Anal. Chem. 2006, 78, 604-608
Electron Paramagnetic Resonance Spectroscopy Studies of Oxidative Degradation of an Active Pharmaceutical Ingredient and Quantitative Analysis of the Organic Radical Intermediates Using Partial Least-Squares Regression Helen Elizabeth Williams,*,† Victoria Catherine Loades,† Mike Claybourn,† and Damien Martin Murphy‡
AstraZeneca, Silk Road Business Park, Macclesfield, Cheshire SK10 2NA, U.K., and School of Chemistry, Cardiff University, Main Building, Park Place, Cardiff, CF10 3AT, U.K.
Electron paramagnetic resonance (EPR) spectroscopy was used to study the radical species formed during the oxidation of an active pharmaceutical ingredient in the solid state. It was found that the extent of radical generation correlated to the formation of an oxidative degradation product. Multifrequency EPR and electron nuclear double resonance spectroscopy gave additional information on the identity of the organic radical species involved in the oxidation process, and a mechanism was proposed for the degradation, involving the formation of both carboncentered and peroxy radicals. The multivariate analysis technique of partial least-squares (PLS) regression was then used to determine the extent of oxidation of the active pharmaceutical ingredient from the EPR spectra. The suitability of this approach was demonstrated from its application to a series of standards. The conventional approach for the quantitative analysis of EPR spectra is to measure the peak height or to perform double integration of the spectral region containing the signal of interest. Both of these methods have intrinsic errors associated with them, particularly for weak EPR signals with a poor signal-to-noise ratio or a sloping background response. The results obtained showed that greatly improved quantitation was obtained using the PLS regression approach. Electron paramagnetic resonance (EPR) spectroscopy is a very sensitive technique that can be used to characterize organic radical species. Electron nuclear double resonance (ENDOR) spectroscopy is an extension to the EPR technique that can yield further structural information about the organic radical in question. In this work, EPR, at both X-band (9 GHz) and Q-band (35 GHz) frequencies, and ENDOR (at X-band) spectroscopies have been used to study oxidative degradation of an amorphous solid active pharmaceutical ingredient (API). Under certain conditions, the API undergoes oxidative degradation through the intermediacy of organic radical species. EPR has been used to monitor the level * Corresponding author: (e-mail)
[email protected]; (phone) 01625 582828; (fax) 01625 510076. † AstraZeneca. ‡ Cardiff University.
604 Analytical Chemistry, Vol. 78, No. 2, January 15, 2006
of the organic radical species in API samples stored at elevated conditions for different lengths of time. EPR has previously been used to study the oxidation of various pharmaceutical compounds, including the polyene antibiotic amphotericin B,1 the antitumor drug 9-hydroxyellipticine,2 and the antimalaria drug 5-hydroxprimaquine,3 but these measurements were made in solution using spin traps and not in the solid state. EPR spectroscopy has not been widely used to study the oxidation of drug molecules in the solid state.4 To the best of our knowledge, EPR spectroscopy has not previously been used to measure or predict the stability characteristics of APIs although it has many applications in the food industry to both measure and predict the oxidative stability of food products. For example, EPR spectroscopy has been used extensively to predict the flavor stability of beer.5-7 An oxidative forcing test is used in which quantification of the radical content is performed by measuring the peak height of the EPR response, which is subsequently used to predict the longer-term stability of the beer.8,9 Similarly, EPR has been used in other areas of the food industry to measure and predict the stability of products such as vegetable oils,10 dairy products,11-14 (1) Lamy-Freund, M. T.; Ferreira, V. F. N.; Schreier, S. J. Antibiot. 1985, 38, 753-757. (2) Auclair, C.; Hyland, K.; Paoletti, C. J. Med. Chem. 1983, 26, 1438-1444. (3) Vasquez-Vivar, J.; Augusto, O. Free Radical Res. Commun. 1990, 9, 383389. (4) Cotelle, P.; Vezin, H. Res. Chem. Intermediates 2003, 29, 365-377. (5) Takemura, O.; Oka, K.; Uchida, M.; Kawasaki, Y.; Kakimi, Y. Proc. Congr.Eur Brew. Conv. 1997, 26th, 545-552. (6) Kaneda, H.; Kano, Y.; Osawa, T.; Kawakishi, S.; Kamada, K. J. Am. Soc. Brew. Chem. 1989, 47, 49-53. (7) Forster, C.; Schwieger, J.; Narziss, L.; Back, W.; Uchida, M.; Ono, M.; Yanagi, K. Monatsschr. Brauwiss. 