NMR and Computational Studies of Chiral ... - ACS Publications

Jan 18, 2007 - Bristol-Myers Squibb Company, 1 Squibb DriVe, New Brunswick, New Jersey ... Morristown, New Jersey 07690, and Department of Chemistry, ...
0 downloads 0 Views 766KB Size
Download PDF
J. Phys. Chem. B 2007, 111, 1189-1198

1189

NMR and Computational Studies of Chiral Discrimination by Amylose Tris(3,5-dimethylphenylcarbamate) Yun K. Ye,† Shi Bai,‡ Shyam Vyas,§ and Mary J. Wirth*,| Bristol-Myers Squibb Company, 1 Squibb DriVe, New Brunswick, New Jersey 08903. Department of Chemistry and Biochemistry, UniVersity of Delaware, Newark, Delaware 19716, Accelrys, Inc., 19 Witherspoon Court, Morristown, New Jersey 07690, and Department of Chemistry, UniVersity of Arizona, 1306 East UniVersity BouleVard, Tucson, Arizona 85721 ReceiVed: June 14, 2006; In Final Form: October 9, 2006

Proton NMR and simulations were combined to study the origin of chiral selectivity by a polysaccharide used in a commercial chromatographic stationary phase: amylose tris(3,5-dimethylphenylcarbamate). This material has unusually high enantioselectivity for p-O-tert-butyltyrosine allyl ester, which is activated by the presence of an acid. Proton NMR spectra agreed with the HPLC in showing that the L-enantiomer interacts much more strongly with the polysaccharide and that acidity switches on the selectivity. 2D NOESY spectra revealed which protons of each enantiomer and the polysaccharide were in proximity, and these spectra revealed folding of the L-enantiomer. Computations generated energy-minimized structures for the polysaccharideenantiomer complexes, independently predicting folding of the L-enantiomer. Molecular dynamics simulations 2 ns in duration, repeated for three different energy-minimized structures, generated pair distribution functions that are in excellent agreement with the 2D NOESY spectra. The modeling studies revealed why acidity switches on chiral selectivity and minimally affects the chromatographic retention time of the unfavored D-enantiomer. The results comprise the first case of a chiral separation by a commercial polysaccharide stationary phase being explained using a combination of 2D NOESY and simulations, providing excellent agreement between experiment and computation and lending detailed molecular insight into enantioselectivity for this system.

Introduction The polysaccharide-based chiral stationary phases (CSPs) have been commercialized for more than a decade and are now the most popular CSPs in the field of chiral separations, yet their chiral discrimination mechanisms remain unknown. Understanding how chiral discrimination occurs will provide valuable insight for predictions such as the magnitude of enantioseparation, suitable CSPs for specific compounds, elution order, and appropriate chromatographic conditions. Understanding the chiral separation mechanism could thus enable method development to become a systematic approach instead of a trial and error process for chiral chromatography. Insight into chiral discrimination at the molecular level for polysaccharide-based CSPs is hindered by the complexity of the polymer, such as the exact stereochemical structure, the geometry of the interaction, the accessible binding sites, and the multiplicity of sites with different affinities for enantiomers. Numerous techniques, such as X-ray crystallography,1-5 NMR,6-11 calorimetric studies,12 IR,13 and computational methods,9-10,14-19 have been used to provide insight into the mechanisms of chiral recognition for other CSPs. NMR is well established as the most powerful tool for the elucidation of chiral recognition mechanisms. The changes in * To whom correspondence should be addressed. E-mail: mwirth@ email.arizona.edu. Fax: (520) 621-8407. † Bristol-Myers Squibb Co. ‡ University of Delaware. § Accelrys, Inc. | University of Arizona.

chemical shifts and line width, as well as signal splitting of the enantiomers in the proton NMR spectrum, provide information about which protons are affected by the interaction between the chiral selector and each enantiomer. In addition, 2D NOESY measurements reveal proximities of protons in the CSP and the analyte to provide structural information for the CSP-analyte complex. These two tools have been used to study chiral recognition by small chiral selectors, including cyclodextrin20 and rationally designed selectors.21 NMR has been used to study commercial stationary phases themselves, rather than only the unbound selector, for small molecules as selectors.22,23 NMR has recently been applied to the study of the polysaccharide-based CSPs, which has been a challenge because these polymers either have low solubility in suitable NMR solvents or have not shown chiral discrimination in solvents in which they are soluble. Yashima et al. found that cellulose tris(4(trimethylsilyl)phenylcarbamate) (CTSP)7,8 is soluble in chloroform and it shows chiral discrimination in 1H NMR spectroscopy that agrees with the elution order in HPLC. The methine proton of trans-2,3-diphenyloxirane was split into two singlet resonances in the presence of CTSP, indicating that CTSP in the NMR-compatible solvent distinguishes the enantiomers. However, the differences in the chemical shifts of the enantiomers in the presence of CTSP were too small to obtain specific structural insight into the chiral recognition through NOE measurements because of weak interactions. Okamoto’s group studied a different polysaccharide, CSP, cellulose tris(5-fluoro2-methylphenylcarbamate) in chloroform, and chose a compound exhibiting high chiral selectivity, 1,1′-bi-2-naphthol (R ) 4.23).9 Their system also exhibited chiral discrimination in

10.1021/jp0637173 CCC: $37.00 © 2007 American Chemical Society Published on Web 01/18/2007

1190 J. Phys. Chem. B, Vol. 111, No. 5, 2007

Ye et al.

