Assignment of the Four Disulfides in the N-terminal Somatomedin B Domain of Native Vitronectin Isolated from Human Plasma*

The primary sequence of the N-terminal somatomedin B (SMB) domain of native vitronectin contains 44 amino acids, including a framework of four disulfide bonds formed by 8 closely spaced cysteines in sequence patterns similar to those found in the cystine knot family of proteins. The SMB domain of vitronectin was isolated by digesting the protein with endoproteinase Glu-C and purifying the N-terminal 1–55 peptide by reverse-phase high performance liquid chromatography. Through a combination of techniques, including stepwise reduction and alkylation at acidic pH, peptide mapping with matrix-assisted laser desorption ionization mass spectrometry and NMR, the disulfide bonds contained in the SMB domain have been determined to be Cys5:Cys9, Cys19:Cys31, Cys21:Cys32, and Cys25:Cys39. This pattern of disulfides differs from two other connectivities that have been reported previously for recombinant forms of the SMB domain expressed in Escherichia coli. This arrangement of disulfide bonds in the SMB domain from native vitronectin forms a rigid core around the Cys19: Cys31 and Cys21:Cys32 disulfides. A small positively charged loop is created at the N terminus by the Cys5: Cys9 cystine. The most prominent feature of this disulfide-bonding pattern is a loop between Cys25 and Cys39 similar to cystine-stabilized α-helical structures commonly observed in cystine knots. This α-helix has been confirmed in the solution structure determined for this domain using NMR (Mayasundari, A., Whittemore, N. A., Serpersu, E. H., and Peterson, C. B. (2004) J. Biol. Chem. 279, 29359–29366). It confers function on the SMB domain, comprising the site for binding to plasminogen activator inhibitor type-1 and the urokinase receptor.

sulfides during refolding yields an ensemble of misfolded enzymes exhibiting less than 1% activity (20,21). In terms of primary sequence, the SMB region is similar to a growing class of proteins known to form cystine knots (22)(23)(24). However, unlike typical examples of the growth factor cystine knot or the inhibitory cystine knot proteins, the SMB domain of vitronectin has four disulfide bridges rather than the more familiar three disulfide arrangement. In this respect, the SMB domain shares sequence features with some members of cystine knot families that block ion channels, have anti-microbial activity and the bone morphogenic protein agonists (24 -27).
Previous efforts to understand the structure of the SMB domain and, specifically, to define the disulfide pairings in this domain have relied upon recombinant surrogates for the native protein (17,19,28). We chose to isolate the N terminus of vitronectin from the native monomeric form of the protein purified from human plasma for the purposes of this study, to avoid any potential pitfalls inherent in using a recombinant protein to determine native structure. Analyses of several representatives of the cystine knot family have shown that disulfide bonds may be eliminated or misfolded compared with those found in the native peptide, while retaining complete or nearly complete biological activity (29,30). Thus, there is no reliable way to assure that a peptide derived from a recombinant source truly represents the disulfide connectivities of the native protein.
Determining the cysteine linkages of proteins that contain closely spaced complex disulfide bonds is a difficult problem. A number of methods have been employed to address the analytical issues posed by these structures, but they all have limitations. Often, a combination of techniques must be recruited to completely characterize the disulfide-bonding pattern in such proteins (31)(32)(33). The primary approach used here was the selective partial reduction of the disulfide bonds in the SMB domain using Tris(2-carboxyethyl)phosphine (TCEP), followed by alkylation of the freed cysteines by N-ethylmaleimide (NEM) under acidic conditions to avoid disulfide scrambling. The SMB region was cleaved from vitronectin by endoproteinase Glu-C, purified by reverse-phase HPLC, and partially reduced with TCEP under acidic conditions. The partially reduced isoforms were subjected to peptide mapping using mass spectrometry, and the polypeptide was characterized by 1 H NMR to identify the individual disulfide pairings. Our results indicate that the domain scaffold with its intricate network of disulfides is important to orient the single ␣-helix within this domain (18) properly for binding of ligands including PAI-1 and uPAR. The SMB domain thus represents a new member of the cystine-stabilized helix (CSH) family (34).

EXPERIMENTAL PROCEDURES
Materials-NEM and TCEP hydrochloride were from Fluka (Buchs, Switzerland). Dithiothreitol (DTT) and trifluoroacetic acid were from Pierce. Iodoacetamide (IAM) and bovine insulin were from Sigma. Guanidine HCl, Tris base, dibasic sodium phosphate, ammonium bicarbonate, ammonium sulfate, acetic acid, phosphoric acid, HPLC grade methanol, and HPLC grade acetonitrile were from Fisher Scientific (Suwanee, GA). Sequencing grades of endoproteinase Glu-C, trypsin, and chymotrypsin were from Roche Applied Science.
Endoproteinase Glu-C Digestion of Vitronectin-Native vitronectin was purified from human blood plasma using a modification of the method developed by Dahlback and Podack (35) and stored as a precipitate under saturated ammonium sulfate. The ammonium sulfate slurry of vitronectin was pelleted in 2-ml Eppendorf® tubes in a Beckman Microfuge at 12,000 ϫ g for 15 min. The pellet was dissolved in 4 M guanidine HCl, 0.02 M iodoacetamide, 0.05 M Na 2 HPO 4 , pH 8.0, at a concentration of 2-4 mg/ml as determined by absorbance at 280 nm. The dissolved vitronectin (5-10 mg) was placed in the dark for 1-2 h at ϳ25°C to block free cysteines. With the free cysteines S-carbamidomethylated, the full-length vitronectin was then transferred to a Slide-A-Lyzer® dialysis cassette from Pierce with a 10,000 molecular weight cut-off, where it was dialyzed against three changes of 1 liter of 4 M guanidine HCl, 0.1 M Na 2 HPO 4 , 0.125 M ammonium bicarbonate, pH 7.4, adjusted with phosphoric acid. Finally, the alkylated vitronectin was dialyzed exhaustively against 0.1 M Na 2 HPO 4 , 0.125 M ammonium bicarbonate, pH 7.4, adjusted with phosphoric acid. The protein was removed from the dialysis cassette and the concentration determined by absorbance at 280 nm. The concentration was adjusted to 1-2 mg/ml using the same buffer. The protease, endoproteinase Glu-C, was added to vitronectin at a 1:100 ratio (w/w), and the digest was performed at 37°C for 24 -30 h.
