- Email: [email protected]
Pharmaceutical analysis in solids using front face fluorescence spectroscopy and multivariate calibration with matrix correction by piecewise direct standardization
Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 103 (2013) 311–318
Contents lists available at SciVerse ScienceDirect
Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy journal homepage: www.elsevier.com/locate/saa
Pharmaceutical analysis in solids using front face fluorescence spectroscopy and multivariate calibration with matrix correction by piecewise direct standardization Julio Cesar L. Alves ⇑, Ronei J. Poppi Institute of Chemistry, State University of Campinas – UNICAMP, P.O. Box 6154, 13083-970 Campinas, SP, Brazil
h i g h l i g h t s
g r a p h i c a l a b s t r a c t
" Determination of pharmaceutical
formulation using EEM data. " Spectra standardization to solving
matrix effects in EEM data. " Accurate results with relative errors
lesser than 5% were obtained.
a r t i c l e
i n f o
Article history: Received 28 August 2012 Received in revised form 29 October 2012 Accepted 31 October 2012 Available online 7 November 2012 Keywords: Front face fluorescence spectroscopy Calibration transfer Piecewise direct standardization Acetylsalicylic acid Paracetamol Caffeine
a b s t r a c t This paper reports the application of piecewise direct standardization (PDS) for matrix correction in front face fluorescence spectroscopy of solids when different excipients are used in a pharmaceutical preparation based on a mixture of acetylsalicylic acid (ASA), paracetamol (acetaminophen) and caffeine. As verified in earlier studies, the use of different excipients and their ratio can cause a displacement, change in fluorescence intensity or band profile. To overcome this important drawback, a standardization strategy was adopted to convert all the excitation–emission fluorescence spectra into those used for model development. An excitation–emission matrix (EEM) for which excitation and emission wavelengths ranging from 265 to 405 nm and 300 to 480 nm, respectively, was used. Excellent results were obtained using unfolded partial least squares (U-PLS), with RMSEP values of 8.2 mg/g, 10.9 mg/g and 2.7 mg/g for ASA, paracetamol and caffeine, respectively, and with relative errors lesser than 5% for the three analytes. Ó 2012 Elsevier B.V. All rights reserved.
Introduction A mixture of the active ingredients acetylsalicylic acid (ASA), caffeine and paracetamol is a widely used pharmaceutical combination with analgesic effects. The most common analytical methods used for determination of this mixture of compounds include extraction followed by UV spectrophotometric analysis and high performance liquid chromatography [1]. These techniques, while perfectly adequate in terms of their accuracy, can be time-consuming ⇑ Corresponding author. Tel.: +55 11 9 9612 7438. E-mail address: [email protected] (Julio Cesar L. Alves). 1386-1425/$ - see front matter Ó 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.saa.2012.10.074
or expensive when implemented in routine quality control laboratories. In recent years, excitation–emission molecular fluorescence spectroscopy has been used in conjunction with chemometric methods for analysis of pharmaceutical formulations [2]. It can be an excellent alternative for analysis of these mixtures, mainly when the fluorescence spectra can be obtained directly from the solid using a front face accessory [3], making it possible to perform the analysis in a few minutes, practically without sample preparation. The front face fluorescence spectra of a solid pharmaceutical preparation containing ASA, caffeine and paracetamol present overlapping bands from these compounds. Because of this, the
312
Julio Cesar L. Alves, R.J. Poppi / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 103 (2013) 311–318
use of univariate calibration is not the best choice for accurate prediction, as previously verified [4], making the use of higher order calibration procedures necessary. The data obtained from excitation–emission fluorescence spectroscopy can be arranged in an I J K three-way array, where the index I refers to the samples, J to the excitation wavelengths and K to the emission wavelengths. Second order calibration methods can be applied to these types of data providing advantages over methods of zero order or first order, since more information is used in model development. There are several algorithms developed to deal with second-order data and development of multi-way models, such as PARAFAC, BLLS, MCR–ALS and NPLS [5]. An alternative to work with this kind of data is to rearrange them into a matrix and apply a first-order algorithm, leading to unfolded-principal component regression (UPCR) and unfolded-partial least squares (U-PLS) [6]. These approaches were first used to deal with second-order data, before true second-order methods were developed, but they are still able to produce satisfactory results [7] and frequently combined with residual bilinearization (RBL) to achieve the second order advantage [8,9]. In model development using multivariate calibration, as already demonstrated in front face fluorescence of solids [2,10], different bulk proportions of the mixture components and their physical properties (such as particle size) and the use of different ingredients (such as different excipients) produce changes in the signals from the active ingredients. In fact, the use of front face fluorescence of solids depends on the amount and type of ingredients used in the pharmaceutical formulation. The effects of the ingredients and their proportions used in the mixtures on the fluorescence can occur due to competing deactivation processes resulting from the specific interaction between the fluorophore and the ingredients present in the sample. The most significant changes observed in these cases are the variation on signal intensities. For this reason, only samples prepared in the laboratory were used in this study. However, in the pharmaceutical industry the ingredients and their proportions are known and the proposed methodology could be applied without restrictions. In this situation, the utilization of PARAFAC [11], applicable even in the presence of unknown interferences and with spectral overlap of the components in the mixture, and already used in pharmaceutical formulations using solid-phase molecular fluorescence [4], is not appropriate. PARAFAC is not able to correct the matrix effects, since in this case the interferences are not caused by a new excipient. The use of different excipients and their ratio can cause a displacement, change in fluorescence intensity or band profile. To overcome these problems, spectra standardization [12– 14] can be applied to convert an excitation–emission matrix into a spectral surface free of this matrix effect. This procedure involves three main steps:
condition Xp and the corresponding spectral window wi containing measurements at a few nearby wavelengths obtained on the secondary instrument or condition. The regression coefficients bi are placed along the diagonal of the transfer matrix F which describes the relation between the spectra from the primary and the secondary instrument or condition. In order to transfer the spectra collected on the secondary instrument or condition, the matrix Xs is multiplied by the transfer matrix F, yielding Xstd:
Xstd ¼ Xs F
ð1Þ
which can then be used for prediction by using calibration models developed on the primary instrument or condition:
ystd ¼ ðXs FÞb
ð2Þ
where ystd is the predicted values obtained with the spectra collected on the secondary instrument or condition after standardization and b is the regression coefficients determined on the primary instrument or condition [15–17]. There are a few recent articles dealing with the joint use of multi-way models and calibration transfer, none of them related to spectrofluorimetric data. Kompany-Zareh and van den Berg [18] have applied a Tucker3 standardization procedure to improve the prediction results obtained from an ordinary instrument by using the calibration model from a high performance instrument. A calibration model from a FT-Raman instrument and a multi-way calibration and standardization approach leads to improve prediction results from a portable CCD based instrument. Jaworski et al. [19] presents a study on the efficiency of the transferability of several multi-way analytical models using PARAFAC/ILS, DTLD/ILS, and N-PLS applied to voltammetric data and direct standardization (DS) and reverse standardization (RS) for calibration transfer. De Zan et al. [20] have overcome matrix effect in LC–DAD data employing PDS and MCR–ALS. It was demonstrated that the use of signal pre-treatment such as baseline correction and standardization, for compensation of the recovery factor that would be applied when the sample is pre-concentrated, improve the quality of second order chromatographic signals and as consequence the performance of resolution by MCR–ALS. In the present work, piecewise direct standardization (PDS) was used in front face fluorescence spectroscopy of solids when different excipients are used in a pharmaceutical preparation based on a mixture of acetylsalicylic acid (ASA), paracetamol and caffeine. After the calibration transfer procedure, multivariate calibration models were developed using U-PLS for prediction of each active ingredient in the pharmaceutical preparation.