Figure 1. 1H NMR spectra for the samples indicated, each with ethanesulfonic acid added: (A) O-tert-butyltyrosine allyl ester alone; (B) D-enantiomer of O-tert-butyltyrosine allyl ester plus ADMPC polymer; (C) L-enantiomer of O-tert-butyltyrosine allyl ester plus ADMPC polymer; (D) ADMPC polymer alone. An asterisk indicates an impurity peak. 1H

and 13C NMR spectroscopies that agreed with HPLC. The binding geometry and dynamics between cellulose tris(5-fluoro2-methylphenylcarbamate) and the enantiomers of 1,1′-bi-2naphthol were described from spin-lattice relaxation times, 1H NMR titrations, and intermolecular NOEs in the presence of cellulose tris(5-fluoro-2-methylphenylcarbamate). These advances demonstrate the promise of using chloroform for studying polysaccharide selectors, but there has not yet been an NMR study of chiral selectivity for a commercialized polysaccharide CSP. One of the most diversified and widely used polysaccharidebased chiral stationary phases is amylose tris(3,5-dimethylphenylcarbamate), ADMPC, which is commercially available as the Chiralpak AD chiral column. Recent reports24-27 demonstrate that low-molecular-weight (DP ≈ 100) ADMPC, prepared by the enzymatic polymerization of R-D-glucose 1-phosphate dipotassium, is soluble in chloroform, which makes the NMR investigation of the chiral discrimination possible for this polysaccharide. Yamamoto et al.10 reported the structural analysis of ADMPC in chloroform using the 2D NMR NOESY technique, along with computer modeling, and the study showed that a left-handed 4/3 helical structure was the most probable one. The binding geometry between ADMPC and the enantiomers of 1-(9-anthryl)-2,2,2-trifluoroethanol was also investigated by 1H NMR titration. On the basis of these results, combined with molecular modeling, a model to explain the chiral discrimination mechanism of 1-(9-anthryl)-2,2,2-trifluoroethanol on ADMPC was proposed. However, intermolecular NOEs for the complex could not be observed because of the low binding strength. To our best knowledge, no other NMR study has been reported for ADMPC CSP with other chiral

analytes, and no successful 2D NOEs were obtained for ADMPC or other commercial polysaccharide CSPs. The purpose of this study is to investigate the chiral recognition between the ADMPC CSP and an analyte that exhibits very large chiral selectivity in HPLC, allowing the use of both 1D proton and 2D NOESY NMR to be combined with molecular modeling. The analyte chosen for these studies is O-tert-butyltyrosine allyl ester, for which R ) 16.29 High enantioselectivity is observed generally for amino acid esters with the ADMPC CSP,28-30 where the enantioselectivity is switched on when the mobile phase is acidic. Acidity increases the retention time of only the L-enantiomer, while the retention of the D-enantiomer is changed only slightly. The very strong binding of the L-enantiomer, combined with the large and specific effect of acidity, makes amino acid esters a particularly opportune system to investigate. Materials and Methods Reagents. All chemicals used in this study were reagent grade or better. The racemic and individual enantiomer compounds studied were obtained from Bachem (King of Prussia, PA), and the structure of this analyte is shown in Figure 1. 1,3,5-Tritert-butylbenzene and CDCl3 (99.8 atom %D) were purchased from Sigma-Aldrich (St. Louis, MO). Ethanesulfonic acid (99%) was purchased from Fluka (Milwaukee, WI). A ChiralPak AD column (ADMPC polymer CSP) was purchased from Chiral Technologies, Inc. (Exton, PA). NMR. The solvent used for NMR in this study is CDCl3. All chemical shifts were reported in parts per million (ppm) with respect to the peak for tetramethylsilane (TMS; 0 ppm) as

Chiral Discrimination by ADMPC the internal reference. The samples were freshly prepared without further degassing. 1H NMR spectra were obtained on a JEOL NMR spectrometer operating at 400 MHz at 297 K. 2D NOESY NMR spectra30a were recorded at 298.0 ( 0.5 K on a Bruker AVANCE 600 MHz spectrometer, operating at a 1H frequency of 600.13 MHz. The 2D NOESY spectrum was carried out in the States-TPPI mode30b,c with a 2048 × 400 data matrix with 16 scans in each t1 experiment. The mixing time of the NOESY spectra was set to 0.80 s. This mixing time was chosen on the basis of the results of the 1H spin-lattice relaxation time (T1) survey experiment of ADMPC, where the proton T1 was found to range from 0.6 to 1.2 s. A spectral width of 8.01 kHz was used for both dimensions. NMRPipe31a was used to process 2D NOESY data, which were presented with Sparky.31b The concentrations for both enantiomers and ADMPC were 1 and 22 mg/mL, respectively. To obtain clean ADMPC for the NMR study, the commercially available Chirlpak AD column, in which ADMPC is physically deposited onto (γ-aminopropyl)silica, was washed with 50 column volumes of hexane at 40 °C using an Agilent HPLC 1100 system. The ADMPC CSP packing material was then emptied from the column and dried at 60 °C under vacuum for 24 h to dispose of hexane. The polysaccharide was extracted from the dried packing material with chloroform. The chloroform solution was filtered using a 4.5 µm syringe filter and evaporated to dryness. The obtained solid was dried at 60 °C under vacuum for 24 h. Molecular Modeling. Generation of Energy-Minimized Structures. The molecular modeling calculations were performed with the Discover molecular mechanics code32 in conjunction with the COMPASS force field.33a-c All visualization and post simulation analysis were done with the MS-Modeling software suite.32 The COMPASS force field was chosen as it has an extensive coverage for most organic functional groups;33c in addition, it has been shown to produce good results for amineand polysaccharide-containing compounds.34a-c Also, to ensure suitability for our systems, we compared the minimized structures of both the monomer of ADMPC and an analyte molecule using COMPASS vs quantum mechanical DFT calculations. The structures agree well, as shown in the Supporting Information (DFT_vs_COMPASS.jpg), and COMPASS is concluded to be a suitable force field for the calculations. The first step was to construct the ADMPC polymer; in the present work a PDB file of a 12-mer of ADMPC was kindly provided by Prof. Okamoto’s group.10 To eliminate so-called finite size effects, which can result when using a relatively short polymer chain, a chain of infinite length was created using periodic boundary conditions. This eliminates the finite size effect without having to include a large number of atoms. A cell 40 Å × 40 Å × 50 Å was used, and the 12-mer was placed parallel to the c-axis. The chain was created by joining the sugar rings of the oligomers in adjacent cells via an ether linkage. The cell size chosen was designed to minimize chain-chain interactions in the a-b direction. The polysaccharide backbone atoms were fixed to the positions determined by Okamoto et al.,10 in accord with their NMR results, but the phenyl side chains were allowed to move freely. This structure was then minimized with the COMPASS force field. We used “quenched annealing” to reduce the possibility of a local minimum. With quenched annealing, the initially minimized complexes were simulated with molecular dynamics at 500 K, 20 different trajectories were sampled, each conformation was then minimized, and the lowest energy configuration was used for the

