Anal. Chem. 1995, 67, 2354-2367
Protein Binding Chiral Discrimination of HPLC Stationary Phases Made with Whole, Fragmented, and Third Domain Turkey Ovomucoid Thomas C. Pinkerton,* W. Jeffrey Howe, Eldon L. Ulrich,t and Joseph P. Comiskey Upjohn Laboratories, Analytical Research & Specification Development and Computer-Aided Drug Discovery, The Upjohn Company, 7000 Portage Road, Kalamazoo, Michigan 49001
Jun Haginaka* and Tokiko Murashima Faculty of Pharmaceutical Sciences, Mukogawa Women's University, 1 1-68 Koshien Kyuban-cho, Nishinomiya 663, Japan
William F. Walkenhorst,i William M. Westler, and John L. Markley National Magnetic Resonance Facility at Madison, Department of Biochemistry, University of Wisconsin, 420 Henry Mall, Madison, Wisconsin 53706-1569
Individual protein domains and two domains in combination were prepared by enzymatic and chemical cleavage of turkey ovomucoid followed by isolation and purification by size-exclusion and ion-exchange chromatography. Silica bonded-phase HPLC columns were made from either whole or isolated domains of turkey ovomucoid. The protein columns were tested for chiral recognition by their abilities to resolve enantiomers among a wide range of racemates. The columns made from whole turkey ovomucoid displayed chiral activity toward many racemates, where as a combination of the first and second domain resolved only a selected number of aromatic weak bases. The first and second domains independently gave no appreciable chiral activity. The turkey ovomucoid third domain exhibited enantioselective protein binding for fused-ringaromatic weak acids. Glycosylation of the third domain did not affect chiral recognition. Titration of the third domain with model compounds in conjunction with NMR measurements enabled the identifiication of the amino acids responsible for binding. Molecular modeling of the ligand-protein complexation provided insights into the ability of a protein surface to discriminate enantiomers on the basis of multiple intermolecular interactions. Over the past decade, many HPLC chiral stationary phases have been developed to separate enantiomers with each phase employing its own unique separation strategy.1,2 Since chiral recognition properties are highly specific, a large number of enantioselective phases have been designed to separate an extensive array of chiral compounds. These phases generate chiral resolution by ligand-exchange complexation, n-donor/ ' Current address: Department of Biochemistry, University of Wisconsin, 420 Henry Mall, Madison, WI 53706-1569. * Visiting Scientist at The Upjohn Co. for 1year through an arrangement with the University of Wisconsin. 8 Current address: Institute for Cancer Research, Fox Chase Cancer Center, 7701 Burholme Ave.. Philadelphia, PA 19111. (1) Armstrong, D. W. LC-GC 1992 , 10 (3), 225. (2) Taylor, D. R.; Maher. K. J. Chromatogr. Sei. 1992,30,67. 2354 Analytical Chemistry, Vol. 67, No. 14, July 15, 1995
acceptor associations (Pirkle), or interaction with chral polymers. Enantioselective macromolecules include synthetic polymers (polyacrylamides, polymethacrylates) , polysaccharides (triacetylcellulose, cyclodextrins), and proteins. Protein bonded-phase HPLC columns exhibit the widest range of enantioselectivity among the chiral stationary phase^.^ It is presumed that this is due to multiple intermolecular interactions afforded by protein surfaces, but exact chiral separation mechanisms for protein bonded-phase columns remains elusive. Commercially available protein-bonded phase chiral columns have been made with bovine serum albumin @SA),human serum albumin (HSA),human al-acid glycoprotein (AGP), and ovomucoid (OM). The advantages of protein bonded-phase columns include the ability to resolve a wide range of polar racemates, the ability to achieve direct chiral recognition without sample derivatization, the use of simple aqueous/organic mobile phases, and high selectivity among some chiral compounds. Disadvantages of protein-bonded columns include low capacity, low efficiency, and limited understanding of enantioselective mechanisms3 The preparation of crude chicken ovomucoid (OMCHI) bonded phases and the chiral recognition properties of OMCHI HPLC columns was first described by Miwa et aL4 A chicken ovomucoid bonded phase is commercially available from Shinwa Chemical Industries (Kyoto, Japan) as the Ultron ESOVM HPLC column. These columns have been utilized for the resolution of acidic, basic, and neutral enantiomers in bulk drugs and formulation^,^-^ and they have been used for the assay of enantiomers in biological fluids.* Kirkland et al.9 reported that the ovomucoid columns provided better long-term stability for repetitive injections com(3) Allenmark, S. In Chiral Separation by HPLC: Applications to Pharmaceutical Compounds; Krustlovic, A. M., Ed.; Ellis Honvood: Chichester. 1989; p 287. (4) Miwa. T.; Ichikawa, M.; Tsuno, M.; Hattori, T.; Miyakawa, T.; Kayano, M.; Miyake, Y. Chem. Pharm. Bull. 1987,35,682. (5) Miwa, T.; Miyakawa. T.; Kayano, M.; Miyake, Y.J. Chromatogr. 1987,408, 316. (6) Iredale, J.; Aubry, A F.; Wainer. I. W. Chromatographia 1991,31,329. (7) Haginaka, J.; Seyama, C.: Yasuda, H.; Takahashi, K.J. Chromatogr. 1992, 598.67. (8) Oda. Y.; Asakawa, N.; Kajima, T.;Yoshida, Y.; Sato, T.J. Chromatogr. 1991, 541,411. 0003-2700/95/0367-2354$9.00/0 0 1995 American Chemical Society
CNBr
n
3rd 1 P
Figure 1. Amino acid sequence of turkey ovomucoid.I2 Arrows at SV8 indicate points of cleavage by staphylococcal proteinase. Arrows at CNBr show points of cleavage by cyanogen bromide. The sawtoothed lines are sites of glycosylation. The lines between amino acids are disulfide bonds.
pared to al-acid glycoprotein columns. The inherent stability of ovomucoid proteins was recognized by early investigators, who could not denature them irreversibly, even when heated to 100 OC.lo More recently, investigators have determined that reversible changes do occur between 40 and 90 OC.ll In aqueous media at ambient temperatures, viscosity measurements indicate that no structural differences exist between ovomucoid at pH 4.6 or pH 7.0; however, in 8 M urea, ovomucoid experiences a two-step reversible unfolding and refolding." Ovomucoid is an egg white protein with a molecular weight of about 28 OOO Da. Structurally, avian ovomucoids are glycoproteins composed of three tandem, homologous domains (Figure 1).l2 Selected domains are potent inhibitors of serine proteinases. The amino acid sequences of ovomucoids from various birds have been determined.12J3 Ovomucoids can be subjected to controlled proteolysis with Staphylococcus aureus strain V8 to isolate the third domain, the second domain, and a combination of the first and second domains.13 Cyanogen bromide can be used to isolate the first domain.12J4 Although some physical properties have been (9)Kirkland, K. M.; Neilson, K. L.; McCombs, D. A J. Chromutogr. 1991,545, 43. (10)Frederick, E.;Deutsch, H. F. J. Biol. Chem. 1949, 181,499. (11) Das, B. K.; Agarwal, S. K.; Khan, M. Y. Biochim. Biophys. Acta 1991,1076, 343. (12)Kato, I.; Schrode, J.; Kohr, W. J.; Laskowski, M., Jr. Biochemisty 1987, 26, 193. (13)Laskowski, M.,Jr.; Kato, I.; Ardelt, W.; Cook, J.; Denton, A; Empie, M. W.; Kohr, W. J.; Park, S. J.; Parks, K.; Schatzley, B. L;Schoenberger, 0. L; Tashiro, M.; Vichot, G.; Whatley, H. E.; Wieczorek, A; Wieczorek, M. Biochemisty 1987,26,202. (14)Das, B. K.;Aganval, S. K.; Khan, M. Y. Biochem. Int. 1990,26 (6),993.