Purification of Intact and Partially Reduced Forms of the SMB Domain-The peptides generated by the endoproteinase Glu-C digestion of vitronectin were separated by reverse-phase HPLC using a linear gradient at 0.5 ml/min from 5 to 55% acetonitrile over 107 min. The gradient was made by mixing buffer A (0.1% trifluoroacetic acid (w/v) in water) and buffer B (acetonitrile with 0.085% trifluoroacetic acid (w/v)) in the appropriate proportions on an Agilent 1100 series quaternary pump HPLC system. A multiple wavelength detector allowed for monitoring of samples at 220 and 280 nm, and a fluorescent detector was used for monitoring for tryptophan fluorescence (excitation 280 nm, emission 356 nm). A C-18 Nucleosil column custom packed by Alltech (5-m particle size, 100 Å pore size, 4.6 ϫ 150 mm) was used. Column temperature was maintained at 50°C. Fractions were collected using a Gilson FC203 B fraction collector for identification by MALDI-MS. Additional purification prior to reduction was performed using the same column and HPLC system with a linear gradient at 0.5 ml/min from 15 to 45% B over 60 min.
Partial Reduction with TCEP and Alkylation with NEM-Partial reductions were carried out with an excess of TCEP and the degree of reduction controlled by stopping the reaction at various times by the addition of NEM. The times and concentrations were adapted from Young et al. (36). The HPLC fraction containing the peptide to be reduced was dried using a Savant Speed-Vac with a dry ice and methanol trap. The dried fractions were dissolved in freshly prepared 0.1% trifluoroacetic acid (w/v), pH 2.8. TCEP was prepared immediately prior to use at a concentration of 0.35 M in 0.1% trifluoroacetic acid (w/v), pH 2.8. NEM was also freshly prepared at a concentration of 0.8 M in 0.1% trifluoroacetic acid (w/v), pH 2.8, and 50% methanol (v/v). Solutions containing TCEP and NEM were protected from the light at all times. A 0.05 M Tris base buffer, pH 7.8, was freshly prepared to titrate the partially reduced samples before alkylation with NEM. Immediately prior to performing the partial reduction, the amount of Tris buffer necessary to titrate a given amount of 0.1% trifluoroacetic acid (w/v), pH 2.8, to pH 6.5 was determined empirically. TCEP was added to peptide to yield a solution with a final concentration of 0.02 M TCEP. Samples were generally withdrawn at 30, 60, 90, and 120 min and sufficient NEM was added to the reaction to create a 5:1 molar ratio of NEM to TCEP. Following the addition of NEM, Tris buffer was added to raise the pH to 6.5 to facilitate the alkylation reaction. Alkylation was allowed to proceed at ambient temperature (ϳ25°C) for at least 2 h. The partially reduced and alkylated peptides were then separated by HPLC as described above and identified by MALDI-MS. Sampling times were adjusted as necessary to enrich for individual isoforms. Additional purification of partially reduced and alkylated forms was performed using a Nucleosil C-18 column and HPLC system with a linear gradient at 0.5 ml/min from 15 to 45% B over 60 min, as described above.
Digestion of Partially Reduced Forms of the SMB Domain-Partially reduced forms of the SMB region of vitronectin were digested with endoproteinase Glu-C, trypsin, and/or chymotrypsin to generate peptides for the determination of individual disulfide bonds. Prior to digestion, partially reduced, alkylated, and HPLC-purified peptides were completely reduced with DTT, and any free cysteines were S-carbamidomethylated with iodoacetamide. Fractions containing the individual partially reduced isoforms were dried on the Savant Speed-Vac and dissolved in 0.1 M ammonium bicarbonate, pH 8.5. Subsequently, DTT (1 M stock) was added to each peptide at a final concentration of 0.01 M, and the sample was incubated for 4 h to completely reduce any remaining disulfide bonds in the partially reduced and alkylated peptide. Iodoacetamide (1 M stock in 0.1 M ammonium bicarbonate, pH 8.5) was added to the fully reduced peptide to give a final concentration of 0.1 M. S-Carbamidomethylation with iodoacetamide was performed in the dark at ambient temperature (ϳ25°C) for 2 h.
Prior to protease digestion, S-carbamidomethylated samples were desalted using a 300-mg Maxi-Clean® C-18 cartridge from Alltech according to manufacturer's instructions, washed on the cartridge with 3% acetonitrile with 0.1% trifluoroacetic acid (w/v), eluted with 50% acetonitrile with 0.1% trifluoroacetic acid (w/v), dried on the Savant Speed-Vac, and dissolved in the appropriate digestion buffer. All buffers for digestion of partially reduced forms were prepared on the day of use, and digests were performed at 37°C for ϳ15 h. Endoproteinase Glu-C digests were performed in 0.1 M ammonium bicarbonate, pH 7.4. Trypsin and chymotrypsin digests were performed in 0.1 M ammonium bicarbonate, pH 8.0. All enzymes were added to give a ratio of ϳ50:1 substrate to enzyme (w/w). In the case of dual digests, the peptide was dried after the first digest using the Savant Speed-Vac, then dissolved in the appropriate buffer for the second digest, and enzyme was added in a 50:1 substrate to enzyme (w/w) ratio. After the digestion sequence was complete, the resulting peptide mixture was dried directly on the Savant Speed-Vac and dissolved in 0.1% trifluoroacetic acid (w/v). Alternatively, the sample was passed over a 300-mg Maxi-Clean® C-18 cartridge from Alltech, washed on the cartridge with 3% acetonitrile and 0.1% trifluoroacetic acid (w/v), eluted with 50% acetonitrile and 0.1% trifluoroacetic acid (w/v), dried as described above, and dissolved in 0.1% trifluoroacetic acid (w/v) for MALDI-MS.