J. Phys. Chem. B, Vol. 111, No. 5, 2007 1191 subsequent studies. A PDB file of this energy-minimized structure is provided in the Supporting Information as ADMPC12mer.pdb. A protonated analyte molecule was placed in a pocket within the polymer; the energy-minimized D- and L-enantiomers were each initially placed in the groove of the main chain of the energy-minimized ADMPC so that the positions of the 3H-bH and 5H-cH pairs of protons were about 5 Å apart. This initial position was chosen to be consistent with the results of 2D NMR. To study the binding interaction between the analyte and ADMPC, we used a two-step process for both enantiomers. First, another quenched annealing process was used, this time to rapidly explore many potential analyte conformations in the polymer pocket. Second, molecular dynamics (MD) simulations were performed on three of the resulting minimized structures to obtain average energies and average interatomic distances. This two-pronged approached was used to explore many possible low-energy configurations without prebiasing the result. In both sets of simulations we use the cell multipole method to account for the nonbonded interactions, and we mimic the effect of the chloroform solvent by setting the dielectric constant to 4.8. In addition, to save computational time, the amylose backbone of the polymer was fixed and only the side chains were permitted to move freely. Specifically, for the quenched annealing process, we performed an initial equilibration MD run for 100 ps at 500 K to allow the molecules to overcome energy barriers to fully explore the potential energy hypersurface. This is followed by a production run of 200 ps, where a frame was saved every 10000 steps, thus generating 20 frames, each of which was subsequently minimized to a convergence criterion of 0.1 kcal mol-1 Å-1. The subsequent minimization process removes any unphysical geometries and finds the nearest minimum-energy structure. To ensure we make efficient use of the hightemperature MD part of the annealing protocol, we applied distance restraints that hold two pairs of atomic distances to be less than 5 Å, which is consistent with experimental NMR data. The restraint was applied to the 5H-cH and 3H-bH distances for both enantiomers. The restraint consists of an extra energy term added to Discover energy calculations for the specified set of protons using a flat-bottomed function, as described below:

E(V e V0) ) (scale)k(V - V0)2 E(V0 < V < V1) ) 0.0 E(V g V1) ) (scale)k(V - V1)2 The scale was set to 1, the force constant k was set to 100, and the maximum derivative was set to 1000. V represents the actual distance, and V0 and V1 represent the distance range of 1 and 5 Å, respectively, for the simulation. No distances other than those of the 5H-cH and 3H-bH protons were restrained. These restraints were then removed for the energy minimization part of the annealing process. They were not used for the subsequent MD studies. MD Studies. After the 20 candidate frames for minimized structures were obtained, 3 were selected for further study with MD. The highest and lowest energy structures of the 20 frames were chosen, as well as a structure of intermediate energy, to explore whether the choice of the initial energy-minimized structure affected the results. The simulations were performed for 2 ns at 298 K, which is the same temperature used in the

1192 J. Phys. Chem. B, Vol. 111, No. 5, 2007

Ye et al.

NMR experiments, using the NVT ensemble (constant number, volume, and temperature). All of the atoms in the system were allowed to move freely. Results and Discussion NMR Studies. Figure 1A shows the 1H NMR spectrum of the analyte O-tert-butyltyrosine allyl ester in CDCl3 in the presence of ethanesulfonic acid. The structure of the analyte is shown at the top, and the spectral assignments are shown with reference to the labels on the structure. Chemical shift assignments of O-tert-butyltyrosine allyl ester were made with the assisance of 1D NMR Expert (a product of ACD Labs). As is shown in Figure 1A, the 1H resonances of the analyte are narrow, as one would expect for a small molecule in solution. Figure 1D shows the spectrum for ADMPC in CDCl3 solution in the presence of ethanesulfonic acid. The chemical shift assignments are provided with reference to the structure of the monomer unit of ADMPC, shown at the bottom of the figure. The chemical shift assignments were first reported by Yamamoto et al.10 and have been further verified by our own 2D COSY experiment (data not shown here). The ADMPC 1H resonances are much broader because of its slower molecular tumbling rate compared with that of small molecules such as the analyte. Parts B and D of Figure 1 show the NMR spectra for the D- and L-enantiomers, respectively, in the presence of ADMPC and ethanesulfonic acid. The sharp 1H NMR peaks of the Denantiomer (cf. Figure 1B) riding on the top of ADMPC broad resonances are slightly broadened, but they still show the spectral characteristics of a small molecule. In other words, most D-enantiomers of the analyte are in the unbound state, instead of in a complex form with ADMPC. However, for the L-enantiomer, the 1H NMR resonances (cf. Figure 1C) are greatly broadened so that almost no sharp peaks of the L-enantiomer can be distinguished from the broad ADMPC resonance. The resonance broadening of the L-enantiomers in the presence of ADMPC and ethanesulfonic acid can be attributed to their strong interactions with the polymeric material ADMPC. The formation of a complex between ADMPC and the L-enantiomer slows the tumbling rate of the small esters, resulting in extreme NMR line broadening. In the absence of ethanesulfonic acid, there was no discernible broadening in the NMR spectrum of either enantiomer with ADMPC (not shown here). These results are consistent with chromatography, which showed that acid was necessary to achieve high enantioselectivity and also that the L-enantiomer increased significantly in retention time while the D-enantiomer changed little in retention time.29 Figure 1 also reveals the upfield shifts (0.05-0.08 ppm) of the 1H resonance frequencies of 3H and especially 5H of ADMPC glucose in the presence of the enantiomer. The fact that no significant shifts for 2H and 4H of ADMPC glucose are observed demonstrates that the analyte approaches ADMPC on the 3H and 5H side of glucose to form a complex. The shifts are larger for the L-enantiomer than the D-enantiomer, implying a stronger average interaction for the L-enantiomer. It is wellknown that resonance frequencies of protons that are situated above the double bond system are shifted upfield, which is often called the magnetic anisotropic effect of the double bond system.36 Therefore, the phenyl or allyl groups of the analyte could cause these upfield shifts; specific information about which protons of the analyte are in proximity to the 3H and 5H protons of glucose can be obtained with 2D NOESY NMR spectroscopy. 2D NOESY spectra, collected and presented at the exact same conditions for the D- and L-enantiomers in the presence of

Figure 2. 2D NOESY NMR spectra of ADMPC and ethanesulfonic acid mixed with the L-enantiomer (green) and D-enantiomer (red).

