Direct photoaffinity labeling by dolastatin 10 of the amino-terminal peptide of beta-tubulin containing cysteine 12.

Tubulin with bound [5-3H]dolastatin 10 was exposed to ultraviolet light, and 8-10% of the bound drug cross-linked to the protein, most of it specifically. The primary cross-link was to the peptide spanning amino acid residues 2-31 of beta-tubulin, but the specific amino acid could not be identified. Indirect studies indicated that cross-link formation occurred between cysteine 12 and the thiazole moiety of dolastatin 10. An equipotent analog of dolastatin 10, lacking the thiazole ring, did not form an ultraviolet light-induced cross-link to beta-tubulin. Preillumination of tubulin with ultraviolet light, known to induce cross-link formation between cysteine 12 and exchangeable site nucleotide, inhibited the binding of [5-3H]dolastatin 10 and cross-link formation more potently than it inhibited the binding of colchicine or vinblastine to tubulin. Conversely, binding of dolastatin 10 to tubulin inhibited formation of the cross-link between cysteine 12 and the exchangeable site nucleotide. Dithiothreitol inhibited formation of the beta-tubulin/dolastatin 10 cross-link but not the beta-tubulin/exchangeable site nucleotide cross-link. Modeling studies revealed a highly favored binding site for dolastatin 10 at the + end of beta-tubulin in proximity to the exchangeable site GDP. Computational docking of an energy-minimized dolastatin 10 conformation at this site placed the thiazole ring of dolastatin 10 8-9 A from the sulfur atom of cysteine 12. Dolastatin 15 and cryptophycin 1 could also be docked into positions that overlapped more extensively with the docked dolastatin 10 than with each other. This result was consistent with the observed binding properties of these peptides.

The subunit protein of microtubules, the ␣␤-tubulin heterodimer, has a number of ligand-binding sites. These include the exchangeable and nonexchangeable GTP-binding sites and at least three reasonably well characterized sites that bind antimitotic drugs. The electron crystallographic model of the tubulin sheet polymer formed in the presence of zinc and paclitaxel and composed of antiparallel protofilaments has provided relatively detailed information about the locations of the nucleotide-binding sites and the paclitaxel site on the ␣␤-dimer (1).
In contrast, drugs that inhibit tubulin polymerization will not bind to structurally normal tubulin protofilaments, so their binding sites have only been approached indirectly through biochemical and genetic techniques. For example, without exception, drugs that bind in the colchicine site inhibit formation of a sulfhydryl cross-link between Cys-239 and Cys-354 of ␤-tubulin (2). These two cysteine residues are also alkylated by A ring chloroacetyl derivatives of thiocolchicine (3,4). Cys-239 and Cys-354 are near the interface between the subunits (1), and such a location for the colchicine site would explain the ability of different photoactive groups attached to the colchicine B ring to preferentially alkylate either the ␣or the ␤-subunit (5,6).
A large number of structurally diverse compounds, including many unusual peptides and depsipeptides, interfere with the binding of vinca alkaloids to tubulin. Despite near total inhibition of vinca alkaloid binding, many of these drugs display noncompetitive patterns when the data they yield are examined by the classic formulas of enzyme kinetic analysis, whereas other inhibitors display competitive patterns. We have suggested that the noncompetitive inhibitors bind near the vinca site, interfering sterically with vinca alkaloid binding to tubulin, and we have proposed that this region of tubulin be called the vinca domain (7). This region of tubulin appears to be close to the exchangeable GTP site on ␤-tubulin, based on two observations. First, a unique sulfhydryl cross-link can be formed in nucleotide-depleted tubulin, between Cys-12 and, probably, Cys-211 (1,2,8). In addition, direct photoaffinity labeling of tubulin by exchangeable site guanosine nucleotide leads to alkylation primarily of Cys-12 (9) and secondarily of Cys-211 (10). All vinca domain drugs that have been examined inhibit the formation of the Cys-12/Cys-211 cross-link, with vinblastine having the weakest effect (2,11,12). Second, all compounds that inhibit vinca alkaloid binding to tubulin also inhibit nucleotide exchange on ␤-tubulin (7,(13)(14)(15). Vinblastine, however, only weakly inhibits nucleotide exchange (16), and rhizoxin is moderately inhibitory (7).
There have been several published studies in which crosslinks were induced between a vinca domain drug and tubulin. By direct photoaffinity labeling with vinblastine (17) and with two photoactive derivatives of vinblastine (18,19), there was greater labeling of ␣-tubulin than ␤-tubulin, ranging from 57 to 75% of the incorporated radiolabel being in the ␣-subunit. A photoaffinity analog of maytansine was incorporated in about a 4:5 ratio into ␣and ␤-tubulin, respectively (20). More specific labeling of ␤-tubulin has also been observed. A photoreactive vinblastine analog specifically labeled a peptide containing res-idues 175-213 (21), and a photoreactive rhizoxin derivative labeled a peptide containing residues 363-379 (22).