MALDI-MS-MALDI mass spectra were obtained in positive ion mode using a PerSeptive Biosystems Voyager DE time-of-flight mass spectrometer, or a Voyager Elite DE reflectron time-of-flight mass spectrometer, each equipped with a nitrogen laser (Applied Biosystems, Framingham, MA). Spectra from the Voyager DE instrument were obtained in linear mode with ϩ20 kV total accelerating voltage, ϩ18.8 kV applied to the grid, ϩ10 V applied to the guide wire, and a 110-ns acceleration delay. Spectra from the Voyager Elite DE instrument were obtained in linear mode with ϩ25 kV total accelerating voltage, ϩ23 kV applied to the grid, ϩ37.5 V applied to the guide wire, and a 100-ns acceleration delay. Spectra are averages of up to 256 individual laser pulses, obtained from several locations on each sample spot. Sample aliquots were applied to a pre-spotted thin-layer matrix, prepared by applying 0.5-l portions of 5 g/liter nitrocellulose (Immobilon NC Pure, Millipore, Bedford MA), 20 g/liter ␣-cyano-4-hydroxycinnamic acid (Aldrich; re-crystallized from ethanol before use) in 1:1 isopropyl alcohol: acetone to the sample plate and allowed to dry (37). External calibration of the m/z axis was performed using gramicidin S and bovine insulin. Expected masses to compare with the observed MALDI-MS results were calculated using Protein Prospector (38).
Electrospray MS-Analysis was also performed by electrospray quadrupole ion trap mass spectrometry (LCQ Deca, ThermoFinnigan, San Jose, CA). Protease digestion samples that were purified by HPLC were desalted using a C-18 ZipTip (Millipore, Bedford, MA), and the peptides were eluted in 10 l of 50:50 acetonitrile:water. A 100-l volume of 48:48:2 acetonitrile:water:acetic acid was added to the desalted sample. The mixture was then infused directly into the electrospray source of the mass spectrometer. The ϩ5 charge state ion of the 1-55 digest fragment, at m/z 1257.3, was isolated using a 5 m/z isolation width and subjected to collision-induced dissociation to produce a tandem mass spectrum. Charge state deconvolution produced a plot with abscissa units of mass (Da).
NMR Spectroscopy-A sample of the SMB domain (residues 1-51 of vitronectin) was prepared for NMR measurements by cyanogen bromide digestion and HPLC purification, as described (18). NMR spectra were collected using ϳ90 M SMB in a 500-l sample volume in both D 2 O and H 2 O containing 10% D 2 O at pH 4.4 and 298 K. All NMR experiments were performed on a 600 MHz Varian INOVA instrument equipped with a single gradient axis and a triple resonance probe for the observation of proton, carbon, and nitrogen nuclei. Two-dimensional NMR data were acquired in phase-sensitive mode using the States-Haberkorn method for quadrature detection in the indirect dimension (39). Spectra of SMB in H 2 O were recorded by using WET (40,41) or WATERGATE sequences (42) for water suppression. Two-dimensional homonuclear NOESY (43) spectra were recorded with mixing times of 150, 200, and 250 ms. TOCSY spectra (44) were recorded using the DIPSI (decoupling in presence of scalar interaction) spin-lock sequence with a 8-kHz radio frequency field and mixing times of 15, 30, 60, and 80 ms. Typically, spectra were acquired with 256 t 1 increments, 2048 data points, a relaxation delay of 1 s, and a spectral width of 8500 Hz. For DQF-COSY spectra (45), 512 t 1 increments were acquired. Spectra were recorded with 64 -96 scans per increment for NOESY, 24 -80 scans per increment for TOCSY, and 56 scans per increment for DQF-COSY. In all NOESY and TOCSY spectra, the data were multiplied by a 60-90°phase shifted sin 2 window function in both dimensions before Fourier transformation.
NMR Data Analysis-NMR data were processed with Felix 2000 and SPARKY (T. D. Goddard and D. G. Kneller, The University of California, San Diego) software operating on a Silicon Graphics Indigo 2 work station or on a Silicon Graphics Origin 300 server. Cross-peak intensities observed in NOE experiments were divided into four categories as strong, medium, weak, and very weak. These intensities were converted into distance restraints as follows: strong, 1.8 -2.7 Å; medium, 1.8 -3.4 Å; weak, 1.8 -4.5 Å, and very weak 1.8 -6.0 Å. An additional 1.0 Å was added to upper limits involving methyl protons. Similarly, an additional 0.5 Å for methylene protons and 2.3 Å for degenerate H ␦ and H ⑀ protons of tyrosines and phenylalanines were added to upper limits. Also, 0.2 Å was added to the upper limits of NOEs involving amide protons. Coupling constants were extracted from the DQF-COSY spectra. Backbone ⌽ angles were restrained to Ϫ120 Ϯ 50°for 3 J HNH␣ ϭ 8 -9 Hz, and Ϫ120 Ϯ 40°for 3 J HNH␣ Ͼ 9Hz. A restraint of Ϫ100 Ϯ 80°was also applied to the ⌽ angle for residues that show stronger NH i -H␣ i-1 NOE than the intraresidue NH-H␣ NOE (46). A total of 329 NOE restraints and 18 ⌽ restraints were used in structure determination. All calculations were carried out using the AMBER force field interfaced with DISCOVER (Accelerys, San Diego, CA) on an Origin300 work station. Random structures were generated by subjecting the peptide to an initial 10,000-step minimization at 298 K. The temperature was then raised gradually to 1,000 K during a 1,000-step dynamics simulation. The peptide was subjected to minimization and a 10-ps dynamics run at 1,000 K. The NMR-derived restraints were then imposed on the peptide and the peptide was slowly annealed to 298 K in a 100-ps trajectory. Finally, the structures were subjected to further minimization at 298 K. The force constant for the distance restraints was 100 kcal/mol Å Ϫ2 and the dielectric constant was 4.

Endoproteinase Glu-C Digestion of Vitronectin and Isolation of the N-terminal
Peptide-Determination of the disulfidebonding pattern of the SMB domain of vitronectin began with isolation by protease digestion and reverse-phase HPLC. The first step in isolating the SMB domain was digesting full-length vitronectin with endoproteinase Glu-C. Vitronectin contains a total of 14 cysteine residues, arranged in six disulfide bonds with the remaining two free sulfhydryls occupying buried positions (47). Four of the six disulfides are found within the first 39 amino acids of the protein in the SMB domain, whereas the other two have been assigned as Cys 137 :Cys 161 and Cys 274 : Cys 453 (16). The free sulfhydryls are at positions 196 and 411 (16). Prior to digestion with endoproteinase Glu-C, these two free cysteines were carboxyamidomethylated with iodoacetamide to avoid any free thiols that could promote disulfide rearrangements. The carboxyamidomethylated vitronectin was then digested with endoproteinase Glu-C and the resulting peptides analyzed by reverse-phase HPLC (Fig. 1A). The key to generating the optimal separation during HPLC analysis and purification of the N-terminal domain was the use of a small pore (100 Å) stationary phase. The size exclusion properties of the small pore matrix provided additional separation that was necessary to purify the N terminus apart from other peptides and full-length vitronectin.
