Accepted Manuscript Multiway analysis methods applied to the fluorescence excitationemission dataset for the simultaneous quantification of valsartan and amlodipine in tablets

Erdal Dinç, Zehra Ceren Ertekin, Eda Büker PII: DOI: Reference:

S1386-1425(17)30360-8 doi: 10.1016/j.saa.2017.04.081 SAA 15134

To appear in:

Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy

Received date: Revised date: Accepted date:

25 January 2017 26 April 2017 29 April 2017

Please cite this article as: Erdal Dinç, Zehra Ceren Ertekin, Eda Büker , Multiway analysis methods applied to the fluorescence excitation-emission dataset for the simultaneous quantification of valsartan and amlodipine in tablets, Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy (2017), doi: 10.1016/j.saa.2017.04.081

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT MULTIWAY ANALYSIS METHODS APPLIED TO THE FLUORESCENCE EXCITATION-EMISSION DATASET FOR THE SIMULTANEOUS QUANTIFICATION OF VALSARTAN AND AMLODIPINE IN TABLETS

T

Erdal Dinç*, Zehra Ceren Ertekin, Eda Büker

US

CR

06100 Tandoğan, Ankara, Turkey

IP

Ankara University, Faculty of Pharmacy, Department of Analytical Chemistry,

AN

ABSTRACT

In this study, excitation-emission matrix datasets, which have strong overlapping bands, were

M

processed by using four different chemometric calibration algorithms consisting of parallel factor

ED

analysis, Tucker3, three-way partial least squares and unfolded partial least squares for the simultaneous quantitative estimation of valsartan and amlodipine besylate in tablets. In analyses,

PT

preliminary separation step was not used before the application of parallel factor analysis

CE

Tucker3, three-way partial least squares and unfolded partial least squares approaches for the analysis of the related drug substances in samples. Three-way excitation-emission matrix data

AC

array was obtained by concatenating excitation-emission matrices of the calibration set, validation set, and commercial tablet samples. The excitation-emission matrix data array was used to get parallel factor analysis, Tucker3, three-way partial least squares and unfolded partial least squares calibrations and to predict the amounts of valsartan and amlodipine besylate in samples. For all the methods, calibration and prediction of valsartan and amlodipine besylate were performed in the working concentration ranges of 0.25-4.50 μg/mL. The validity and the performance of all the proposed methods were checked by using the validation parameters. From 1

ACCEPTED MANUSCRIPT the analysis results, it was concluded that the described two-way and three-way algorithmic methods were very useful for the simultaneous quantitative resolution and routine analysis of the related drug substances in marketed samples. Keywords: Excitation-emission matrix; Three-way analysis; Quantification; Valsartan,

IP

T

Amlodipine besylate

CR

*Corresponding author: Erdal Dinç

US

e-mail: [email protected]

AN

phone: +90 312 203 31 76

AC

Abbreviations

CE

PT

ED

M

fax: +90 312 213 10 81

3W-PLS: three-way partial least squares, ALS: alternating least squares, AML: amlodipine besylate, ANOVA: one-way analysis of variance, DAD: diode array detector, EEM: excitationemission matrix, GRAM: generalized rank annihilation method, HCl: hydrochloric acid, HPLC: high performance liquid chromatography, LC-MS: liquid chromatography-mass spectrometer, LOD: limit of detection, LOQ: limit of quantitation, MCR: multivariate curve resolution, PARAFAC: parallel factor analysis, PCA: principal component analysis, PLS: partial least squares, PRESS: prediction residual error sum of squares, UPLC: ultra-performance liquid chromatography, VAL: valsartan 2

ACCEPTED MANUSCRIPT 1. Introduction

In analytical chemistry and related scientific fields, multiway analysis methods have been used for the decomposition of three-way or higher-order arrays into triads, which are loadings and scores corresponding to pure contributions of analyzed compounds. Multiway analysis

IP

T

methodologies offer direct resolution of measured matrices, obtained as a function of different

CR

scales (for example excitation and emission or time and wavelength etc.), into the essential original contributions from individual components. Besides, these chemometric tools provide a

US

simple interpretation and short experimental period without requiring long and tedious pretreatment of analyzed samples to remove interferences with the second-order advantage (analyte

AN

quantitation in the presence of uncalibrated interference).

M

Three-way array (or higher-order array), also called tensor, can be obtained from different

ED

analytical instruments e.g. high performance liquid chromatography-diode array detector (HPLCDAD), liquid chromatography-mass spectrometer (LC-MS), fluorescence spectrophotometer, etc.