1999, 52, 86-93. (8) Uchida, M.; Suga, S.; Ono, M. J. Am. Soc. Brew. Chem. 1996, 54, 205211. (9) Stasko, A.; Rapta, P.; Malik, F. Monatsschr. Brauwiss. 2000, 53, 4-7. (10) Velasco, J.; Andersen, M. L.; Skibsted, L. H. Food Chem. 2004, 85, 623632. (11) Kristensen, D.; Andersen, M. L.; Skibsted, L. H. Milchwissenschaft 2002, 57, 255-258. (12) Kristensen, D.; Kroger-Ohlsen, M. V.; Skibsted, L. H. ACS Symp. Ser. 2002, 807, 114-125. (13) Thomsen, M. K.; Vedstesen, H.; Skibsted, L. H. J. Food Lipids 1999, 6, 149-158. (14) Thomsen, M. K.; Kristensen, D.; Skibsted, L. H. J. Am. Oil Chem. Soc. 2000, 77, 725-730. 10.1021/ac051697f CCC: $33.50
© 2006 American Chemical Society Published on Web 12/09/2005
and processed pork.15 A method has also been reported using EPR to predict the color stability of acrylic resins used in dentistry.16 The conventional approaches for quantitative analysis of EPR data are to measure the peak height or to take the double integral to determine the peak area. These approaches are appropriate for a well-defined EPR response with a good signal-to-noise ratio and a flat background baseline. Double integration and peak height measurements are routinely used to quantify EPR spectra, including some pharmaceutical applications.17-20 However, as discussed by Nagy,21 there are many possible causes of error in quantitative EPR spectroscopy, including poor standardization of sample position in the cavity, signal measurements under conditions of microwave saturation, insufficiently wide integration range, rapid scan times, and poor calibration of the instrument variables. Further errors are also introduced when a reference sample is used to determine the number of electron spins, since the spectroscopic response characteristics of the reference material may not match those of the analyte. These additional errors are compounded by differences between the reference and sample shapes, dielectric constants, and inequality of the probabilities of the transitions producing the spectral lines used. An objective of this work was to use multivariate calibration to improve quantitation and prediction of the percentage level of oxidation of the API based on the EPR spectra, compared to the traditional quantification methods of peak height and double integration. There have been several applications of chemometrics techniques to EPR data, some of which will be discussed here. It should be noted that publications in this area are not as prolific as for other spectroscopies. In 1977, Koskinen and Kowalski22 published the benefits of simple chemometric techniques of Fourier transform followed by stepwise regression analysis to model the effect of temperature and the amount of lipid present from the EPR spectra and to provide a tool for the analysis of spin labels in model membrane systems. Since then, principal component analysis has been applied to EPR data on a number of occasions. Examples include classification of the provenance of marbles23 and identification of pure component spectra from complex mixtures.24 Karlstro¨m et al.,25 extended this work by applying partial least-squares (PLS) regression26 to EPR spectra, to correlate the spectral data to various recorded chemical and physical properties of the samples. The data presented in this paper are based on the solid-state EPR measurements of an API. The EPR signal is a nonsymmetrical broad spectrum. In this work, PLS regression will be used to (15) Carlsen, C. U.; Andersen, M. L.; Skibsted, L. H. Eur. Food Res. Technol. 2001, 213, 170-173. (16) Hasegawa, A.; Hamano, T.; Miwa, M.; Nagasaka, S. Dent. Mater. J. 1999, 18, 207-217. (17) Basly, J. P.; Duroux, J. L.; Bernard, M. Int. J. Pharm. 1996, 139, 219-221. (18) Basly, J. P.; Longy, I.; Bernard, M. Pharm. Res. 1997, 14, 1192-1196. (19) Basly, J. P.; Basly, I.; Bernard, M. J. Pharm. Biomed. Anal. 1998, 17, 871875. (20) Ambroz, H. B.; Kornacka, E. M.; Marciniec, B.; Ogrodowczyk, M.; Przybytniak, G. K. Radiat. Phys. Chem. 2000, 58, 357-366. (21) Nagy, V. Appl. Magn. Reson. 1994, 6, 259-285. (22) Koskinen, J. R.; Kowalski, B. R. ACS Symp. Ser. 1977, 52, 117-126. (23) Attanasio, D. Appl. Magn. Reson. 1999, 16, 383-402. (24) Steinbock, O.; Neumann, B.; Cage, B.; Saltiel, J.; Muller, S. C.; Dalal, N. S. Anal. Chem. 1997, 69, 3708-3713. (25) Karlstro ¨m, H.; Norden, B.; Wikander, G. Soil Sci. 1994, 157, 300-311. (26) Geladi, P.; Kowalski, B. R. Anal. Chim. Acta 1986, 185, 1-17.