ADMPC, are shown in Figure 2 in an overlaid mode. The negative NOE enhancement, which is consistent with the slowed molecular tumbling of the complexes, was observed for some internuclear pairs between resonances of the analyte and ADMPC. For a polymer the size of ADMPC, a NOESY crosspeak is expected for internuclear dipole-dipole interactions within 5 Å of one another due to the r-6 dependence on the internuclear distance.37,38 The interpretation of the NOE measurement can be complicated by other nuclear spin interactions, such as spin diffusion, which may also contribute to the crosspeak volume. In this study, experimental conditions such as concentrations, temperatures, and mixing times are maintained to be exactly the same for D- and L-enantiomer mixtures. Therefore, we can assume that the differences observed in NOESY cross-peaks represent differences in the interactions of the D- and L-enantiomers with ADMPC. The NOESY cross-peaks reveal that the especially large upfield shift discussed earlier for 5H of the ADMPC sugar ring in the 1H spectra is due to close proximity to the phenyl ring of

Chiral Discrimination by ADMPC

J. Phys. Chem. B, Vol. 111, No. 5, 2007 1193

TABLE 1: Binding Energy and Total Energy Results for Minimization Frames from Quenched Annealing Studiesa L-enantiomer

total energy

binding energy

D-enantiomer

frame no.

frame no.

total energy

binding energy

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 ave

-1255.3 -1250.6 -1249.8 -1246.5 -1248.4 -1247.6 -1248.9 -1247.0 -1247.5 -1248.6 -1250.8 -1251.6 -1251.8 -1254.5 -1257.5 -1247.2 -1244.7 -1245.9 -1254.0 -1246.4 -1249.7

-46.3 -45.8 -45.5 -45.8 -44.7 -45.4 -45.1 -45.2 -45.8 -45.7 -45.1 -46.1 -47.3 -47.3 -47.1 -46.4 -43.1 -44.7 -45.3 -46.4 -45.7

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 ave

-1223.2 -1225.3 -1228.0 -1227.4 -1226.1 -1225.7 -1227.7 -1226.4 -1228.9 -1234.6 -1234.6 -1237.8 -1233.6 -1241.9 -1233.6 -1231.4 -1221.6 -1223.2 -1221.0 -1223.0 -1228.8

-36.3 -35.9 -39.2 -38.4 -40.1 -38.3 -38.9 -35.1 -39.6 -31.5 -31.5 -41.5 -39.8 -35.2 -37.1 -34.8 -35.8 -35.5 -33.9 -37.7 -36.8

a All energies are in units of kcal/mol. Those marked in italics were used as starting structures for the MD studies. A dielectric constant of 4.8 was used to simulate the solvent effect of chloroform.

the analyte. Specifically, NOESY cross-peaks are especially prominent for cH-5H for both enantiomers, as shown in Figure 2 for the cross-peak (upper left). As discussed later, the proximity of cH and 5H is used to set the initial placement of the analyte near ADMPC for the computations. The major differences in NOESY cross-peaks between the D- and L-enantiomers involve the phenyl protons cH and bH of the tyrosine moiety. For the D-enantiomer, intermolecular NOE cross-peaks between cH of the tyrosine phenyl ring and 5H of the AMDPC sugar ring, as well as bH and 3H of these same moieties, were observed, which clearly shows that the phenyl and glucose rings are in close proximity. For the L-enantiomer, more intermolecular NOE cross-peaks were observed for more internuclear pairs between its tyrosine phenyl protons and the glucose protons of ADMPC. These correlations are between the internuclear pairs bH-2H, bH-3H, bH-5H, cH-2H, cH3H, and cH-5H. Qualitatively, the greater intensity of the NOE cross-peaks between the phenyl protons and glucose ring protons implies that a stronger complex between the L-enantiomer and ADMPC exists than for the D-enantiomer. The greater number and intensity of the NOESY cross-peaks for the L-enantiomer are in agreement with the 1H spectra in showing that the L-enantiomer is positioned more closely, on average, to ADMPC than is the D-enantiomer. For the computational studies, the bH3H and cH-5H proximities, but no others, were used for the initial placement of the enantiomer in the ADMPC complex. Molecular Modeling Studies. Energy Minimization. Table 1 shows the binding energy and total energy for each of the 20 frames from the quenched annealing studies for both the protonated D- and L-enantiomers. The bottom entry in each column gives the average binding energy to the polymer, which is shown to be much more favorable for the L-enantiomer compared to the D-enantiomer. The energy difference is about 9 kcal mol-1, which is considerably larger than the spread in the binding energies among the 20 frames for a given enantiomer. The energy difference correlates with the NMR measurements just discussed, and it also correlates with HPLC measurements of the protonated enantiomers,29 which showed that the L-enantiomer elutes significantly later than the D-enantiomer.

Figure 3. Energy-minimized structures of complexes between each enantiomer and the chiral ADMPC polymer. The enantiomers perturb the structure of the polymer differently. The image shows the L-enantiomer (green) interacting with the ADMPC polymer (brown) and the D-enantiomer (red) interacting with the ADMPC polymer (blue). The D-enantiomer is from frame 6, and the L-enantiomer is from frame 19. The PDB files of these structures are provided as Supporting Information (D_6.pdb and L_19.pdb).