We have been studying the interaction of the marine peptide dolastatin 10 ( Fig. 1) with tubulin. Besides potent cytotoxicity, dolastatin 10 inhibits microtubule assembly, induces formation of ring and spiral polymers of tubulin, noncompetitively inhibits the binding of vinca alkaloids to tubulin, and inhibits nucleotide exchange on ␤-tubulin. With radiolabeled dolastatin 10, we showed that the peptide bound tenaciously to tubulin, with negligible dissociation during gel filtration chromatography. The nominal K d value for binding of drug to tubulin was about 25 nM. The binding was reversible, however, because an active analog could displace radiolabel from tubulin. The ability of dolastatin 10 to induce aberrant assembly reactions was so potent that we were not able to demonstrate binding of the radiolabeled drug to any tubulin species smaller than 200 kDa (7,23).
Besides the natural product, a number of peptide analogs of dolastatin 10 have equivalent activity in both cell-based and tubulin-based assays. One of the more interesting, which we also prepared in a radiolabeled form, is auristatin-PE (24) (Fig.  1), also prepared independently as TZT-1027 (25). We began to study the potential of direct photoaffinity labeling with the [ 3 H]dolastatin 10 1 and the [ 3 H]auristatin-PE to provide information about the binding site for these peptides on tubulin. Our findings are presented here.

EXPERIMENTAL PROCEDURES
Materials-Purified bovine brain tubulin with nonradiolabeled GDP (26)  Synthesis of  H]Dolastatin 10 -Dolastatin 10 was first converted to 5-bromo-dolastatin 10. The dolastatin 10 (30 mg) was dissolved in 0.2 ml of acetic acid with 40 mg of silver trifluoroacetate. A solution of bromine in acetic acid (0.114 mmol in 0.6 ml) was added, and the mixture was stirred for 18 h at ambient temperature. Chloroform (2 ml) was added, and the solution was filtered through celite. The solvents were removed by evaporation under reduced pressure. The residue was chromatographed on a DuPont Zorbax ODS column (0.75 ϫ 26 cm), which was developed with 10 mM triethylamine in 75% methanol at 1 ml/min. The column effluent was monitored at 240 nm. The solvent fraction containing the largest peak, with a retention time of 33 min, was concentrated to dryness, and 2.6 mg of product was obtained. The 1 H NMR spectrum, when compared with that of dolastatin 10, showed collapse of a doublet at ␦ 7.4, corresponding to the hydrogen atom at C-4, to a singlet, indicating bromination at C-5 on the thiazole ring. The 5-bromo-dolastatin 10 was dissolved in 0.3 ml of ethyl acetate containing 25 l of triethylamine and 2 mg of 10% paladium on carbon. The reaction mixture was exposed to carrier-free tritium gas at 630 mm Hg for 2 h at ambient temperature. The catalyst was removed by filtration, and the filtrate was exchanged three times with ethanol. The crude product was chromatographed on a 5 ϫ 20 cm silica gel 60F plate with 3:2 acetone/hexane. The dolastatin 10 band was eluted from the silica with 1:1 CHCl 3 /ethanol. The solvent was removed by evaporation under vacuum, and the residue was redissolved in 50 ml of ethanol. The product was 96% pure and its mobility identical to that of dolastatin 10 by TLC and HPLC. Specific activity was 5.4 Ci/mmol, and the yield was 20.7 mCi. The 3 H NMR spectrum showed a doublet at ␦ 7.29, J ϭ 3 Hz, indicating that the tritium was at C-5 in the thiazole ring. Back exchange for 24 h at ambient temperature in pH 7 phosphate buffer showed 0.2% exchangeable tritium.
Synthesis of [phenyl-4-3 H]Auristatin-PE-An analog of auristatin-PE (2.5 mg), with a chlorine atom at C-4 in the phenyl ring (prepared as described in Ref. 24), was mixed with 1.5 mg of 10% paladium on carbon and 2 l of triethylamine in 0.4 ml of ethyl acetate under nitrogen at ambient temperature and exposed to carrier-free tritium gas for 4 h. The catalyst was removed by filtration, and the solution was back-exchanged four times with ethanol on a vacuum line. The crude material was chromatographed on an analytical 20 ϫ 20-cm silica gel 60F TLC plate with acetone/hexane/methanol 5:4:1. The band corresponding to auristatin-PE was collected and eluted from the silica with 1:1 CHCl 3 /ethanol. The TLC mobility of the radiolabeled material was identical with that of nonradiolabeled auristatin-PE. Material from two identical procedures was combined, and HPLC indicated that the radiopurity of the combined product was 81%. The pooled material was applied to a Waters 8 ϫ 10 C 8 Novapak radial compression column, which was developed with 10 mM triethylamine in 75% methanol. Column flow rate was 1 ml/min, and the effluent was monitored at 230  10. The mixtures were incubated for 5 min at 22°C, and 10 M ͓ 3 H͔dolastatin 10 was added. After an additional 15 min at 22°C, the reaction mixtures were exposed to UV light on ice as described in the text for 5 min. A 0.2-ml aliquot of each reaction mixture was mixed with 0.3 ml of 8 M guanidine HCl. Triplicate 0.15-ml aliquots from the original reaction mixture and from the aliquot mixed with the denaturant were applied to syringe columns prepared in 0.1 M Mes, 0.5 mM MgCl 2 or in 4 M guanidine HCl, as appropriate. Averages obtained from three independent experiments are shown. In these experiments the average total binding of ͓ 3 H͔dolastatin 10 was 0.52 mol/mol tubulin and of covalently bound ͓ 3 H͔dolastatin was 0.040 mol/mol tubulin. In some experiments, as indicated, other components were included in reaction mixtures, and incubations prior to exposure to UV light were as described for individual experiments. Reaction mixtures with volumes up to 0.25 ml, with volumes between 0.25 and 1.0 ml, and with volumes over 1.0 ml were placed in wells in Costar polystyrene tissue culture plates, with well diameters of 1.5, 2.5, and 3.5 cm, respectively. Samples were placed on ice at a distance of 10 cm from the UV lamp and exposed to 254 nm light for times as indicated. Light intensity at 10 cm was 2.5 mW/cm 2 .
