http://informahealthcare.com/ddi ISSN: 0363-9045 (print), 1520-5762 (electronic) Drug Dev Ind Pharm, Early Online: 1–5 ! 2014 Informa Healthcare USA, Inc. DOI: 10.3109/03639045.2014.922571

Drug Dev Ind Pharm Downloaded from informahealthcare.com by Karolinska Institutet University Library on 01/28/15 For personal use only.

RESEARCH ARTICLE

Application of chemometric methods to differential scanning calorimeter (DSC) to estimate nimodipine polymorphs from cosolvent system Akhtar Siddiqui, Ziyaur Rahman, and Mansoor A. Khan Division of Product Quality Research, Office of Testing and Research, OPS, CDER, FDA, Silver Spring, MD, USA

Abstract

Keywords

The focus of this study was to evaluate the applicability of chemometrics to differential scanning calorimetry data (DSC) to evaluate nimodipine polymorphs. Multivariate calibration models were built using DSC data from known mixtures of the nimodipine modification. The linear baseline correction treatment of data was used to reduce dispersion in thermograms. Principal component analysis of the treated and untreated data explained 96% and 89% of the data variability, respectively. Score and loading plots correlated variability between samples with change in proportion of nimodipine modifications. The R2 for principal component regression (PCR) and partial lease square regression (PLS) were found to be 0.91 and 0.92. The root mean square of standard error of the treated samples for calibration and validation in PCR and PLS was found to be lower than the untreated sample. These models were applied to samples recrystallized from a cosolvent system, which indicated different proportion of modifications in the mixtures than those obtained by placing samples under different storage conditions. The model was able to predict the nimodipine modifications with known margin of error. Therefore, these models can be used as a quality control tool to expediently determine the nimodipine modification in an unknown mixture.

Chemometrics, DSC, nimodipine, PCS, PLS, polymorphs

Introduction With the advancements in computer science, data storage ability and the advent of powerful statistical packages that added amazing data mining capabilities, data analysis and online monitoring of the output of many conventional analytical instruments have become more efficient than ever before. Chemometrics is a data-driven method to extract information about the state of a system by employing statistical or mathematical means1. To name a few, chemometric methods has been employed and modeled for data obtained from near infrared, infrared, chemical imaging and E-tongue for evaluating drug bitterness2–5. Although chemometrics may be applied to other instrumental data, its application to thermal data is not common. Differential scanning calorimetry (DSC) is a rapid thermoanlytical technique, which is extensively used to study various solid state properties of materials6. Because it requires small sample quantity to obtain material-relevant information by controlled thermal treatment to the sample, it has found many novel applications in the field of food science, chemistry, pharmaceuticals and biologics7–11. Despite its frequent utility in different scientific disciplines to perform material characterization, very few studies employing chemometrics to the DSC data have been

Address for correspondence: Mansoor A. Khan, FDA, CDER, DPQR, White Oak, LS Building 64, Room 1070, New Hampshire Avenue, Silver Spring, MD 20993-002, USA. Tel: +1 301 796 0016. Fax: +1 3017969816. E-mail: [email protected]

History Received 27 November 2013 Revised 2 May 2014 Accepted 4 May 2014 Published online 26 May 2014

reported in food and pharmaceutical field6,7,11,12. Application of chemometrics to DSC is particularly interesting in terms of the rapid quantification of the chemical component of interest using predictive models. By using chemometrics, the cumbersome process of calculating crystallinity of each sample through heat of fusion, and crystallization has been simplified13. In the field of food science, predictive chemometric models based on PCR and PLS regressions of the DSC data have been applied to rapidly quantitate the percentage of fatty acid in extra-virgin olive oil. These studies further demonstrated close agreement between values predicted by the models and high pressure liquid chromatography (HPLC) Mass spectrometry (MS) or Gas Chromatography14,15. Therefore, the objective of the current study was to evaluate the applicability and performance of chemometrics in predicting the proportion of nimodipine modifications present in recrystallized mixtures from formulation using DSC data. In this study, nimodipine was selected as a model drug. Generic product of this drug has recently been recalled due to its recrystallization issues from the formulation in the capsules16. Such recrystallization of drug from solution can negatively impact the treatment outcome. Nimodipine is a Biopharmaceutical Classification System (BCS) Class II drug displaying low solubility, high permeability and exhibits two modifications17. The two modifications significantly differ in their physical properties. Notably, the solubility of modifications (Mod) I and Mod II at 25  C in water are 0.036 and 0.018 mg/ml17. Additionally, their melting points differ as the onset melting

2

A. Siddiqui et al.

Drug Dev Ind Pharm, Early Online: 1–5

points of Mod I and Mod II are 124  C and 116  C, respectively17. This thermal characteristic has been frequently used to identify polymorphism18,19.