maximize the covariance between the EPR spectra and the reference values of the percentage level of oxidation. The procedure will be first applied to a series of standards in solution and then repeated for the API samples. EXPERIMENTAL SECTION EPR and ENDOR Analysis. The experimental studies carried out involved analysis of a batch of an API at various time points over a 12-month period after storage at 25 °C/60% relative humidity (RH) and 40 °C/75% RH. The EPR spectra were recorded at room temperature on a cw-Bruker EMX X-band spectrometer operating at 100 kHz field modulation and equipped with an ER4119HS high-sensitivity cavity. Q-band EPR spectra of a degraded API sample were recorded at variable temperatures on a cwBruker ESP 300E series spectrometer operating at 100 kHz field modulation fitted with a Q-band Bruker ENDOR resonator. X-band ENDOR spectra of a degraded sample of the API were recorded at 120 K on a cw-Bruker ESP 300E series spectrometer operating at 12.5 kHz field modulation and equipped with an ESP360 DICE ENDOR unit in an EN-801 ENDOR cavity using 8 dB power from an ENI A-300 amplifier. Quantitative Analysis. Each X-band EPR spectrum was scaled depending on the sample weight. A relative measure of the amount of radical species present in each sample was then determined by peak height measurement (PH), double integration (DI), and PLS regression. Peak height measurements were taken from the peak maximums to peak minimums. Baseline subtraction followed by double integration was performed using the commercially available Bruker WINEPR System v 2.11 software. PLS regression was performed using MatlabR14 (Mathworks Ltd.) with the PLS•Toolbox v 3.5 (Eigenvector Research Inc.). For PLS calibration, the data were mean centered and the sample spectra smoothed. The number of latent variables (LVs) used to model the data was decided based on the lowest error of prediction of cross validation, with a sensible number compared to known variables in the samples. For consistency, leave-one-out cross validation was applied to the data from each quantitation method (PH, DI, and PLS regression). For cross validation (CV), a sample is removed from the data set, the model calculated and then used to predict the removed sample. This continues until all samples have been excluded and predicted. The error in CV (ECV)27 and correlation coefficient (R2) from the actual versus predicted graphs were used to compare the different quantitation methods.
x
I
∑(c -
ECV )
i
cv
cˆi)2
i)1
I
where ci is the actual value,cvcˆi is the predicted value from cross validation, and I is the number of times CV has been repeated. To demonstrate proof of principle for the use of PLS regression to quantify EPR spectral responses, a series of 2,2,6,6-tetramethylpiperidinyoxyl free radical, hereafter abbreviated as TEMPO (Sigma-Aldrich) in toluene (Fisher Scientific), standards were (27) Brereton, R. G. Chemometrics: Data Analysis for the Laboratory and Chemical Plant, 1st ed.; Wiley: Chichester, 2003.
Analytical Chemistry, Vol. 78, No. 2, January 15, 2006
605
Figure 2. cw 1H ENDOR spectrum of degraded API recorded at 120 K. The spectrum was recorded at a static magnetic field position of 3515 G. No angular variation in the couplings was detected at other magnetic field positions.