Measurement of Ligand Bound to Tubulin (Total Bound and Covalently Bound)-The centrifugal gel filtration method, as described previously (32), was used, except that all centrifugations were for 4 min at 2,000 rpm in a Beckman Allegra 6KR centrifuge equipped with a GH-3.8A horizontal rotor. The Sephadex G-50 (superfine) columns were prepared in tuberculin syringes. When total binding of a ligand was measured, aliquots of the reaction mixture were applied directly to the pre-centrifuged Sephadex, which had been swollen in a solution containing 0.1 M Mes (pH 6.9) and 0.5 mM MgCl 2 . When covalently bound ligand was measured, the reaction mixture was mixed with 1.5 parts of 8 M guanidine HCl, and aliquots of the resulting solution were applied to syringe columns containing Sephadex that had been swollen in 4 M guanidine HCl. Each assay mixture was evaluated in triplicate, and both radiolabel and protein in column filtrates were measured.
Peptide Sequencing-Samples for sequencing were radiolabeled as described above, although in some experiments the tubulin and [ 3 H]dolastatin 10 concentrations were increased to 25 M. Reaction mixtures were initially incubated for 15 min at room temperature (22°C) and subsequently irradiated for 5-10 min on ice. Two volumes of ethanol were added to precipitate the tubulin. This removed the noncovalently bound, soluble dolastatin 10. The protein was harvested by centrifugation, washed twice with 70% ethanol, and either dried under vacuum for later use or immediately dissolved in 75% formic acid. Formic acid digestion (96 h at 37°C in the dark), cyanogen bromide digestion, peptide recovery by lyophilization, peptide separation by SDS-PAGE, transfer of resolved peptides to PVDF membranes, and automated Edman degradation were performed as described previously (4), except that after formic acid digestions the peptides were dissolved in the Tricine/SDS sample buffer solution instead of concentrated Tris buffer. PVDF membranes were sprayed with EN 3 HANCE, which was allowed to dry for 15 min prior to exposure to the Biomax MR film. Autoradiograms were prepared over a 5-7-day exposure at Ϫ70°C.
Molecular Modeling-Modeling consisted of identifying acceptable conformations for the docked ligands, scanning the solvent-accessible surface of the tubulin dimer (1) for potential binding sites, and computational docking of the candidate ligands to these sites. Conformations of dolastatin 10, dolastatin 15, and cryptophycin 1 were obtained by sampling the lowest energy geometries derived from in vacuo minimization and molecular dynamics using the CVFF force field of the Dis-cover2002 tool in the Accelrys molecular package (BIOSYM, MSI). A sample set of the five lowest energy conformations for each molecule was selected for docking studies. Candidate docking sites were identified using a method developed for assigning docking sites of ligands against their Protein Data Bank crystal structures (see Ref. 33 for details). Briefly, the method scans the solvent-accessible surface of the electron crystallographic structure of the tubulin dimer (1) using a set of residue-based molecular probes that had been identified previously as the crystal packing geometries of proteins within the Protein Data Bank (34). Each surface point is scored according to the energy associated with docking each of the molecular probes at that point. Calculated binding strengths are then scored by summing the interaction energies for the most favorable binding interactions within this suite of molecular probes. By using this method, potential binding sites are identified for subsequent docking. The interaction energies of this method are derived from extensive study of residue-residue-based potentials (35) and, as such, represent coarse assessments of candidate binding sites.

