, to minimize ambiguity and subjectivity, extra care ought to be exercised when "ruling in" or "ruling out" structural constraints for energy calculations. X-ray crystallography clearly enjoys the upper hand in this particular case.