aswrtained for individual and combination OM domains,14chiral recognition properties have not been determined. The first domain of turkey ovomucoid is glycosylated at two sites (Figure l), and its carbohydrate structures have been elucidated by proton NMRI5 The third domain of turkey ovomucoid (OMTKY3) is by nature about 50%glycosylated at Asn 175 (Figure 1). The structure of the unglycosylated third domain has been thoroughly elucidated by NMRI6-l8 and single crystal X-ray.19 The objective in this investigation is to gain information regarding the enantioselective recognition mechanism of ovomucoid proteins. Since the ovomucoid consists of three tandem domains, it is of interest to know whether chiral recognition is expressed by all three domains, by only one domain, or by a combination of domains. To understand a protein binding recognition process, one needs to ask the following questions: (a) Do chiral binding sites exist on each domain of the protein; do the domains act independently or do they act in combination? (b) If multiple binding sites exist, are they enantioselective to the same class of compounds or different classes? (c) Is glycosylation necessary for chiral recognition? (d) Can amino acids participating in any of the binding sites be identified, and can enantiomer orientations be modeled? (e) Do impurities that may accompany a protein contribute to chiral recognition? In order to address the first four questions, an investigation was conducted by isolating, purifying, and characterizing the individual domains of turkey ovomucoid; by preparing HPLC columns with single domains, a combination of two domains, and whole turkey ovomucoid; and by testing the enantioselectivity of the columns against a range of chral compounds. Turkey ovomucoid (OMTKY) was selected for this study because NMR assignments of the third domain (OMTKY3) had been determined.16 This enabled the interactive amino acids on OMTKY3 to be idenmed from NMR measurements on protein-ligand complexes with model compounds. Given binding site locations, chiral recognition mechanisms could be studied by molecular modeling since crystal structure coordinates for third domain ovomucoids are k n 0 ~ n . l ~ EXPERIMENTAL SECTION
Isolation of 0M"KY Domains. OMTKY was isolated from turkey egg whites by column chromatography using a modified form of the method reported by Kat0 et a l l 2 After cleavage by the protease from S. aurew strain V8 (SV8), the OMTKY domains were isolated and purified by high-performance preparative chromatography, using the scheme outlined in Figure 2. The OMTKY domains isolated, after SV8 cleavage, included unglycosylated third domain (OMTKY3), glycosylated third domain (OMTKY3S), glycosylated second domain (OMTKY2), and a Combination of the glycosylated fist and second domains (OMTKY[l+21). The first domain isolated following SV8 cleavage could not be used because it contained internally clipped loops (Figure 1);therefore, OMTKYl was prepared by CNBr cleavage, using a method described elsewhere.I2 (15)Risley, J. M.; VanEtten, R L. Carbohydr. Res. 1986,147,21. (16)Krezel, A M.; Darba, P.; Robertson, A. D.; Frejzo, J.; Macura, S.: Markley, J. L. J. Mol. Biol. 1994 242,203. (17)Robertson, A D.;Westler, W. M.; Markley, J. L. Biochemisty 1988,27, 2519. (18)Ogino, T.; Croll D. H.; Kato, I.;Markley, J. L. Biochemistty 1982,21,3452. (19)Bode, W.; Epp, 0.;Huber, R.; Laskowski, M.. Jr.; Ardelt, W. Eur. J. Bichem. 1985,147,387.
Analytical Chemistty, Vol. 67, No. 14, July 15, 1995
2355
Turkey Ovomucoid (OMTKY)
I 1
Domain [1+2]
1
Native (uncleaved)
SV8 Cleavage Size Exclusion Chromatography
2
3
Domain 2 Domain 3s
Domain 3
J
g410=nqyh. Fh i:.pt
I
Anion Exch. (pH = 7.9)
OMTKY[1+2]
OMTKY 3s
OMTKY 3
21170 Da (+1)
8670 Da (-1)
6021 Da (-1)
OMTKY 2 9321 Da (-2)
Figure 2. Chromatographic scheme for the isolation and purification of turkey ovomucoid domains. Mass in daltons (Da) determined by ESI-MS on pure fractions.
Preparative Chromatography. The preparative size-exclusion chromatography system consisted of a Dupont Zorbax GF250 XL (21 mm i.d. x 250 mm L) column run with a mobile phase of 0.1 M ammonium carbonate at a flow of 2 mL/m with optical detection at 280 nm. The anion-exchange chromatographic isolation of OMTKY3 was achieved by isocratic elution with a 10 mM ammonium formate (PH 7.9) mobile phase run at a flow rate of 6 mL/m through a Bio-Gel TSK DEAE5PW (21 mm i.d. x 150 mm L) column @io-Rad Labs, Hercules, CA) attached to a 280 nm optical detector. The OMTKY2 and OMTKY3S were isolated with the same anion-exchange system using a mobile phase gradient of from 10 to 100 mM ammonium formate @H 7.9). The combination domain OMTKY[1+21 from the SV8 cleavage and the OMTKYl from the CNBr cleavage were isolated with a BioGel TSK SP-5PW (7.5 mm i.d. x 75 mm L) sulfyl propyl cationexchange column @io-Rad Labs) using ammonium formate (PH 4.4) mobile-phase gradients at a flow rate of 0.8 mL/m. The ionexchange packings in these TSK columns are made from macroporous G5000PW hydroxylated polyether supports with sizeexclusion limits over several million daltons. The desired protein fractions from each ion-exchange system were collected and lyophilized for several days to remove water and volatile salts. Analytical Chromatography. Two analytical-scale chromatographic systems were used during the study. The first HPLC system consisted of an HP 1050 pump (Hewlett Packard, Palo Alto, CA), a WISP Model 712 autosampler (Waters, Milford, MA), and a Lamda-Max Model 481 LC spectrophotometer (Waters). The second HPLC system was a fully integrated HP1090 liquid chromatograph, equipped with a diode array detector (Hewlett Packard). Chromatographic data were collected with a PE Nelson 941 interface, transmitted to a VAX computer, and integrated using PE Nelson’s AccessChrom chromatography software. To evaluate the purity of whole ovomucoid protein, reversedphase chromatography was conducted using the following col2356 Analytical Chemistry, Vol. 67,No. 14, July 15, 7995
umns: Bio-Rad RP304 C4 (4.6 mm i.d. x 250 mm L), Bio-Rad TSK-Phenyl (4.6 mm i.d. x 75 mm L), and Alltech Macrosphere 300 C18 (4.6 mm i.d. x 250 mm L). These separations used gradient elution with 0.1% aqueous trifluoroacetic acid (”A) to water:acetonitrile (40:60) containing 0.1%TFA over periods from 30 to 60 m at flow rates from 1.0 to 1.5 mL/m with detection at 220 or 280 nm. To evaluate the purity of the isolated protein domains, two types of HPLC were performed. The first separation used a reversed-phase gradient elution of 0.1%TFA to water:acetonitrile (40:60) containing 0.1%TFA over a period of 60 min on an Alltech Macrosphere 300 C18 (4.6 mm i d . x 250 mm L) column run at 1.5 mL/m with detection at 220 nm. A second separation utilized a macroporous TSK DEAE-5PW (7.5 mm i.d. x 75 mm L) anionexchange column run isocratically with 10 mM ammonium acetate (PH 7.9) at a flow rate of 0.8 mL/m with detection at 280 nm. To investigate the enantioselectivity of whole OMTKY and the isolated domains, protein bonded phases (vide infra) were packed in either 4.6 mm i.d. x 100 mm L or 2 mm i.d. x 100 mm L HPLC columns. Isocratic elution of various test racemates was conducted with eluents consisting of 20 mM phosphate buffer (PH 6.9) combined with 2-propanol in volume/volume ratios ranging from 100/3 to 90/10, respectively. Flow rates ranged from 0.1 to 0.8 mL/m with detection at 220 nm. The test racemates consisted of the following therapeutic agents: clorazepate, lorazepam, oxazepam, temazepam, lormethazepam, hydroxyzine, glutethimide, mephobarbital, hexobarbital, trimipramine, ibuprofen, fenoprofen, ketoprofen, flurbiprofen, pranoprofen, pindolol, propranolol, oxprenolol, tolperisone, chlorpheniramine, promethazine, bupivacain, methylphenidate, verapamil, prenylamine lacate, mepenzolate bromide, trihexyphenidyl, meclizine, homochlorcyclizine, benzoin, and folinic acid; in addition to a collection of investigative benzodiazepines, profens, and antioxidants indicated herein by Upjohn “U-numbers”. Trypsin and Chymotrypsin Inhibitory Activitiesof OMTKY Domains. The discrimination of isolated turkey ovomucoid domains is readily accomplished because the OMTKY2 domain inhibits trypsin while the OMTKY3S and OMTKY3 domains inhibit chymotrypsin. By contrast, the third domain of chicken ovomucoid does not inhibit chymotrypsin, even though the second domain inhibits trypsin. The third domain of chicken ovomucoid does weakly inhibit Streptomyces griseus protease B. The presence or absence of the turkey ovomucoid second or third domain in chromatographically collected fractions was initially indicated by enzymatic inhibitory activities. To test for the presence of OMTKY2 in a chromatographic fraction, 100pL of mobile phase containing protein was incubated at room temperature for 5 s with 100 pL of 0.1 mg/mL trypsin (Sigma Chemical Co.) in 0.001 M HC1 in 0.02 M CaC12, combined in 1.8 mL of 0.1 M Tris-HC1buffer (PH 7.8) in 0.02 M CaC4. After this short incubation, 1.0 mL of 1.0 mg/mL N-benzoyl-Arg-pnitroanilide (BAPNA) (Sigma Chemical Co.) substrate was added, and the absorbance at 405 nm was monitored with a HewlettPackard Model 8452A diode array spectrophotometer. If the change in absorbance with time for the chromatographic fraction was significantly less than that of a substrate control, then the inhibitory activity of OMTKYZ was indicated. To test for the presence of OMTKY3 or OMTKY3S in a chromatographic fraction, 100 pL of mobile phase containing protein was incubated at room temperature for 5 s with 100 pL of
0.5 pg/mL a-chymotrypsin (Sigma) in 0.001 M HCl, combined 1.8 mL of 0.1 M Tris-HC1 buffer @H 7.8) in 0.02 M CaC12. A substrate solution was prepared by dissolving 52.5 mg of Nsuccinyl-Ala-Ala-ProPhefmitroanilide(Sigma Chemical Co.) in 2 mL of 1-methyl-2-pyrrolidione (Aldrich Chemical Co.); this was diluted with 50 mL of 0.1 M Tns buffer. After the short incubation, 1.0 mL of the 1 mg/mL substrate solution was added, and the absorbance was monitored at 405 nm with a Hewlett-Packard Model 8452A diode array spectrophotometer. If the change in absorbance with time for the chromatographic fraction was signifkantly less than that of a substrate control, then the inhibitory activity of OMTKY3 or OMTKY3S was indicated. N-Terminal Sequencing of OMTKY Domains. The Nterminal sequencing of the first 15-20 amino acids was conducted on each isolated OMTKY domain by using an AB1 470A protein sequencer. Samples were prepared by reconstituting 500 pg of protein in 0.5 mL of purified water @&J Labs. Inc.) and then by combining 25 pL of the sample with 25 pL of 50%acetic acid. A 3@pL portion of the final dilution was spotted on a solid support for analysis.