FIG. 2. Coomassie Blue-stained membrane (track A) and its autoradiogram (track B). Tubulin with bound [ 3 H]dolastatin 10 was
irradiated, harvested, and digested with formic acid. Following removal of the formic acid by evaporation, the peptide mixture was dissolved and subjected to SDS-PAGE. Peptide from about 50 g of tubulin was applied to the gel. The separated peptides were transferred electrophoretically to a PVDF membrane, which was stained, sprayed with EN 3 HANCE, and autoradiographed for 1 week at Ϫ70°C. See text for further details. 10-containing peptide derived from formic acid digestion Fifteen cycles were performed on an Applied Biosystems model 494A protein sequencer. The first cycle also yielded minor peaks corresponding to the residues in parentheses. Cysteine is destroyed during the Edman procedure, as indicated by the X at cycle 12. The sequence obtained is that of the amino terminus of ␤-tubulin (46  This method proved effective in correctly assigning binding sites in subsequent studies (36,37). Docking studies were completed with a computational method developed for placing candidate ligands into possible binding sites. Previous calibration of this procedure found the correct binding position for over 93% of the known crystal complexes studies at the time of analysis (38,39). Docking is achieved in three successive steps, each with increasing demands for scoring acceptable binding positions. The initial step is sufficiently crude, so that all candidate binding sites on the tubulin dimer can be scanned for docking. Following this step, candidate binding sites with the greatest predicted binding strength are passed along for more exact placement into each potential binding site. Completion of all computational steps for docking yields the minimum energy position for each test molecule into its most favorable binding location. 2

Direct Photoaffinity Labeling of Tubulin by [ 3 H]Dolastatin 10 and Identification of the Predominant Radiolabeled Pep-
tide-We decided to explore the potential of the direct photoaffinity labeling technique for providing information about the dolastatin 10-binding site on tubulin based on the successful studies with colchicine (17), paclitaxel (40), and GTP (9). Only in the latter case was a specific amino acid residue, Cys-12 of ␤-tubulin, identified, and Shivanna et al. (9) proposed that exposure of tubulin to ultraviolet radiation resulted in generation of sulfhydryl free radicals as the reactive species. Table I summarizes our initial experiments, in which we found that about 8% of the bound [ 3 H]dolastatin 10 reacted covalently with tubulin, and this covalent reaction, as well as the total binding reaction, was substantially inhibited by preincubating the tubulin with a 5-fold molar excess of two highly active analogs of dolastatin 10, auristatin-PE (24,25) and (19aR)-isodolastatin 10 (12,41). The cross-linking reaction therefore appeared to be largely specific and to require initial binding of the peptide to the protein.
In the experiments summarized in Table I and elsewhere in this study, the stoichiometry of cross-linking ranged from 0.040 to 0.054 mol of dolastatin 10/mol of tubulin. Whereas this stoichiometry is relatively low compared with the results obtained by direct photoaffinity labeling with exchangeable site 2 The molecular coordinates for all binding conformations described in this paper are available on request to D. G. Covell at covell@ncifcrf.gov.

FIG. 3. Effect of irradiation time on subsequent ligand binding by tubulin.
For each ligand study, a 2.5-ml reaction mixture was prepared containing 10 M tubulin, 0.1 M Mes (pH 6.9), and 0.5 mM MgCl 2 . The reaction mixture was irradiated as described in the text. At each of the indicated time points, a 0.4-ml aliquot was removed from the reaction mixture. When all timed aliquots were removed from the irradiated reaction mixtures, one of the following radiola-   (17) all yielded crosslinking stoichiometries of less than 0.05 mol of ligand/mol of tubulin. Initial attempts to identify the tubulin subunit cross-linked to dolastatin 10 were unsuccessful. We found it difficult to resolve the tubulin subunits after UV irradiation either by SDS-PAGE or by hydrophobic chromatography on decyl-agarose (43). There was, however, no significant intersubunit crosslink formation during the brief irradiation period, as no higher molecular weight species were visualized on the gels. The usual tubulin doublet was replaced by a broad band, with the greatest amount of protein in the usual ␤-tubulin position (data not shown).
Therefore, unresolved tubulin that had been irradiated in the presence of [ 3 H]dolastatin 10 was subjected to formic acid digestion. Fig. 2 presents a typical PVDF membrane blot transferred from a polyacrylamide gel, with track A showing the Coomassie Blue-stained peptides and track B the autoradiograph of the same membrane. The primary cleavage site for formic acid is an aspartyl-proline bond (44). There are only three such sites in tubulin. One is located in the ␣-subunit (positions 306 -307) (45) and two in the ␤-subunit (positions 31-32 and positions 304 -305) (46). In our previous studies with decyl-agarose-purified ␤-tubulin cross-linked to colchicine analogs (3,4), we had noted a number of secondary cleavage sites derived from the initially formed larger peptides. This undoubtedly accounts for the multiple bands in the Coomassie Blue-stained track A, but the striking feature of the corresponding autoradiograph (track B) is that there is only one prominent band. This corresponds to the small A1 peptide (in the terminology of Rao et al. (47)) at the amino terminus of ␤-tubulin, as shown when the peptide was eluted from the membrane and subjected to sequential Edman degradation (Table II).
We next attempted to localize the radiolabel to a specific residue, but the bond formed between the [ 3 H]dolastatin 10 and ␤-tubulin was unstable to the harsh conditions of sequential Edman degradation. A large amount of radiolabel was recovered only in the first cycle (it is unlikely that the initial methionine was the residue linked to the drug, see below), even when 31 digestive cycles were performed, and little residual radiolabel was found when the membrane itself was counted after being removed from the sequencing apparatus.