Experimental (a) Estimation of the spectral difference between the reference (primary) situation in relation to spectra obtained in a different (secondary) situation. For this, samples in the two situations are measured. (b) Calculation of the standardization parameters in order to estimate the differences. (c) Validation of the standardization parameters. In the standardization approach, a mathematical model reproduces the spectrum Xp of a sample measured in a reference instrument (primary), from a spectrum Xs, of an identical sample, measured in different equipment or condition (secondary). A widely used standardization method is piecewise direct standardization (PDS) that is based on a moving window that scans across the variable range. A PLS regression is built between each spectral measurement collected on the primary instrument or
Materials The measurements were performed on a Perkin Elmer LS-55 fluorescence spectrometer equipped with a front surface accessory with an incidence angle of the excitation radiation of 22.5°. The conditions of the measurements were: excitation and emission slit widths selected as 7 nm, photomultiplier voltage set to 775 mV and monochromator scan rate at 500 nm/min. The chemicals used were: microcrystalline cellulose (VETEC), soluble starch PA – ACS (Synth), lactose PA – ACS (Nuclear) for excipients and caffeine PA (VETEC), acetylsalicylic acid PA (VETEC) and paracetamol USP (Synth) for actives. For homogenization of the mixtures an analytical mill from IKA, Model A 11, was used. All calculations were performed using the software MATLAB 6.5 and the PLS Toolbox 4.2 from Eigenvector Inc. [21].
313
Julio Cesar L. Alves, R.J. Poppi / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 103 (2013) 311–318
0
Table 1 Values of RMSEC, RMSECV and RMSEP for the calibration models and its validation set #1 and transformed validation set #2 predictions.
1.0
0.20
0.80
0.40
0.60
RMSECV (mg/g)
RMSEP (mg/g) validation set #1
RMSEP (mg/g) transformed validation set #2
3.7 5.0 1.2
7.8 10.5 2.6
5.0 6.7 1.7
8.2 10.9 2.7
Paracetamol (%)
Caffeine (%) 0.60
ASA Paracetamol Caffeine
0.40
0.80 1.0 0
RMSEC (mg/g)
0.20 0
0.20
0.40
0.60
0.80
1.0
ASA (%) Fig. 1. Active ingredients mixture design for calibration (d) and validation ( ) sample sets. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
Procedure In order to obtain the concentrations used commercially for this pharmaceutical preparation, samples were prepared containing mixtures of ASA, caffeine, paracetamol and excipient based on starch/cellulose or lactose with a total weight of 1 g. The amount of excipient was set at 200 mg or 20% of total weight that corresponds to a proportion commonly used in commercial production of this type of mixture [22]. The concentrations of the active ingredients ASA, caffeine and paracetamol were in analytical intervals ranging from 316 to 484 mg/g, 52 to 108 mg/g and 208 to 432 mg/g, respectively, for which a linear behavior was found in the calibration curve obtained in previous work [4] with the same spectral ranges and a PARAFAC model. The central point of these intervals provides a mixture with composition similar to that found in several commercial pharmaceutical preparations: 400 mg/g, 80 mg/g and 320 mg/g for ASA, caffeine and paracetamol, respectively. The amount of active ingredients changes of ±10.5%, ±3.5% and ±14% for ASA, caffeine and paracetamol, respectively, from the central point of its analytical intervals, as illustrated in Fig. 1 which show the mixture design considering the relative amount of each active ingredient in relation to the total amount of active ingredients. The concentrations of the three active ingredients were independently varied and the calibration and validation samples have different concentrations. The samples
were reduced to powder with homogeneous particle size in a mortar and, after that, the mixture was homogenized in an analytical mill for 20 seconds. The measurements were performed with excitation wavelengths ranging from 265 to 405 nm (5 nm intervals) and emission wavelengths from 300 to 480 nm (1 nm intervals). The calibration models for ASA, caffeine and paracetamol were built using 25 samples and five levels of concentration, with five replicates for each level for PLS model and 15 samples and five levels of concentration, with three replicates for each level for U-PLS models. Two sets of fifteen validation samples one of them using 1:1 mixture of starch/cellulose and the other using lactose as excipient were prepared for models validation, with concentrations at five levels with three replications for each level. Standardization was built with 10 samples, five of them containing an excipient based on a 1:1 mixture of starch/cellulose and five containing an excipient based on lactose.
Results and discussion ASA has an intense emission band with a maximum around 335 nm and an excitation maximum at 265 nm. Caffeine presents two bands in this region, one with an emission maximum around 367 nm, which was not used in the model development due to its complete overlap with the ASA band, and another with a strong emission at 400 nm that had an excitation maximum at 355 nm [23]. Paracetamol presents only one emission band around 390 nm with the excitation maximum at 320 nm. A software tool provided by Bro [24] for emission scatter removal was used. In this program, interpolated values are added in the areas from which scatter was removed such that there are no missing values in the fluorescence surface.
intensity (a.u.)
400
300 200
100
260 285 310 450 335 400
360
excitation (nm)
350
385 410
emission (nm)
300
Fig. 2. Excitation and emission spectra of the three active ingredients in the pharmaceutical preparation.
314
Julio Cesar L. Alves, R.J. Poppi / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 103 (2013) 311–318
a
405
700
600
385
excitation (nm)
500 360 400 335 300 310
200
100
285
325
b
350
375
400
425
450
475
0
emission (nm) 405
700
600
385
excitation (nm)
500 360 400 335 300 310
200
100
285
325
c
350
375
400
425
450
475
0
emission (nm) 405
700
600
385
excitation (nm)
500 360 400 335 300 310
200
100
285
325
350
375
400
425
450
475
0
emission (nm) Fig. 3. Surfaces excitation–emission for the same pharmaceutical preparation. (a) Excipient based on the mixture starch/cellulose; (b) excipient based on lactose before standardization; (c) excipient based on lactose after standardization.