A comparison between the computed energy difference and the chromatographic selectivity is made in a later section after thermalized energies are computed using MD simulations. The minimized structures for the polymer complexes with the protonated L- and D-enantiomers are shown in Figure 3. The ADMPC structure upon binding each enantiomer is rendered with a different color to distinguish them. The ADMPC structure

1194 J. Phys. Chem. B, Vol. 111, No. 5, 2007

Figure 4. Close-up views from Figure 3 for each enantiomer interacting with the chiral ADMPC polymer. The distances listed in Table 2 are also shown for the (A) L-enantiomer and (B) D-enantiomer.

is shown to change slightly upon binding for each enantiomer to achieve minimum energy, which is expected. Figure 4 shows the same structures of the two complexes, but in close-up detail, allowing one to examine the hydrogen bonding that these minimized structures report between the protonated amine group and the electronegative carbonyl oxygen. In the COMPASS force field there are no explicit terms for hydrogen bonding. Hydrogen bonding is seen to arise as a result of the nonbonded interaction, which consists of the electrostatic interaction and vdW interactions. We use the convention that a hydrogen bond is formed when the distance between H and O is less than 3 Å and the angle among the donor atom, the hydrogen, and the acceptor atom is more than 90°. The amine group of the L-enantiomer has a strong coordination with the three carbamate oxygens in this minimumenergy structure. Such interactions are present for the Denantiomer, as one can see in Figure 4, but only one of these interactions exists between the protonated amino group and carbamate oxygen groups; i.e., only one is less than 3 Å. In addition, the L-enantiomer has a hydrogen bond between its ether linkage and the NH group on the ADMPC carbamate side chain. These stronger interactions are possible for the L-enantiomer because it is able to fit better into the ADMPC pocket by folding over itself to maximize the contacts. The D-enantiomer, on the other hand, remains relatively linear because the energetic cost of maximizing the amine-carbonyl interaction is greater. These stronger hydrogen-bonding interactions for the L-enantiomer are seen for all 20 of the minimized structures. These were not imposed by the restraints. Table 2 lists the intermolecular distances between the carbonyl group on the side chains and the protons of the amine groups, as well as the distances for

Ye et al. the carbamate-ether interactions. The average distances for these hydrogen-bonding interactions, shown at the bottom of the table, are much shorter for the L-enantiomer than for the D-enantiomer, and the differences are significantly smaller than the spread of the numbers in the 20 frames. The table also shows that not only are there fewer strong hydrogen bonds between the D-enantiomer and the carbamate side chains, but also these interactions are weaker; the average bond length of the single strong interaction is 2.75 Å, while for the L-enantiomer this is the upper limit of the amine-carbamate interaction. Figure 4 shows that the folded structure of the L-enantiomer brings its allyl protons close to its phenyl protons, while the D-enantiomer remains nearly linear. These structures are consistent with the NOESY data: using the labels denoted in Figure 1, the fH proton approaches the phenyl protons bH and cH to give the fH-bH and fH-cH NOESY peaks of Figure 2 for the L-enantiomer only. These proximities were not imposed by the initial restraints. The agreement between experiment and computation for the folded structure of the L-enantiomer is seen in all 20 frames, implying independence from the starting conditions and lending credence to the minimized structures. Molecular Dynamics Simulations. MD simulations mimic the thermal conditions of HPLC and NMR. These simulations enable calculation of intermolecular distances for comparison with NOESY cross-peaks. They also enable calculation of the enantioselectivity for comparison with HPLC data. In addition, the thermal effect allows molecules to move more freely and overcome energy barriers to more stable equilibrium structures. Compared to the minimized energies, the average energies from MD simulations are less subject to bias by improbable local minima. Three MD simulations were run for 2 ns at 298 K on the minimized structures corresponding to the italicized entries in Table 1 for each of the protonated enantiomers. The results are shown in Table 3, which lists the average binding energy of the three structures for both of the protonated isomers over the time scale of the MD simulation. The overall average of the three averages is shown in italics for each enantiomer to facilitate comparison. The individual average energy for each of the three MD simulations is uncorrelated with the initial, minimized binding energy, and the results are similar for each of the three runs, indicating that unusual local minima are not being probed. These results indicate that the L-enantiomer has a more favorable binding to ADMPC than the D-enantiomer. The energy differences are slightly smaller than those done with energy minimization, between 7 and 9 kcal/mol. This would be expected given that we are now including the thermal distributions which would work against the attractive forces. The difference in binding energy between the two enantiomers corresponds to a chromatographic selectivity factor of 12, which is comparable to the chromatographically observed value of 16, reported earlier,29 which necessarily includes both enthalpy and entropy. The MD simulations report on the positions of the protons after thermalization, which allows a check for consistency with the 2D NMR results. Figure 5 shows the pair distribution functions (PDFs) for the same proton pairs observed to couple in the 2D NOESY measurements. The graphs are arranged in the same order as in the 2D NMR spectrum of Figure 2 to allow for convenient comparison. The 5H-cH pair gave the most intense of the isolated NOESY peaks, showing a significant NOE for both enantiomers in the upper left of Figure 2, and Figure 5 for this pair agrees by showing that the PDF falls within 5 Å for both enantiomers. The computation thus places the aromatic group of the analyte under the sugar ring of the

Chiral Discrimination by ADMPC

J. Phys. Chem. B, Vol. 111, No. 5, 2007 1195

TABLE 2: Hydrogen Bond Distance between the Enantiomer and the ADMPC Polymer L-enantiomer

frame no. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 ave

ether ether amine amine amine O-carbamate D-enantiomer amine amine amine O-carbamate NH1-carbonyl NH2-carbonyl NH3-carbonyl NH frame no. NH1-carbonyl NH2-carbonyl NH3-carbonyl NH 2.04 2.10 2.09 2.04 2.05 2.07 2.09 2.11 2.00 2.16 2.03 2.13 2.06 2.05 2.05 2.13 2.09 2.00 2.01 2.03 2.07