Preparative amounts of the digest were injected (1-1.5 mg/ injection) and purified on the same small pore reverse-phase matrix. The resulting fractions were analyzed by MALDI-MS. A peak eluting at ϳ47-48 min was identified by mass as the fully oxidized 1-47-amino acid N-terminal fragment of vitronectin (Fig. 1A). The expected average mass-to-charge ratio (m/z) was 5332.9 ([M ϩ H] ϩ , isotopically averaged) and the observed m/z was 5336.1. The peak eluting at ϳ53-54 min was identified by mass spectrometry as the fully oxidized 1-55 amino acid N-terminal fragment of vitronectin (Fig. 1A). The expected m/z was 6281.9 ([M ϩ H] ϩ , isotopically averaged) and the observed m/z was 6284.2. Identity of these peptides also was confirmed by Edman sequencing (data not shown). The two N-terminal fragments, vitronectin 1-47 and vitronectin 1-55, were dried and re-purified by reverse-phase HPLC prior to partial reduction. The sample comprising residues 1-55, shown in Fig. 1B following purification and re-chromatography by HPLC to evaluate purity, was used for the remainder of the analyses on disulfide bonds described in this work. Fig. 1C shows the MALDI-MS spectrum of this sample. Fig. 1D shows an experiment performed on this sample by electrospray quadrupole ion trap MS using a Finnigan LCQ Deca instrument. From multiple charge states, the average mass of the peptide representing the SMB region (residues 1-55 with 4 disulfides) was measured as 6281.2, in excellent agreement with the calculated value of 6280.9 for the peptide. Modern MS technology allows for isolation of a peak of interest in the electrospray spectrum to confirm the peptide by sequence tag identification. The ϩ5 charge state ion at m/z 1257.3 was isolated using a 5 m/z isolation width, subjected to collision-induced dissociation, and the resulting MS/MS (or tandem MS) spectrum was obtained (Fig. 1D). The sequence tag derived from the spectrum, corresponding to species differing in length by a single amino acid residue, agrees with the known vitronectin sequence, thus confirming that the m/z 1257.3 parent ion is the ϩ5 charge state of the partial digest fragment containing residues 1-55.
Partial Reduction with TCEP and Alkylation of Free Cysteines with N-Ethylmaleimide-The second step in determining the disulfide bonding structure of the N-terminal domain of vitronectin was the partial reduction and alkylation of the liberated free thiols under acidic conditions. TCEP, the reducing agent, and NEM, the primary alkylating agent, were specifically chosen because the reactions may be performed under acidic conditions to avoid disulfide scrambling. The purified N-terminal 1-55 fragment of vitronectin was dried and dissolved in 0.1% trifluoroacetic acid, pH 2.8. The solution was adjusted to a final concentration of 0.02 M TCEP, and samples were withdrawn every 30 min over a period of 120 min to capture all of the possible reaction products. Based upon bovine insulin injected as a standard on the HPLC for comparison to the 1-55 fragment at 220 nm prior to partial reduction, the average concentration of the N-terminal peptide in the reaction mixture was estimated to be ϳ8 M. The reduction rate of the peptide at this concentration and pH in the presence of 0.02 M TCEP was reproducible. Higher pH and/or higher concentrations of TCEP resulted in unacceptably fast reaction times, making it difficult to capture the single and doubly reduced reaction products.
The partially reduced samples were alkylated upon addition of NEM to a final concentration of 0.1 M NEM; sufficient Tris base was added to bring the pH to 6.5 to facilitate the alkylation reaction with NEM, which effectively stops further reduction by TCEP. The 30-and 60-min samples were analyzed by MALDI-MS to determine the partial reduction products (Fig.  2). All five possible reduction products ranging from fully oxidized to fully reduced were observed, confirming that the N terminus of native vitronectin contains four cystines that can be selectively reduced over time. Fig. 3 is a schematic that depicts this strategy of partial reduction and NEM blocking, with the three reduction products of interest for further analysis, the SMB domain corresponding to one, two, or three disulfides open and alkylated (SMB1O, SMB2O, and SMB3O, respectively). These reduced and alkylated peptides generated during the partial reduction were purified by reverse-phase HPLC (Fig. 4A) and the resulting fractions analyzed by MALDI-MS. Small shoulders observed on the HPLC peaks (Fig. 4A) from the partial reduction were confirmed by MALDI-MS to be isoforms reflecting varying amounts of alkylation with NEM. Selected fractions containing the individual reaction products from the partial reduction of the SMB domain corresponding to one, two, or three disulfides open and alkylated (SMB1O, SMB2O, and SMB3O) were re-purified by HPLC (Fig. 4, B-D) and carried forward for further analysis to determine individual disulfide pairs.
Protease Digestion and Analysis of Alkylated Partial Reduction Products-The strategy for determining the identity of the liberated sulfhydryls and remaining disulfides in the partially reduced and alkylated forms of vitronectin 1-55 is presented in the schemes outlined in Figs. 5-7. Two approaches were used. In some cases, the samples were digested directly and analyzed by MALDI-MS. Alternatively, the partially reduced and NEMblocked samples were treated further with DTT to achieve complete reduction, and alkylation with IAM differentially tagged the residues that still existed as cystines after TCEP treatment. Subsequent enzymatic digestion and mass analysis was then pursued, and the presence of NEM or CAM at individual sites was used to determine which cysteine residues were oxidized and reduced in the samples with one cystine, two cystines, or three cystines reduced by TCEP (denoted SMB1O, SMB2O, and SMB3O in Figs. 3-7).
First, the isoform of the SMB domain that was partially reduced leaving three disulfides intact and one disulfide pair reduced and alkylated with NEM (SMB1O, Fig. 5) was digested directly with trypsin, and the peptides were analyzed by MALDI-MS (Table I, Supplemental Materials Fig. 1). A peptide product was identified at m/z of 2063.4. This corresponds to the predicted m/z of 2061.3 for two peptides linked through a Cys 5 : Cys 9 disulfide bond (SMB1O-T1/2 in Fig. 5). Specifically, this product corresponds to vitronectin residues 1-6 (DQESCK) disulfide bonded to vitronectin residues 7-18 (GRCTEGFN-VDKK). This disulfide bond was the first identified in our analysis, and its presence was confirmed repeatedly during the course of this work (see Supplemental Materials Table I).
