Determination of Enantiomeric Excess in Samples of Chiral Molecules

Development, LLC, Spring House, Pennsylvania 19477, and BioTools Inc., 950 North Rand Road, ... need to scan the VCD spectrum in time to monitor the %...
0 downloads 0 Views 202KB Size
Anal. Chem. 2004, 76, 6956-6966

Determination of Enantiomeric Excess in Samples of Chiral Molecules Using Fourier Transform Vibrational Circular Dichroism Spectroscopy: Simulation of Real-Time Reaction Monitoring Changning Guo,† Rekha D. Shah,‡ Rina K. Dukor,§ Xiaolin Cao,† Teresa B. Freedman,† and Laurence A. Nafie*,†,§

Department of Chemistry, Syracuse University, Syracuse, New York 13244, Johnson & Johnson Pharmaceutical Research & Development, LLC, Spring House, Pennsylvania 19477, and BioTools Inc., 950 North Rand Road, Suite 123, Wauconda, Illinois 60084

The first use of Fourier transform vibrational circular dichroism (FT-VCD) to follow changes in the percent enantiomeric excess (% EE) of chiral molecules in time using a flow cell sampling apparatus is reported. FT-VCD, as opposed to dispersive scanning VCD, eliminates the need to scan the VCD spectrum in time to monitor the % EE at more than one spectral location. The first use of partial least-squares chemometric analysis to determine % EE values from kinetic sets of VCD spectral data is also reported. These two advances have been used to monitor simultaneously changes in the fractional composition and the % EE of a mixture of two different chiral molecules. This simulates the progress of the chemical reaction from a chiral reactant to a chiral product where the % EE of both molecules can change with time. For the molecules studied, r-pinene, camphor, and borneol, the accuracy of following % EE changes for one species alone is ∼1%, while for simultaneously following % EE changes in two species is ∼2% for 10-20-min sampling periods at 4 cm-1 spectral resolution. This accuracy can be increased for the same collection times or maintained for shorter periods of collection by lowering the spectral resolution. These findings demonstrate the potential for VCD to be used for real-time monitoring of the composition and % EE of chemical reactions involving the synthesis chiral molecules. Vibrational circular dichroism (VCD)1-5 is the extension of electronic circular dichroism6,7 from the visible-ultraviolet region, where electronic transitions occur, into the mid-infrared (mid-IR) and near-infrared (near-IR) regions where vibrational transitions * Corresponding author. E-mail: [email protected]. † Syracuse University. ‡ Johnson & Johnson Pharmaceutical Research & Development. § BioTools Inc. (1) Nafie, L. A.; Freedman, T. B. Enantiomer 1998, 3, 283-297. (2) Dukor, R. K.; Nafie, L. A. In Encylopedia of Analytical Chemistry: Instrumentation and Applications; Meyers, R. A., Ed.; John Wiley and Sons: Chichester, U.K., 2000; pp 662-676. (3) Nafie, L. A.; Freedman, T. B. In Infrared and Raman Spectroscopy of Biological Materials; Gremlich, H.-U., Yan, B., Eds.; Marcel Dekker: New York, 2000; pp 15-54.

6956 Analytical Chemistry, Vol. 76, No. 23, December 1, 2004

occur. These vibrational transitions can be seen either as fundamental bands that are predominant in the mid-IR region or overtones and combination bands that occur in the near-IR.8-11 VCD is defined as the difference in the absorbance intensity of a vibrational transition in a molecule for left versus right circularly polarized radiation.12,13 Most VCD studies to date have involved vibrational transitions in the mid-IR region from 800 to 2000 cm-1 where commercially available Fourier transform VCD (FT-VCD) spectrometers operate,14 although the earliest studies of VCD,15-17 using scanning dispersive instruments, were confined to the region of hydrogen and deuterium stretching modes from 2000 to 4000 cm-1. In addition to VCD, there is a Raman analogue called Raman optical activity (ROA)18,19 that is defined as the difference in Raman intensity for right versus left circularly polarized radiation, either incident, scattered, or both.20-22 VCD and ROA together comprise the field known as vibrational optical activity (VOA).12,13 (4) Keiderling, T. A. In Circular Dichroism: Principles and Applications, 2nd ed.; Berova, N., Nakanishi, K., Woody, R. W., Eds.; Wiley-VCH: New York, 2000; pp 621-666. (5) Stephens, P. J.; Devlin, F. J. Chirality 2000, 12, 172-179. (6) Salvadori, P.; Ciardelli, F. In Optical Rotatory Dispersion and Circular Dichroism; Ciardelli, F., Salvadori, P., Eds.; Heyden and Sons: Ltd.: London, 1973; pp 3-24. (7) Berova, N., Nakanishi, K., Woody, R. W., Eds. Circular Dichroism: Principles and Applications, 2nd ed.; Wiley-VCH: New York, 2000. (8) Abbate, S.; Longhi, G.; Givens, J. W.; Boiadjiev, S. E.; Lightner, D. A.; Moscowitz, A. Appl. Spectrosc. 1996, 50, 642-643. (9) Abbate, S.; Longhi, G.; Boiadjiev, S.; Lightner, D. A.; Bertuccic, C.; Salvadori, P. Enantiomer 1998, 3, 337-347. (10) Abbate, S.; Longhi, G.; Kwon, K.; Moscowitz, A. J. Chem. Phys. 1998, 108, 50-62. (11) Abbate, S.; Longhi, G.; Santina, C. Chirality 2000, 12, 180-190. (12) Nafie, L. Appl. Spectrosc. 1996, 50, A14-A26. (13) Nafie, L. A. Annu. Rev. Phys. Chem. 1997, 48, 357-386. (14) Nafie, L. A.; Long, F.; Freedman, T. B.; Buijs, H.; Rilling, A.; Roy, J.-R.; Dukor, R. K. In Fourier Transform Spectroscopy: 11th International Conference; Haseth, J. A. d., Ed.; American Institute of Physics: Woodbury, NY, 1998; Vol. 430, pp 432-434. (15) Nafie, L. A.; Cheng, J. C.; Stephens, P. J. J. Am. Chem. Soc. 1975, 97, 3842. (16) Nafie, L. A.; Keiderling, T. A.; Stephens, P. J. J. Am. Chem. Soc. 1976, 98, 2715-2723. (17) Nafie, L. A.; Diem, M. Acc. Chem. Res. 1979, 12, 296-302. (18) Barron, L. D.; Bogaard, M. P.; Buckingham, A. D. J. Am. Chem. Soc. 1973, 95, 603-605. (19) Hug, W.; Kint, S.; Bailey, G. F.; Scherer, J. R. J. Am. Chem. Soc. 1975, 97, 5589-5590. 10.1021/ac049366a CCC: $27.50