2.07 2.15 2.11 2.09 2.14 2.14 2.12 2.15 2.04 2.17 2.10 2.15 2.10 2.22 2.16 2.15 2.13 2.06 2.02 2.09 2.12

2.56 2.68 2.78 2.89 3.14 2.81 2.84 2.60 2.99 2.48 2.84 2.47 2.70 2.46 2.47 2.59 2.87 3.06 2.88 3.10 2.76

2.52 2.32 2.48 2.34 2.58 2.30 2.85 2.70 2.88 2.63 2.55 2.41 2.58 2.15 2.54 2.82 2.91 2.46 2.87 2.85 2.59

TABLE 3: Average Binding Energies (kcal/mol) from Three Different MD Frames of 2 ns Each, for Protonated and Deprotonated L- and D-Enantiomersa frame no. frame 15 frame 17 frame 20 aVe l D frame 7 D frame 12 D frame 19 aVe d L L L

a

protonated

deprotonated

total electrostatic vdW

total electrostatic vdW

-37.39 -39.06 -36.43 -37.63 -29.03 -31.64 -30.2 -30.29

-7.25 -7.53 -5.37 -6.72 -2.07 -3.11 -3.68 -2.95

-30.14 -31.53 -31.07 -30.91 -26.96 -28.53 -26.51 -27.33

-32.67 -29.87 -31.97 -31.50 -33.65 -30.51 -28.11 -30.75

-0.35 -0.21 -0.27 -0.28 -0.41 -0.10 -0.24 -0.25

-32.32 -30.07 -31.70 -31.36 -33.24 -30.41 -27.86 -30.51

Averages of the three frames are also shown.

amylose polymer, in agreement with NOESY. The other isolated peaks in the NOESY spectrum also agree well with the PDF results: experimentally there is an NOE only for the Lenantiomer for the peaks 5H-bH, fH-cH, fH-bH, 2H-cH, hH-cH, and 3H-cH, while computationally the PDF results agree that only the L-enantiomer is within the 5 Å distance in each case. The significant intensities from the fH-cH, fHbH, and hH-cH pairs indicate that only the L-enantiomer is folded, and the MD simulation correctly predicts that only the L-enantiomer is folded under these conditions. For the NOE peaks that have overlap with a nearby large resonance, trends can be discerned and the PDF results show the same trends. Specifically, the distances for the protons of the L-enantiomer are closer than those of the D-enantiomer for all three cases: 2H-bH, hH-bH, and 3H-bH. In addition, hH-bH shows the least NOE intensity for the D-enantiomer, and the PDF agrees in showing that the distance for the D-enantiomer is more than 5 Å, which shows more detail about how the L-enantiomer is folded. Further, the NOE for the 3H-bH pair is greater than for the 2H-bH pair for the L-enantiomer, consistent with the distances in the PDF results being closer for the 3H-bH pair (4 Å) than for the 2H-bH pair (5 Å). This again places the aromatic ring of the analyte underneath the sugar ring of the amylose polymer. While computations today can be estimates at best, the pair distributions support the interpretation that these computed structures reasonably represent the actual structures of the complexes between ADMPC and each enantiomer.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 ave

2.44 2.67 2.44 2.34 2.48 2.78 2.56 2.78 2.39 3.54 3.54 2.14 2.46 3.03 3.06 3.02 2.64 2.84 2.65 3.26 2.75

4.71 4.37 4.38 3.72 4.33 3.76 4.87 3.96 4.56 5.91 5.91 2.54 2.57 4.96 3.76 3.52 3.86 3.92 4.39 4.60 4.23

6.60 5.70 6.10 4.63 6.06 5.29 6.97 4.90 5.78 7.75 7.10 5.94 7.43 7.18 5.77 5.68 4.99 5.69 6.40 6.00 6.10

5.74 6.72 5.41 5.78 4.92 3.75 5.82 4.02 5.50 5.64 5.64 3.95 5.88 4.81 3.40 5.39 3.82 3.54 3.81 4.88 4.92

Insight into the enantioselectivity can be gained from the average energies generated from the MD simulations and listed in Table 3. For the protonated enantiomers, the van der Waals interactions contribute almost as much as electrostatic interactions to the enantioselectivity: both contributions are about 3.5 kcal-1 mol-1 more favorable for the L-enantiomer than for the D-enantiomer. This raises the question of why acid switches on the selectivity, rather than merely augmenting the enantioselectivity. The role of acid was studied computationally by removing the proton from the analyte and running each of the three 2 ns MD simulations for these deprotonated enantiomers. The results are included in Table 3. Deprotonation is shown to make the electrostatic contribution practically zero for both enantiomers, which is not surprising. The van der Waals interactions change little for the L-enantiomer, but they become significantly more favorable for the D-enantiomer. The latter is attributed to the D-enantiomer now being able to optimize its van der Waals interactions rather than trading them off for electrostatic interactions. The total binding energy thus becomes virtually identical for the two enantiomers upon deprotonation, which agrees with the chromatographic result that there is little selectivity without acid. In addition to offering an explanation for why acid switches on selectivity, theses results also predict correctly that the retention time of the D-enantiomer changes little with addition of acid.29 Without the computation, one might have falsely concluded that neither the electrostatic nor the van der Waals binding interactions of the D-enantiomer were affected by acid. Instead, the computation reveals that both of these interactions change, but with offsetting effects. The computation provides unique insight, unavailable from experiment, to show why the retention time of the L-enantiomer is strongly affected by acidity while that for the D-enantiomer is not. Long run times were needed to gain such insight into enantioselectivity. Figure 6 illustrates this with plots of the thermal fluctuating binding energies from the MD simulations. This figure charts how the binding energies evolve over the 2000 ps run times within each of the three frames for the protonated and deprotonated cases of the L- and D-enantiomers. The plots show that the electrostatic interaction energies are readily distinguishable between the two protonated enantiomers within any of the three 2 ns frames, and also the electrostatic energy becomes distinctly smaller and indistinguishable upon

1196 J. Phys. Chem. B, Vol. 111, No. 5, 2007

Ye et al.