PLS modeling Firstly, for simultaneous determination of the three active ingredients in the pharmaceutical preparation using front face
excitation–emission fluorescence, a calibration model based on a PLS algorithm for a fixed excitation wavelength was tested. In this case, the excitation wavelength was 320 nm, a region where it is possible to obtain signals from the three compounds of interest.
Julio Cesar L. Alves, R.J. Poppi / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 103 (2013) 311–318
315
2500 2000
1
1500
Scores on PC 2 (0.23%)
4
5 2
1000 500
6 8
7
10
14
9
12 11
0
13 15
16
−500 17 20
−1000
18 19
25
21
24 22
23
−1500 1.6
1.7
1.8
1.9
2
2.1
2.2
Scores on PC 1 (99.73%)
2.3
2.4
2.5 x 104
Fig. 4. First versus second PCA scores, where samples used in standardization procedure (3, 10, 13, 19 and 23) are emphasized.
Fig. 2 shows the excitation and emission spectra of the three active ingredients in the pharmaceutical preparation considering the medium level of concentration used in this work. In the development of the PLS model the preprocessing methods: mean-centering, baseline correction and standard normal variate (SNV) were tested and the number of latent variables was defined by a leave-one-out cross-validation procedure which allow choose the number of latent variables that gives a minimal prediction residual sum of squares (PRESS) [25]. The final model was mean centered and three latent variables where used, explaining 99.31% of the data variance in X. The prediction values were calculated as the average of the three replicates in each level, where the results for several samples presented relative errors near or above 10%. The RMSEC and RMSEP values were 12.6 and 29.8 mg/g, 16.8 and 40.3 mg/g, 4.2 and 9.8 mg/g for ASA, paracetamol and caffeine, respectively. Also, the relative standard deviation (RSD) was calculated, resulting in values up to 14%, characterizing a method with poor precision. These results indicate that, in this specific case, for dealing with a complex mixture in front face fluorescence spectroscopy, with strong spectral overlapping and low emission intensity, as can be seen in Fig. 2, measurements at only one fixed excitation wavelength do not supply enough information to develop a first order multivariate calibration model in practice. So, in order to obtain a good level of information to build a calibration model that provides better results in terms of precision and accuracy of prediction a second order calibration was used. Now, a whole surface of the excitation–emission fluorescence spectra for each sample was employed in model development. U-PLS modeling PARAFAC has also been used in this quantification with excellent results [4], but not in situations of different composition of excipients in the formulations. When PARAFAC was applied in samples with different excipients in calibration and validation, a displacement, change in fluorescence intensity or band profile takes place, no calibration was possible and very poor prediction results were obtained, with relative errors above 20% for the three compounds. Then, U-PLS was chosen as the second order multivariate calibration method, since it has been demonstrated that it can produce suitable models to be used in practice and a standardization procedure to correct the matrix effect can be implemented. In previous studies [26] for simultaneous determination of the active
ingredients without changes in excipients, a statistical comparison of the prediction results using PARAFAC and U-PLS was performed using an F-test and the obtained RMSEP values and the U-PLS model provided better results for ASA and caffeine. The critical F14,14 value at 99% confidence level is 3.66 and the calculated F-values for ASA, paracetamol and caffeine are 7.83, 3.50 and 13.29, respectively. For the determination of the concentrations of active ingredients using U-PLS, from an excitation–emission fluorescence surface, it was necessary to unfold the three-dimensional array of data (I J K) for attainment of the matrix of data where I is the number of samples and J and K are the numbers of variables in the second (excitation) and third (emission) dimension, respectively, generating a matrix of dimensions I JK. The same samples used in the PLS calculations showed early were employed in the calibration set for development of the UPLS models with excipient based on a mixture of starch/cellulose. The validation set, now called #1, was formed from the same fifteen samples previously used, also with excipient based on the mixture of starch/cellulose. In the construction of the U-PLS models the data set was mean centered and cross-validation (leave one out) was used to find the best number of latent variables. Four latent variables, that explained 99.34% of data variance in X, were selected based on the same criterion used for PLS modeling. The number of latent variables has increased probably due the high overlapping of the excitation–emission spectra. However the increase in spectral data leads to better model accuracy and good prediction results were obtained with close values of RMSEC and RMSEP for each compound suggesting a good model fit. The values of RMSEC, RMSECV and RMSEP for the calibration models and its predictions for the validation set #1 are shown in Table 1. In these models the relative errors were below 3% and the RSD were below to 2%. The U-PLS models provided better results than PLS with a fixed excitation wavelength and it can be used to predict the active ingredients of the pharmaceutical preparation with good reproducibility and accuracy. Standardization The effect of excipient on the excitation–emission fluorescence surface is presented in Fig. 3a and b where, for the same pharmaceutical mixture, different fluorescence signals for each excipient are observed. These results indicate the necessity of a standardization strategy to predict appropriately with different types of excipient.
316
Julio Cesar L. Alves, R.J. Poppi / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 103 (2013) 311–318
a
500
ASA predicted concentrations (mg/g)
validation set #1 validation set #2 transformed validation set #2 450
400
350
300
250 250
300
350
400
450
500
ASA actual concentrations (mg/g)
b paracetamol predicted concentrations (mg/g)
550
500
450
400
350
300
250
validation set #1 validation set #2 transformed validation set #2 250
300
350
400
450
500
550
paracetamol actual concentrations (mg/g)
c
110
caffeine predicted concentratios (mg/g)
100
validation set #1 validation set #2 transformed validation set #2
90
80 70
60 50
40 30 30
40
50
60
70
80
90
100
110
caffeine actual concentrations (mg/g) Fig. 5. Actual concentrations versus predicted concentrations for: (a) ASA, (b) paracetamol and (c) caffeine.
Another validation set, named #2, was formed with fifteen samples in five different levels, with identical concentration to those used in validation set #1, but using an excipient based on the lactose.