Electrospray Ionization Mass Spectrometry (FBI-MS) of OMTKY Domains. The isolated OMTKY domains were analyzed by loop flow injection electrospray ionization mass spectrometry using an in-house designed interface based on the capillary drift tube concept.2O A mobile phase of acetonitrile/water (1:l) containing 0.1%TFA acid was delivered to the injector at a flow rate of 2 mL/m. Approximately 10 pg of each sample was dissolved directly in 20 pL of mobile phase, and a portion of this solution was injected. The sample stream was conveyed to the ESI interface via a 1OOO:l eluent splitter delivering 2 pL for mass analysis. The ESI aerosol probe needle voltage was 2200 V with collar, drift tube, and source temperatures of 70, 150 and 200 "C, respectively. Spectra were recorded using a Finigan TSQ-70 triple quadruple spectrometer scanning repetitively from 200 to 4000 amu every 3 s. Sourceinduced collisional activation was employed for analyte/solvate declustering via drift tube bias voltages of 2575 v.
Preparation of OMTKY and OMTKY Domain-Bonded Columns. OMTKY and its domains were bonded to an aminopropyl-silica gel (Ultron NH2,5 pm, 120 A pore diameter; Shinwa Chemical Industries, Kyoto, Japan) via the NJV'-disuccinimidyl carbonate (DSC) reaction reported by Miwa et al.* The DSCactivated aminopropyl-silicagel was slumed in 20 mM phosphate buffer (PH 6.8) and combined with protein in the weight ratio of 3:l gekprotein by dropwise addition of protein in buffer over a 2-h period. Further, the mixture was slowly rotated at 30 "C for 15 h, filtered, and washed with water and methanol. The isolated material was dried in vacuo over PZOSat 40 "C for 6 h. Bondedphase packing was made with whole turkey ovomucoid (OMTKY), the third domain unglycosylated (OMTKY3), the glycosylated third domain (OMTKY3S), the second domain (OMTKY2), and the combination of the first and second domains (OMTKY [1+21). The materials were slurry packed into a 100 mm L x 4.6 mm i.d. or 100 mm L x 2.0 mm i.d. HPLC stainless steel columns. N M R Studies. To investigate the stereoselective mechanism of the ovomucoid third domain, three model compounds were (20) Whitehouse, C. M.; Dreyer, R N.; Yamashita, M.; Fenn, J. B. Anal. Chem. 1985, 57, 675.
titrated with protein accompanied by NMR measurements.21 These compounds included clorazepate (7chloro-2,3-dihydro2,2dihydroxy-5phenyl-~-l,4benzodiazepine3~arboxylic acid) (Sigma Chemical, St. Louis, MO), pranoprofen (a-methyl-5H-[11benzopyrano[2,3lpyidine7-aceticacid) (Yoshitomi Pharmaceutical Co., Osaka, Japan), and U-80,413 (a-methyl-4-oxo-2-phenyl-4H-1benzopyran-&aceticacid) and aJ3CH&80,413 synthesized by The Upjohn Co. (vide infra}. Racemic clorazepate and pranoprofen were titrated with recombinant chicken ovomucoid third domain (OMCHI3) with NMR measurements conducted after each addition of protein. The recombinant OMCHI3 was overexpressed as a fusion protein in E. coli and purified, as described elsewhereaZ1 Titrations with OMCHI3 were conducted at pH 7.0 in 0.1 M potassium phosphate buffer at 27 "C. Racemic mixtures of clorazepate, pranoprofen, and U-80,413 were titrated with turkey ovomucoid third domain (OMTKY3). The 13C-labeled pure enantiomers of U-80,413 were titrated with OMTKY3. Titrations with the OMTKY3 were carried out at pH 8.0 in 0.1 M potassium phosphate buffer at 27 "C. During the titrations, the protein and ligand concentrations were varied between 0.3 and 4.5 mM. NMR data were collected on Bruker AM400, AM500, and AM600 spectrometers. Nuclear Overhouser effect spectroscopy WOESY) and isotopeedited @E)NOESV2spectra of the proteinligand complexes were collected in a phasesensitive mode.21The spectral range typically covered 12 ppm ('H) with 4096 t2 points and 512 blocks (tl)of 64 transients. IE-NOESY spectra for each enantiomer of a-13CHJJ-80,413were collected with 160 scans per block. For two-dimensional spectra, protein and ligand concentrations were varied between 3 and 14 mM, with ratios typically near 1:l. Mixing times of 100-150 ms were used for NOESY spectra. Synthesis and Pudication of 13CH3-U-80,413.A racemic mixture of a-13CH&80,413 was synthesized through methylation of 4oxo-2-phenyl-4H-l-benzopyran-8-acetic acid at the a-carbon with 13CH31and sodium hydride in dimethoxyethane. The racemic a-13CH3-U-80,413was purified by silica gel chromatography, recrystallized from ethanol/water, and dried in a vacuum desiccator. Structural identity was confirmed by mass spectrometry and NMR. A portion of the a-13CH3-U-80,413was separated into its pure enantiomers by using a chiral Welk-0 I preparative column (Regis Chemical Co., Morton Grove, IL) with a mobile phase of 10%IPA/90%hexane/O.l% acetic acid. The enantiomeric punty was determined by a Chiralpak AD column with a mobile phase of 20800.1 ethanol/hexane/TFk The isolated a-13CH3U-80,413 enantiomer mobile-phase fractions were rotary evaporated to near dryness at 43 "C, reconstituted in aqueous 2-propanol, desalted on a Hamilton PRP-1 column with acetonitrile elution, diluted with water, and lyophilized. Protein Binding Simulation of Pranoprofen and U-80,413 with OMTKY3. All molecular modeling was conducted on an in-house computational chemistry system developed by The Upjohn Co., called Mosaic. The Mosaic system runs on either Evans & Sutherland or Silicon Graphics hardware connected to VAX or UNM servers. Energy minimization programs in Mosaic used the AMBER* force fields as implemented in BatchMin (21) Walkenhorst. W. F. Structural Investigations of Ovomucoid Third Domains
by Nuclear Magnetic Resonance Methods. Ph.D. Thesis, University of Wisconsin-Madison, Madison, WI, 1993. (22) Griffey, R H.; Jarema. M. A; Kunz. S.; Rosevear. P. R.; Redfield. A G. J. Am. Chem, SOC. 1985, 107,711.
Analytical Chemistry, Vol. 67,No. 14, July 15, 7995
2357
version 3.5,23 developed at Columbia University, New York, NY. AMBER* is a modified form of authentic AMBER, developed by Kollman et alSZ4 Enantioselective protein binding was computer simulated by first constructing models of each enantiomer for pranoprofen and U-80,413 and energy minimizing in the absence of protein. The coordinates of silver pheasant ovomucoid third domain were taken from the Brookhaven Protein Data BankZ5entry 20V019 and converted to turkey ovomucoid third domain (OMTKY3) by changing Met 18 to Leu. The protein structure was further moditied by adding protons to heteroatoms. Mosaic's Autodock program, which performs rigid ligand dockings against a protein surface, was used to generate in excess of 6000 orientations associating each pranoprofen and U-80,413 enantiomer with the surface of OMTKY3. In the next step, the 200 highest scoring orientations of each ligand docking were further refined to optimize ligand geometric and energetic interaction with the protein. The refinement was camed out by energy minimizing the 200 orientations from each set, using Batchmin with the AMBER* united atom force field and PRCG minimizer, allowing ligand flexibility while keeping the protein rigid. Although the protein atoms were held tixed during the minimizations, nonbonded interactions between protein and ligand atoms were included in the calculations. Degenerate states were removed, and the lowest energy 100 orientations of each were saved as working sets. The 10 orientations with lowest energy located the model compounds in same region of the protein composed of amino acids, which created a surface indentation with Leu 23 at the bottom, and amino acids adjacent to and surrounding Leu 23 were clockwise as follows: Arg 21, Pro 22, Phe 53, Val 6, and Lys 34. Independently, interpretation of NMR data on the ligandprotein complexes suggested interactions with the same amino acids without knowledge of the molecular modeling results. In addition, NMR experiments implicated the potential involvement of other amino acids, including Val 41, Val 42, Leu 48, and Leu 50. Orientations of the lowest energy dockings were believed to be biased by the placement of the carboxylate anion of the model compounds between Lys 34 and Lys 13, because of overemphasis on ionic interactions. In order to select a more optimum orientation for each enantiomer among the 100 dockings of each set, carbon-carbon distances were calculated between atoms of the ligands and amino acids implicated by NMR data. Seventyeight distances were calculated between the following amino acid carbon atoms and carbon atoms of U-80,413 (see Figure 17) for all 100 orientations of each enantiomer: Val 6 a to c and d; Val 6 y l and y2 to f, g, and a-methyl; Pro 22 p to c-e; Pro 22 y to d; Leu 23 61 and 62 to e, f, and methyl; Lys 34 y to d and methyl; Lys 34 6 to e and Val 41 y l to a-g; Val 42 y l and y2 to a, e, and f; Leu 48 y and 62 to Leu 50 62 to e and f; and Phe 53 61,62, €1,and €2 to a, e, a-carbon, and a-methyl. When it was realized that some amino acids constituted a second region, carboncarbon distances between the protein and U-80,413 were also calculated for the following reduced set: Val 41 y l to a-c, e, and (23) Mohamadi, F.; Richards, N. G. J.; Guida, W. C.; Liskamp, R.; Lipton, M.; Caufield, C.; Chang, G.; Hendrickson, T.; Still, W. C. J. Comput. Chem. 1990, 11, 440.