Materials and methods Materials Nimodipine was obtained from LeapChem (Hangzhou, China) and Super refinedÕ PEG 400-(LQ)-(MH) (polyethylene glycol 400 NF) was bought from CRODA (Edison, NJ). Acetonitrile, methanol, glycerol and monobasic potassium hydrogen phosphate were procured from Fisher Scientific Co. (Norcross, GA). All other chemicals and solvents used were of analytical grade. Drug Dev Ind Pharm Downloaded from informahealthcare.com by Karolinska Institutet University Library on 01/28/15 For personal use only.

Methods Preparation of Mod I and Mod II Nimodipine Mod II was prepared using method described by Guo et al.20 Briefly nimodipine Mod I was dissolved in ethanol and the solvent was evaporated to dryness. The crystals obtained were powdered, passed through USP sieve no. 80, and characterized by thermal and spectroscopic methods. Nimodipine polymorphs I and II were mixed in the following ratios (% w/w); 0:100, 5:95, 10:90, 20:80, 30:70, 40:60, 50:50, 60:40, 70:30, 90:10, 95:5 and 100:0 by hand-shaking each mixture 50 times in longitudinal and transverse directions. Preparation of samples with unknown nimodipine modifications Cosolvent formulation was prepared by dissolving 26 mg of nimodipine (Mod I as supplied) in PEG400, glycerol and purified water in a ratio referred in Table 1 and allowed to crystallize by storing the formulation at various temperatures. Briefly, nimodipine (26 mg) was added to the mixture of PEG400 and glycerol Table 1. Composition of the cosolvent formulation of nimodipine. Recrystallized nimodipine samples were obtained by placing the formulation at different stability conditions. %W/W of ingredients (nimodipine weight 26.0 mg)

Formulation

Water

Glycerin

PEG400

20.4

29.1

50.6

Figure 1. Representative thermogram of the Mod I and Mod II of nimodipine. Melting endotherm (onset) for Mod I and Mod II are 124  C and 116  C.

(Table 1) and the mixture was sonicated (15 min). Clear solution was then obtained by stirring the solution after adding purified water. These formulations were kept at 25  C, 15  C and 5  C for 4 weeks for nimodipine to crystallize. Nimodipine crystals were separated from the cosolvent system by centrifuging the formulation at 4500 rpm for 30 min. The separated crystals were then washed with water and air dried for 24 h followed by vacuum drying at 30  C for another 24 h. Differential scanning calorimetry Differential scanning calorimetry (DSC Q2000: TA Instruments, New Castle, DE) was employed to evaluate thermal properties of the nimodipine Mod I and Mod II after calibrating the instrument for baseline and temperature using indium. About 4–5 mg of nimodipine Mod I and Mod II were hermetically sealed in an aluminum pan and heated at the rate of 10  C/min from 10  C to 150  C maintaining an inert nitrogen atmosphere by keeping the nitrogen flow rate of 50 ml/min to avoid oxidative degradation of the samples. Statistical analysis Multivariate data analysis of the sample data was carried out by UnscramblerX chemometrics software (version 10.1, Camo Process, Oslo, Norway). The heat-data used were normalized to 1 mg of the drug in developing models and prediction. Furthermore, data from 110  C to 140  C were taken in the analysis as these temperature ranges were found to represent most of the thermal events in the polymorphs.

Result and discussion Characterization of the nimodipine modifications The purity of the nimodipine modifications was verified by spectroscopic (Fourier transform infrared spectroscopy, Raman, Near infrared spectroscopy), powder X-ray diffractrometry (XRPD) and DSC method, and have been extensively discussed by Rahman et al.21 Of particular relevance to this work is DSC showing two distinct endotherms with an onset at 124  C and 116  C for Mod I and Mod II, respectively (Figure 1). These values agree with the data reported previously17. Other than the two modifications, nimodipine has been reported to occur either as eutectic mixtures or pure forms, which displays melting points at 122  C and 134  C. The thermogram of the various

DOI: 10.3109/03639045.2014.922571

proportions of the two modifications mixed for developing calibration displayed two endotherms. The size of these endotherms with an onset at 116  C and 124  C was found to vary according to the relative amount of nimodipine modification in the mixtures (Figure 2a). Such systematic variation in endothermic property due to the proportion of nimodipine modification present can be exploited to build a chemometric model, which can be used as a predictor of the modifications obtained from the crystallized cosolvent samples.