Figure 3. Proposed oxidation mechanism of the API. Figure 1. Room-temperature cw-EPR spectra of API (a) X-band and (b) Q-band frequencies.
prepared and the acquired spectra also quantified using the three methods described. RESULTS AND DISCUSSION A broad singlet peak was detected in each API sample analyzed by EPR at X-band frequencies, as shown in Figure 1a. A g-value of 2.004 was determined for the singlet, with an approximate line width of 8 G. Owing to the amorphous nature of the API sample, the inhomogeneous EPR line width was significantly broadened (compared to an isotropic radical species in fluid solution) due to the random orientation of the paramagnetic species with respect to the direction of the applied magnetic field. The EPR response was detected in each sample with increasing intensity relating to an increasing level of oxidation over the 12-month period. The Q-band EPR analysis of the degraded API sample provided enhanced resolution and revealed that the broad response seen at X-band is actually due to two separate paramagnetic species with g-values of 2.0063 and 2.0038, as shown in Figure 1b. The more intense high-field component is likely due to a carbon-based radical while the less intense low-field feature is likely due to an oxygen-based radical, possibly a peroxy radical. In this amorphous solid, a peroxy radical would be expected to give an anisotropic spectrum with unresolved hyperfine components producing broad line widths. The g-value obtained is consistent with an average of gx and gy values previously reported for immobilized peroxy radicals28 that can be assumed to appear as one feature in this spectrum due to the broad line widths. For a similar reason, the gz component, which is normally the weaker feature, cannot be detected from the baseline noise in this amorphous sample. A degraded sample of the API was also analyzed by X-band ENDOR spectroscopy. A weak signal was obtained centered at (28) Attwood, A. L.; Edwards, J. L.; Rowlands, C. C.; Murphy, D. M. J. Phys. Chem. A 2003, 107, 1779-1782.
606 Analytical Chemistry, Vol. 78, No. 2, January 15, 2006
14.96 MHz, corresponding to the proton nuclear Lamore frequency at the detection field (3515 G), as shown in Figure 2. The spectrum obtained was very weak due to the relatively low level of radical species present, but at least two separate proton couplings could be seen with a coupling of 1.65 and 0.5 MHz, suggesting that at least two different proton environments exist. It should be noted that at X-band frequencies the ENDOR spectrum would have contributions from both radical species, which were detected and easily resolved at Q-band. Unfortunately, Q-band ENDOR measurements were not possible due to limitations in sample size. The EPR and ENDOR spectra support a proposed degradation mechanism for the oxidation of the API with both alkyl (carbonbased) and peroxy-based radical intermediates (Figure 3). This mechanism is consistent with previously reported oxidation mechanisms of APIs in the solid state.29-33 EPR spectroscopy can therefore be used to determine the extent of oxidation from the organic radical content of the API. The level of the oxidation product is routinely quantified by high-performance liquid chromatography (HPLC) and has been identified by HPLC-mass spectroscopy. Determining the level of oxidative degradation by HPLC in a sample of the API is a time-consuming and laborintensive process. It is also a destructive technique and uses a significant volume of solvents. EPR would have significant benefits to determine the level of oxidation including improved sensitivity, faster analysis, reduced sample size, and its nondestructive nature. Quantitation. If an EPR spectrum has a good signal-to-noise ratio and a symmetric response, then comparison between peak height measurements of the same radical response in different samples of the same compound can give relative results with reasonable accuracy. If a radical response has a weak intensity (29) Hovorka, S. W.; Scho ¨neich, C. J. Pharm. Sci. 2001, 90, 253-269. (30) Smith, G. B.; DiMichele, L.; Colwell, L. F., Jr.; Dezeny, G. C.; Douglas, A. W.; Reamer, R. A.; Verhoeven, T. R. Tetrahedron 1993, 49, 4447-4462. (31) Ingold, K. U. Acc. Chem. Res. 1969, 2, 1-9. (32) Stewart, P. J.; Tucker, I. G. Aust. J. Hosp. Pharm. 1985, 15, 111-117. (33) Waterman, K. C.; Adami, R. C.; Alsante, K. M.; Hong, J.; Landis, M. S.; Lombardo, F.; Roberts, C. J. Pharm. Dev. Technol. 2002, 7, 1-32.