The modeling studies described below and the finding of Shivanna et al. (9) that ultraviolet light specifically activates sulfhydryl residues indicate that a likely covalent interaction between dolastatin 10 and ␤-tubulin could be between Cys-12 and the thiazole ring of the drug, perhaps even with formation of a disulfide bond and disruption of the thiazole ring. We therefore performed additional protein degradation studies in attempts to better define the amino acid residue cross-linked to the peptide. Cyanogen bromide digestion allowed us to exclude Met-1 as the reactive residue (the next methionine residues in 3 S. B. Horwitz, personal communication. 4 J. Wolff, personal communication.  (46)), because this technique yielded large radiolabeled peptides (data not shown). Multiple attempts at enzymatic digestions with trypsin, and the endopeptidases Glu-C and Lys-C failed to yield a radiolabeled peptide that could be sequenced. This could be due to the low stoichiometry of cross-linking between dolastatin 10 and ␤-tubulin and/or to the small size of the anticipated radiolabeled peptides (17-19 amino acid residues).

Indirect Studies Suggesting That the Cross-link between [ 3 H]dolastatin 10 and ␤-Tubulin
Occurs at Cys-12-Shivanna et al. (9) reported, and we confirmed (10), that direct photoaffinity labeling of exchangeable site guanine nucleotide occurs primarily at Cys-12. Moreover, dolastatin 10 potently inhibits nucleotide exchange on ␤-tubulin (7). When we modeled the energy minimized conformation of dolastatin 10 into the most promising potential binding pocket on the electron crystallographic model of tubulin (1), the sulfur atom of Cys-12 and the thiazole ring of dolastatin 10 were in reasonable proximity (see below). These observations suggested several experiments that might strengthen the case for UV-mediated cross-link formation between Cys-12 and dolastatin 10. A simple prediction was that auristatin-PE could not form a similar cross-link to ␤-tubulin. Despite lacking the thiazole ring, this analog of dolastatin 10 has activity comparable with the natural product in all biochemical, cytological, and in vivo antitumor systems where the two compounds have been examined (24,25,29,48). In an earlier study, using a photolabeling methodology similar to that used here, we found that cross-link formation between [8-14 C]GDP bound in the exchangeable site (prepared by performing two cycles of assembly/disassembly with [8-14 C]GTP (27)) and ␤-tubulin occurred rapidly. With tubulin at 5 mg/ml, the covalent reaction was 75% complete within 5 min (42). Therefore, if Cys-12 were the amino acid that formed the covalent bond with dolastatin 10, brief prior irradiation of tubulin to form the GDP-Cys-12 cross-link (9) should inhibit subsequent cross-link formation with dolastatin 10. This proved to be the case, and irradiation of tubulin also dramatically reduced the ability of dolastatin 10 to even bind to tubulin (Fig. 3). With about 3 min of prior irradiation the amount of dolastatin 10 bound to tubulin was reduced by 50%. The ability of a second irradiation to cause formation of the dolastatin 10-␤-tubulin cross-link showed a similar time course for the inhibitory effect of the first irradiation period when this was done in the absence of the drug (data not presented). This, like the inhibitory effect of dolastatin 10 analogs shown in Table I, confirmed that binding of dolastatin 10 to tubulin was a requirement for cross-link formation. Prior irradiation had almost an identical effect on the binding of exogenous [8][9][10][11][12][13][14] C]GTP to tubulin (Fig. 3), as would be expected if the endogenous GDP in the exchangeable site had been cross-linked to ␤-tubulin. In contrast, prior irradiation of tubulin had a significantly smaller effect on the binding of either [ 3 H]vinblastine or [ 3 H]colchicine to tubulin (Fig. 3).
The converse was also true. As shown in Table IV, dolastatin 10 strongly inhibited covalent bond formation between [8-14 C]GDP and ␤-tubulin when we used the tubulin preparation with the radiolabeled GDP prebound in the exchangeable site (27). We also examined the effects of a number of other vinca domain drugs that strongly inhibit nucleotide exchange on covalent bond formation between tubulin and the prebound exchangeable site [8-14 C]GDP (Table IV). Of the compounds examined, only the dolastatin 10 analogs (19aR)-isodolastatin 10 and auristatin-PE had a similar inhibitory effect. Even other peptide antimitotic agents that competitively inhibit the binding of [ 3 H]dolastatin 10 to tubulin were either noninhibitory (phomopsin A, cryptophycin 1) or less inhibitory than dolastatin 10 and its analogs (hemiasterlin). 5 In fact, there was an apparent stimulation of covalent bond formation by phomopsin A and cryptophycin 1, as well as by maytansine and halichondrin B. 6 This difference between dolastatin 10 and the other three antimitotic peptides was unexpected, because they all inhibit nucleotide exchange on ␤-tubulin and bind tightly to the protein as indicated by the equivalent persistence of apparent drug-tubulin complexes during gel filtration HPLC (7,15,49). Such a result may imply significant differences in contact points between the different peptides and ␤-tubulin (i.e. different proximities than dolastatin 10 to Cys-12 or, possibly, Cys-211).