A sample of the SMB1O form of the domain having a single pair of cysteines alkylated with NEM was fully reduced with DTT, and the newly released cysteines were carboxyamidomethylated with IAM, as outlined in Fig. 5. The resulting peptide was then digested with chymotrypsin and analyzed by MALDI-MS (Supplemental Materials Figs. 2 and 3). Three peptide products were identified (Table I). A peak of m/z 2964.6 was observed that corresponds to the expected m/z of 2965.3 for residues 1-24 with all four cysteines (Cys 5 , Cys 9 , Cys 19 , and Cys 21 ) carboxyamidomethylated (peptide SMB1O-Ch1, Fig. 5).
The presence of the CAM tag indicates that all of these cysteines were disulfide bonded in the original SMB1O peptide generated by TCEP and NEM treatment. This is consistent with the initial observation of the Cys 5 :Cys 9 bond assignment.
Second, a peptide with an observed m/z of 1576.1 was also seen in the spectrum of this digest. This agrees with the expected m/z, 1575.7, of a peptide containing residues 25-35 in which two of the cysteines were carboxyamidomethylated and one was alkylated with maleimide (peptide SMB1O-Ch2, Fig.  5). Also, seen in this reaction is a further cleavage product of SMB1O-Ch2, a peptide spanning residues 28 -35 with both cysteines carboxyamidomethylated (peptide SMB1O-Ch3, Fig.  5). This peptide yields an observed m/z of 1096.4, which corresponds to the expected m/z of 1097.2 for this sequence. From comparison of the results on peptides SMB1O-Ch2 and SMB1O-Ch3, it is deduced that Cys 25 in peptide SMB1O-Ch2 must be the residue alkylated with NEM. Thus, Cys 25 was a half-cystine released during the partial reduction with TCEP that yielded the SMB1O form. All of the other 6 cysteine residues identified from this analysis were CAM labeled, indicating that these 6 residues remain paired in the initial reduction to form SMB1O. By the process of elimination, the only remaining cysteine residue, Cys 39 , should be the other half-cystine partner for Cys 25 . Unfortunately, the MALDI-MS spectrum for this digest did not show the peptide comprising residues 36 -55 in whole or part. Whereas these results, combined with those for the tryptic digest of the SMB1O form of vitronectin 1-55, suggest that Cys 25 :Cys 39 forms a disulfide pair, Cys 39 alkylated with NEM was not recovered to confirm this inference.
Definitive evidence for the Cys 25 :Cys 39 assignment was gained from analysis of the SMB domain that was partially reduced to yield four free cysteines alkylated with NEM (SMB2O, Fig. 6). This partial reduction product was fully reduced and alkylated with IAM, so that the cystines remaining FIG. 3. Strategy for partial reduction of SMB, followed by blocking of freed cysteines with NEM. Shown schematically are the starting material, fully oxidized SMB, and partially reduced species that have either one, two, or three cystine bonds opened. NEM was used to block the freed cysteines that are liberated in each of these species. The partially reduced, blocked intermediates that were analyzed further in this work are labeled as SMB1O, SMB2O, and SMB3O for the partially reduced and NEM-blocked species with either 1, 2, or 3 cystines opened with TCEP, respectively. Gray circles represent cysteines in a disulfide linkage, aqua circles represent TCEP-reduced and NEM-blocked cysteines, and red lines represent disulfide bridges in the SMB domain.
after TCEP treatment can be identified with the CAM tag. The doubly tagged peptide was digested with chymotrypsin, and the products were analyzed by MALDI-MS (supplemental Fig. 4). Three peptides of interest were observed (Table I). The first (SMB2O-Ch1, Fig. 6), with an observed m/z of 1677.4, agrees with the expected m/z of 1676.9 for a peptide containing residues 36 -49 with the sole cysteine (Cys 39 ) alkylated by NEM. The second peptide, SMB2O-Ch2 (Fig. 6) is identified as containing residues 14 -24, with both cysteines tagged with CAM. Finally, the third m/z observed was 1097.0, which corresponds to residues 28 -35 with both cysteines tagged with CAM, with a calculated m/z 1097.2 (SMB2O-Ch3, Fig. 6). Thus, Cys 19 , Cys 21 , Cys 31 , and Cys 32 observed here in the CAM form are, therefore, disulfide bonded to each other in SMB2O. Cys 39 was observed in peptide SMB2O-Ch1 to be alkylated by NEM, indicating that Cys 39 was a half-cystine released during the partial reduction that disrupted two disulfides. Thus, the combined information derived from all three digests of the partial reduction products from both the SMB1O and SMB2O isoforms supports the conclusion that Cys 25 :Cys 39 forms a disulfide pair and are the first pair released during the partial reduction.
Other data supporting the Cys 25 :Cys 39 and Cys 5 :Cys 9 assignments are summarized in supplemental Table I.