© 2004 American Chemical Society Published on Web 11/02/2004

One of the simplest applications of VCD, or ROA, is to determine the enantiomeric excess (EE) of a sample consisting of some mixture of enantiomers of a chiral molecule. The percent EE (% EE) of a sample is defined as the excess amount of one enantiomer over the other relative to the total amount of both enantiomers. The expression for the % EE of enantiomer A relative to that of enantiomer B is given by

% EE(A) ) (NA - NB) × 100%/(NA + NB)

(1)

where NA is some measure of the moles of enantiomer A and NB the corresponding quantity for enantiomer B. Thus, the % EE(A) for an optically pure sample of enantiomer A is 100%, while the value for the racemic mixture is 0% and that for an optically pure sample of enantiomer B is -100%. The motivation to use VCD or ROA for the determination of % EE derives from the multiplicity of spectral bands that can be measured simultaneously in the spectrum. These bands serve not only as source of VOA intensity, but they also carry structural specificity of the chiral molecule being measured, even permitting the simultaneous determinations of % EE of more than one chiral species in a given sample. By contrast, optical rotation, for example, provides only a single datum per sample, regardless of the number of species present. The most accurate measure of % EE currently available, for measurements on the order of ∼1 h in duration, is chiral chromatography. This is achieved by separation of the two enantiomers from one another and measuring peak areas from a detector monitoring the output of the chromatographic column. The sensitivity of chiral chromatography can be extended by using either optical rotation or CD to detect the output of the column.23 It is not uncommon, however, that the enantiomers of a chiral molecule are not easily or completely separated from one another on a column, if at all. Furthermore, chromatography measurements cannot easily be carried out in real time to monitor the course of a reaction of chiral molecules. Chromatography often can take a long time compared to direct spectroscopic measurements and, in particular, the case of VCD considered here.23 The magnitude of the VCD spectrum of enantiomer A, or the magnitude of any other form of optical activity, scales directly and linearly with the % EE of enantiomer A. Specifically, the VCD has its maximum value for a sample of 100% EE, is zero across the entire spectrum for the racemic mixture, and is the negative of its maximum value for a sample with -100% EE (a pure sample of the opposite enantiomer). One of the earliest publications of EE determinations using VCD was published from our laboratory just over 10 years ago.24-26 These were based on measurements using a dispersive scanning (20) Barron, L. D.; Blanch, E. W.; Bell, A. F.; Syme, C. D.; Day, L. A.; Hecht, L. In 219th ACS National Meeting Book of Abstracts; American Chemical Society: San Francisco, 2000. (21) Barron, L. D.; Hecht, L. In Circular Dichroism: Principles and Applications, 2nd ed.; Berova, N., Nakanishi, K., Woody, R. W., Eds.; Wiley-VCH: New York, 2000; pp 667-701. (22) Nafie, L. A. In Encyclopedia of Spectroscopy and Spectrometery; Lindon, J., Tranter, G., Holmes, J., Eds.; Academic Press Ltd.: London, 1999; pp 19761985. (23) Shah, R. D.; Nafie, L. A. Curr. Opin. Drug Discuss. Dev. 2001, 4, 764-775. (24) Spencer, K. M.; Cianciosi, S. J.; Baldwin, J. E.; Freedman, T. B.; Nafie, L. A. Appl. Spectrosc. 1990, 44, 235-238. (25) Cianciosi, S. J.; Ragunathan, N.; Freedman, T. B.; Nafie, L. A.; Baldwin, J. E. J. Am. Chem. Soc. 1990, 112, 8204-8206.