The five transference samples (with excipient starch/cellulose) had been selected from 25 calibration samples used in the first order calibration, based on their distribution in the first two principal component spaces, as illustrated in Fig. 4. The five transference
317
Julio Cesar L. Alves, R.J. Poppi / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 103 (2013) 311–318
ASA
40 set #1
Caffeine transformed validation set #2
set #2
30
Paracetamol
set #2
set #1
transformed validation set #2
transformed validation set #2
set #2
set #1
relative error (%)
20
10
0
−10
−20
−30
−40 0
5
10
15
20
25
30
35
40
45
sample number Fig. 6. Relative errors for validation samples. Samples 1–5: ASA set #1; samples 6–10: ASA set #2; samples 11–15: ASA transformed set #2; samples 16–20: caffeine set #1; samples 21–25: caffeine set #2; samples 26–30: caffeine transformed set #2; samples 31–35: paracetamol set #1; samples 36–40: paracetamol set #2; samples 41–45: paracetamol transformed set #2.
samples were selected so that they spanned all of the PC space and in order to select representative samples of each level of concentration the approach used has considered that PC space discriminate five groups related to different concentration levels and each one with five samples, the samples nearest the central point of each group were selected. Also five samples were prepared with the same concentrations as the transference samples, using now the excipient based on lactose. Fig. 3c shows the excitation–emission surface with excipient based on lactose after standardization using the PDS method. A great similarity is observed between Fig. 3a and c, indicating the success of the standardization strategy. The calibration set was used to develop independent U-PLS models for each active ingredient in the pharmaceutical preparation. These models had been used for prediction of the samples of validation set #2 without standardization and the transformed validation set #2 (after standardization). The U-PLS models produced excellent results for validation set #1 and for transformed validation set #2, with relative errors that do not exceed 5%, while for the validation set #2 without standardization, the errors are above 20%. Also predictions of validation set #2 before transformation present systematic behavior, with all negative or positive errors. The values of RMSEC, RMSECV and RMSEP for the calibration models and its predictions for the transformed validation set #2 are shown in Table 1. For each compound, it is possible to observe the similarities of the RMSEP values from validation set #1 and RMSEP values from transformed validation set #2, showing that the standardization procedure has solved the problem of matrix effects due to different excipients. A F-test was calculated using the RMSEP values of these two validation sets. The critical F14,14 value at 99% confidence level is 3.66 and the calculated F-values for ASA, paracetamol and caffeine are 2.69, 2.65 and 2.52, respectively. There is statistical evidence at 99% confidence level that similar results are obtained. The results are illustrated in Figs. 5 and 6. Fig. 5 shows the actual concentrations versus predicted concentrations for ASA, paracetamol and caffeine considering the validation set #1, validation set #2 and transformed validation set #2, providing an adequate visualization of the good model fit and of the PDS application. The upper and lower predictions values for the
replicates are show by an error bar. Fig. 6 presents the relative errors for samples of validation set #1, validation set #2 and transformed validation set #2. Conclusion An analytical methodology was developed to perform the determination of the active ingredients in the front face fluorescence that is fast and practically without sample preparation. The use of the standardization procedure based on the PDS algorithm and U-PLS using the whole excitation–emission data provided better results than PLS with a fixed excitation wavelength. It makes possible to overcome the problem of matrix effects due to different excipients that is a critical step when developing a model using front face fluorescence of solids. Conclusive evidence was found that the prediction results for two sets of validation with different excipients are identical, even when using the calibration model developed with one of the excipients. Acknowledgements The authors thank CNPq (Conselho Nacional de Desenvolvimento Científico e Tecnológico), CAPES (Coordenação de Aperfeiçoamento de Pessoal de Nível Superior) and FAPESP (Fundação de Amparo à Pesquisa do Estado de São Paulo), for financial support and fellowships. References [1] British Pharmacopeia, vol. 2. HMSO, London, 2002. [2] A.B. Moreira, I.L.T. Dias, G. Oliveira Neto, E.A.G. Zagatto, M.M.C. Ferreira, L.T. Kubota, Talanta 67 (2005) 65–69. [3] J.R. Lakowicz, Principles of Fluorescence, third ed., Springer, New York, 2006. [4] J.C.L. Alves, R.J. Poppi, Anal. Chim. Acta 642 (2009) 212–216. [5] G.M. Escandar, N.M. Faber, H.C. Goicoechea, A.M. de la Pena, A.C. Olivieri, R.J. Poppi, Trends Anal. Chem. 26 (2007) 725–765. [6] S. Wold, P. Geladi, K. Esbensem, J. Öhman, J. Chemom. 1 (1987) 41–56. [7] A. Espinosa-Mansilla, A.M. de La Peña, F. Cañada-Cañada, D.G. Gómez, Anal. Biochem. 347 (2005) 275–286. [8] A. García-Reiriz, P.C. Damiani, A.C. Olivieri, Talanta 71 (2007) 806–815.
318
Julio Cesar L. Alves, R.J. Poppi / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 103 (2013) 311–318
[9] M.J. Culzoni, H.C. Goicoechea, A.P. Pagani, M.A. Cabezón, A.C. Olivieri, Analyst 131 (2006) 718–723. [10] A.B. Moreira, H.P. Oliveira, T.D.Z. Atvars, I.L. Dias, G. Oliveira Neto, E.A.G. Zagatto, L.T. Kubota, Anal. Chim. Acta 539 (2005) 257–261. [11] R. Bro, Chemom. Intell. Lab. Syst. 38 (1997) 149–171. [12] Y. Wang, D.J. Veltkamp, B.R. Kowalski, Anal. Chem. 63 (1991) 2750–2756. [13] F.D. Barbosa, R.J. Poppi, Anal. Bioanal. Chem. 377 (2003) 695–701. [14] O.E. de Noord, Chemom. Intell. Lab. Syst. 25 (1994) 85–97. [15] J. Lin, Appl. Spectrosc. 52 (1998) 1591–1596. [16] E. Bouveresse, D.L. Massart, Chemom. Intell. Lab. Syst. 32 (1996) 201–213. [17] Y. Wang, B.R. Kowalski, Appl. Spectrosc. 46 (1992) 764–771. [18] M. Kompany-Zareh, F. van den Berg, Analyst 135 (2010) 1382–1388. [19] A. Jaworski, H. Wikiel, K. Wikiel, Anal. Chim. Acta 656 (2009) 42–50. [20] M.M. De Zan, M.D.G. García, M.J. Culzoni, R.G. Siano, H.C. Goicoechea, M.M. Galera, J. Chromatogr. A 1179 (2008) 106–114.
[21] B.M. Wise, N.B. Gallagher, R. Bro, J.M. Shaver, W. Windig, R.S. Koch, PLS Toolbox Version 4.2 for Use with MATLAB. Eingenvector Research, Manson, 2005. [22] A. Wade, P.J. Weller, Handbook of Pharmaceutical Excipients, second ed., The Pharmaceutical Press, London, 1994. [23] D.D. McKemy, W. Welch, J.A. Airey, Cell Calcium 27 (2000) 117–124. [24] M. Bahram, R. Bro, C. Stedmon, A. Afkhami, J. Chemom. 20 (2006) 99–105. [25] P. Geladi, B.R. Kowalski, Anal. Chim. Acta 185 (1986) 1–17. [26] J.C.L. Alves, Aplicação de métodos quimiométricos de calibração de segunda ordem e transferência de calibração na determinação simultânea de misturas de fármacos utilizando espectroscopia de fluorescência molecular em fase sólida, M.Sc. Dissertation, Universidade Estadual de Campinas, Campinas, 2009.