(24) Weiner, S. J.; Kollman, P. A.; Case, D. A,; Chandra-Singh, U.; Chio. C.; Alagona, G., Profeta. S.; Weiner, P. J. Am. Chem. SOC. 1984,106,765. (25) Bemstein. F. C.; Koetzle, T. F.; Williams, G. J. B.; Meyer, E. F.; Brice, M. D.; Rodgers, J. R.; Dennard, 0.;Shimanouchi, T.; Tasumi, M. J Mol. Biol. 1977,112. 535.
2358 Ana/ytica/ Chemistry, Vol. 67, No. 14, July 15, 1995
E Val 42 y l and y2 to e and E Leu 48 p and 62 to a-c,
e, and f; Leu 50 62 to e and E Glu 10 to methyb and Pro 12 p to methyl. For pranoprofen (see Figure 16), carbon-carbon distances were calculated only for amino acids of the first region: Leu 23 61 to a, b, and g; Val 6 y l to a and c; Lys 34 y to a-c and g; Arg 21 y to a, c, and g; Pro 22 y to a and g; Lys 34 6 to a; and Phe 53 a to g. The carbon-carbon distances were summed for each 100 orientations of a set to ascertain which dockings were in the closest general proximity of the amino acids implicated. Si dockings from each set with the lowest sum of distances were visualized in three dimensions, and the best orientation for each enantiomer was selected on the basis of logical intermolecular interactions and constraints imposed by NMR data. RESULTS AND DISCUSSION
Enzymatic Cleavage of OMTKY. Whole turkey ovomucoid (OMTKY) is composed of three independent domains connected by two linked peptides (Figure 1). The sawtoothed solid lines indicate glycosylation sites on each domain. The first and second domains are 100%glycosylated, whereas the third domain is about 50% g1yco~ylated.l~Methods for isolation of each domain by enzymatic and chemical cleavage have been described by Laskowski and c o - w ~ r k e r s . ~The ~ J ~third domain can be cleaved enzymatically from the second domain by several methods. The S. aureus strain V8 protease (SV8), which splits the peptide bonds at the C-terminal side of exposed glutamic acid, was chosen for this study because of the high selectivity for cleavage at Glu 130 between the second and third domains (Figure 1). Secondary preference by SV8 is given to cleavage at Glu 64 between the first and second domains (Figure 1), thus overall cleavage with this protease yields a mixture of OMTKY3, OMTKY3S, OMTKY2, internally clipped OMTKY1, and OMTKY[1+21 domains. Since the first domain contains glutamic acid residues, internal cleavage occurs (Figure l), thus OMTKYl from the SV8 cleavage was not used to make chromatography columns. Instead, OMTKYl was isolated after selective cleavage at Met 68 with cyanogen bromide (Figure 1). The OMTKY[2+31 combination obtained from the CNBr cleavage was not used to make columns because of the expected clipping at Met 84 in the second domain. Chromatographic Isolation and Purification of Protein Domains. The mixture of OMTKY domains produced from SV8 cleavage was separated and purified according to the chromatographic scheme illustrated in Figure 2. A high-performance sizeexclusion chromatography profile generates three gross fractions (Figure 3). On the basis of size, one expects to find the native OMTKY and the OMTKY[1+2] in the first fraction. In like manner, the OMTKY2 and OMTKY3S is expected in the second fraction, because each is glycosylated and larger than the nonglycosylated OMTKY3, which is found in the third fraction. The gross fractions were collected by preparative size-exclusion chromatography and subjected to ion-exchange chromatography. The cation-exchange chromatography of the first size-exclusion fraction yielded two broad peaks: the OMTKY[1+2] combination domain and the native OMTKY. This cation-exchange elution at pH 4.4 is consistent with protonation of about half of the Glu and Asp amino acids, which leaves a slight positive charge on the protein. The anion-exchange HPLC separation of the second sizeexclusion fraction at pH 7.9 produced many components (Figure 4). By collecting peaks and testing each for enzyme inhibitory activity, several were identified as OMTKY2 and some as the glycosylated third domain OMTKY3S (Figure 4). Anion-exchange
F3
1 2 mAu
I
20 mAu
I
t 0
, I
I
4
I
I 12
8
I6
lime (min 1
Figure 3. Size-exclusion chromatography of SV8 cleaved turkey ovomucoid on a TSK 2000 SWXL (7.5 mm i d . x 300 mm L) column with a 100 mM ammonium carbonate buffer mobile phase at a flow rate of 1.O m u m and detection at 280 nm. The lines dividing F1, F2, and F3 indicate collected fractions.
1 4 mAu
I 0
IO
20
30
40
50
60
70
80
90
lime (min )
Figure 4. Anion-exchange HPLC separation of the second sizeexclusion fraction (F2) on a TSK DEAE-5PW (7.5 mm id. x 75 mm L) column with a gradient of 10 mM to 100 mM ammonium formate (pH 7.9) over a period of 120 m at a flow rate of 0.8 mUm and detection at 280 nm. Fraction A consisted of glycosylatedOMTKY3S, and fraction B consisted of glycosylated OMTKY2.
HPLC of the third size-exclusion fraction indicated it contained the nonglycosylated third domain (OMTKY3) (Figure 5). The elution order on the anion-exchange column was OMTKY3, OMTKY3S, and OMTKY2, respectively, with increasing ammonium formate concentration at constant pH. This elution order is consistent with the charge on the domains at pH 7.9 and the hydrogen bonding interactions expected for the glycosylation of OMTKY3S and OMTKY2. The OMTKY3 would have a charge of - 1 at this pH with three Lys, one Arg, three Glu, and two Asp. The OMTKY2 should have a charge of -2 with five Lys, two Arg, five Glu, and four Asp. Microheterogeneity in glycosylation explains the multiple peaks occurring for the OMTKY2 (Figure 4). The OMTKY2, OMTKY3S, and OMTKY3 peaks were collected from these separations, as shown in Figures 4 and 5, for preparation of protein columns. Characterization of Whole OMTKY and Isolated Protein Domains. The isolated protein fractions were first identified as either containing second or third domains according to their
0
5
10
15
20
lime (min )
Figure 5. Anion-exchange HPLC separation of the third sizeexclusion fraction (F3) on a TSK DEAE-5PW (7.5 mm i.d. x 75 mm L) column with isocratic elution with a 5 mM ammonium formate (pH 7.7) mobile phase at a flow rate of 0.8 mUm and detection at 280 nm. The bars denote the collected fraction of OMTKY3.
proteinase inhibitory activities toward trypsin or chymotrypsin, respectively.26 Next, N-terminal sequencing was conducted on the fkst 15-20 amino acids of each domain. The results were consistent with the known sequence of each domain generated by SV8 cleavage with OMTKY2 starting at Ala 65 and OMTKY3 starting at Leu 131. In addition to the primary sequence, the OMTKY3 was found to contain a very small amount of a secondary sequence of eight amino acids from a clipped section of the first domain (Figure 1) corresponding to Cryr 50-Glu 57). In addition to the primary sequence, three secondary sequences were found in the OMTKY[1+2] combination domain. These sequences started with either Gly 16, Asp 25, or Tyr 50, indicating that additional cleavage at the glutamic acid positions of the first domain by the SV8 (Figure 1) had created some clipped protein. The OMTKY2 and OMTKY3S fractions exhibited only primary sequences. The purities of OMTKY (whole), OMTKYl,OMTKY2, OMTKY3, OMTKY3S, and OMTKY[1+2] determined by reversedphase HPLC were 90%,86%,96%,98%,97%,and 72%,respectively. The purities of OMTKY, OMTKY3, and OMTKY3S were confirmed with anion-exchange chromatography. The macroporous reversed-phase HPLC of whole OMTKY is shown in Figure 6. The late eluting component comprised the 10% impurity in OMTKY. The whole OMTKY was used without further purification. After SV8 cleavage of OMTKY, this impurity peak appeared in the first size-exclusion fraction. The reversed-phase HPLC indicated that the OMTKY[1+21 combination isolated by cationexchange chromatography of the first fraction contained roughly 20% of an unresolved ovomucoid component, which may have been either OMTKY or clipped protein as indicated by the N-terminal sequencing. The late eluting impurity peak shown in Figure 6 was not found in the OMTKY[1+2] isolated protein.