VCD spectrometer in the carbon-hydrogen or carbon-deuterium stretching region. They were measurements of a sample not changing in time across a band of frequencies based on first recording the VCD spectrum of the pure enantiomer and then measuring the VCD of a sample of trans-dideuteriocyclopropane that had undergone thermal racemization as well as trans to cis geometrical isomerization. VCD intensities were analyzed by determining the area of the VCD spectrum relative to a predetermined baseline. IR intensities alone were used determine the degree of geometrical isomerization, since geometrical isomers have different IR spectra, while the magnitude and sign of the VCD intensities could be used to determine the % EE of the chiral (trans) species. Each of these measurements required a great deal of time for even one relatively narrow spectral region; there was no opportunity to follow the course of this reaction in time, say, for example, while thermal degradation was occurring. The first multiplex measurements of VCD, using FT methodology, were developed in our laboratory ∼25 years ago.27-30 The first multiplex measurements of ROA, which appeared soon thereafter, were developed in the laboratory of Hug.31,32 The first measurements of % EE using multichannel (simultaneous multispectral measurements) detection of vibrational optical activity were carried out with an ROA spectrometer equipped with a diode array or charge-coupled device detector. 33-35 Subsequently, the equivalent VCD multiplex measurements of % EE were carried out with FT-VCD instruments.14,36,37 More recently, Polavarapu compared the relative merits of optical rotation and VCD for the determination of % EE.38 Three chiral systems were investigated. In the majority of cases, % EE determination from optical rotation was more accurate than that from VCD. However, in cases where either the optical rotation is very small or more than one chiral species is present as a mixture, the % EE determination from VCD provides definite advantages with reasonable accuracy. Average areas of several VCD bands were used in their analysis. Because the baseline in VCD spectra is not completely flat at zero, using band areas that depend on the exact location of the baseline may contribute to an increase in the errors inherent in % EE determinations. All of the measurements of VCD and ROA to date have been carried out with samples fixed in conventional sample cells. To (26) Cianciosi, S. J.; Ragunathan, N.; Freedman, T. B.; Lewis, D. K.; Glenar, D. A.; Baldwin, J. E. J. Am. Chem. Soc. 1991, 113, 1864-1866. (27) Nafie, L. A.; Diem, M. Appl. Spectrosc. 1979, 33, 130-135. (28) Nafie, L. A.; Diem, M.; Vidrine, D. W. J. Am. Chem. Soc. 1979, 101, 496498. (29) Nafie, L. A.; Lipp, E. D.; Zimba, C. G. In Proceedings of the 1981 International Conference on Fourier Transform Infrared Spectroscopy; Sakal, J., Ed.; SPIE: Bellingham, WA, 1981; Vol. 289, pp 457-468. (30) Lipp, E. D.; Zimba, C. G.; Nafie, L. A. Chem. Phys. Lett. 1982, 90, 1-5. (31) Hug, W.; Kamatari, A.; Srinivasan, K.; Hansen, H. J.; Sliwka, H. R. Chem. Phys. Lett. 1980, 76, 469-474. (32) Hug, W. In Raman Spectroscopy; Lascombe, J., Ed.; Wiley: Chichester, U.K., 1982; pp 3-12. (33) Spencer, K. M.; Edmonds, R. B.; Ruah, R. D.; Carrabba, M. M. Anal. Chem. 1994, 66, 1269-1273. (34) Hecht, L.; Phillips, A. L.; Barron, L. D. J. Raman Spectrosc. 1995, 26, 727732. (35) Spencer, K. M.; Edmonds, R. B.; Rauh, R. D. Appl. Spectrosc. 1996, 50, 681-685. (36) Buijs, H.; Nafie, L. A. Book of Abstracts for Pittcon ′97 ;1997; Abstr. 1206. (37) Dukor, R. K.; Roy, J.-R.; Nafie, L. A. In Proceeding of the 12th International Conference on Fourier Transform Spectroscopy; Tasumi, M., Itoh, K., Eds.; Waseda University Press: Tokyo, 1999; pp 263-264. (38) Zhao, C.; Polavarapu, P. L. Appl. Spectrosc. 2001, 55, 913-918.

Analytical Chemistry, Vol. 76, No. 23, December 1, 2004

6957

develop the capability to follow changes in % EE with time, a flow cell apparatus in the sample chamber of our FT-VCD spectrometer can be employed. Flow cells have been incorporated into many branches of analytical spectroscopy where they are used routinely to monitor chemical change with time in a dynamic chemical system. This technique has many important practical applications, especially in quantitatively following chemical reactions. It should be emphasized, however, that use of a flow cell is limited to single point spectral measurements at a particular point in time if a spectrometer must be scanned to measure more than one spectral point. FT-VCD and multichannel ROA are currently the only techniques in all of optical activity where multiwavelength spectral regions can be measured simultaneously. This is a very significant advantage relative to optical rotation and electronic CD, the conventional measures of optical activity. In an attempt to improve upon the conventional method of determination of % EE using peak areas, for the reasons mentioned above concerning the uncertainty of the VCD baseline, a statistical method known as partial least squares (PLS) was implemented to analyze our % EE VCD data. One of the advantages of PLS analysis is that one can carry out “whole spectra” analysis. Instead of selecting individual peaks, a large spectral region is used and raw spectra can be employed for PLS analysis directly, without solvent subtraction and baseline correction. Compared with alternative analysis methods, such as principle component regression, which first decomposes the spectral matrix into a set of eigenvectors and scores and performs regression against the concentrations as a separate step, PLS actually uses the concentration information during the decomposition process. As a result, spectra containing higher constituent concentrations are weighted more heavily than those with low concentrations. The main goal of PLS is to get as much concentration information as possible into the first few loading vectors. Thus, the PLS method may be considered a nearly ideal tool for VCD quantitative analysis. In this paper, we take full advantage of the spectral multiplicity of FT-VCD to measure the % EE of chiral molecules in both onecomponent solutions and two-component mixtures. We report here the first use of a flow cell technique for VCD spectral measurements and the first application of chemometrics in general, and in particular PLS, to analyze time-dependent VCD % EE data. The flow cell-VCD method developed was based on the use of a simple sampling arrangement with inexpensive cells. The flow cell method permits changing the mole fraction and EE of sample mixtures by injecting prepared solutions of a chosen mole fraction with preselected % EE into the system during measurements. As a result, this study demonstrates the potential of VCD to monitor the composition and % EE of chemical reactions of chiral species using a flow cell. The determination of % EE in real chemical reactions changing with time will be reported in future publications. EXPERIMENTAL SECTION A dual-source, dual-polarization-modulated VCD spectrometer developed from a commercial FT-VCD spectrometer, the ChiralIR (ABB Bomem-BioTools, Quebec, PQ, Canada and BioTools, Inc., Wauconda, IL) was used to carry out IR and VCD measurements in the spectral range 800-2000 cm-1. All spectra were collected at room temperature with a resolution of 4 cm-1. A closed flow cell system employed for sample introduction inside the VCD 6958