Contents lists available at SciVerse ScienceDirect
Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy journal homepage: www.elsevier.com/locate/saa
Pharmaceutical analysis in solids using front face fluorescence spectroscopy and multivariate calibration with matrix correction by piecewise direct standardization Julio Cesar L. Alves ⇑, Ronei J. Poppi Institute of Chemistry, State University of Campinas – UNICAMP, P.O. Box 6154, 13083-970 Campinas, SP, Brazil
h i g h l i g h t s
g r a p h i c a l a b s t r a c t
" Determination of pharmaceutical
formulation using EEM data. " Spectra standardization to solving
matrix effects in EEM data. " Accurate results with relative errors
lesser than 5% were obtained.
a r t i c l e
i n f o
Article history: Received 28 August 2012 Received in revised form 29 October 2012 Accepted 31 October 2012 Available online 7 November 2012 Keywords: Front face fluorescence spectroscopy Calibration transfer Piecewise direct standardization Acetylsalicylic acid Paracetamol Caffeine
a b s t r a c t This paper reports the application of piecewise direct standardization (PDS) for matrix correction in front face fluorescence spectroscopy of solids when different excipients are used in a pharmaceutical preparation based on a mixture of acetylsalicylic acid (ASA), paracetamol (acetaminophen) and caffeine. As verified in earlier studies, the use of different excipients and their ratio can cause a displacement, change in fluorescence intensity or band profile. To overcome this important drawback, a standardization strategy was adopted to convert all the excitation–emission fluorescence spectra into those used for model development. An excitation–emission matrix (EEM) for which excitation and emission wavelengths ranging from 265 to 405 nm and 300 to 480 nm, respectively, was used. Excellent results were obtained using unfolded partial least squares (U-PLS), with RMSEP values of 8.2 mg/g, 10.9 mg/g and 2.7 mg/g for ASA, paracetamol and caffeine, respectively, and with relative errors lesser than 5% for the three analytes. Ó 2012 Elsevier B.V. All rights reserved.
Introduction A mixture of the active ingredients acetylsalicylic acid (ASA), caffeine and paracetamol is a widely used pharmaceutical combination with analgesic effects. The most common analytical methods used for determination of this mixture of compounds include extraction followed by UV spectrophotometric analysis and high performance liquid chromatography [1]. These techniques, while perfectly adequate in terms of their accuracy, can be time-consuming ⇑ Corresponding author. Tel.: +55 11 9 9612 7438. E-mail address: [email protected] (Julio Cesar L. Alves). 1386-1425/$ - see front matter Ó 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.saa.2012.10.074
or expensive when implemented in routine quality control laboratories. In recent years, excitation–emission molecular fluorescence spectroscopy has been used in conjunction with chemometric methods for analysis of pharmaceutical formulations [2]. It can be an excellent alternative for analysis of these mixtures, mainly when the fluorescence spectra can be obtained directly from the solid using a front face accessory [3], making it possible to perform the analysis in a few minutes, practically without sample preparation. The front face fluorescence spectra of a solid pharmaceutical preparation containing ASA, caffeine and paracetamol present overlapping bands from these compounds. Because of this, the
312
Julio Cesar L. Alves, R.J. Poppi / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 103 (2013) 311–318
use of univariate calibration is not the best choice for accurate prediction, as previously verified [4], making the use of higher order calibration procedures necessary. The data obtained from excitation–emission fluorescence spectroscopy can be arranged in an I J K three-way array, where the index I refers to the samples, J to the excitation wavelengths and K to the emission wavelengths. Second order calibration methods can be applied to these types of data providing advantages over methods of zero order or first order, since more information is used in model development. There are several algorithms developed to deal with second-order data and development of multi-way models, such as PARAFAC, BLLS, MCR–ALS and NPLS [5]. An alternative to work with this kind of data is to rearrange them into a matrix and apply a first-order algorithm, leading to unfolded-principal component regression (UPCR) and unfolded-partial least squares (U-PLS) [6]. These approaches were first used to deal with second-order data, before true second-order methods were developed, but they are still able to produce satisfactory results [7] and frequently combined with residual bilinearization (RBL) to achieve the second order advantage [8,9]. In model development using multivariate calibration, as already demonstrated in front face fluorescence of solids [2,10], different bulk proportions of the mixture components and their physical properties (such as particle size) and the use of different ingredients (such as different excipients) produce changes in the signals from the active ingredients. In fact, the use of front face fluorescence of solids depends on the amount and type of ingredients used in the pharmaceutical formulation. The effects of the ingredients and their proportions used in the mixtures on the fluorescence can occur due to competing deactivation processes resulting from the specific interaction between the fluorophore and the ingredients present in the sample. The most significant changes observed in these cases are the variation on signal intensities. For this reason, only samples prepared in the laboratory were used in this study. However, in the pharmaceutical industry the ingredients and their proportions are known and the proposed methodology could be applied without restrictions. In this situation, the utilization of PARAFAC [11], applicable even in the presence of unknown interferences and with spectral overlap of the components in the mixture, and already used in pharmaceutical formulations using solid-phase molecular fluorescence [4], is not appropriate. PARAFAC is not able to correct the matrix effects, since in this case the interferences are not caused by a new excipient. The use of different excipients and their ratio can cause a displacement, change in fluorescence intensity or band profile. To overcome these problems, spectra standardization [12– 14] can be applied to convert an excitation–emission matrix into a spectral surface free of this matrix effect. This procedure involves three main steps:
condition Xp and the corresponding spectral window wi containing measurements at a few nearby wavelengths obtained on the secondary instrument or condition. The regression coefficients bi are placed along the diagonal of the transfer matrix F which describes the relation between the spectra from the primary and the secondary instrument or condition. In order to transfer the spectra collected on the secondary instrument or condition, the matrix Xs is multiplied by the transfer matrix F, yielding Xstd:
Xstd ¼ Xs F
ð1Þ
which can then be used for prediction by using calibration models developed on the primary instrument or condition:
ystd ¼ ðXs FÞb
ð2Þ
where ystd is the predicted values obtained with the spectra collected on the secondary instrument or condition after standardization and b is the regression coefficients determined on the primary instrument or condition [15–17]. There are a few recent articles dealing with the joint use of multi-way models and calibration transfer, none of them related to spectrofluorimetric data. Kompany-Zareh and van den Berg [18] have applied a Tucker3 standardization procedure to improve the prediction results obtained from an ordinary instrument by using the calibration model from a high performance instrument. A calibration model from a FT-Raman instrument and a multi-way calibration and standardization approach leads to improve prediction results from a portable CCD based instrument. Jaworski et al. [19] presents a study on the efficiency of the transferability of several multi-way analytical models using PARAFAC/ILS, DTLD/ILS, and N-PLS applied to voltammetric data and direct standardization (DS) and reverse standardization (RS) for calibration transfer. De Zan et al. [20] have overcome matrix effect in LC–DAD data employing PDS and MCR–ALS. It was demonstrated that the use of signal pre-treatment such as baseline correction and standardization, for compensation of the recovery factor that would be applied when the sample is pre-concentrated, improve the quality of second order chromatographic signals and as consequence the performance of resolution by MCR–ALS. In the present work, piecewise direct standardization (PDS) was used in front face fluorescence spectroscopy of solids when different excipients are used in a pharmaceutical preparation based on a mixture of acetylsalicylic acid (ASA), paracetamol and caffeine. After the calibration transfer procedure, multivariate calibration models were developed using U-PLS for prediction of each active ingredient in the pharmaceutical preparation.