Drug Dev Ind Pharm Downloaded from informahealthcare.com by Karolinska Institutet University Library on 01/28/15 For personal use only.

Preparation of multivariate calibration model Principal component analysis (PCA) is the preliminary step in developing multivariate models, wherein new variables (Latent Variable) are defined from the linear combination of the original variables. These new variables or principal components (PC) are orthogonal to each other; the first PC usually explains maximum variability in the data, and the remaining variability is explained by successive components. Under this analysis, two trends from the overall data matrix are obtained, i.e. visualizing relationship between sample/s and displaying relationship between variables. Then, based on these observations, trend in the samples is

Chemometric of DSC data for nimodipine polymorphs

3

explained by variable largely responsible for such trends. This vital information is displayed through score and loading plots. In this study, various mathematical pretreatment to the raw data (heat flow as a function of temperature) yielded no improvement in the variability explained by PC except baseline correction (Figure 2b). When linear base line correction was employed to the data (Figure 2a and b), the offset in the base line as seen in untreated data was found to be compressed. The PCA analysis was performed on the treated and untreated (raw) data after mean centering the data, and the PCA model was validated by test methods in which the calibration samples were divided into calibration (blue dot in Figure 3) and validation set (red dot in Figure 3). The two principal components in the PCA analysis were found to explain maximum variability in the data set. The overall variability explained by the two components of the treated and untreated data was found to be 96% and 89%, respectively (Figure 3). Furthermore, PC1 of the treated data explained the variability better than the PC1 of untreated data. The PC1 of treated data was found to explain 81% variability in the data set whereas PC1 of untreated data explained only 75% variability in the data. In the score plot of both the treated and untreated data, the PC1 score was found to increase with an increase in the

Figure 2. Representative (a) raw and (b) linear base line-corrected thermogram of the Mod I and Mod II mixtures used in building model.

Figure 3. Representative PCA Score plot of (a) the raw and (b) base line-corrected data. Blue and red dot represents samples used in calibration and validations, respectively.

4

A. Siddiqui et al.

Drug Dev Ind Pharm, Early Online: 1–5

Drug Dev Ind Pharm Downloaded from informahealthcare.com by Karolinska Institutet University Library on 01/28/15 For personal use only.

Figure 4. Representative PCA loading plot of base line-corrected data (a) PC1 and (b) PC2.

proportion of Mod I in the samples or vice versa. Therefore, PC1 was ascribed to nimodipine modification. PC2, which explains 15% and 14% variability in treated and untreated data, appears to be introduced by differences in crystal size, morphology and purity of the modifications. Therefore, PC2 was assigned to crystal geometry. The PC1 loading plot for both treated and untreated data demonstrated two peaks: one negative peak at 119  C and another positive peak at 127  C, which reflects the endotherm of Mod II and Mod I, respectively (Figure 4). These two oppositely correlated peaks at 119  C and 127  C further corroborated that PC1 explains the type of nimodipine modification present in the sample. Furthermore, given the fixed total amount, increasing the proportion of one modification decreases the others proportion in the overall mixture of the sample. The loading score of PC2 indicated two negatively correlated peaks: one negatively correlated peaks at 116  C and 119  C and another at 126  C and 129  C. These observations might reflect the purity of the crystals forms as described by Grunenberg et al.17 After PCA, regression analysis using PCR and PLS techniques was carried out using treated and untreated data. Overall, the goodness of fit for the PLS was found to be 1% more than the PCR. The R2 for PCR and PLS was found to be 0.91 and 0.92, respectively. In other words, PCR and PLS models can explain 91% and 92% of the variability in the data. The overall difference in actual and predicted data was found to be higher in the untreated than the treated data (Table 2). Furthermore, RMSE values for calibration samples were found to be lower than the validation samples. The RMSE of the treated and untreated samples for calibration and validation in PCR were found to be 9.94, and 9.97; 10.15, and 10.22, respectively. Similarly, the RMSEs of the treated and untreated samples for calibration and validation in PLS were found to be 9.17, and 9.76; 9.13, and 9.33, respectively. There are different potential sources for introducing error in data, which include sample to sample weight variation and variation in packing of the sample in the pan. In summary, treatment of the data has improved the predictive ability of the model.