Electrospray Ionization Mass Spectrometry (ESI-MS). The ESI mass spectrum of the nonglycosylated OMTKY3 contains ions of charge states +2, +3, and +4, yielding masses of 1506, 2016, and 3011 amu, respectively. From these masses, an average molecular weight of 6021 was calculated for OMTKY3,which was (26) Kato, I.; Schrode, J.; Wilson, K A: Laskowski, M., Jr. Pept. Bid. Fluids 1976, 23. 235.
Analytical Chemistry, Vol. 67, No. 14, July 15, 1995
2359
I
A
40 mAu
I
W I
0 0
IO
30
20
40
50
60
Time (min )
Figure 6. Reversed-phase chromatography of whole turkey ovomucoid with an Alltech Macrosphere 300 C18 (4.6 mm i.d. x 250 mm L) column, gradient elution from 0.1% TFA to 60:40 acetonitrile: 0.1% TFA for 60 m at a flow rate of 1.5 m u m and detection at 220 nm.
consistent with theory. The ESI-MS of the glycosylated OMTKY3S produced charge states of f 3 and +4, yielding an average molecular weight of 8670, which is 2649 higher than the molecular weight of OMTKY3. Each major peak was accompanied by a family of multiplets, indicating significant glycosylated microheterogeneity. The glycosylated OMTKY2 ESI-MS spectra exhibited f 4 and +5 charge states. Based on an intense 1864 m a d c h a r g e ion (charge state +5), the calculated molecular weight was 9321. This exceeds the unglycosylated weight by 1816. The ESI-MS of OMTKY[1+2] was composed of unresolved overlapping peaks for each charge states, likely due to protein fragments and the microheterogeneity in glycosylation. A molecular weight was estimated at 21 170, which exceeds the unglycosylated molecular weight of OMTKY[ 1+2] by 7000, roughly the amount expected for carbohydrates on the first and second domains. The ESI-MS spectra provide insights into the nature of the carbohydrates on the OMTKY isolated domains. The glycosylation of ovomucoid domains has been reported by other investigaGlycosylation at Asn 53 on the turkey first domain was elucidated by Risley and VanEtten using NMR15 They found that the microheterogeneity could be divided into three carbohydrates: (GlcNAc)&ex)s, (GlcNAc),j(Hex)s, and (GlcNAc)&Iex)7 in amounts of 16%,23%,and 6196, respectively; where GlcNAc is N-acetylglucosamine and Hex is a hexose, either manose or galactose. Assuming homology, one can surmise from the ESIMS spectra that the predominant carbohydrate on OMTKY3S is (GlcNAc)g(Hex)s. Differences in masses observed between spectral signals correspond to a hexose or GlcNAc-Hexunit, thus additional G1cNAc:Hex carbohydrate ratios of 9:4, 85, 8:4,and 8:3 can be postulated for OMTKY3S. Similarly for glycosylated OMTKY2, a primary (M 5) ion in ESI-MS spectra yields 9320 amu, which corresponds to OMTKY2 with (GlcNAc):,(Hex)s at Asn 75. Heterogeneity at this site is evident from the ESI-MS spectra, which exhibits additional signals at 9691, 9161,8785, and 8629 amu; logically corresponding to (GlcNAc):(Hex) ratios of 6:6, 5:4, 4:3, and 4:2, respectively. Chiral Recognition of OMTJW and Isolated DomainBonded Columns. Silica bonded-phase HPLC columns were
+
(27) Wang. R.; Cotter, R. J.; Lin, T.Y.; Laskowski, M., Jr. Rapid Commun. Mass Spectrom. 1988,2, 71.
2360 Analytical Chemistry, Vol. 67, No. 74, July 75, 7995
IO
20
30
h e (mm)
40
50
0
5
IO lime t m n
15
)
Figure 7. Separation of racemic lorazepam on columns made with The mobile phase for the OMTKY OMTKY (A) and OMTKY3 (6). column consisted of 20 mM phosphate buffer (pH 6.9)/2-propanol 1005 v:v, while the OMTKY3 column was eluted with the same mobile phase components at a ratio of 100:3 v:v; the flow rate was 0.8 m U m. and detection was at 220 nm.
made with whole OMTKY and the isolated domains of OMTKY3, OMTKY3S, OMTKY2, OMTKY1, and OMTKY[1+2]. The columns were tested for chiral recognition using many racemic compounds, including 31 commercially available therapeutic agents and a variety of investigative compounds produced at Upjohn. As necessary, the racemates were tested for achiral purity by reversed-phase HPLC. In addition, many separations were conducted with diode array UV-visible detection at multiple wavelengths to af6rm spectral integrity of enantiomers. In spite of the wide diversity of racemic mixtures tested, the OMTKY3 columns exhibited chiral recognition for only selected benzodiazepines and fused-ring profens, whereas the columns made with whole OMTKY protein resolved the enantiomers of many chiral compounds. Among racemic benzodiazepines, OMTKY resolved most compounds, whereas OMTKY3 resolved only selected compounds (Table 1). Figure 7 shows the resolution of lorazepam enantiomers on OMTKY and OMTKY3-bonded columns. By contrast, clorazepate and U-85,182 were resolved on the OMTKY3 column, but were only slightly or not resolved on the OMTKY column under comparable mobile-phase conditions (Ta.ble 1). This implies that isolation of the OMTKY3 enhances chiral recognition for some racemates. The chiral separations of several additional benzodiazepine racemates on the OMTKY column are shown in Figure 8. Results in Table 1 demonstrate that modification of the & group at the chiral carbon of a series of benzodiazepines greatly enhances resolution on OMTKY and enables weak chiral recognition on OMTKY3 for U-30,662 and U-32,069but not for U-38,621 or U-36,105. Placement of a methyl group at R1inhibits chiral recognition on OMTKY3 and eliminates enantioselectivity on OMTKY for U-33,836 and U-33,323 (Table 1) but greatly enhances resolution of U-31,628 compared to U-32,069 (Figure 9). The difference in chiral recognition between compounds that vary only by a chlorine atom at group & suggest that subtle substitutions can markedly affect enantioselectivity (T.able 1). Although ibuprofen and flurbiprofen could be resolved on OMTKY, such simple aromatic profens could not be resolved on OMTKY3. However, a collection of fused-ring profens did exhibit enantioselective resolution on both OMTKY and OMTKY3 columns, including pranoprofen (Figure 10) and U-80,413 (Figure
Table I.Structures of Benzodiazepines and Enantioseparation Factors on OMTKY and OMTKYS (:orllpoulld
R:,
'R,
1
€I I1 €1 I1
R,
OMTKY"
OMTKY3b
a
a
CI CI CI CI CI
weak 1.9
1.o 1.0 1.1 1.1 1.o
OH CH:,
€I
OCOCH, ocociI,cocII, CH,CH2C€i,COOH
€1
ternazqiaiii
OII
_-c
1J-31,628 1J-33,836 IJ-33,323 1J -33,206
CI CI
2.1 1.o 1.o
OH
I1 II II €1 I1
CI
OCOCH, OCOCH,COC€I, CH,CH,COCIf,
C1
1.4
1.0 1.0 1.o 1.o 1.0
3.0 31
1.3 12
1.0
1.1 1.1
oxazepam
IJ-38,621 IJ-32,069 1J-30,662 IJ-36,105
11
II I1 I1
c1
IJ-85,196 IJ-85,283 IJ-85,182 lorazepain
I1 I1 I1
011
c1
OCOCII,
(2
IJ
011
C1
11 Ii I1 C1
lormethazepam
CIi, I1
OH C0OIi
CI I1
Cl CI
cIorazapated
c1
cI
1.1 2.0 1.1
1.5 --c
1.o
1.0
1.2
a The separation factors (a) were determined on an OMTKY column with a mobile phase of 90:10, 20 mM phosphate buffer (PH 6.9)/2propanol. The separation factors (a) were determined on an OMTKY3 column with a mobile phase of 97:3, 20 mM phosphate buffer (PH 6.9)/ 2-propanol. An a of 1.0 means no separation was observed. Temazepam and lormethazepam were not tested during this survey, but were later strongly resolved on a column made with the combination domains OMTKY[1+2]. Neither was resolved on OMTKY3. Clorazapate has (OH)2 on the ring carbon between RI and R2 in place of the carbonyl.
0
5
IO Time (min )
15
20
0
5
10
15
20
Time (mln )
0
IO
20 lime (min )
30
I 40
I
0
5
10
15
20
lime (min )
Figure 8. Separation of racemic U-38,621 (A) and U-30,662 (B) on an OMTKY column. The mobile phase was 20 mM phosphate buffer (pH 6.9)/2-propanoI90:10v:v; the flow rate was 0.8 mum, and detection was at 220 nm.
Figure 9. Separation of racemic U-31,628 (A) and U-32,069 (B) on an OMTKY column. The mobile phase was 20 mM phosphate buffer (pH 6.9)/2-propanoI90:10 v:v; the flow rate was 0.8 mum, and detection was at 220 nm.