Analytical Chemistry, Vol. 76, No. 23, December 1, 2004

spectrometer included a pump and an IR flow cell with a 100-µm path length and BaF2 windows (both from International Crystal Laboratories, Garfield, NJ) and a 40-mL solution reservoir. For sample transport, Teflon tubing with 1/8-in. outer diameter and 1/ -in. inner diameter was used. Samples of (1R)-(+)-R-pinene, 16 (1S)-(-)-R-pinene, (1S)-(-)-camphor, (1R)-(+)-camphor, [(1S)endo]-(-)-borneol, [(1R)-endo]-(+)-borneol, and CCl4 were purchased from Aldrich Chemical Co. and used without further purification. The pure chiral molecules used have EE values in the range of 98-99+%. In this paper, for the convenience of calculation and comparison, we assumed that they all had an optical purity of 100.00%. The general procedure for carrying out flow cell VCD measurements was as follows. The starting solution (8-10 mL) was placed in the reservoir beaker and pumped into the flow cell system. The VCD spectrometer was set to continuously measure 20-24 blocks, each a 10- or 20-min measurement. At the beginning of the even-numbered blocks (2nd, 4th, 6th, 8th...), 0.5-3 mL of a second, add-in solution was injected into the reservoir beaker, thereby changing the solution being pumped through the flow cell. The add-in and original solutions were allowed to mix and flow through the system for the remainder of the even-numbered block. The subsequent odd-numbered block was used to measure the IR and VCD spectra of the newly mixed solution. All the spectra of odd-numbered blocks were then used in the analysis. In each block, 10% of the time was used for the IR measurement and 90% of time was used for the VCD measurement. The PLS analyses were carried out with the PLSplus/IQ module in Grams/32 AI (6.00) (Galactic Industries, Inc., Salem, NH). A cross-validation method was employed for prediction in this work. In this method, for a data set with N spectra, N - 1 spectra were used as a training set to predict the spectrum that was left out. All the spectra used in PLS analysis were raw data without any processing such as solvent subtraction or baseline correction. Statistical accuracy of the calibration models is described by the correlation coefficient (R2) and the root-meansquare error of cross validation (RMSECV) using the leave-oneout procedure. The optimum number of PLS factors for each component was determined by use of the predictive residual error sum of squares calculation. For concentrations CA and CB of the A and B enantiomers of a chiral molecule in a solution, the EE of the sample is (CA CB)/(CA + CB) or (CB - CA)/(CA + CB), depending on the enantiomer is selected for reference. The sum CA + CB can be obtained from the IR measurement, and the difference CA - CB can be obtained from the VCD measurement. Thus, the enantiomeric excess can be obtained by the combination of IR and VCD measurements. RESULTS AND DISCUSSION One-Component Solutions. A one-component solution, which includes only the enantiomers of a single chiral molecule, is the simplest case. Here, R-pinene and camphor were selected as molecules to test the accuracy of the VCD technique. Figure 1 shows the IR and VCD spectra of 3.146 M (1R)-(+)-R-pinene in CCl4 solution (50.00 vol %). For the flow cell experiment, the original solution was 8.000 mL of 3.146 M (1R)-(+)-R-pinene solution. The add-in solution was 3.146 M (1S)-(-)-R-pinene CCl4 solution, with consecutive add-in volumes of 0.500, 0.500, 0.500,

Figure 2. VCD spectra of R-pinene flow cell experiment. From bottom to top, the % EEs of (1R)-(+)-R-pinene are 100.00, 88.24, 77.78, 68.42, 60.00, 52.38, 45.45, 33.33, 23.08, 14.29, 6.67, and 0.00. Figure 1. IR and VCD spectra of (1R)-(+)-R-pinene (3.14 M CC14 solution).

0.500, 0.500, 0.500, 1.000, 1.000, 1.000, 1.000, and 1.000 mL at the even-numbered blocks. Each block was a 10-min measurement (1 min for the IR absorbance measurement and 9 min for the VCD measurement). The VCD spectra of the flow cell experiment are shown in Figure 2. During this experiment, the IR spectra did not change, while the decrease in enantiomeric excess was monitored by the changes in the VCD spectra. The VCD spectra collected were analyzed by PLS for the range 900-1350 cm-1. The results of the PLS analysis for this flow cell experiment are listed in Table 1 and plotted in Figure 3. The number of loading eigenvectors (factors) was 1, and the R2 was 0.9984. The RMSECV of % EE prediction was 1.21%. In this flow cell experiment, only one measurement was recorded at the maximum % EE value (100.00%). During the leaveone-out cross validation, when this sample was left out for prediction and other samples were used to build the calibration model, the sample value was out of the range of the calibration set and the prediction result should not be trusted. The same thing occurred for the sample with the minimum % EE value (0.00%). Thus, when we calculated RMSECV, the predicted values for these two samples were taken out from the data set as outliers. The same procedures were employed for all the RMSECV calculations in this paper. The second one-component system studied was camphor. For this flow cell experiment, the original solution was 10.000 mL of a 1.000 M (1S)-(-)-camphor solution in CCl4. The add-in solution was 1.000 M (1R)-(+)-camphor CCl4, with consecutive add-in volumes of 0.500, 0.500, 1.000, 1.000, 1.000, 1.000, 1.000, 1.000, 1.000, and 2.000 mL at even-numbered blocks. Each block was a 20-min measurement (2 min for the IR measurement and 18 min for the VCD measurement). The magnitude of the VCD bands in

Table 1. % EE PLS Analysis Results of the r-Pinene Flow Cell Experiment actual % EE

predicted % EE

variance (%)