PT

Two-way analysis methods, which provide simple resolution and rapid results, are based on the

CE

use of unfolded data obtained from multiway data arrays. These methods, such as multivariate curve resolution – alternating least squares (MCR-ALS) [1-2] and unfolded partial least squares

AC

(unfolded PLS) [2-4] have been applied to resolve complex data system of analyzed samples. The most popular multiway analysis methods are parallel factor analysis (PARAFAC) [5], generalized rank annihilation method (GRAM) [6], multiway partial least squares (multiway PLS) [7] and Tucker3 [8]. In analyses of complex samples, multiway data analysis methods, particularly Tucker3 and PARAFAC models are powerful, flexible and versatile tools to process three-way data arrays for the qualitative and quantitative analysis in different scientific fields. Main preferable reasons of multiway methods in analytical chemistry are second-order advantage 3

ACCEPTED MANUSCRIPT and simple interpretation of results, without using any chemical pre-treatment with short analysis period and low-cost. In previous researches, some interesting applications of three-way (or multiway) data analysis methods were reported for the spectrofluorimetric [9-10] spectrophotometric [11-12], and chromatographic [13-15] analysis of chemicals and pharmaceuticals.

IP

T

The antihypertensive drugs valsartan (VAL) and amlodipine besylate (AML) lower the blood pressure with different pathways and are commonly used in combination. VAL is an angiotensin

CR

II receptor antagonist and AML is a calcium channel blocker. Single-pill combinations of VAL

US

and AML are more effective than monotherapy of each drug and are commonly used for the

AN

patients with mild to moderate hypertension. [16-17]

There are several methods including ultraviolet spectrophotometry [18-19], high performance

M

liquid chromatography-ultraviolet detector, [20-22], high performance liquid chromatography-

ED

fluorescence detector [23-25], liquid chromatography tandem mass spectrometry [26] and spectrofluorimetry [27-30] for the determination of VAL and AML in pharmaceutical

PT

preparations and biological samples.

CE

In previous studies on the spectrofluorimetric analysis of VAL and AML, some chemical pretreatments, e.g. micellar, surfactants, nanoparticles were used to enhance the fluorescence

AC

intensity of the related drugs. These preliminary chemical procedures require the additional effort and chemical experiments. Although both VAL and AML have native fluorescence, especially AML gives weak fluorescence intensity. In our case, the overlapping spectrofluorimetric bands in emission wavelength region were reported for the analysis of VAL and AML in their mixture. Due to the above mentioned drawbacks, it is difficult to simultaneously determine VAL and AML using classical spectrofluorimetric approach, which is based on the selection of one 4

ACCEPTED MANUSCRIPT excitation wavelength and the use of measurements at one emission wavelength. Preliminary studies showed that classical PLS application to overlapping emission fluorescence dataset, was not useful to get quantitation of the VAL and AML. In order to overcome the complexity of the analysis, four different multivariate calibration methods, consisting of PARAFAC, Tucker3,

T

three-way partial least squares (3W-PLS) and unfolded PLS were applied to the excitation-

IP

emission matrix (EEM) dataset for the simultaneous spectrofluorimetric quantitation of VAL and

CR

AML in a solid dosage form. General validation parameters in terms of recovery, standard addition, inter-day and intra-day studies were used for the evaluation of the applied methods. The

US

implementation of the validated PARAFAC, Tucker3, 3W-PLS and unfolded PLS methods in the

AN

analysis of tablets containing VAL and AML indicated that multiway data analysis methods were

ED

M

very useful and effective techniques for the drug analysis and drug quality control.

2.1. PARAFAC model

PT

2. Three-way Calibration Methods

CE

PARAFAC model is one of the most common techniques for the multiway data analysis. In

AC

three-mode component analysis, the algorithmic PARAFAC approach was conceived by Harshman and was named as the parallel factor model [31]. At the same time, this technique was independently introduced by Carroll and Chang with the name of canonical decomposition [32]. In addition, the model was formulated Cattel in the form of principal of parallel proportional profiles [33]. PARAFAC model has been applied for the resolution of overlapping bands in fluorescence

excitation-emission

analyses

of

chemical

analytes

[34-38].