11). Moreover, clorazepate, U-85,182, and U-80,413 all exhibited higher enantioselective resolution on columns made with OMTKY3 compared to OMTKY. Table 2 shows a comparison of capacity factors for the first eluted enantiomer and enantioseparation factors of selected benzodiazepine and profen derivatives on the OMTKY3 and glycosylated OMTKY3S columns. The OMTKY3 and OMTKY3S columns gave comparable capacity factors for each compound tested, while the OMTKY3 column gave slightly higher enantioseparation factors than the OMTKY3S column. This demonstrates that the sugar moiety is not necessary for chiral recognition of these compounds on the turkey ovomucoid third domain. On the OMTKY[lf21-bonded column, chiral recognition was realized with a variety of compounds. Although only partially resolved, enantioselectivity could be observed with lorazepam,
oxazepam, and benzoin using a mobile phase of 20 mM phosphate (PH 6.9) with 3%2-propanol. With a slightly stronger mobile phase of 20 mM phosphate with 5%2-propanol, enantioselective resolution of hydroxyzine is slight, while enantiomeric separation of temazepam, lormethazepam, and verapamil are quite distinct. With 20% of 2-propanol in the mobile phase, enantioselectivity of the strongly retained promethazine and prenylamine lactate are observed. All of these molecules have some structural attributes in common: (i) with the exception of benzoin, all of the chiral carbons are attached to a nitrogen as a ternary or secondary amine; (ii) all of the compounds have more than one aromatic ring; (iii) all of the chiral carbons have an a hydrogen; (iv) none of the compounds have an acidic group; and (v) each chiral carbon has attached to it a hydrogen bond acceptor, amine, or hydroxyl group. This implies that a chiral binding site for multi-aromatic Analytical Chemistry, Vol. 67, No. 74, July 75, 7995
2361
Figure 12. R-Enantiomer structure of the U-78,517 racemate.
I
I
0
10
5
15
0
5
15
IO
lime Win )
Time (min 1
Figure 10. Separation of racemic pranoprofen on columns made The mobile phase was 20 mM with OMTKY (A) and OMTKYB (6). phosphate buffer (pH 6.9)/2-propanol 100:3 v:v; the flow rate was 0.8 mum, and detection was at 220 nm.
15
10
5
0
0
lime (min )
5
15
10 lime (min
)
Figure 11. Separation of racemic U-80,413 on columns made with The mobile phase was 20 mM OMTKY (A) and OMTKYB (6). phosphate buffer (pH 6.9)/2-propanol 1003 v:v; the flow rate was 0.8 mum, and detection was at 220 nm. Table 2. Comparlson of Enantloselectivity between UnglycosylatedOMTKY3 and Glycosylated OMTKY3S
OMTKY3
OMTKY3S
racemate
ki'
U
ki'
U
lorazepam clorazepate pranoprofen
2.00 1.53 3.14 3.14
1.12 1.14 1.04 1.10
1.94 1.57 3.34 3.08
1.08 1.12 1.00 1.05
U-80,413
kl' is the capacity factor of the first eluting enantiomer. u is the separation factor defined as the ratio of capacity factors between the enantiomers. The separations were conducted with OMTKY3 and OMTKY3S columns packed into 2 mm id. x 100 mm L columns eluted with 20 mM phosphate buffer (PH 6.9)/2-propanol 1003 v:v mobile run at a flow rate of 0.1 mL/m with 220 nm detection.
weak bases is present on the columns made with the OMTKY[1+2] protein fraction. Since the first and second domains are glycosylated, interaction with the carbohydrate may play a role in enantioselectivity with these compounds. It is important to emphasize that clorazepate, pranoprofen, and U-80,413, which were resolved by OMTKY3, were not resolved on columns made with the OMTKY[1+2]. This supports the conclusion that a chiral recognition site for these three weak acids is on the third domain. This also means that the amount of OMTKY present as an 2362
Analytical Chemistry, Vol. 67,No. 74, July 75, 7995
impurity in the OMTKY[l+2] fraction was insufficient to express chiral activity for these aromatic weak acids. No chiral recognition was realized on columns made with OMTKY2, although achiral retention was observed. Given the high purity of the OMTKY2 isolated protein, one can conclude that the second domain is void of chiral activity. This is supported by previous work by Miwa et aLZ8who treated whole chicken ovomucoid columns with trypsin in an attempt to determine if the enantioselective binding site was related to the enzyme inhibition site on the second domain. The trypsin-treated columns still exhibited chiral recognition, suggesting that the chiral site was not related to the trypsin site on the second domain. The OMTKYl columns exhibited activity for only a few aromatic weak bases, demonstrating a distinct lack of chiral activity compared to the OMTKY, OMTKY[1+21, and OMTKY3 columns. With the exception of the pure OMTKY3, which clearly can act independently, one must conclude that either the domains act in combination to express the high degree and wide range of enantioselectivity or that unknown components in the OMTKY and OMTKY[1+2] contribute to chiral recognition. This loss of enantioselectivity among the isolated domains compared to the whole OMTKY is best exemplified by the resolution of an investigative racemate shown in Figure 12. The enantiomers of this compound can be separated on a commercial Ultron ESOVM column made with whole chicken (OMCHI) ovomucoidZ9and the whole OMTKY columns, with resolutions (RJ of 4.7 and 3.6, respectively. In spite of this extremely high resolution on columns made with the whole ovomucoid, the U-78,517 enantiomers could not be resolved on the columns made with the isolated turkey domains. This implies that something is unique about the OMCHI and OMTKY columns made with whole ovomucoid. Either the domains act in combination or a protein impurity is bound along with the whole ovomucoid, accounting for much of the of enantioselectivity. As noted above, the whole turkey ovomucoid contained an addition component (Figure 6). Analogous protein impurities are also found in chicken ovomucoid (Figure 13). Follow-up research has demonstrated that one of these unknown glycoprotein impurities in chicken ovomucoid exhibits significant chiral By analogy, one may suggest that the missing amount of chiral recognition of the OMTKY column compared to the OMTKY3 and OMTKY[1+21 resides on a protein impurity. Unfortunately, insufficient resources have been available to test this hypothesis with the turkey ovomucoid. Although this impurity was found in the first size-exclusion fraction, it was (28) Miwa, T.: Kuroda. H.; Sakashita, S. J. Chromatogr. 1990,511, 89. (29) Pinkerton, T. C.; Koeplinger, K A,; Haginaka. J. Chiral Separations of a Macamine-Tocopherol on an Ovomucoid Bonded-Phase HPLC Column. Presented at the Fourth International Symposium on Pharmaceutical and Biomedical Analysis, Baltimore, MD. April 18-21, 1993. (30) Haginaka, J.; Seyama. C.; Kanasugi, N . Anal. Chem., submitted for publication.
I 0
5
[Pranopmfen] : [OMTKY3]
20 mAu
10
15
20
25
30
35
40
45
50
lime (mln.)
Figure 13. Reversed-phase chromatography of whole chicken ovomucoid using an Alltech Macrosphere 300 C18 (4.6 mm i.d. x 250 mm L) column with gradient elution from 0.1% TFA to 60:40 acetonitrile/O.l% TFA for 60 m at a flow rate of 1.5 m u m with detection at 220 nm.
removed on isolation of the OMTKY[1+21, so it was not isolated with OMTKY[1+21; thus, this impurity could not be responsible for the chiral activity of the OMTKY[l+21 column. NMR Studies on the Third Domain. In spite of the fact that OMTKY3 exhibited chiral recognition for only a few benzodiazepines and weak acid fused ring profens, the availability of proton NMR assignments and crystal structure coordinates for ovomucoid third domain afforded a unique opportunity to study an enantioselectiveprotein binding mechanism. An NMR investigation was conducted on the purified ovomucoid third domain in order to identify the binding site amino acids responsible for the enantioselective discrimination among weak acid benzodiazepines and fused ring profens.21 The experiments involved the titration of racemic clorazepate and pranoprofen with recombinant chicken ovomucoid third domain (OMCHI3); the titration of racemic clorazepate, pranoprofen, and racemic U-80,413 with OMTKY3; and the titration with OMTKY3 of pure enantiomers of U-80,413, which had been Wlabeled at the profen a-methyl DOUP. The first experimentation was conducted by titrating clorazepate with OMCHB, while monitoring changes in proton chemical shifts. Although chicken and turkey ovomucoid differ at 29 amino acid residues, only three of these amino acids reside on the third domain. OMTKY3 and OMCHI3 d ~ e only r at residues at 145, 148, and 150, which are respectively Ala (A), Leu Q, and Tyr 0 in OMTKY3 (Figure 1) and Asp @), Ala (A), and Asp 0) in OMCHIS. For purposes of this discussion, the amino acids of the third domain 131-186 (Figure 1) are renumbered from 1to 56, so the three amino acids above in OMTKY3 are Ala 15, Leu 18, and Tyr 20. During the titration of racemic clorazepate with OMCHI3, aromatic proton resonances due to clorazepate shifted downfield and broadened as the OMCHI3 to clorazepate ratio increased, indicating interactions between clorazepate and the protein. The free and bound ligand were in fast exchange on the NMR time scale, so observed chemical shifts for both protein and ligand resonances represented weighted averages of free and bound populations. NOESY spectra of the clorazepate-OMCHI3 complex revealed that several clorazepate aromatic proton signals did not overlap with protons from the aromatic amino acids. This
1.