In addition, if irradiation produces a disulfide bond between 5 The data showing a competitive inhibitory pattern obtained with phomopsin A are by R. Bai, unpublished observations. 6 The most likely explanation for the stimulatory effect of phomopsin A, cryptophycin 1, maytansine, and halichondrin B is that they minimized dissociation of [8-14 C]GDP from the exchangeable site when the stock tubulin solution was diluted to its final concentration of 10 M. Cys-12 and the sulfur atom of the thiazole ring of dolastatin 10, dithiothreitol should release radiolabel derived from dolastatin 10 from the tubulin. In contrast, if, as mass spectrometry data indicate (9,42), the cross-link formed by irradiation between cysteine residues and guanosine nucleotides results from attack by a sulfhydryl-derived free radical on C-5 of the guanine moiety with formation of C-S bonds, dithiothreitol should not release nucleotide-derived radiolabel from tubulin. This was in fact what was observed when dithiothreitol was added to the reaction mixtures after irradiation (Table V, treatment 1). Moreover, adding dithiothreitol to reaction mixtures prior to irradiation produced the same result; a sharp reduction in the amount of [ 3 H]dolastatin cross-linked by UV light to tubulin, but little effect on the amount of exogenously added [8-14 C]GTP cross-linked to tubulin (Table V, treatment 2).
Molecular Modeling Analysis-The electron crystallographic model of tubulin derived from paclitaxel-stabilized, zinc-induced sheets of antiparallel protofilaments (1) was used in this analysis. Currently, there is no analogous structural information about the unpolymerized ␣␤-heterodimer that is probably the actual binding species for dolastatin 10 and other inhibitors of tubulin assembly. Initially we performed a global scanning of the solvent-accessible surface of ␤-tubulin in the isolated ␣␤heterodimer for potential ligand-binding sites. Two areas were identified in this step as candidate binding sites (Fig. 4).
One area was close to the exchangeable nucleotide site at the ϩ end of the ␣␤-heterodimer. Fig. 4A shows this potential binding domain in a surface rendition, with areas of high potential for ligand interaction shown in red and low potential in blue. The sulfur atom of Cys-12 is shown in yellow. In this diagram the GDP bound in the exchangeable site is shown in black and ␣-tubulin in green. Fig. 4B displays the best docked position of dolastatin 10 (shown in light blue) in the proposed binding site. For additional orientation, the position of paclitaxel, derived from the electron crystallographic analysis, is shown in orange. It must be stressed, however, that it is unlikely that dolastatin 10 and paclitaxel ever bind simultaneously to the same ␣␤-heterodimer. Dolastatin 10 has high affinity for unpolymerized tubulin and aberrant ring and spiral polymers, whereas paclitaxel has high affinity for polymers composed of linear protofilaments, such as sheets and microtubules.
The second area of ␤-tubulin that merited consideration was on the "lateral" surface thought to be critical for protofilament interactions (50). This is shown in Fig. 4C, with ␤-tubulin on the right and ␣-tubulin on the left. Dolastatin 10 is shown in light blue in its best docked position in this potential binding site (the ϩ end binding site is barely visible from this view and is indicated by the pink and white surface at the far right of the image). Binding in a site required for interprotofilament contacts could explain the potent inhibition of microtubule assembly by dolastatin 10. However, the calculated binding strengths for all ligands at this binding site are lower than the values obtained at the ϩ end binding site shown in Fig. 4B. Furthermore, the lateral binding site is distant from the exchangeable nucleotide site clearly affected by dolastatin 10 and also distant (see below) from the peptide spanning residues 2-31 to which dolastatin 10 can be cross-linked by UV light.
Therefore, the binding model as shown in Fig. 4B was examined in greater detail, with a higher resolution image presented in Fig. 5. Here the polypeptide backbone of ␤-tubulin is shown in white, nonhydrogen atoms of the exchangeable site GDP in light blue, and nonhydrogen atoms of the Cys-12 side chain and of dolastatin 10 in yellow for sulfur, green for carbon, red for oxygen, and purplish-blue for nitrogen. Arrows specifically indicate the Cys-12 sulfur atom, the C-5 carbon atom of GDP, and the thiazole ring of dolastatin 10. This binding conformation readily explains the suppression of nucleotide exchange by dolastatin 10, for the drug and exchangeable nucleotide are in close proximity. The high reactivity of the Cys-12 residue with the guanine ring is also readily understandable. In our hands the stoichiometry of cross-link formation between Cys-12 and exchangeable site GDP can be as high as 30 -40% (42), and the interatomic distance between the Cys-12 sulfur atom and C-5 of GDP is 3.43 Å. In this binding conformation for dolastatin 10, the distance between the Cys-12 sulfur atom and the thiazole FIG. 6. Positions of three polypeptide regions of ␤-tubulin that have been photocross-linked to vinca domain drugs in relationship to the two potential dolastatin 10 binding sites shown in Fig. 4. The dolastatin 10-binding sites are indicated by atomic models (hydrogen atoms not shown) of dolastatin 10 placed in the lateral binding site (left side of diagram) and the ϩ end binding site (center of diagram), with the sulfur atom shown in yellow, carbon atoms in green, oxygen in red, and nitrogen in purplishblue. Only the polypeptide backbone portion of ␤-tubulin is shown, except for the Cys-12 side chain (arrow), with its carbon atom green and its sulfur atom yellow. The peptide spanning residues 2-31 is shown in magenta, that spanning residues 175-213 in orange, and that spanning residues 367-379 in light blue. The remainder of the ␤-tubulin peptide backbone is shown in white.