Figure 5. Pair distribution profiles over the 2000 ps of molecular dynamics simulation for the protonated L- and D-enantiomers at 298 K with chloroform. The panels are arranged to correspond to Figure 2 to facilitate comparison.

deprotonation for any of the 2 ns frames. Any 2 ns run time would thus have been sufficient for probing electrostatic interactions. The plots of the van der Waals interactions show that these differences in binding energy are barely discernible in any of the frames. Overall, the two plots, parts a and b of Figure 6, graphically illustrate that sufficient run times were used to support the conclusions that protonation imparts enantioselectivity and deprotonation removes the enantioselectivity. One of the advantages of MD over many techniques, both experimental and computational, is that it allows the study of molecular phenomena in atomic detail over a period of time. Thus, we can examine the migration of molecules along the polymer backbone or other effects. Parts a and b of Figure 7 show, for the L- and D-enantiomers, respectively, the distances

between each of the three protons of the analyte amine and the ADMPC carbamate oxygen as a function of time. Parts c and d of Figure 7 show the distances between the analyte ether and proton of the ADMPC carbamate as a function of time, again for each enantiomer. These used the molecular dynamics runs from frame 17 for the L-enantiomer and frame 12 for the D-enantiomer. The scales have been kept identical to facilitate comparison. It is clear from these four panels that the Lenantiomer stays tightly bound in the pocket over the 2 ns time period, with all of the distances remaining very similar to each other. The L-enantiomer seems to go through some type of twisting motion periodically; between 350 and 600 ps and 9200 and 10500 ps, when two of the hydrogen bonds break, after which they are re-established, as shown in Figure 7a. During these two periods, the ether-carbamate hydrogen bond also

Chiral Discrimination by ADMPC

J. Phys. Chem. B, Vol. 111, No. 5, 2007 1197

Figure 6. Electrostatic and van der Waals energy vs time during the three MD simulations for each enantiomer: (a) each enantiomer was protonated, and (b) each enantiomer had its proton removed prior to simulation.

Figure 7. Distance vs time for atom pairs of interest: (a) for the L-enantiomer, distances between amino protons of the analyte and the oxygen of the ADMPC carbamate group and (b) the same for the d-enantiomer; (c) for the L-enantiomer, distances between the ether of the analyte and the proton of the ADMPC carbamate group (black) and the distance between c5 of the analyte aromatic ring and H5 of the ADMPC glucose ring (blue) and d) the same for the D-enantiomer.

breaks, while there is a jump in the cH-5H distance, which corresponds to the aromatic ring of the analyte and the glucose ring of ADMPC. These groups also return to their positions. Also other features are noticeable in Figure 7a, including the twisting of the amine group in the pocket, and at about 1100 and 1600 ps there is a swap between two of the amine distances. For the D-enantiomer, the amine group is further away from the carbamate, with distances much longer than were shown in the energy-minimized structure of Figures 3 and 4. Parts c and d of Figure 7 show the distances between the cH proton and the H5 protons on the adjacent glucose ring. The L-enantiomer is much closer to the backbone than the Denantiomer, with the distances from the nearest neighbors being much closer. For the L-enantiomer, the cH-H5 distance shows periods of moving away from and then closer to the backbone,

between 400 and 600 ps and 900 and 1200 ps. Analysis of the trajectory files during these time periods shows that the molecule is following a fairly complex motion, combining a spiral rotation around the chain with torsional changes in the molecule. For the D-enantiomer, the ether-carbamate distance of Figure 7d shows a somewhat oscillatory motion, implying the molecule is free to rotate. Overall, the shorter distances between the L-enantiomer and the ADPMPC polymer correspond to the more favorable electrostatic and van der Waals interactions compared to those of the D-enantiomer. Conclusions The high enantioselectivity of the commercial polysaccharide amylose tris(3,5-dimethylphenylcarbamate) for O-tert-butylty-

1198 J. Phys. Chem. B, Vol. 111, No. 5, 2007 rosine allyl ester in the presence of ethanesulfonic acid in chloroform is explained through a combination of 1D and 2D proton NMR experiments and MM and MD simulations. The experiments and computations agree well. They show that the polysaccharide favors stronger electrostatic and van der Waals interactions with the L-enantiomer than with the D-enantiomer. The MD simulations explain that the retention time of the D-enantiomer is unchanged because the more favorable electrostatic interactions are offset by the less favorable van der Waals interactions. The system studied here was amenable to NMR and MD methods because of its enormous enantioselectivity, combined with its high sensitivity to acidity, which served as a consistency test. The binding site might be the same for other amino acid esters because their enantioselectivities were also shown to be greatly enhanced by acid.29 Acknowledgment. We thank Professors Yamamoto and Professor Okamato of Nagoya University for sharing their computed ADMPC structure with us. This work was supported in part by the National Science Foundation under Grant CHE0433779. Supporting Information Available: Three structures, as described in the text, the polysaccharide oligomer (ADMPC12mer.pdb), the D-enantiomer in an energy-minimized binding pocket (D_6.pdb), and the L-enantiomer in an energy-minimized binding pocket (L_19.pdb), and images comparing the minimized structures for the COMPASS force field vs density functional theory for both a monomer unit and the analyte (DFT_vs_COMPASS.jpg). This material is available free of charge via the Internet at http://pubs.acs.org. References and Notes (1) Shibata, T.; Okamoto, I.; Ishii, K. J. Liq. Chromatogr. 1986, 9, 313-340. (2) Steinmeier, H.; Zugenmaier, P. Carbohydr. Res. 1987, 164, 97105. (3) Vogt, U.; Zugenmaier, P. Ber. Bunsen-Ges. Phys. Chem. 1985, 89, 1217-1224. (4) Francotte, E.; Wolf, R. M.; Lohmann, D. J. J. Chromatogr. 1985, 347, 25-37. (5) Shibata, T.; Sei, T.; Nishimura, H.; Deguchi, K. Chromatographia 1987, 24, 552-554. (6) Danhelka, J.; Netopilik, M.; Bohdanecky, M. J. Polym. Sci., Part B: Polym. Phys. 1987, 25, 1801- 1815. (7) Yashima, E.; Yamada, M.; Okamoto, Y. Chem. Lett. 1994, 3, 579582. (8) Yashima, E.; Yamada, M.; Yamamoto, C.; Nakashima, M.; Okamoto, Y. Enantiomer 1997, 2, 225-240. (9) Yashima, E.; Yamamoto, C.; Okamoto, Y. J. Am. Chem. Soc. 1996, 118, 4036-4048. (10) Yamamoto, C.; Yashima, E.; Okamoto, Y. J. Am. Chem. Soc. 2002, 124 (42), 12583-12589. (11) Okamoto, Y.; Ohashi, T.; Kaida, Y.; Yashima, E. Chirality 1993, 5, 616-621.