88.24 77.78 68.42 60.00 52.38 45.45 33.33 23.08 14.29 6.67 RMSECV

87.19 75.89 69.74 61.33 52.16 45.32 32.82 24.39 15.41 4.92 1.21

1.05 1.89 -1.32 -1.33 0.22 0.13 0.51 -1.31 -1.12 1.75

camphor are a factor of ∼2 smaller than that of R-pinene, and hence, longer scan times were selected to achieve the desired signal-to-noise levels. The spectral region selected for the PLS analysis was 1220-1350 cm-1. The number of loading eigenvectors (factors) was 1, and the correlation coefficient (R2) was 0.9991. The PLS analysis results for this flow cell experiment are presented in Table 2. The RMSECV of % EE prediction was 0.88%. Compared with the R-pinene flow cell experiment, the longer collection time in the camphor flow cell experiment (20 vs 10 min) resulted in a better prediction even though camphor has a weaker overall VCD spectrum than that of R-pinene. Two-Component Solutions. A mixture of two chiral molecules, camphor and borneol, was investigated to simulate a chemical reaction. Camphor and borneol differ only in one functional group, carbonyl for camphor and hydroxyl for borneol. The reduction of a ketone to an alcohol was thereby simulated in this study as shown in Figure 4. A comparison of the IR and VCD spectra of 1.000 M (1S)-(-)-camphor and 1.000 M [(1S)-endo](-)-borneol in CCl4 solution is presented in Figure 5. For all the two-component solutions in this paper, the VCD measurements Analytical Chemistry, Vol. 76, No. 23, December 1, 2004

6959

Figure 3. Results of % EE PLS analysis for (1R)-(+)-R-pinene in a VCD flow cell experiment. Table 2. % EE PLS Analysis Results of Camphor Flow Cell Experiment actual % EE

predicted % EE

variance (%)

90.48 81.82 66.67 53.85 42.86 33.33 25.00 17.65 11.11 RMSECV

89.54 82.09 66.21 53.05 43.92 34.04 23.80 18.80 11.95 0.88

0.94 -0.27 0.46 0.80 -1.06 -0.71 1.20 -1.15 -0.84

Figure 4. Simulated chemical reaction of (1S)-(-)-camphor to [(1S)endo]-(-)-borneol.

were set up as a 20-min collection (2 min for IR absorbance and 18 min for VCD). Model Building. Sixteen CCl4 solutions with various (1S)(-)-camphor and [(1S)-endo]-(-)-borneol compositions were prepared and measured by the flow cell sampling method and comprised a calibration set for both IR absorbance and VCD models. The details of the compositions of the calibration samples are presented in Table 3 and Figure 6. The concentrations of (1S)(-)-camphor and [(1S)-endo]-(-)-borneol in the calibration samples varied throughout the expected range of subsequent samples. The calibration samples covered the lower left-hand corner of the graphic square in Figure 6. No samples were expected to fall into the upper right-hand corner mainly because of solubility limitations. The optical purities of camphor or borneol in the 16 calibration samples were all 100%. 6960 Analytical Chemistry, Vol. 76, No. 23, December 1, 2004

Figure 5. Comparison of IR and VCD spectra of (1S)-(-)-camphor and [(1S)-endo]-(-)-borneol (both 1.000 M CC14 solution).

The spectral regions selected for both the IR absorbance and VCD PLS analyses were 1350-1070 and 1035-900 cm-1. Five samples (samples 3, 4-6, and 13) in the calibration set had a large peak in the 1070-1035-cm-1 region with IR absorbance intensity over 1.0, which was due to borneol. When an IR absorption intensity in a spectral region is greater than 1, the single-to-noise ratio for the corresponding VCD will decrease dramatically and in some cases the VCD in that region cannot be measured. Thus, the 1070-1035-cm-1 region was not used for the chemometric analysis. The cross-validation results for IR absorbance and VCD analysis are presented in Table 3. For the IR absorbance model, the optimal numbers of factors were 5 for camphor and 4 for borneol. The correlation coefficients were 0.9997 for camphor and 0.9996 for borneol. The RMSECVs of concentration prediction were 0.0050 M for camphor and 0.0076 M for borneol. For the parallel VCD model, the optimal numbers of factors were 5 for (1S)-(-)-camphor and 3 for [(1S)-endo]-(-)-borneol. The correlation coefficients were 0.9985 for camphor and 0.9976 for borneol. The RMSECV of concentration prediction were 0.0126 M for camphor and 0.0158 M for borneol. For the same calibration set, the IR absorbance model had a better accuracy due to the IR absorbance spectra having much better signal-to-noise ratios than the VCD spectra. Combining the cross-validation results of the IR absorbance and VCD models, “predicted” EE values of the chiral components in the calibration samples were calculated and shown in Table 3. The predicted EE values of sample 7 have much larger errors than other samples. Considering sample 7 had the lowest concentrations of all the calibration samples except the pure solvent (sample 1), four sources may contribute to the unusual errors. First, IR absorbance and VCD spectra of a same sample

Table 3. Calibration Set for Two-Component Solutions and the Cross-Validation Results of the PLS Modelsa camphor sample number

actual value

IR prediction

VCD prediction

1 2 3 4 5 6 7b 8 9 10 11 12 13 14 15 16 RMSECV

0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.2000 0.3000 0.4000 0.4000 0.5000 0.6000 0.6000 0.8000 1.0000 1.2000

-0.0106 0.0017 -0.0041 0.0000 -0.0016 0.0073 0.1950 0.3031 0.3999 0.4031 0.5051 0.6040 0.5870 0.8071 1.0019 1.1870 0.0050

0.0100 0.0041 0.0033 0.0190 0.0036 -0.0353 0.2146 0.2958 0.3961 0.3874 0.5175 0.5873 0.5771 0.7915 1.0135 1.2072 0.0126

borneol predicted % EE

110.08 97.59 99.05 96.11 102.47 97.23 98.31 98.07 101.15 101.70 2.33

actual value

IR model prediction

VCD prediction

0.0000 0.4000 0.6000 0.8000 1.0000 1.2000 0.2000 0.3000 0.0000 0.4000 0.5000 0.0000 0.6000 0.0000 0.0000 0.0000

0.0072 0.3815 0.5991 0.8020 1.0090 1.1914 0.2147 0.3061 -0.0003 0.3984 0.5059 -0.0014 0.5945 -0.0014 -0.0006 -0.0005 0.0076

0.0312 0.3887 0.5990 0.7809 0.9746 1.2122 0.1812 0.2921 0.0187 0.3910 0.5047 0.0260 0.6236 0.0038 0.0011 -0.0338 0.0158

predicted % EE 101.90 99.98 97.36 96.59 101.74 84.42 95.43 98.14 99.76 104.89

2.97

a Samples 1, 6, and 16 were out of PLS model prediction ranges in cross validation and not used for RMSECV calculation. b Outlier for % EE prediction.