Experimental (a) Estimation of the spectral difference between the reference (primary) situation in relation to spectra obtained in a different (secondary) situation. For this, samples in the two situations are measured. (b) Calculation of the standardization parameters in order to estimate the differences. (c) Validation of the standardization parameters. In the standardization approach, a mathematical model reproduces the spectrum Xp of a sample measured in a reference instrument (primary), from a spectrum Xs, of an identical sample, measured in different equipment or condition (secondary). A widely used standardization method is piecewise direct standardization (PDS) that is based on a moving window that scans across the variable range. A PLS regression is built between each spectral measurement collected on the primary instrument or
Materials The measurements were performed on a Perkin Elmer LS-55 fluorescence spectrometer equipped with a front surface accessory with an incidence angle of the excitation radiation of 22.5°. The conditions of the measurements were: excitation and emission slit widths selected as 7 nm, photomultiplier voltage set to 775 mV and monochromator scan rate at 500 nm/min. The chemicals used were: microcrystalline cellulose (VETEC), soluble starch PA – ACS (Synth), lactose PA – ACS (Nuclear) for excipients and caffeine PA (VETEC), acetylsalicylic acid PA (VETEC) and paracetamol USP (Synth) for actives. For homogenization of the mixtures an analytical mill from IKA, Model A 11, was used. All calculations were performed using the software MATLAB 6.5 and the PLS Toolbox 4.2 from Eigenvector Inc. [21].
313
Julio Cesar L. Alves, R.J. Poppi / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 103 (2013) 311–318
0
Table 1 Values of RMSEC, RMSECV and RMSEP for the calibration models and its validation set #1 and transformed validation set #2 predictions.
1.0
0.20
0.80
0.40
0.60
RMSECV (mg/g)
RMSEP (mg/g) validation set #1
RMSEP (mg/g) transformed validation set #2
3.7 5.0 1.2
7.8 10.5 2.6
5.0 6.7 1.7
8.2 10.9 2.7
Paracetamol (%)
Caffeine (%) 0.60
ASA Paracetamol Caffeine
0.40
0.80 1.0 0
RMSEC (mg/g)
0.20 0
0.20
0.40
0.60
0.80
1.0
ASA (%) Fig. 1. Active ingredients mixture design for calibration (d) and validation ( ) sample sets. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
Procedure In order to obtain the concentrations used commercially for this pharmaceutical preparation, samples were prepared containing mixtures of ASA, caffeine, paracetamol and excipient based on starch/cellulose or lactose with a total weight of 1 g. The amount of excipient was set at 200 mg or 20% of total weight that corresponds to a proportion commonly used in commercial production of this type of mixture [22]. The concentrations of the active ingredients ASA, caffeine and paracetamol were in analytical intervals ranging from 316 to 484 mg/g, 52 to 108 mg/g and 208 to 432 mg/g, respectively, for which a linear behavior was found in the calibration curve obtained in previous work [4] with the same spectral ranges and a PARAFAC model. The central point of these intervals provides a mixture with composition similar to that found in several commercial pharmaceutical preparations: 400 mg/g, 80 mg/g and 320 mg/g for ASA, caffeine and paracetamol, respectively. The amount of active ingredients changes of ±10.5%, ±3.5% and ±14% for ASA, caffeine and paracetamol, respectively, from the central point of its analytical intervals, as illustrated in Fig. 1 which show the mixture design considering the relative amount of each active ingredient in relation to the total amount of active ingredients. The concentrations of the three active ingredients were independently varied and the calibration and validation samples have different concentrations. The samples
were reduced to powder with homogeneous particle size in a mortar and, after that, the mixture was homogenized in an analytical mill for 20 seconds. The measurements were performed with excitation wavelengths ranging from 265 to 405 nm (5 nm intervals) and emission wavelengths from 300 to 480 nm (1 nm intervals). The calibration models for ASA, caffeine and paracetamol were built using 25 samples and five levels of concentration, with five replicates for each level for PLS model and 15 samples and five levels of concentration, with three replicates for each level for U-PLS models. Two sets of fifteen validation samples one of them using 1:1 mixture of starch/cellulose and the other using lactose as excipient were prepared for models validation, with concentrations at five levels with three replications for each level. Standardization was built with 10 samples, five of them containing an excipient based on a 1:1 mixture of starch/cellulose and five containing an excipient based on lactose.
Results and discussion ASA has an intense emission band with a maximum around 335 nm and an excitation maximum at 265 nm. Caffeine presents two bands in this region, one with an emission maximum around 367 nm, which was not used in the model development due to its complete overlap with the ASA band, and another with a strong emission at 400 nm that had an excitation maximum at 355 nm [23]. Paracetamol presents only one emission band around 390 nm with the excitation maximum at 320 nm. A software tool provided by Bro [24] for emission scatter removal was used. In this program, interpolated values are added in the areas from which scatter was removed such that there are no missing values in the fluorescence surface.
intensity (a.u.)
400
300 200
100
260 285 310 450 335 400
360
excitation (nm)
350
385 410
emission (nm)
300
Fig. 2. Excitation and emission spectra of the three active ingredients in the pharmaceutical preparation.
314
Julio Cesar L. Alves, R.J. Poppi / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 103 (2013) 311–318
a
405
700
600
385
excitation (nm)
500 360 400 335 300 310
200
100
285
325
b
350
375
400
425
450
475
0
emission (nm) 405
700
600
385
excitation (nm)
500 360 400 335 300 310
200
100
285
325
c
350
375
400
425
450
475
0
emission (nm) 405
700
600
385
excitation (nm)
500 360 400 335 300 310
200
100
285
325
350
375
400
425
450
475
0
emission (nm) Fig. 3. Surfaces excitation–emission for the same pharmaceutical preparation. (a) Excipient based on the mixture starch/cellulose; (b) excipient based on lactose before standardization; (c) excipient based on lactose after standardization.