Figure 4. Room-temperature cw-EPR spectra for a series of TEMPO standards and corresponding quantitation results: (a) raw spectra; (b) preprocessed spectra; (c) latent variables; (d) peak height results; (e) double-integral results; (f) PLS regression results.
with a poor signal-to-noise ratio, then the peak height measurement process is hindered and can become inaccurate. When comparing peak heights between samples to quantify a relative EPR response, it is assumed that the line width in each spectrum is constant. In this work, this assumption is not valid and leads to errors in the peak height measurements. This error is also affected by the nonsymmetric nature of the response and the sloping background seen in this study. The double-integral approach for quantitation gives a measure of the total peak area of an EPR response by integrating the first-derivative spectrum twice. Before the integration can be performed, a baseline subtraction is carried out. This baseline subtraction and double-integration procedure can introduce additional errors into the data, especially for a weak signal as seen in this work, which can lead to nonreproducible results. The application of PLS regression to quantify the EPR data set was therefore investigated. The EPR spectra of a series of TEMPO in toluene standard solutions, in the concentration range (1-7) × 10-6 M, were quantified using the three different methods described. The spectra and results are displayed in Figure 4. The PLS regression method, with the solution concentration as the Y-variable, used two latent variables to model the data. Cross validation was applied to all the quantitation methods using the leave-one-out method, and the resulting predicted solution concentration results were plotted against the actual results for each calibration method. Due to the symmetric nature and good signal-to-noise ratio of the TEMPO spectra, the peak height and double-integration measurements gave a good correlation with solution concentration. An improved correlation coefficient (R2) was determined, however, for the PLS regression method compared to the peak height and double-integration methods (0.98 vs 0.93 and 0.87, respectively).
Table 1. Quantitative Method Statistical Results R2
ECV
PH DI PLS
TEMPO 0.933 0.866 0.978
6.54 × 10-7 8.36 × 10-7 3.04 × 10-7
PH DI PLS
API Samples 0.896 0.633 0.966
0.10 0.65 0.06
The statistical results for the quantitation methods are shown in Table 1. This work demonstrates the applicability of PLS regression to correlate EPR spectral data to an external Y-variable. The level of oxidative degradation product in each API sample was determined by HPLC, using external calibration standards. These results were then correlated to the X-band EPR spectra, i.e., the summation of the concentration of both radical species, using the peak height and double-integral methods. Both the calibration models were then cross validated using the leave-oneout method, and the resulting predicted percentage oxidation results were plotted against the actual results for each calibration method, as seen in Figure 5d and e. Although a general linear trend can be seen, the correlation coefficients (R2) obtained by these quantitation techniques show a relatively poor correlation between the EPR response and the percentage of oxidation of the samples (0.90 and 0.63, respectively). This is due to the errors present in both these procedures as discussed previously. In addition to the traditional peak height and double-integration methods, PLS regression was used for quantitation of the X-band Analytical Chemistry, Vol. 78, No. 2, January 15, 2006
607
Figure 5. Room-temperature cw X-band EPR spectra for a series of API samples and corresponding quantitation results: (a) raw spectra; (b) preprocessed spectra; (c) latent variables; (d) peak height results; (e) double-integral results; (f) PLS regression results.
EPR spectra for the API samples. A PLS model, following the leave-one-out cross validation method, was calculated (Figure 5). Three LVs were used to model the data. The loadings, demonstrating the regions of spectrum used to model the data, can be seen in Figure 5c. To demonstrate the quantification ability of the PLS model, the actual percentage of oxidation determined by HPLC was plotted against the predicted values following cross validation of the PLS model (Figure 5f). The statistical results for all the quantitation methods are shown in Table 1. The PLS model reported low ECV values and high correlation coefficient between the actual and predicted values. The plot of actual against predicted percentage degradation demonstrates good correlation for the PLS regression model, with a correlation coefficient of 0.97, when compared to the traditional quantitation techniques. CONCLUSIONS It has been shown that EPR spectroscopy can be used to monitor the extent of oxidative degradation in solid-state samples
608
Analytical Chemistry, Vol. 78, No. 2, January 15, 2006
of an active pharmaceutical ingredient and that the quantitation of the EPR response can be improved by using partial leastsquares regression, compared to the traditional methods of peak height and double-integral measurements. Q-band EPR and X-band ENDOR spectroscopy also gave additional information to differentiate the organic radical species involved in the oxidation process. ACKNOWLEDGMENT The authors thank Dr. Margaret Henderson from AstraZeneca for supplying the HPLC data used in this publication.
Received for review September 22, 2005. Accepted November 9, 2005. AC051697F