Figure 6. Concentration of [(1S)-endo]-(-)-borneol versus the concentration of (1S)-(-)-camphor for the calibration samples.

were not measured at the same time. Thus, the efficiency of the sample mixing in the flow cell procedure may affect the measurements, especially in low-concentration situations. Second, EE values were calculated from the ratio of two predicted values. The errors of quotients tend to become larger when the values of the numerator and denominator become smaller. Third, for lower concentrations, the signal-to-noise ratio of a VCD spectrum becomes worse. Finally, various kinds of random errors may have occurred when the sample 7 was measured. Excluding sample 7 as an outlier, the root-mean-squared errors of prediction (RMSEP) were 2.33% for camphor and 2.97% for borneol. Reaction Simulation 1. Two flow cell experiments were designed to simulate a chemical reaction from (1S)-(-)-camphor

to [(1S)-endo]-(-)-borneol in which the reaction proceeds from reactant to product at 100% EE for both molecules. For the first experiment, the original solution was 10.000 mL of 1.000 M 100.00% EE (1S)-(-)-camphor CCl4 solution. The add-in was 1.000 M 100.00% EE [(1S)-endo]-(-)-borneol CCl4 solution, with consecutive add-in volumes of 1.000, 1.000, 1.000, 2.000, 2.000, and 3.000 mL. For the second flow cell experiment, the original solution and add-in solution was reversed and same procedure was executed. The results from these two experiments were combined to obtain a set of data to simulate a chemical reaction from camphor to borneol or vice versa. During the experiment, the sum of the concentrations of (1S)(-)-camphor and [(1S)-endo]-(-)-borneol did not change. The only changes were the mole fractions of the (1S)-(-)-camphor and [(1S)-endo]-(-)-borneol. Because both the reactant and product are 100.00% EE during the experiment, either the IR or the VCD can be used to monitor this “reaction” separately. Figure 7 shows the IR absorbance and VCD spectra for the flow cell experiment. The concentrations of the reactant and product during the reaction were predicted by the PLS models obtained above and are presented in Table 4 and Figure 8. For the IR absorbance analysis, the RMSEP were 0.0046 M for camphor and 0.0070 M for borneol. For the VCD analysis, the RMSEP were 0.0074 M for camphor and 0.0101 M for borneol. Enantiomeric excesses of the reactant and product were calculated after the IR and VCD prediction, and the results are shown in Table 4. The RMSEP were 1.42% for camphor and 2.57% for borneol. Reaction Simulation 2. Experiments were next carried out in which camphor and borneol were mixed at different mole fractions and different enantiomeric excesses. These measurements were carried out to determine whether the VCD flow cell technique can be used to monitor the conversion of a reactant to a product when both the mole fraction and the % EE of both species are changing with time. The original solution was 9.000 mL of 0.600 M at 100.00% EE [(1S)-endo]-(-)-borneol and 0.400 Analytical Chemistry, Vol. 76, No. 23, December 1, 2004

6961

Figure 7. IR and VCD spectra for the chemical reaction simulation experiment. The mole fraction of (1S)-(-)-camphor relative to [(1S)-endo](-)-borneol, from spectra labeled 1-13 in Table 3: 100.00, 90.91, 83.33, 76.92, 66.67, 58.82, 50.00, 41.18%, 33.33, 23.08, 16.67, 9.09, and and 0.00%. Table 4. PLS Analysis Results of Camphor-Borneol Reaction Simulation Experiment I camphor

borneol

sample number

actual value (M)

IR prediction (M)

VCD prediction (M)

predicted % EE

actual value (M)

IR prediction (M)

VCD prediction (M)

predicted % EE

1 2 3 4 5 6 7 8 9 10 11 12 13 RMSEP

1.0000 0.9091 0.8333 0.7692 0.6667 0.5882 0.5000 0.4118 0.3333 0.2308 0.1667 0.0909 0.0000

0.9916 0.9076 0.8350 0.7729 0.6684 0.5951 0.5017 0.4177 0.3411 0.2349 0.1676 0.0926 0.0040 0.0046

0.9978 0.9063 0.8496 0.7720 0.6601 0.5794 0.4975 0.4277 0.3387 0.2315 0.1649 0.0919 0.0013 0.0074

100.63 99.86 101.75 99.89 98.77 97.35 99.15 102.41 99.30 98.57 98.44 99.29

0.0000 0.0909 0.1667 0.2308 0.3333 0.4118 0.5000 0.5882 0.6667 0.7692 0.8333 0.9091 1.0000

-0.0092 0.0873 0.1647 0.2328 0.3414 0.4177 0.5002 0.5939 0.6775 0.7799 0.8423 0.9118 0.9914 0.0070

-0.0055 0.0908 0.1585 0.2245 0.3497 0.4245 0.5193 0.5768 0.6682 0.7832 0.8365 0.9122 1.0059 0.0101

104.04 96.27 96.42 102.44 101.62 103.82 97.11 98.63 100.43 99.31 100.04 101.46 2.57