3.8 mM :0.5 mM
4
k
1
I
I
I
I
I
I
8.2
8.0
7.8
7.6
7.4
7.2
7.0
PPm Figure 14. Shift in proton NMR resonances between 7.0 and 8.2
ppm during titration of racemic pranoprofen with OMTKYB. See Figure 15 for the structure of pranoprofen and its proton designations.
division of resonances held true with the pranoprofen and U-80,413 as well, thus facilitating the ease of monitoring chemical shifts from model compounds and protein. These initial NMR results suggested that clorazepate interacted with Val 6, Val 41, and Lys 34 on the third domain. Subsequent titrations of the clorazepate with OMTKY3 gave results similar to that described below for pranoprofen and U-80,413. The second set of experiments involved the titration of racemic pranoprofen with OMTKY3. Figure 14 illustrates the proton spectra between 7.0 and 8.2 ppm during the titration. The spectrum between 3.5 and 4.1 ppm is shown in Figure 15, along with the pranoprofen structure. As more pranoprofen was combined with OMTKY3, the 6 and E ring protons of Phe 53 exhibited a strong upfield shift, while the 6 protons of Tyr 20 showed small downfield shifts compared to the OMTKY3 spectrum (not shown). The c, d, and a-methyl protons of pranoprofen shifted upfield, while the a proton shifted downfield. The methylene protons at position d divided into two signals (Figure 15). Since the sample was racemic, this implies that the environment of one enantiomer has changed with respect to the other on binding. A similar division of signals was observed with the c proton, whereby the doublet progressed toward a triplet (overlap ping doublets) as the sets of signals for each enantiomer shifted on binding (Figure 14). These interactions were further elucidated by NOESY spectra (Figure 161, which clearly demonstrate the existence of two environments for protons c and d when protein bound. The data also suggest that proton c l from one Analytical Chemistry, Vol. 67,No. 14, July 75, 7995
2363
I
-
b
C
d
e
2
'l
bHCOOH CH, [U-80,413] : [OMlKY3] 0.64 mM : 2.6 mM
[Pranoprofen]:[OMTKY3] 3.2 mM : 2.0 mM
I
I
I
I
I
8.0
7.8
7.6
7.4
7.2
I
7.0
I
6.8
ppm
Figure 17. Structure of U-80,413 and shifts in proton NMR resonances during titration of racemic U-80,413 with OMTKY3. Dashed lines indicate the shifts of the 6 and e protons of phenylanine 53 upon binding of U-80,413; and the corresponding shifts of the noninteractive tyrosine 11 are shown for reference. I
I
I
I
I
I
I
3.9
3.7 3.5 PPm Figure 15. Structure of pranoprofen and division of NMR resonances of methylene protons d on racemic pranoprofen during titration with OMTKY3. 4.1
a K34
R21
c l c2
-=e I -\-,k3= I
0
0
0
4
1- 1.5 0
ou
0 '
0
n
'
20
35
C
T
d
e
.
CHCOO I CH3
4.0
' Pranoprofen 7.6 PPn' Figure 16. Portion of NOESY data for the complex between racemic pranoprofen and OMTKY3 at concentrations of 7 and 10 mM, respectively, in 100 mM phosphate at pH 8. Data were collected on a Bruker AM400 NMR spectrophotometer as 512 blocks with 64 scans per block over period of 22.5 h with a NOESY mixing time of 150 ms.
enantiomer may interact with the y protons of Arg 21, while proton c2 (on the other enantiomer) may interact with the y protons of Lys 34. In addition, NOEs with protons a and b were observed with the &methyl groups of Leu 23, as well as protons on other amino acids (Figure 16). This second set of experiments with racemic pranoprofen further revealed the binding region on the third domain by demonstrating interactions with Phe 53, Leu 23, Arg 21, Pro 22, and Lys 34. Similar titrations were conducted with pranoprofen and OMCHI3, whereby comparable results were observed, except for the contrasting shift in the c proton of pranoprofen, which moved downfield rather than upfield. Since the OMCHI3 has Asp at position 20 rather than Tyr, this further supports the location of the c proton for one enantiomer of pranoprofen near amino acids 20 and 21. 2364 Analytical Chemistry, Vol. 67, No. 14, July 15, 1995
The third set of experiments encompassed the titration of racemic U-80,413 and pure enantiomers of aJ3CH3-U-80,413 with OMTKY3. During the titration of racemic U-80,413 with OMTKY3, the Phe 53 6 and 6 proton signals shifted downtield on binding (Figure 17), as did the signals for Leu 23 (not shown). In contrast to pranoprofen, Tyr 20 was unaffected by the titration. The aromatic proton signals of racemic U-80,413shifted downtield and split into doublets of doublets. The d proton divided into two signals, in a manner analogous to the methylene protons of pranoprofen. Again, this is an indication that each enantiomer is experiencing distinctly different environments at this position. Titration of the pure enantiomers confirmed these shifts, giving individual signals for the d proton on binding. Chemical shifts and NOESY spectra of U-80,413 with OMTKY3 indicated interactions with most of the same amino acids observed with clorazepate and pranoprofen including Val 6, Arg 21, Pro 22, Leu 23, Lys 34, Val 41, Val 42, Leu 48, and Phe 53. Signals from Val 42 and Leu 48 were weak and could be a result of spin diffusion. The isotopeedited NOEs of the pure enantiomers of aJ3CH3-U-80,413 with OMTKY3 exhibited very similar spectra and showed interaction of the a-13CH3with Val 6 and Leu 23 as well as weaker signals with other amino acids. This result was interpreted to mean that the a-methyl group from each enantiomer was located in the similar environment near Val 6 and Leu 23 on association with the protein. One can conclude from these NMR experiments that clorazepate, pranoprofen, and U-80,413 each interact with the same protein surface region of OMCHI3 and OMTKY3. By measuring the shifts in proton resonances for Phe 53 (6 and E ) , e, c, a, and d over a OMTKY3/U-80,413 ratio range of 0.03-14.5 by six steps in 0.1 M phosphate buffer at 27 "C, pH = 8.0, average binding association constants were estimated as 2300 and 1400 M-l for the strongly retained (Kz) and weakly retained enantiomer (K:), respectively.21 One can compare these binding constants to that of L-and D-tryptophan with BSA, which are 2600 and 960 M-I, respectively, or (3-and @)-ibuprofen with HSA, which are 1400 and 840 M-l, re~pectively.~~ In each case, these enantiomers are easily resolved on bonded-phase column made from the albumin protein noted. -~
(31) Cheruvallath. 1'. K.: Narayanan, S. R.: Lindenbaum S. A Microcalorimetri Investigation of the Selectivity of Proteins for Chiral Drugs. Presented at the 7th Annual Meeting of the American Association of Pharmaceutical Scientists, San Antonio, TX, 1991.
The bmdmg constants for U-80,413enantiomers with OMTKY3 indicate ample resolving power for the fused-ring profens. The separation factor (a) is related to these thermodynamic quantities as follows:
a = K2/Kl and A(&) = R T l n a where A(&) is the difference in the chemical potential of binding. R is the gas constant of 1.988 cal/mol K, and T i s temperature in K These binding association constants would yield an a of 1.6 and a difference in chemical potential of roughly 300 cal/mol. The fact that a 1.6 separation factor is not observed on the OMTKY3 columns reflects the difference between the conditions of the titration and the mobile phase (20 mM phosphate in 3%2-propanol, pH 6.9). Tnis available chemical potential is consistent with the amount of energy theoretically predicted as necessary for enantiomeric resolution on chiral stationary phases.” Molecular Modeling of the Ovomucoid Third Domain. Molecular modeling of the enantioselective bmdmg process of pranoprofen and U-80,413 with OMTKY3 was possible because crystal structures have been determined for nonglycosylated ovomucoid third domains in several avian species. The Upjohn Company‘s Mosaic Autodock program was used to allow a model of each enantiomer to search the surface of the OMTKY3 for low energy binding orientations while keeping both the ligand and the protein rigid. The Autodock program employs a modification of the geomehic hard sphere approach to docking developed by Kuntz.” In Mosaic’s Autodock, an extended radius dot surface is first generated over the entire protein. Dots are pruned away according to a procedure, which retains dots that represent potential ligand atom positions that would interact with more than one protein atom. The retained dots are then used as ligand atom target positions in a Kuntz clique detection-based” rigid overlaying of the ligand shucture on as many dots as possible. Each docking is scored by summing the van der Waals and electrostatic interactions of the ligand atoms with the protein. When docking the U-80.413 and pranoprofen enantiomers with OMTKY3. the Autodock program generated about 6500 orientations. Energy minimization of the best 200 dockings and graphics display of the lowest energy 100 orientations revealed most of the enantiomers clustered around Arg 21. Observation of the 10 lowest energy dockings for each enantiomer revealed localization near the same protein surface cavity with Leu 23 at the bottom and surrounded by the adjacent amino acids Arg 21, Pro 22, Phe 53, Val 6, and Lys 34, respectively, in clockwise order. Independently, this Same binding region was identified by NMR conducted at a separate research site without knowledge of the modeling results (vide supra). In order to identify the orientations best representing the chual recognition process, a sum of distances was calculated (for each of the 100 docking orientations) between the carbon atoms of the enantiomers and the amino acids on OMTKY3 identified during NMR titrations (see Experimental Section for details). These amino acids included Val 6, Arg 21, Pro 22, Leu 23, Lys 34, Val 41, Val 42. Leu 48, Leu 50, and Phe 53. The six orientations for each enantiomerwith the lowest sum of distances were visualized in three dimensions and studied relative to discriminating inter(32) Iipkowifz,K B.;Demeter, D.A Parish, C. A And. Cham. 1987.59.1731. (33) Kunfz, 1. D.; Blaney, 1. M.:Oatley, S.I.:Langridge. R Ferrin, T.E.J. Mol Bid. 1982, 161, 269.