Table 2. Performance of regression model using raw and treated data. Calibration Model/treatment

The PCR and PLS models developed using DSC data were applied to the unknown samples (crystals from solubilized drug) obtained after 4-week exposure of the cosolvent formulations to 5  C, 15  C and 25  C (Table 3). The prediction results indicated that storage condition does affect the crystallization of one modification in preference to others from the cosolvent formulation, where drug is present as solution. For example, Mod II was predominantly crystallized from the sample at 5  C and 25  C whereas crystallization of Mod I was more than Mod II from the same formulation at 15  C (Table 3). Furthermore, prediction of the modification was found to better using treated model than

Validation

RMSEC

SEC

R

2

RMSEP

SEP

Modification I PCR/UT PCR/T PLS/UT PLS/T

0.91 0.91 0.92 0.92

9.97 9.94 9.13 9.17

10.19 10.15 9.33 9.37

0.90 0.90 0.92 0.92

10.22 10.15 9.85 9.76

10.7 10.6 10.1 9.93

Modification II PCR/UT PCR/T PLS/UT PLS/T

0.91 0.91 0.92 0.92

9.97 9.94 9.13 9.17

10.19 10.15 9.33 9.37

0.90 0.90 0.92 0.92

10.22 10.15 9.85 9.76

10.67 10.61 10.5 10.7

RMSEC, root mean square error for calibration; RMSEP, root mean square error for prediction; SEC, standard error for calibration; SEP, standard error for prediction; PCR, principal component regression, PLS, partial least square regression; T, baseline corrected; UT, base line not corrected.

Table 3. Quantitative prediction of the nimodipine modifications using PCR/PLS regression model with and without baseline correction. The recrystallized samples were obtained after exposing the cosolvent system at 5 C, 15 C and 25 C for 4 weeks. Crystallized nimodipine from cosolvent formulation PCR/UT

PCR/T

PLS/UT

PLS/T

Modification

Modification

Modification

Modification

I

II

I

II

I

II

I

II



5 C 2.66

102.66

7.25

107.25

2.38

102.38

6.57

106.57

53.19

46.81

55.42

44.58

95.63

0.55

100.55



15 C 55.32

Estimation of nimodipine Mod I and Mod II in unknown recrystallized samples

R

2

44.68

57.56

42.44 

25 C 5.38

94.62

0.26

99.74

4.37

T, Baseline corrected; UT, Base line not corrected.

untreated model. The PLS-treated prediction model for samples at 25  C demonstrated that formulation contains 100.55% of Mod II whereas the untreated model for the same showed 95.63% of Mod II. When the same formulation was exposed to 5  C, the PLStreated model indicated that samples contain 6.57% and 106.57% Mod I and Mod II, respectively. The PLS-untreated model of the same sample exhibited 2.38% and 102.38% Mod I and Mod II, respectively. Similarly, at 15  C, it was found to

DOI: 10.3109/03639045.2014.922571

contain 55.42% and 44.58% Mod I and Mod II, respectively, when sample was evaluated using treated prediction model. When untreated prediction model was used, it showed 53.19% and 46.81% Mod I and Mod II, respectively. These results, which were consistent with our findings by other spectroscopic methods22, showed the impact of formulation and sample storage conditions on the nature of precipitated polymorphs. Overall, the chemometric models using the DSC data were found to be able to discriminate the two nimodipine modifications in the powdered mixtures with known margin of errors.

Drug Dev Ind Pharm Downloaded from informahealthcare.com by Karolinska Institutet University Library on 01/28/15 For personal use only.

Conclusions Chemometric models for nimodipine to evaluate its modifications in the recrystallized nimodipine from cosolvent mixtures were developed. Linear baseline correction improved the overall predictability of nimodipine modifications in the mixture. No significant difference in the predictability of the nimodipine modifications using the PCR and PLS models was observed. These models were able to discriminate the proportion of nimodipine modification in the recrystallized sample from the cosolvent system with a known margin of error. Therefore, this approach could be employed as a quality control approach to make a quick estimation of the two modifications in an unknown sample. A future study might entail the verification of this model by attaching a module to the computer to feed the real time data from DSC instrument to the model to get percent polymorphs present in an unknown sample.

Acknowledgements The authors would like to thank Medical Counter Measure Initiative (MCMi) for infrastructure support and Oak Ridge Institute for Science and Education (ORISE) for supporting post-doctoral research program.

Declaration of interest The findings and conclusions in this article have not been formally disseminated by the Food and Drug Administration and should not be construed to represent any Agency determination or policy.

References 1. Wise BM, Gallagher NB. The process chemometrics approach to process monitoring and fault detection. J Process Control 1996;6: 329–48. 2. Kona R, Qu H, Mattes R, et al. Application of in-line near infrared spectroscopy and multivariate batch modeling for process monitoring in fluid bed granulation. Int J Pharm 2013;452:63–72. 3. Rajalahti T, Kvalheim OM. Multivariate data analysis in pharmaceutics: a tutorial review. Int J Pharm 2011;417:280–90. 4. Puchert T, Lochmann D, Menezes JC, Reich G. Near-infrared chemical imaging (NIR-CI) for counterfeit drug identification – four-stage concept with a novel approach of data processing (Linear Image Signature). J Pharm Biomed Anal 2010;51:138–45.