sulfur atom of the drug is over twice as great, 8.75 Å. Distances from the cysteine sulfur to the other atoms of the thiazole ring are similar, ranging from 7.70 Å for C-5 to 9.30 Å for C-2. This greater distance from the cysteine sulfur atom to the thiazole ring is consistent with the lower stoichiometry of cross-linking of the amino acid to dolastatin 10 (8 -10%) as compared with its covalent bond formation with the exchangeable site GDP. But, in addition, it should be pointed out that there are multiple, alternative, and energetically less favorable but still possible binding conformations of dolastatin 10 to this binding region of ␤-tubulin. In some of these the distance between the Cys-12 sulfur atom and the thiazole ring is about 4 Å. DISCUSSION We have described here direct photolabeling by [ 3 H]dolastatin 10 of the ␤-tubulin peptide spanning amino acid residues 2-31, and it is highly likely that UV irradiation of the tubulindrug complex creates sulfur free radicals at the cysteine residues of tubulin (9). Consequently, Cys-12, the only cysteine residue in the target peptide, is the probable site of covalent bond formation with the radiolabeled drug. Because the potent dolastatin 10 analog auristatin-PE, which lacks the thiazole ring, has little ability to form an UV-induced covalent bond with tubulin, we propose that this moiety of dolastatin 10 reacts with Cys-12, perhaps through formation of a disulfide bond. Alternatively, the sulfur free radical could disrupt one of the double bonds of the thiazole ring. A number of observations were consistent with this hypothesis, including sensitivity of covalent bond formation to dithiothreitol and mutual inhibition by dolastatin 10 and GDP of formation of their UV-induced cross-links to ␤-tubulin.
By using the electron crystallographic model of tubulin derived from zinc-induced sheets stabilized by paclitaxel (1), we found that the quantitatively most likely binding site for dolastatin 10 was very close to the exchangeable nucleotide site and to Cys-12. This readily explains two important biochemical features of the dolastatin 10 interaction with tubulin: strong inhibition both of cross-link formation (12) induced by N,Nethylene(bis)iodoacetamide between Cys-12 and probably Cys-211 (1,2,8,10) of ␤-tubulin and of nucleotide exchange without displacement of prebound GDP from the exchangeable site (7).
How does the binding site near Cys-12 described here relate to previous observations with indirect photoaffinity probes? The indirect method involves placement of a relatively bulky photoactive substituent on a ligand molecule. For such agents to be useful they must bind with adequate affinity to their target and be specific in that both the binding and the crosslinking reactions are inhibited by the natural ligand. This creates two problems inherent in the technique. First, the bulky ligand introduces an additional distance between target amino acid and the substituted atom of the ligand. Second, it can be argued that ipso facto the substituted atom cannot be in close contact with an important part of the binding site, because good activity is observed when it bears a bulky substituent. In addition, it should be pointed out that, whereas dolastatin 10 is a noncompetitive inhibitor of vinca alkaloid binding to tubulin (7), all published indirect photoaffinity studies were performed with derivatives of competitive inhibitors (7, 51) of vinca alkaloid binding: vinca alkaloid analogs (18,19,21), a maytansine analog (20), and a rhizoxin analog (22).
In only two of these studies was a specific reactive peptide identified, both in ␤-tubulin. Wolff and co-worker (21) found that a photoactive vinblastine analog labeled a peptide containing residues 175-213. Iwasaki and co-workers (22) found that a photoactive rhizoxin analog labeled a peptide containing residues 363-379. To correlate these findings with our direct photoaffinity labeling of the peptide containing residues 2-31, the model proposed here is reiterated as a ribbon diagram in Fig. 6, with these three peptide regions shown in different colors. The figure also shows dolastatin 10 bound in both of the sites explored in Fig. 4, with the Cys-12 side chain also shown (Fig.  6, arrow). Peptide 2-31 is colored magenta, 175-213 is orange, and 367-379 7 is light blue. Dolastatin 10 modeled into the lateral binding site (shown on the left in the Fig. 6 image) is relatively distant from the three target peptides. In contrast, the dolastatin 10 modeled into the ϩ end binding site adjacent to the exchangeable nucleotide site (shown in the center of the Fig. 6 image) is close to portions of the 175-213 peptide as well as the 2-31 peptide but distant from the 367-379 peptide. We have not modeled vinblastine or rhizoxin into the potential ϩ end binding domain shown in Fig. 4, A and B,  atively weak inhibitors of nucleotide exchange on ␤-tubulin (7,16).