Having confirmed that the first two disulfide bonds disrupted by TCEP treatment are Cys 25 :Cys 39 and Cys 5 :Cys 9 , respectively, the final assignments required discerning the arrangement of two other disulfides among residues Cys 19 , Cys 21 , Cys 31 , and Cys 32 . The partially reduced form of the SMB domain that resulted in three reduced disulfides (6 free cysteines) labeled with NEM after TCEP treatment, and a single remaining disulfide bond, was purified (SMB3O, Fig. 7). The residual disulfide bond was reduced with DTT and the newly freed cysteines were carboxyamidomethylated with IAM. This peptide was initially digested with chymotrypsin and analyzed by MALDI-MS. Two prominent peptides were observed from this analysis. One, with a m/z of 2648.0, corresponded to the expected core fragment comprising residues 18 -35 with three cysteines tagged with NEM and only two (from the last disulfide to remain unreduced with TCEP) tagged with CAM, with a calculated m/z of 2648.9. This data in itself was not definitive, because there are 5 cysteines within this core peptide and they cannot be identified without sequencing data. Although it had FIG. 4. Purification of individual partially reduced isoforms of vitronectin 1-55. A shows the reversephase HPLC separation of the isoforms of vitronectin 1-55 created during the partial reduction and alkylation. SMB is the fully oxidized SMB domain of vitronectin. As outlined schematically in Fig. 3, SMB1O is the singly reduced isoform, shown here in both singly and doubly alkylated forms. SMB2O is the doubly reduced isoform that has been fully alkylated. SMB3O is the triply reduced isoform, and SMB4O is the fully reduced isoform shown here with one free cysteine as well as the fully alkylated form. A linear gradient was used at 0.5 ml/min from 5 to 55% buffer B over 107 min, as described under "Experimental Procedures." B-D show the reverse-phase HPLC profiles of the partially purified isoforms representing 1, 2, and 3 open disulfides, respectively. B, the re-purification of SMB1O, which used a gradient from 15 to 45% B over 50 min at a flow rate of 0.5 ml/min. C, the re-purification of SMB2O, which used a gradient from 18 to 45% B over 50 min at a flow rate of 0.5 ml/min. D, the re-purification of SMB3O, which used a gradient of 20 to 50% B over 50 min at a flow rate of 0.5 ml/ml. Shown beside panels B-D are schematic figures depicting SMB1O, SMB2O, and SMB3O, respectively (see Fig. 3). already been determined that Cys 25 was paired with Cys 39 (outside this core peptide), the three possible arrangements between the remaining four cysteine residues are not obvious.
Therefore, the SMB3O sample was digested sequentially with two enzymes, endoproteinase Glu-C followed by chymotrypsin (Fig. 7). This combination of enzymes is expected to cleave at positions 22 or 23, intermediate between the Cys 19 -Gln-Cys 21 region and the Cys 31 -Cys 32 sequence. Mass identification of each of the two separate fragments (containing Cys 19 and Cys 21 or Cys 31 and Cys 32 ) with both NEM and CAM labels would indicate that proximal cysteines do not pair and that disulfides must form in a non-linear fashion. In fact, this result is observed; each of two peptides, SMB3O-GCh1 and SMB3O-GCh2 (Fig. 7, Table I), contained one NEM-modified residue and one CAM-modified cysteine (supplemental Figs. 5 and 6).
The MALDI-MS data are therefore consistent with Cys 19 bonding with either Cys 31 or Cys 32 , and, likewise, Cys 21 bonding with the other between the same two residues (see also supplemental Table I). Thus, the data indicate one of only two disulfide arrangements in the core region, i.e. a combination of Cys 19 :Cys 31 and Cys 21 :Cys 32 or a pairing between Cys 19 :Cys 32 and Cys 21 :Cys 31 . Although final distinction between these two possible disulfide assignments was not possible from this approach alone, the analysis ruled out the unlikely possibility that cysteines Cys 31 and Cys 32 are in a vicinal disulfide. Such vicinal arrangements are extremely rare in protein structures and do not appear in the SMB domain.
NMR Approach to Assign the Disulfides in the Core Region Spanning Residues 19 -35-As the tactic using peptide mapping and mass spectrometry was in progress, we initiated a study to determine the three-dimensional structure of the SMB domain isolated from human plasma vitronectin using 1 H FIG. 6. Strategy for proteolysis and MALDI-MS analysis on digestion products from SMB2O. This scheme outlines the experimental procedures that were performed to identify free cysteines and disulfide bridges in the SMB domain with two cystines opened with TCEP, and the four liberated cysteines blocked with NEM (SMB2O). This SMB2O fragment was fully reduced with DTT and the newly liberated cysteines were blocked with IAM, as described under "Experimental Procedures." This fully reduced, and differentially blocked SMB isoform, was then digested with chymotrypsin, and digest products were analyzed by MALDI-MS. Three fragments were identified, denoted SMB2O-Ch1, SMB2O-Ch2, and SMB2O-Ch3. Table I lists expected masses and observed masses for all fragments depicted in this scheme.

FIG. 5. Strategy for proteolysis and MALDI-MS analysis on digestion products from SMB1O.
This scheme outlines the experimental procedures that were performed to identify free cysteines and disulfide bridges in the SMB domain with one cystine opened with TCEP, and the two liberated cysteines blocked with NEM (SMB1O). This SMB isoform was treated in two ways. In one strategy, the SMB1O was digested with trypsin, and the digest products were identified by mass analysis using MALDI-MS. A unique peak was identified that corresponded to two cleavage fragments connected by a disulfide, denoted as SMB1O-T1/2. In the second approach, the SMB1O fragment was fully reduced with DTT and the newly liberated cysteines were blocked with IAM, as described under "Experimental Procedures." This fully reduced, and differentially blocked SMB isoform, was then digested with chymotrypsin, and digest products were analyzed by MALDI-MS. Three fragments were identified, denoted SMB1O-Ch1, SMB1O-Ch2, and SMB1O-Ch3. Table I lists expected masses and observed masses for all fragments depicted in this scheme.
NMR. The source material for the NMR analysis was the Nterminal 51 amino acids from vitronectin, isolated by cyanogen bromide cleavage, rather than the 55-residue fragment isolated by endoproteinase Glu-C digestion that was used for the peptide mapping/MS analyses. Because the four cystines are confined to residues 5-39, the two samples were suitable for parallel studies on the disulfide bonds. The procedure for determining the overall three-dimensional fold for the SMB domain was standard, involving sequence assignments for 1 H resonances and two-dimensional NOESY data to assign interactions between protons that occur at medium to long range and thus dictate the fold of the domain (18). Simulated annealing calculations were used with iterative energy minimizations to arrive at the final structure (18). Initially the structural calculations were pursued without any disulfide restraints, yielding results that were consistently in agreement with the Cys 5 :Cys 9 and Cys 25 :Cys 39 assignments (18). Thus, an NMR approach was taken to distinguish the two possibilities for the final disulfide assignments in the 19 -35 core region.