Figure $8. Molecular modeling simulation of U-80,413 enantiomers bound to OMTKY3. The white curved graphic represents the protein backbone. Selected protein side chains are also shown in white. Ligands are labeled according to their Ror Schirality and numbered according to their position among the 100 lowest energy minimized binding orientations.
molecular interactions and conformance to constraints imposed by NMR observations. The first finding of the molecular modeling was that on the average the R-enantiomers were docked closer to the protein and occupied lower energy states than the Senantiomers. Second, the amino acids were clustered into two groups with one surface region defined by Val 6, Arg 21, Pro 22, Leu 23, Lys 34, and Phe 53 and a second region occupying amino acids Val 41, Val 42, Leu 48, and Leu 50. With the 6rst binding region, all of the amino acids in the group were adjacent to one another and created a pocket with Leu 23 at the bottom and the remaining amino acids making up the walls. The second group of amino acids resided on another side of the protein with a large high, protruding ridge separating the two binding regions. It was immediately obvious from surface distances that an enantiomer could not interact with both regions at the same time. Further, it was noticed that the second region was composed of mainly hydrophobic amino acids and afforded only limited opportunities for hydrogen bonding or ionic interactions. Figure 18illustrates the binding orientations of the enantiomers of U-80.413 in each of the two surface regions identified on the OMTKY3. The first group of amino acids mentioned above is on the left, while the second group is on the right The white tubular structure represents the protein backbone. The peptide strand is wrapped around an a-helix and held in place by three disulfide bonds (shown in yellow). The N-terminal is at the top right and the C-terminal is in the back and to the left where a vertical strand comes down to meet a horizontal strand. At this point the Cterminus is attached to the other strand by a disulfide bond. The glycosylation Asn 45, attached to the right end of the a-helix, would extend away from the protein, thus explaining its noninvolvement in the enantioselective discrimination. The R-enantiomen of U-80,413 are shown in green while the Senantiomers are in magenta. The dockings in the second site to the right were of higher energy and without apparent points of interaction that could produce chiral recognition. This nonspecific binding site Analytical Chemistry, Vol. 67,NO. 14,J U I ~15,1995 2365
onlo lnonotlon
Figure 19. Enantiomers of U-80,413 in enantioselective binding site of OMTKY3 (see text lor description).
(Le., only hydrophobic interactions) likely accounts for the NOEs to Val 41. One cannot rule out, however, that this might be a site for chiral discrimination on OMTKY3 for the neutral benzodiazepines Kable l). Among the three closest dockings for each of the R- and Senantiomers of U-80,413. all appear in the binding region to the left. A closer view of that region and the selected binding model for each enantiomer is shown in Figure 19. One can see similarities and differences in orientation and intermolecular interactions between the R- and S-enantiomers of U-80.413. The carboxyl groups of each enantiomer engage in electrostatic interactions with the positive charge on Arg 21. The l3C-Iabeled methyl groups on the chiral carbon of each enantiomer are directed toward the protein pocket, thus accounting for similarities in the isotope edited NOEs. The carbonyl group on U-80,413's central ring share a hydrogen bond with NH3+ group of Lys 34. The distinguishing difference between the enantiomers is the proximity of the phenyl group of R-enantiomer and Phe 53. One must keep in mind that these models were generated from docking against a rigid protein structure in the absence of solvent. When an energy minimization is conducted on the R-enantiomer with the inclusion of solvent effects, the three intermolecular interaction hold the R-enantiomer in place. When the S-enantiomer is given the same treatment, the molecule rotates clockwise about an axis between the ionic and hydrogen bonded interactions and drifts from the surface on each iteration, as it searches for minima. The orientation depicted in Figure 19 is likely the most probable for the more tightly bound R-enantiomer. The orientation of the S-enantiomer, however, might be thought of as a preferred limit if the polar and ionic interactions predominate. In reality, the more loosely held Senantiomer likely assumes more than one orientation during the dynamic partitioning process, as any two interactions compete for binding, thus resulting in shorter resident times. As an example, one might imagine the aromatic and ionic interactions predominating for an association event that would orient the Senantiomer differently and perhaps still account for the NMR data given the flexibility of the protein in solution and spin diffusion energy transfer. At this juncture, it was decided to depict only those orientations arrived at by the modeling approach described. Additional chiral recognition mechanisms may be uncovered as more detailed NMR analyses progress and alternative modeling approaches are attempted. The orientations illustrated in Figure 19 are largely consistent with the NMR data. The interaction of the U-80,413 phenyl ring 2366 Analytical Chemisty, Vol. 67, No. 14. July 15, 1995
would account for the downEeld shift of Phe 53 protons. The proximity to Leu 23 would explain the downfield shift in Leu 23 protons and the NOEs observed between Leu 23 d-protons. Further, the closeness ofVal 6 (in the front of the pocket between Lys 34 and Phe 53) and Lys 34 explain NOEs to protons on these amino acids. Protons a. c. d. and e on each enantiomer occur in different environments, thus accounting for differences in chemical shifts on binding between the enantiomers and the division into multiplets for the racemic mixture (Figure 17). The enantiomers of pranoprofen were modeled in the same manner as described above. The enantioselective binding site for pranoprofen was the same as that for U-80,413.but the favored orientations for the enantiomers were in different positions on the protein surface. The carboxyl groups of both enantiomers exhibited charged interactions with Arg 21, but the fused aromatic rings were pointed in opposite directions. The Senantiomer of pranoprofen was directed toward Phe 53 and Lys 34, while the R-enantiomer was pointed in an opposite direction towards Tyr 20. These orientations would be consistent with NOEs of the c protons of the separate enantiomers to L p 34 or Arg 21 (Figure 16). Other orientations about Arg 21 were plausible for the Renautiomer, several being near Tyr 20. CONCLUSION
A chiral binding site has been identified on the third domain of turkey ovomucoid and characterized by chromatography, NMQ and molecular modeling. The protein binding is enantioselective toward fused ring aromatic weak acids. The glycosylated group on the third domain is not needed for this chiral recognition. HPLC columns made with a combination of the first and second domains of turkey ovomucoid exhibit chual recognition for a small group of aromatic weak bases. The absence of any significant chiral recognition on the isolated first and second domains infers that the domains may act in Combination to express enantioselectivity. It is known from NMR studies that the third domain of turkey ovomucoid interacts with the rest of the protein.I6 The unaccounted for chiral recognition in the whole turkey ovomucoid suggests that an impurity in the whole ovomucoid may be responsible for much of the columns chiral discrimination. Given the amino acid differencesbetween avian ovomucoids, this work does not imply that isolated chicken third domain or even puriried whole chicken ovomucoid would exhibit the same chiral recognition. ACKNOWLEDGMENT The authors wish to express their appreciation to Frank Crow,
Russ Robins, Scott Plaisted. and Greg Cavey at Upjohn for mass spectrometry and amino acid analysis of the protein domains. Special thanks are extended to Jim Tustin at Upjohn for synthesis of the 13C-labeled U-80,413 and to Dale Wieber at Upjohn for assistance in preparing the tigures. Hiroo Wada and Hiroya Fujima of Shinwa Chemical Industries, Kyoto, Japan, are thanked for their generous help in packing the protein bonded-phase columns and for donating the aminopropyl-silica gel. Particular gratitude is expressed to The Upjohn Co. for financial support to J.H. for a 1-year stay at Upjohn as a Visiting Scientist under an arrangement with the University of Wisconsin and for external grant funding which enabled the continuance of chromatography research at Mukogawa Women's University, Nishinomiya. Japan. The Upjohn Co. is also thanked for the unrestricted funds provided
to J.L.M. at the University of Wisconsin in appreciation of this collaborative research endeavor. In addition, the work was supported in part by NIH Grant GM-35976 to J.L.M. NMR studies were carried out at the National Magnetic Resonance Facility at by the NIH Biomedical Research Madison (operation Technology Program under Grant RR02301; equipment funded by the NIH Biomedical Research Technology Program under Grant RR02301, the University of Wisconsin, the NSF Biological Instrumentation Program under Grant DMB-8415048, the NIH
Shared Instrumentation Program under Grant RR02781, and the US. Department of Agriculture). Received for review December 16, 1994. Accepted March 18, 1995.@ AC941222G @Abstractpublished in Advance ACS Abstracts, May 15, 1995.
Analytical Chemisfry, Vol. 67,No. 14, July 75, 7995
2367