Ye et al. (12) Francotte, E.; Wolf, R.; Lohmann, D.; Mueller, R. J. Chromatogr., A 1985, 347, 25-37. (13) O’Brien, T.; Crocker, L.; Thompson, R.; Thompson, K.; Toma, P. H.; Conlon, D. A.; Feibush, B.; Moeder, C.; Bicker, G. and Grinberg, N. Anal. Chem. 1997, 69, 1999-2007. (14) Yashima, E.; Yamada, M.; Kaida, Y.; Okamoto, Y. J. Chromatogr., A 1995, 694, 347-354. (15) Yamamoto, C.; Yamada, M.; Yashima, E.; Okamoto, Y. Polym. Prep. Jpn. 1995, 44, 3029-3030. (16) Wolf, R. M.; Francotte, E. and Lohmann, D. J. Chem. Soc., Perkin Trans. 2 1988, 6, 893-901. (17) Booth, T. D.; and Wainer, I. W. J. Chromatogr., A 1996, 737, 157169. (18) Booth, T. D. and Wainer, I. W. J. Chromatogr., A 1996, 741, 205211. (19) Booth, T. D.; Azzaoui, K. and Wainer, I. W. Anal. Chem. 1997, 69, 3879-3883. (20) Bednarek, E.; Bocian, W.; Poznanski, J.; Sitkowski, J.; SadlejSosnowska, N.; Kozerski, L. J. Chem. Soc., Perkin Trans. 2002, 2, 9991004. (21) Pirkle, W. H.; Murray, P. G.; Rausch, D. J.; McKenna, S. T. J. Org. Chem. 1996, 61, 4769-4774. (22) Skogsberg, U.; Handel, H.; Sanchez, D.; Albert, K. J. Chromatogr., A 2004, 1023, 215-223. (23) Hellriegel, C.; Skogsberg, U.; Albert, K.; Lammerhofer, M.; Maier, N. M. Lindner, W. J. Am. Chem. Soc. 2004, 126, 3809-3816. (24) Enomoto, N.; Furukawa, S.; Ogasawara, Y.; Akano, H.; Kawamura, Y.; Yashima, E.; Okamoto, Y. Anal. Chem. 1996, 68, 2798-2804. (25) Cori, G. T.; Cori, C. F. J. Biol. Chem. 1940, 135, 733-756. (26) Pfannemuller, B.; Burchard, W. Makromol. Chem. 1969, 121, 1-17. (27) Kitamura, S.; Yunokawa, H.; Mitsuie, S.; Kuge, T. Polym. J. 1982, 14, 93-99. (28) Ye, Y. K.; Lord, B. S.; Stringham, R. W. J. Chromatogr., A 2001, 945, 139-146. (29) Ye, Y. K.; Stringham, R. W.; Wirth, M. J. J. Chromatogr., A 2004, 1057, 75-82. (30) (a) Macura, S.; Huang, Y.; Suter, D.; Ernst, R. R. J. Magn. Reson. 1981, 43, 259-281. (b) States, D. J.; Haberkorn, R. A.; Ruben, D. J. J. Magn. Reson. 1982, 48, 286-292. (c) Marion, D.; Wu¨thrich, K. Biochem. Biophys. Res. Commun. 1983, 113, 967-974. (31) (a) Delaglio, F.; Grzesiek, S.; Vuister, G. W.; Zhu, G.; Pfeifer, J.; Bax, A. J. Biomol. NMR 1995, 6, 277-293. (b) Goddard, T. D.; Kneller, D. G. Sparky 3; University of California: San Francisco; www.cgl.ucsf.edu/ home/sparky. (32) MS Modeling, version 3.2; Accelrys, Inc.: San Diego, CA. (33) (a) Rigby, D.; Sun, H.; Eichinger, B. E. Polym. Int. 1997, 44, 311330. (b) Sun, H.; Fried, J. R.; Ren, P. Comput. Theor. Polym. Sci. 1998, 8, 229-246. (c) Classical simulation theory. MS Modeling Help Text, version 3.2; Accelrys, Inc.: San Diego, CA. (34) (a) Ren, Z.; Zeng, X.; Yang, X.; Ma, D.; Ling Hsu, S. Polymer 2005, 46 (26), 12337-12347. (b) Eichhorn, S. J.; Young, R. J.; Davies, G. R. Biomacromolecules 2005, 6 (1), 507-513. (c) Heddleson, R. A.; Perdon, A. A.; Marston, B.; Hite, H. E.; Sung, S. S. U.S. Patent 20030147993. (35) Zimmerman, S. C.; Wu, W.; Zeng, Z. J. Am. Chem. Soc. 1991; 113, 196-201. (36) Gunther, H. NMR Spectroscopy, 2nd ed.; John Wiley & Sons: New York, 1992; pp 82-83. (37) Ernst, R. R.; Bodenhausen, G.; Wokum A. Principles of Nuclear Magnetic Resonance in One and Two Dimension; Oxford Scientific Publications: Oxford, U.K., 1986. (38) Cavanagh, J.; Fairbrother, W. J.; Palmer, A. G., III; Skelton, N. J. Protein NMR Spectroscopy, Principles and Practice; Academic Press Inc.: New York, 1996.