For this analysis, the two disulfides generated from the peptide mapping/MS work were imposed to refine the structure of the SMB domain, and two possible alternatives were then pursued separately in the structure determination: Cys 19 :Cys 31 and Cys 21 :Cys 32 pairs or Cys 19 :Cys 32 and Cys 21 :Cys 31 pairs. The disulfide bridge restraints were set for the two alternatives (48), and both were subjected to simulated annealing separately. Fig. 8 shows the results of these analyses, which clearly assign the core disulfides as Cys 19 :Cys 31 and Cys 21 :Cys 32 . Plotted in Fig. 8 are two arrays that compare sets of simulated annealing calculations for the two possible arrangements of disulfide pairs. The results are color coded to represent the calculated root mean square deviation relating pairs of structures in the matrix. For each set of calculations, the color range is yellow (0 -2.5 Å), blue (2.5-5 Å), and cyan (Ͼ5 Å). From this visual display of the matrix of structures generated, it is obvious that calculations with the second alternative involving disulfides Cys 19 :Cys 32 and Cys 21 :Cys 31 did not yield acceptable converged structures. Thus, the disulfide arrangement for the two cystine linkages in the core of the SMB domain is Cys 19 : Cys 31 and Cys 21 :Cys 32 .

DISCUSSION
In the study presented here, we have selectively reduced, in a stepwise fashion, the native disulfide bonds of the SMB domain of vitronectin and then alkylated the freed cysteines. Using peptide mapping, mass spectrometry, and NMR, we have determined that the disulfide bonds in the SMB domain of vitronectin are Cys 5 :Cys 9 , Cys 19 :Cys 31 , Cys 21 :Cys 32 , and Cys 25 : Cys 39 . Apart from the first two cysteine residues that form the Cys 5 :Cys 9 bond, the order of disulfides in this small domain is reminiscent of the arrangement found in a large number of proteins that fall into the "cystine knot" motif (22,24,26,49). That is, the remaining six cysteines are bonded in an "n:n ϩ 3" pattern, i.e. the first in the series forms a disulfide with the fourth (Cys 19 :Cys 31 ), the second with the fifth (Cys 21 :Cys 32 ), and the third with the sixth (Cys 25 :Cys 39 ). Perhaps the best recognized members of this cystine knot family are the growth factor-type cystine knots, which contain six cysteines paired in a similar pattern to form a "ring" structure through which part of the peptide chain is threaded in this small domain (22,50). Despite this interesting comparison, this analogy with cystine knots should not be over-interpreted. For one thing, many of the growth factor-type cystine knots have an extra unpaired cysteine residue. Also, a major difference between this structure and cystine knots is apparent from the solution structure of the SMB domain (18), which contains a single ␣-helix as the only element of secondary structure, whereas cystine knots are often rich in mini-␤-sheets (22). Furthermore, the local context and amino acid spacing between cysteine residues is important in directing the overall fold of a cystine knot. A close inspection of the primary sequence of the cysteine-rich N-terminal portion of vitronectin reveals that it contains four disulfides within contexts highly similar to cystine knot proteins (27) (Fig. 9A). However, the ordering of recognized sequence patterns surrounding individual cysteines along the SMB polypeptide chain differs from many of the known homologues. For example, key common patterns in cystine knot proteins are represented in the form of CXGXC and CXC, where X may be any amino acid other than Cys. In six-membered and eight-membered cystine knots, such as observed with the EGF fold (27) or bone morphogenic protein agonists (24), respectively, these four cysteines occur in a well defined order and form the ring structure of the classical cystine motif. Typically, the order is C2XGXC3 and C5XC6 with disulfide pairings of Cys 2 :Cys 5 and Cys 3 :Cys 6 (24) (Fig. 9A). However, in the SMB domain of vitronectin, the cysteines occur in the order C1XGXC2 and C3XC4 with disulfide pairings of Cys 1 :Cys 2 , Cys 3 :Cys 6 , and Cys 4 :Cys 7 resulting in a pseudo-knot (51) (Fig. 9B). Thus, this domain assumes a different bonding pattern from the typical knotted protein and the ring motif is lacking. FIG. 7. Strategy for proteolysis and MALDI-MS analysis on digestion products from SMB3O. This scheme outlines the experimental procedures that were performed to identify free cysteines and disulfide bridges in the SMB domain with three cystines opened by TCEP, and the six liberated cysteines blocked with NEM (SMB3O). This SMB3O fragment was fully reduced with DTT and the newly liberated cysteines were blocked with IAM, as described under "Experimental Procedures." This fully reduced, and differentially blocked isoform, was then digested with two enzymes, endoproteinase Glu-C and chymotrypsin, and digest products were analyzed by MALDI-MS. Two fragments were identified, denoted SMB3O-GCh1 and SMB3O-GCh2. Table I lists expected masses and observed masses for the fragments depicted in this scheme.   1-6. b Sequences of peptides from the SMB domain are given, with the corresponding numbers in the amino acid sequence. Cysteines labeled with CAM (carbamidomethyl) or NEM (N-ethylmaleimide) are shown by subscripts. Boxes in the bottom row indicate that one cysteine in the peptide is labeled with CAM and the other with NEM, but it is not possible to distinguish these possibilities form MALDI data alone.
c Nomenclature for identified peptides correlates with the schemes shown in Figs. 5-7. d Observed m/z ratios by MALDI are given mean Ϯ range. The number of replicate samples measured is shown in parentheses. e Boxes indicate that one cysteine is modified with CAM and the other with NEM, but it is not possible to identify which from MS data alone.
In this way, we were able to show that the SMB domain of vitronectin does not form a classical disulfide knot. Nevertheless, it does possess features with a high degree of similarity to cystine knot proteins. Most prominent among these is a loop formed by the disulfide pairing between Cys 25 and Cys 39 , which is reminiscent of structures classified as cystine-stabilized ␣-helices (CSH motifs). Such structures are known to function in extracellular ligand interactions (27). Interestingly, this loop contains residues that have been shown to be involved in binding PAI-1 (13) and uPAR (15), and the region from residues 26 to 30 forms the sole ␣-helix in this small domain (17,18). This ligand-binding loop has a calculated pI of 3.7 and is oriented toward the solvent by the tightly structured core formed by the Cys 19 :Cys 31 and Cys 21 :Cys 32 disulfide pairings. On the far N terminus is a small positively charged loop formed by the Cys 5 :Cys 9 disulfide bond, suggesting that the SMB domain of vitronectin has a small positively charged tail opposite a large negatively charged ligand-binding nose.