PLS modeling Firstly, for simultaneous determination of the three active ingredients in the pharmaceutical preparation using front face
excitation–emission fluorescence, a calibration model based on a PLS algorithm for a fixed excitation wavelength was tested. In this case, the excitation wavelength was 320 nm, a region where it is possible to obtain signals from the three compounds of interest.
Julio Cesar L. Alves, R.J. Poppi / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 103 (2013) 311–318
315
2500 2000
1
1500
Scores on PC 2 (0.23%)
4
5 2
1000 500
6 8
7
10
14
9
12 11
0
13 15
16
−500 17 20
−1000
18 19
25
21
24 22
23
−1500 1.6
1.7
1.8
1.9
2
2.1
2.2
Scores on PC 1 (99.73%)
2.3
2.4
2.5 x 104
Fig. 4. First versus second PCA scores, where samples used in standardization procedure (3, 10, 13, 19 and 23) are emphasized.
Fig. 2 shows the excitation and emission spectra of the three active ingredients in the pharmaceutical preparation considering the medium level of concentration used in this work. In the development of the PLS model the preprocessing methods: mean-centering, baseline correction and standard normal variate (SNV) were tested and the number of latent variables was defined by a leave-one-out cross-validation procedure which allow choose the number of latent variables that gives a minimal prediction residual sum of squares (PRESS) [25]. The final model was mean centered and three latent variables where used, explaining 99.31% of the data variance in X. The prediction values were calculated as the average of the three replicates in each level, where the results for several samples presented relative errors near or above 10%. The RMSEC and RMSEP values were 12.6 and 29.8 mg/g, 16.8 and 40.3 mg/g, 4.2 and 9.8 mg/g for ASA, paracetamol and caffeine, respectively. Also, the relative standard deviation (RSD) was calculated, resulting in values up to 14%, characterizing a method with poor precision. These results indicate that, in this specific case, for dealing with a complex mixture in front face fluorescence spectroscopy, with strong spectral overlapping and low emission intensity, as can be seen in Fig. 2, measurements at only one fixed excitation wavelength do not supply enough information to develop a first order multivariate calibration model in practice. So, in order to obtain a good level of information to build a calibration model that provides better results in terms of precision and accuracy of prediction a second order calibration was used. Now, a whole surface of the excitation–emission fluorescence spectra for each sample was employed in model development. U-PLS modeling PARAFAC has also been used in this quantification with excellent results [4], but not in situations of different composition of excipients in the formulations. When PARAFAC was applied in samples with different excipients in calibration and validation, a displacement, change in fluorescence intensity or band profile takes place, no calibration was possible and very poor prediction results were obtained, with relative errors above 20% for the three compounds. Then, U-PLS was chosen as the second order multivariate calibration method, since it has been demonstrated that it can produce suitable models to be used in practice and a standardization procedure to correct the matrix effect can be implemented. In previous studies [26] for simultaneous determination of the active
ingredients without changes in excipients, a statistical comparison of the prediction results using PARAFAC and U-PLS was performed using an F-test and the obtained RMSEP values and the U-PLS model provided better results for ASA and caffeine. The critical F14,14 value at 99% confidence level is 3.66 and the calculated F-values for ASA, paracetamol and caffeine are 7.83, 3.50 and 13.29, respectively. For the determination of the concentrations of active ingredients using U-PLS, from an excitation–emission fluorescence surface, it was necessary to unfold the three-dimensional array of data (I J K) for attainment of the matrix of data where I is the number of samples and J and K are the numbers of variables in the second (excitation) and third (emission) dimension, respectively, generating a matrix of dimensions I JK. The same samples used in the PLS calculations showed early were employed in the calibration set for development of the UPLS models with excipient based on a mixture of starch/cellulose. The validation set, now called #1, was formed from the same fifteen samples previously used, also with excipient based on the mixture of starch/cellulose. In the construction of the U-PLS models the data set was mean centered and cross-validation (leave one out) was used to find the best number of latent variables. Four latent variables, that explained 99.34% of data variance in X, were selected based on the same criterion used for PLS modeling. The number of latent variables has increased probably due the high overlapping of the excitation–emission spectra. However the increase in spectral data leads to better model accuracy and good prediction results were obtained with close values of RMSEC and RMSEP for each compound suggesting a good model fit. The values of RMSEC, RMSECV and RMSEP for the calibration models and its predictions for the validation set #1 are shown in Table 1. In these models the relative errors were below 3% and the RSD were below to 2%. The U-PLS models provided better results than PLS with a fixed excitation wavelength and it can be used to predict the active ingredients of the pharmaceutical preparation with good reproducibility and accuracy. Standardization The effect of excipient on the excitation–emission fluorescence surface is presented in Fig. 3a and b where, for the same pharmaceutical mixture, different fluorescence signals for each excipient are observed. These results indicate the necessity of a standardization strategy to predict appropriately with different types of excipient.
316
Julio Cesar L. Alves, R.J. Poppi / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 103 (2013) 311–318
a
500
ASA predicted concentrations (mg/g)
validation set #1 validation set #2 transformed validation set #2 450
400
350
300
250 250
300
350
400
450
500
ASA actual concentrations (mg/g)
b paracetamol predicted concentrations (mg/g)
550
500
450
400
350
300
250
validation set #1 validation set #2 transformed validation set #2 250
300
350
400
450
500
550
paracetamol actual concentrations (mg/g)
c
110
caffeine predicted concentratios (mg/g)
100
validation set #1 validation set #2 transformed validation set #2
90
80 70
60 50
40 30 30
40
50
60
70
80
90
100
110
caffeine actual concentrations (mg/g) Fig. 5. Actual concentrations versus predicted concentrations for: (a) ASA, (b) paracetamol and (c) caffeine.
Another validation set, named #2, was formed with fifteen samples in five different levels, with identical concentration to those used in validation set #1, but using an excipient based on the lactose.