We have, however, performed initial modeling studies with dolastatin 15 and cryptophycin 1 at the ϩ end binding site. We examined these two antimitotic depsipeptides because of the seemingly contradictory effects of cryptophycin 1 on the binding of radiolabeled dolastatins 10 and 15 to tubulin. Radiolabeled dolastatin 10 was shown to bind with high affinity to tubulin, with an apparent K d value of about 25 nM (23). Cryptophycin 1 strongly inhibits the binding of dolastatin 10 to tubulin in a competitive manner (49), whereas dolastatin 15 is noninhibitory (23). The inability of dolastatin 15 to inhibit dolastatin 10 binding probably derives from its low affinity for tubulin, because the apparent K d value of the binding interaction is about 30 M (52). Although the binding of cryptophycin 1 to tubulin has not been quantitatively measured, its close analog cryptophycin 52 binds to tubulin with an apparent K d value of 0.1-0.45 M (53). When the effects of dolastatin 10 and cryptophycin 1 on the weak binding of dolastatin 15 to tubulin were examined (52), puzzling results were obtained. Dolastatin 10 strongly inhibited dolastatin 15 binding, completely displacing dolastatin 15 from tubulin, whereas cryptophycin 1 had a much weaker inhibitory effect. There was no displacement of bound dolastatin 15 from tubulin by equimolar cryptophycin 1, and about 67% displacement when the concentration of cryptophycin 1 was 6-fold higher than that of dolastatin 15.
One possible explanation for these seemingly contradictory effects of dolastatin 10, dolastatin 15, and cryptophycin 1 on each other's binding to tubulin is that the drugs actually bind to overlapping regions in the vinca domain. We asked whether comparing the energetically optimal models for the binding of the three drugs to the ϩ end binding domain would be in accord with the inhibitory patterns that had been observed. Fig. 7 demonstrates that the favored binding modes of the three drugs are in fact consistent with the inhibition patterns. Fig.  7A compares the most favorable orientations of dolastatin 10 (blue) and cryptophycin 1 (red). The substantial overlap in the binding orientations is consistent with the observed strong competitive inhibition by cryptophycin 1 of dolastatin 10 binding to tubulin. Fig. 7B compares the most favorable orientations of dolastatin 10 (blue) and dolastatin 15 (white). The intertwining of the superimposed structures is consistent with the observation that the tightly binding dolastatin 10 can totally inhibit the interaction of the weakly binding dolastatin 15 with tubulin. Finally, Fig. 7C compares the most favorable orientation of cryptophycin 1 (red) and dolastatin 15 (white). There is relatively little overlap between the superimposed structures, consistent with the apparent simultaneous binding of both compounds to tubulin predicted by the failure of equimolar cryptophycin 1 to inhibit dolastatin 15 binding despite its Ͼ100-fold higher K d value.
We should also note the recent modeling study of Barbier et al. (54), in which cryptophycin 52 was modeled into the ␣␤-heterodimer and into the ring oligomers that the drugtubulin complex forms. Although the model proposed by these workers placed cryptophycin 52 in the ϩ end binding domain, the depsipeptide was not as close to the exchangeable site as the model shown in Fig. 7, A and C (the relative locations of cryptophycin 1 and the exchangeable site GDP in the model presented here are shown in Fig. 8). The model proposed by Barbier et al. (54) showed the cryptophycin 52 molecule in close contact with the peptide spanning residues 204 -225, with specific interactions between the drug and Tyr-208, Cys-211, Phe-212, and Tyr-222. Fig. 8 attempts to correlate the positions of the two proposed binding sites for cryptophycins by reiterating the energetically favored binding of cryptophycin 1 (red) against the ␤-tubulin polypeptide backbone (white), with the 204 -225 sequence shown in orange, the side chains of Tyr-208, Cys-211, Phe-212, and Tyr-222 shown in green, yellow, dark blue, and magenta, respectively, and the exchangeable site GDP in light blue. The only interaction the two models seem to have in common is of a phenyl ring of the drug with Tyr-222, but even this involves different phenyl rings of the cryptophycins. Although the modeling methods used by Barbier et al. (54) do not differ greatly from those used here, nuances in methodology clearly affect the precise orientation selected by the computational methods used to obtain docking results.
In summary, we have demonstrated significant cross-linking of radiolabeled dolastatin 10 to the ␤-tubulin peptide spanning residues 2-31 when the drug-tubulin complex is exposed to UV light. Indirect evidence strongly indicates that the cross-link is between the sulfur atom of Cys-12 and the thiazole ring of dolastatin 10. Modeling of dolastatin 10 into the electron crystallographic model of tubulin is highly consistent with the  al. (54). The polypeptide backbone of ␤-tubulin is shown in white, except for the peptide spanning residues 204 -225, which is shown in orange. The exchangeable site GDP is shown in light blue, and cryptophycin 1, as modeled in Fig. 7, is shown in red. The peptide side chains of Tyr-208, Cys-211, Phe-212, and Tyr-222 are shown in green, yellow, dark blue, and magenta, respectively. biochemical data. The energetically favored binding site is in intimate contact with the exchangeable nucleotide site and with Cys-12.