This sequence of disulfide pairings is in agreement with the solution structure from NMR determined recently by our laboratory (18) (Fig. 9B). These results are, however, at odds with previously published results for recombinant counterparts of the SMB domain of vitronectin (17,28). Kamikubo et al. (28) employed similar partial reduction methods with TCEP to discern the disulfide connectivities of a form of the SMB domain isolated from a fusion protein expressed in E. coli. They identified Cys 5 :Cys 9 , Cys 19 :Cys 21 , Cys 25 :Cys 31 , and Cys 32 :Cys 39 as the cystine pattern in their recombinant protein (rSMB) (28) (Fig. 9C). The underlying assumption made by Kamikubo et al. (28) was that rSMB contained the native disulfide framework of vitronectin because it was functionally active in their in vitro assays and it had comparable PAI-1 binding activity to multimeric vitronectin. Also, mutagenesis of half-cystines within two of their proposed disulfides resulted in a complete loss of biological activity. However, work on other proteins displaying a cystine knot motif has shown that biological activity may often be misleading, in that forms with one or more disulfides missing or disrupted can maintain full biological activity (30,(52)(53)(54).
Whereas not previously observed as a native structure among proteins harboring the familiar cystine knot motif, the structure proposed by Kamikubo et al. in 2002 (28) and further elaborated upon in 2004 (19) is familiar from the literature on cystine knots where it has been characterized as a highly stable folding intermediate (55). In a study describing non-native disulfide bonding patterns in oxidative folding intermediates of the Amaranthus ␣-amylase inhibitor, a member of the inhibitory cystine knot family, a major folding intermediate is identified as a peptide structure with a "linear uncrossed pattern" (55). The NMR structure for the major folding intermediate of the ␣-amylase inhibitor showed that it appears as compact as the native knotted protein, presumably because of conformational restraints imposed by a vicinal disulfide bridge. Other disulfide intermediates are short-lived compared with the major folding intermediate with its linear bonding pattern (56). Another example of a linear arrangement of disulfides observed en route to the correct final fold comes from work by Wilken and Bedows (57) on chorionic gonadotropin ␤-subunit. In this work, a transient disulfide forms in vivo between Cys 3 and Cys 4 (nearest neighbors, although not vicinal, in the sequence). This disulfide is rearranged to form a typical Cys 3 :Cys 6 linkage in the secreted, mature form of the protein. Thus, it appears that the linear disulfide assignments reported by Kamikubo et al. (28) on rSMB represent a stable folding intermediate rather than the native arrangement of cystines.
The first report of a three-dimensional structure for the SMB domain by Zhou et al. (17) also used a recombinant SMB domain that was co-crystallized with the active recombinant mutant PAI-1. The disulfide-bonding pattern observed in their study was Cys 5 :Cys 21 , Cys 9 :Cys 39 , Cys 19 :Cys 32 , and Cys 25 : Cys 31 (Fig. 9D). This matches neither our pattern for native SMB (Fig. 9B), nor the original linear pattern reported for a recombinant SMB by Kamikubo et al. (28) (Fig. 9C). More recently Kamikubo et al. (19) have determined a three-dimensional structure for their recombinant SMB and have suggested that there are numerous disulfide patterns that are consistent with their fold. We emphasize by contrast the unique assignment of disulfides for the SMB domain isolated in our work from human plasma vitronectin, for which we have no evidence for heterogeneity or multiple disulfide arrangements.
Although the solution structure from our NMR work (18) and the structures on the recombinant SMB domains (17,19) differ in overall fold, they are similar in one critical feature, the presence of the single ␣-helix that houses PAI-1-and uPARbinding residues (17,28). Despite different disulfide scaffolds, both structures form what appears to be a conserved helix reminiscent of the CSH motif observed in traditional cystine knot proteins. This curious result may stem from the demonstrated malleability of CSH structures to accommodate cysteine substitutions, yet retain biological activity (29,30). Nevertheless, the different disulfide pairs in the recombinant forms of the SMB domain do not represent the proper arrangement in the physiological scenario. In similar types of knotted structures, folding studies have shown that the energetic differences FIG. 9. Pattern of disulfides in somatomedin B from human plasma vitronectin. A schematic of the sequence of the vitronectin SMB domain with assigned disulfides is shown for comparison with the typical disulfide pattern for the growth factor type cystine knots (A (27)). Local sequence motifs that are common between the growth factor type cystine knots and the SMB domain are color coded, revealing a different ordering between the two sequences. The assignment of disulfides determined in this work for the SMB domain from human plasma vitronectin is depicted in B. This assignment of cystines agrees with the solution structure determined on the native domain from human plasma vitronectin (18). 1, the arrangement of disulfides originally proposed for a recombinant form of the SMB domain by Kamikubo et al. (28) is shown in C. This group has recently published work that proposes that at least 4 other disulfide patterns for rSMB are compatible with their NMR structure (19). 2, the disulfide pattern observed by Zhou and co-workers (17) for a recombinant SMB in a co-crystal with PAI-1 is shown in D.
between native and non-native folds is not large (55,56,58). In a study on disulfides in vascular endothelial growth factor, it was observed that removal of disulfides leads to folding anomalies and a lower T m for unfolding, although the overall thermodynamic stability for the various mutants was comparable (59). Also, other work has suggested that a native disulfide arrangement may be more significant for the biological half-life and proper secretion of the polypeptide, as opposed to activity (60). In this regard, comparisons among several disulfide arrangements in gonadotropin ␣-subunit showed that, despite the fact that all were nominally active in a bioassay, only the native disulfide cross-links lend normal kinetics of secretion from cells (53). Such studies suggest that the native fold is more kinetically favored, regardless of small energetic differences between native and misfolded disulfide knots.
The SMB domain of vitronectin possesses sequence homologies that are tantalizing in their similarity to the cystine knot motif. In comparison to classic cystine knots, it appears that a permutation in the arrangement of sequence microdomains surrounding the cysteines in the SMB domain nevertheless results in a structure that mimics these models in terms of folding and stability. Adopting the fold of a pseudo-knot, the SMB domain of vitronectin shares common attributes with true cystine knot proteins. Whereas studies on recombinant SMB domains have contributed valuable information regarding some features that are required for ligand binding to vitronectin, the SMB domain isolated from native monomeric vitronectin, having disulfide bonds between Cys 5 :Cys 9 , Cys 19 :Cys 31 , Cys 21 :Cys 32 , and Cys 25 :Cys 39 , represents the biologically relevant form of the domain with a CSH motif.