The five transference samples (with excipient starch/cellulose) had been selected from 25 calibration samples used in the first order calibration, based on their distribution in the first two principal component spaces, as illustrated in Fig. 4. The five transference
317
Julio Cesar L. Alves, R.J. Poppi / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 103 (2013) 311–318
ASA
40 set #1
Caffeine transformed validation set #2
set #2
30
Paracetamol
set #2
set #1
transformed validation set #2
transformed validation set #2
set #2
set #1
relative error (%)
20
10
0
−10
−20
−30
−40 0
5
10
15
20
25
30
35
40
45
sample number Fig. 6. Relative errors for validation samples. Samples 1–5: ASA set #1; samples 6–10: ASA set #2; samples 11–15: ASA transformed set #2; samples 16–20: caffeine set #1; samples 21–25: caffeine set #2; samples 26–30: caffeine transformed set #2; samples 31–35: paracetamol set #1; samples 36–40: paracetamol set #2; samples 41–45: paracetamol transformed set #2.
samples were selected so that they spanned all of the PC space and in order to select representative samples of each level of concentration the approach used has considered that PC space discriminate five groups related to different concentration levels and each one with five samples, the samples nearest the central point of each group were selected. Also five samples were prepared with the same concentrations as the transference samples, using now the excipient based on lactose. Fig. 3c shows the excitation–emission surface with excipient based on lactose after standardization using the PDS method. A great similarity is observed between Fig. 3a and c, indicating the success of the standardization strategy. The calibration set was used to develop independent U-PLS models for each active ingredient in the pharmaceutical preparation. These models had been used for prediction of the samples of validation set #2 without standardization and the transformed validation set #2 (after standardization). The U-PLS models produced excellent results for validation set #1 and for transformed validation set #2, with relative errors that do not exceed 5%, while for the validation set #2 without standardization, the errors are above 20%. Also predictions of validation set #2 before transformation present systematic behavior, with all negative or positive errors. The values of RMSEC, RMSECV and RMSEP for the calibration models and its predictions for the transformed validation set #2 are shown in Table 1. For each compound, it is possible to observe the similarities of the RMSEP values from validation set #1 and RMSEP values from transformed validation set #2, showing that the standardization procedure has solved the problem of matrix effects due to different excipients. A F-test was calculated using the RMSEP values of these two validation sets. The critical F14,14 value at 99% confidence level is 3.66 and the calculated F-values for ASA, paracetamol and caffeine are 2.69, 2.65 and 2.52, respectively. There is statistical evidence at 99% confidence level that similar results are obtained. The results are illustrated in Figs. 5 and 6. Fig. 5 shows the actual concentrations versus predicted concentrations for ASA, paracetamol and caffeine considering the validation set #1, validation set #2 and transformed validation set #2, providing an adequate visualization of the good model fit and of the PDS application. The upper and lower predictions values for the
replicates are show by an error bar. Fig. 6 presents the relative errors for samples of validation set #1, validation set #2 and transformed validation set #2. Conclusion An analytical methodology was developed to perform the determination of the active ingredients in the front face fluorescence that is fast and practically without sample preparation. The use of the standardization procedure based on the PDS algorithm and U-PLS using the whole excitation–emission data provided better results than PLS with a fixed excitation wavelength. It makes possible to overcome the problem of matrix effects due to different excipients that is a critical step when developing a model using front face fluorescence of solids. Conclusive evidence was found that the prediction results for two sets of validation with different excipients are identical, even when using the calibration model developed with one of the excipients. Acknowledgements The authors thank CNPq (Conselho Nacional de Desenvolvimento Científico e Tecnológico), CAPES (Coordenação de Aperfeiçoamento de Pessoal de Nível Superior) and FAPESP (Fundação de Amparo à Pesquisa do Estado de São Paulo), for financial support and fellowships. References [1] British Pharmacopeia, vol. 2. HMSO, London, 2002. [2] A.B. Moreira, I.L.T. Dias, G. Oliveira Neto, E.A.G. Zagatto, M.M.C. Ferreira, L.T. Kubota, Talanta 67 (2005) 65–69. [3] J.R. Lakowicz, Principles of Fluorescence, third ed., Springer, New York, 2006. [4] J.C.L. Alves, R.J. Poppi, Anal. Chim. Acta 642 (2009) 212–216. [5] G.M. Escandar, N.M. Faber, H.C. Goicoechea, A.M. de la Pena, A.C. Olivieri, R.J. Poppi, Trends Anal. Chem. 26 (2007) 725–765. [6] S. Wold, P. Geladi, K. Esbensem, J. Öhman, J. Chemom. 1 (1987) 41–56. [7] A. Espinosa-Mansilla, A.M. de La Peña, F. Cañada-Cañada, D.G. Gómez, Anal. Biochem. 347 (2005) 275–286. [8] A. García-Reiriz, P.C. Damiani, A.C. Olivieri, Talanta 71 (2007) 806–815.
318
Julio Cesar L. Alves, R.J. Poppi / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 103 (2013) 311–318
[9] M.J. Culzoni, H.C. Goicoechea, A.P. Pagani, M.A. Cabezón, A.C. Olivieri, Analyst 131 (2006) 718–723. [10] A.B. Moreira, H.P. Oliveira, T.D.Z. Atvars, I.L. Dias, G. Oliveira Neto, E.A.G. Zagatto, L.T. Kubota, Anal. Chim. Acta 539 (2005) 257–261. [11] R. Bro, Chemom. Intell. Lab. Syst. 38 (1997) 149–171. [12] Y. Wang, D.J. Veltkamp, B.R. Kowalski, Anal. Chem. 63 (1991) 2750–2756. [13] F.D. Barbosa, R.J. Poppi, Anal. Bioanal. Chem. 377 (2003) 695–701. [14] O.E. de Noord, Chemom. Intell. Lab. Syst. 25 (1994) 85–97. [15] J. Lin, Appl. Spectrosc. 52 (1998) 1591–1596. [16] E. Bouveresse, D.L. Massart, Chemom. Intell. Lab. Syst. 32 (1996) 201–213. [17] Y. Wang, B.R. Kowalski, Appl. Spectrosc. 46 (1992) 764–771. [18] M. Kompany-Zareh, F. van den Berg, Analyst 135 (2010) 1382–1388. [19] A. Jaworski, H. Wikiel, K. Wikiel, Anal. Chim. Acta 656 (2009) 42–50. [20] M.M. De Zan, M.D.G. García, M.J. Culzoni, R.G. Siano, H.C. Goicoechea, M.M. Galera, J. Chromatogr. A 1179 (2008) 106–114.
[21] B.M. Wise, N.B. Gallagher, R. Bro, J.M. Shaver, W. Windig, R.S. Koch, PLS Toolbox Version 4.2 for Use with MATLAB. Eingenvector Research, Manson, 2005. [22] A. Wade, P.J. Weller, Handbook of Pharmaceutical Excipients, second ed., The Pharmaceutical Press, London, 1994. [23] D.D. McKemy, W. Welch, J.A. Airey, Cell Calcium 27 (2000) 117–124. [24] M. Bahram, R. Bro, C. Stedmon, A. Afkhami, J. Chemom. 20 (2006) 99–105. [25] P. Geladi, B.R. Kowalski, Anal. Chim. Acta 185 (1986) 1–17. [26] J.C.L. Alves, Aplicação de métodos quimiométricos de calibração de segunda ordem e transferência de calibração na determinação simultânea de misturas de fármacos utilizando espectroscopia de fluorescência molecular em fase sólida, M.Sc. Dissertation, Universidade Estadual de Campinas, Campinas, 2009.