Chemometric of DSC data for nimodipine polymorphs

5

5. Rohman A, Man YBC. The chemometrics approach applied to FTIR spectral data for the analysis of rice bran oil in extra virgin olive oil. Chemo Intell Lab Sys 2012;110:129–34. 6. Maggio RN, Cerretani L, Barnaba C, Chiavaro E. Application of differential scanning calorimetry-chemometric coupled procedure to the evaluation of thermo-oxidation on extra virgin olive oil. Food Biophys 2012;7:114–23. 7. Artiaga Rn, Lopez-Beceiro J, Tarrio-Saavedra J, et al. Estimating the reversing and non-reversing heat flow from standard DSC curves in the glass transition region. J Chemometrics 2011;25:287–94. 8. Garbett NC, Mekmaysy CS, Helm CW, et al. Differential scanning calorimetry of blood plasma for clinical diagnosis and monitoring. Exp Mol Pathol 2009;86:186–91. 9. Yoshihashi Y, Yonemochi E, Maeda Y, Terada K. Prediction of the induction period of crystallization of naproxen in solid dispersion using differential scanning calorimetry. J Therm Anal Calorim 2010; 99:15–9. 10. Rahman MS, Al-Saidi G, Guizani N, Abdullah A. Development of state diagram of bovine gelatin by measuring thermal characteristics using differential scanning calorimetry (DSC) and cooling curve method. Thermochim Acta 2010;509:111–19. 11. Zidan AS, Rahman Z, Sayeed V, et al. Crystallinity evaluation of tacrolimus solid dispersions by chemometric analysis. Int J Pharm 2012;423:341–50. 12. Miltyk W, Antonowicz E, Komsta L. Recognition of tablet content by chemometric processing of differential scanning calorimetry curves – an acetaminophen example. Thermochim Acta 2010;507– 508:146–9. 13. Pan X, Julian T, Augsburger L. Quantitative measurement of indomethacin crystallinity in indomethacin-silica gel binary system using differential scanning calorimetry and X-ray powder diffractometry. AAPS PharmSciTech 2006;7:E72–8. 14. Maggio R, Barnaba C, Cerretani L, et al. Study of the influence of triacylglycerol composition on DSC cooling curves of extra virgin olive oil by chemometric data processing. J Therm Anal Calorim 2013;1–8. 15. Cerretani L, Maggio RM, Barnaba C, et al. Application of partial least square regression to differential scanning calorimetry data for fatty acid quantitation in olive oil. Food Chem 2011;127:1899–904. 16. FDA.Safety. Availble from: http://www.fda.gov/Safety/Recalls/ ucm318118.htm [last accessed 20 Nov 2013]. 17. Grunenberg A, Keil B, Henck JO. Polymorphism in binary mixtures, as exemplified by nimodipine. Int J Pharm 1995;118:11–21. 18. Papageorgiou G, Bikiaris D, Karavas E, et al. Effect of physical state and particle size distribution on dissolution enhancement of nimodipine/PEG solid dispersions prepared by melt mixing and solvent evaporation. AAPS J 2006;8:E623–31. 19. Otsuka M, Kato F, Matsuda Y, Ozaki Y. Comparative determination of polymorphs of indomethacin in powders and tablets by chemometrical near-infrared spectroscopy and X-ray powder diffractometry. AAPS PharmSciTech 2003;4:58–69. 20. Guo Z, Ma M, Wang T, et al. A kinetic study of the polymorphic transformation of nimodipine and indomethacin during high shear granulation. AAPS PharmSciTech 2011;12:610–19. 21. Rahman Z, Siddiqui A, Khan MA. Assessing the impact of nimodipine devitrification in the ternary cosolvent system through quality by design approach. Int J Pharm 2013;455:113–23. 22. Siddiqui A, Rahman Z, Sayeed VA, Khan MA. Chemometric evaluation of near infrared, fourier transform infrared, and raman spectroscopic models for the prediction of nimodipine polymorphs. J Pharm Sci 2013;102:4024–35.

Application of chemometric methods to differential scanning calorimeter (DSC) to estimate nimodipine polymorphs from cosolvent system.

The focus of this study was to evaluate the applicability of chemometrics to differential scanning calorimetry data (DSC) to evaluate nimodipine polym...
319KB Sizes 4 Downloads 3 Views