M at 100.00% EE (1S)-(-)-camphor mixture in CCl4 solution. The add-ins were 0.400 M at 0.00% EE [(1S)-endo]-(-)-borneol and 0.600 M at 0.00% EE (1S)-(-)-camphor CCl4 solution. The volume of add-in was 1.000 mL for each injection. In total, 10 samples were prepared and measured by this flow cell sampling method and their detailed composition data are presented in Table 5. For this system, to obtain the % EE of camphor and borneol at each step, both IR and VCD spectra are required, since during the experiment, both the mole fractions and the enantiomeric excesses of the two components change. The IR alone can be used to monitor the changes in mole fraction and the VCD normalized by the IR for concentration can be used to monitor simultaneously the changes in the % EE of both chiral molecules present. Due to the similarity of camphor and borneol and the small change of mole fraction, the IR and VCD spectra during the flow cell 6962 Analytical Chemistry, Vol. 76, No. 23, December 1, 2004

1.42

experiment do not change dramatically (see Figure 5). However, according to the experimental design, the % EE of camphor and borneol will have large changes during the experiment. Because the predicted % EE will come from both the IR and VCD analyses, the prediction of % EE during compositional change is quite challenging. For the flow cell experiment, the % EE of both camphor and borneol can be obtained simultaneously from the apparent predicted VCD intensities of camphor and borneol, influenced by both concentration and the % EE of each species, divided by the actual predicted IR intensity for camphor and borneol, influenced only by the concentration of each species. The results are listed in Table 5 and Figure 9. The RMSEP of EE prediction were 2.73% for camphor and 2.28% for borneol.

Figure 8. PLS analysis results of camphor-borneol reaction simulation experiment I.

The results presented here demonstrate that FT-VCD can be used to monitor simultaneously the % EE of multiple chiral molecules as a function of time. Although the intrinsic signal-tonoise ratio of VCD is not as high as other traditional monitors of the kinetics of chiral molecules, there are a number of advantages of FT-VCD that arise from its information content and from the multiplex nature of its measurement process. FT-VCD spectra typically contain many bands representing different vibrational modes from all portions of the molecule. For VCD, there is no concern about the concept of chromophore since all molecules have “vibrational chromophores” representing all structural locations in the molecule. Since different molecules have different vibrational bands and different vibrational frequencies for these bands, it is often possible to identify individual peaks in a mixture that belong to particular molecules. In the case of monitoring the % EE of a single species, the multiplex advantage of FT-VCD means that each point in the spectrum, of which there may be hundreds or thousands across the spectrum, represents an independent measure of the % EE. Combining these data points using PLS, with higher weighting accorded to regions of higher

signal-to-noise ratio, the disadvantage of relative low signal-to-noise ratio at individual spectral locations is in large measure overcome. Accuracies in using VCD to follow the % EE of a single species over a set of samples of differing known % EE is shown to be in the range of 1-2% for the spectral resolution chosen, namely 4 cm-1. This accuracy, though not explicitly demonstrated in this paper, can be increased up to below 1% by using lower resolution and the same signal collection time. We are currently testing the lower limit of resolution for typical vibrational spectra in the midIR region, but preliminary results suggests that 8 or 16 cm-1 may be close to an optimal value from which we would expect to improve our % EE prediction accuracy. Lowering the resolution further to 32 or even 64 cm-1 erodes the spectral information present and, in the case of VCD, leads to cancellation of adjacent positive and negative VCD bands and, hence, loss of information. We expect the optimum resolution to be influenced by the nature of the VCD by the degree that lowering the resolution reduces spectral bands by cancellation of adjacent positive and negative VCD intensities. Analytical Chemistry, Vol. 76, No. 23, December 1, 2004

6963

Table 5. PLS Analysis Results of Camphor-Borneol Reaction Simulation Experiment II IR

VCD

sample number

actual value (M)

predicted value (M)

1 2 3 4 5 6 7 8 9 10 RMSEP

0.4000 0.4200 0.4364 0.4500 0.4615 0.4714 0.4800 0.4875 0.4941 0.5000

0.3966 0.4193 0.4349 0.4508 0.4628 0.4729 0.4820 0.4913 0.5013 0.5060 0.0026

1 2 3 4 5 6 7 8 9 10 RMSEP

0.6000 0.5800 0.5636 0.5500 0.5385 0.5286 0.5200 0.5125 0.5059 0.5000

0.6006 0.5809 0.5650 0.5490 0.5381 0.5266 0.5183 0.5112 0.5035 0.4982 0.0015

actual value (M) Camphor 0.4000 0.3600 0.3273 0.3000 0.2769 0.2571 0.2400 0.2250 0.2118 0.2000 Borneol 0.6000 0.5400 0.4909 0.4500 0.4154 0.3857 0.3600 0.3375 0.3176 0.3000

Lowering the resolution also permits us to do short-time VCD measurements with much better signal-to-noise ratios, which is very important for real world applications such as process control. Preliminary results suggest that decent accuracies can be achieved with 1-min VCD measurements with resolution of 8 or 16 cm-1. In contrast, for the same time of measurement, the VCD signal obtained with a resolution of 4 cm-1 is deeply immersed in the noise, thus preventing quantitative analysis at reasonable levels of accuracy. The tradeoffs between resolution, % EE accuracy, and collection time will be the subject of a future paper. In the more interesting case of simultaneously monitoring the % EE of two or more species, FT-VCD has a distinct advantage over all traditional single spectral point monitors of chirality. Given the distinctness of the IR and VCD spectra from different chemical species, chemometric methods, such as PLS, can follow the fractional composition of multiples species present in a sample using FT-IR spectroscopy. What is new in this paper is the demonstration that FT-VCD can extend the capabilities of traditional FTIR-PLS analysis to include simultaneous monitoring of the % EE of multiple species (in the present case two species). All that is needed is to recognize that the apparent fractional composition of each contributing species to the total VCD spectrum is its fractional composition, identical to that in the IR spectrum, times the % EE of that species. If each species present in the total FT-VCD spectrum is first normalized by its actual fractional composition obtained from the concurrently measured FTIR spectrum, the only remaining weighting factor is the % EE. To obtain the absolute values of the % EE of each species, the value of the measured VCD intensity of each species must be normalized by the magnitude of the VCD spectrum of that species for the value 100% EE. This in turn can be obtained from a single VCD measurement, for a particular value of the IR absorbance spectrum, of a sample of the species in question having a known 6964 Analytical Chemistry, Vol. 76, No. 23, December 1, 2004

% EE predicted value (M)

actual value

predicted value

0.3995 0.3735 0.3330 0.3143 0.2911 0.2712 0.2529 0.2522 0.2142 0.2086 0.0034

100.00 85.71 75.00 66.67 60.00 54.55 50.00 46.15 42.86 40.00

100.74 89.08 76.57 69.74 62.90 57.34 52.46 51.33 42.73 41.23 2.67

0.6129 0.5324 0.4910 0.4329 0.4036 0.3944 0.3376 0.3250 0.3036 0.2969 0.0126

100.00 93.10 87.10 81.82 77.14 72.97 69.23 65.85 62.79 60.00

102.03 91.66 86.90 78.84 75.00 74.90 65.13 63.57 60.29 59.59 2.28

value, typically 100% or a large % EE value for reasons of signalto-noise ratio. Given these preliminary calibrations, reaction kinetics involving simultaneous changes in mole fraction and % EE can be followed by simultaneous time-dependent measurements of FTIR and FT-VCD. The last condition is trivial to meet since all FT-VCD measurements are always accompanied by simultaneous FTIR measurements. Single point measures of optical activity, such as optical rotation or electronic CD at a fixed wavelength point, cannot compete, when more than one chiral species is present, with multiplefrequency (multiplex) measures of IR and VCD that are obtainable using FT methods. While these methods allow one to monitor a change in the optical rotation or electronic CD at a fixed wavelength, they cannot resolve the sources of the change and, as a result, cannot fully monitor the course of the reaction of the chiral molecules. In particular, for a two-component system, sources of the observed change in optical rotation or single-point electronic CD are as follows: (1) reduction in mole fraction of the chiral reactant molecule, (2) change (racemization) in the % EE of the remaining reactant, (3) increase in the mole fraction of the chiral product molecule, and (4) change (racemization) in the % EE of the product. While an electronic CD spectrum could be scanned, the kinetic process would have to halted while the scan took place, and one would not be able to signal average very long at any one location. Further, electronic CD spectra do not possess the structural richness of VCD and one needs further to worry about whether the molecule of interest has a chromophore in the UV-visible region to make it suitable for monitoring purposes. The advantage of electronic CD is intrinsically high signal-to-noise ratio when a chromophore is present and the potential in some cases to rapidly scan the CD spectrum when the signal quality is sufficiently high.

Figure 9. PLS analysis results of camphor-borneol reaction simulation experiment II.

For VCD, the more difficult cases arise when the molecule is large and has several populated conformational states at room temperature and when the molecule dissolves readily only in solvents unfavorable for IR measurements, such as water or methanol. Multiple conformational states for VCD tend to have opposing VCD spectra in some regions of the spectrum, and this can lead to lower overall VCD intensity, signal-to-noise ratio, and accuracy in % EE studies. Solvents that are highly absorbing in the IR either greatly attenuate the available IR radiation needed for VCD measurements or completely block given regions of the spectrum. This does not mean that VCD % EE analysis cannot be carried out, since even in the case of difficult solvents such as water, with appropriate concentrations, sufficiently short path lengths, and use of available spectral windows, effective analyses can be carried out. FT-VCD % EE determination can also be compared to chiral chromatography and NMR using chiral shift reagents. The power of the FT-VCD approach is that no additional chiral agent needs

to be involved in the % EE measurement. Although chiral chromatography is much more accurate than VCD for % EE determination (typically an accuracy of 0.1% can be achieved), a physical separation of the two enantiomers is required, the interaction with a second chiral species is needed as is time for the chromatographic separation, and there is considerable expense associated with maintaining a column and use of eluting solvents over time. For NMR, a chiral shift reagent must be found and used, and the technique is also time-consuming. Neither chromatography nor NMR can be used to monitor simultaneously more than one species as a function of time, and neither of these techniques can be considered as a real-time monitor of the kinetics of the reactions of chiral molecules. CONCLUSIONS We demonstrate here, for the first time, a flow cell VCD technique for time-dependent % EE measurements followed by PLS analysis for the final % EE analysis. The RMSECV and RMSEP Analytical Chemistry, Vol. 76, No. 23, December 1, 2004

6965

of these % EE determinations are approximately 0.8-3.0% at 4-cm-1 resolution, which is a reasonable level of accuracy. The simultaneous monitoring of mole fraction composition and % EE for two chiral species (camphor and borneol mixture) is shown to be successful. Thus, the FT-VCD technique can be employed to follow the kinetics of mixtures of chiral species, as well as individual species. Thus, we demonstrate that FT-VCD is an appropriate analytical tool to monitor the kinetics of reactions involving chiral molecules. Since optical rotation is not useful for chiral mixtures and both chromatography and NMR need selection of different chiral reagents for different chiral molecules, VCD provides a convenient nondestructive approach for the time-

6966

Analytical Chemistry, Vol. 76, No. 23, December 1, 2004

dependent determination of the % EE of individual components in a reaction mixture containing chiral molecules. ACKNOWLEDGMENT Support for this work from Johnson & Johnson Pharmaceutical Research & Development, LLC and BioTools, Inc. is gratefully acknowledged.

Received for review April 29, 2004. Accepted September 13, 2004. AC049366A