PARAFAC

decomposition of the data array X into a set of trilinear components and a residual array gives 5

ACCEPTED MANUSCRIPT three loading matrices, A, B and C with elements of aif, bjf, and ckf. From three loadings, two correspond to excitation and emission spectra of analyzed substances, and one corresponds to relative concentration profiles of analytes. The PARAFAC decomposition model of a three-way

denotes the intensity of fluorescence obtained from analyzed sample at excitation

US

where

CR

IP

T

array is given below:

wavelength and emission wavelength, aif is the spectral profiles of analytes at excitation

AN

wavelengths, bjf is the spectral profiles of analytes at emission wavelengths and ckf is the relative concentration profiles of analytes in samples, F is the number of PARAFAC components and

ED

M

represents a residual error term.

The main difference between PARAFAC and bilinear principal component analysis (PCA) is that

PT

PARAFAC decomposes a third-order dataset into triads (trilinear components) which are one score vector and two loading vectors, while bilinear PCA gives one score vector and one loading

CE

vector. PARAFAC gives a unique solution to get the pure contributed profiles whereas bilinear

AC

PCA has the problem of rotational freedom to reach pure profiles of the analyzed components. Theoretical details, application and validation of PARAFAC model can be found in references [5, 31-39]

In practice, PARAFAC decomposition of EEM array X was computed using alternating least squares (ALS) method under Kruskal condition defined by k’ + k2 + k3 ≥ 2 +2 providing unique solution [5].

6

ACCEPTED MANUSCRIPT

2.2. Tucker3 model

IP

T

Tucker3 model is the most well-known and commonly applied multiway data analysis tool to identify the pure spectra of analytes from measurements of mixtures of chemicals [39-41].

CR

Tucker3 is a generalisation of bilinear factor analysis to higher-order dataset. The following

AN

US

expression indicates the application of a Tucker3 model to a three-way array X.

M

The equation (2) can be formulated in matrix notation as:

ED

where A, B and C are the component matrices, which correspond to first, second and third is

PT

modes, respectively. G is the core array and E is the residual array of the model,

is used for the Kronecker product in Tucker3 algorithm.

AC

symbol

CE

matricized data array and E indicates the residual variation not described by the model. The

2.3. Three-way PLS model

Three-way PLS (or multiway PLS) is a generalization of two-way PLS algorithm to a multiway case. As in ordinary two-way PLS, multiway PLS is based on the decomposition of the independent data (X) and the dependent data (Y) to reach maximal covariance between two block. 3W-PLS model can be expressed in the form given below: 7

ACCEPTED MANUSCRIPT (4) (5) (6)

T

here T and U are scores for the dependent and independent blocks, W and Q are the PLS weights

IP

for X and Y, respectively. C is the matrix of regression coefficients. In the multiway PLS

CR

approach, when Y is a matrix, this is multiway PLS2 algorithm. In the vector Y case, this

US

approach is called three-way PLS1. As in 3W-PLS2, the scores “T” obtained by least squares regression of X onto the weights WJ (in second mode) and WK (in third mode) for each

AN

component has a maximum covariance with Y vector. For the validation of the results obtained from training set, the model is applied to the external test samples. Finally, the 3W-PLS

M

modelling is applied for the prediction of the quantities of the analytes in unknown samples.

ED

Details of the algorithmic 3W-PLS model and chemical applications were described in the

CE

PT

literature [7,39, 42-43].

AC

2.4. Unfolded partial least squares model

Unfolded PLS is the application of ordinary PLS (or two-way PLS) to two-way matrices obtained by unfolding three-way array X. The details of the ordinary PLS algorithm was firstly introduced by Wold and co-worker [44]. The unfolded PLS method is very practical to analyze complex mixture systems. Ordinary two-way PLS calibration is constructed by unfolding three-way data array into a two-way matrix [3-4]. In ordinary two-way PLS calibration, PLS decomposes the two-way data matrix X and matrix or vector Y. The PLS decomposition can be given as below: 8

ACCEPTED MANUSCRIPT X = T x PT + E

(7)

Y = U x QT + F

(8)

Where T and U are the scores of two-way matrix and Y is vector of concentration set, P and Q

CR

IP

T

are the loadings for X and Y, and E and F are the residuals of X and Y, respectively.

3. Material and Methods

US

3.1. Experimental setup

AN

Fluorescence measurements of the analyzed samples were done using an Agilent Cary Eclipse

M

Fluorescence Spectrometer (Agilent, USA) connected to a desktop computer with Cary Eclipse

ED

software. Fluorescence spectra were recorded at the excitation wavelength from 238 to 418 nm with a step of 4 nm and the emission wavelength from 400 to 640 nm with a step of 2 nm. In

PT

order to scan the fluorescence excitation and emission, bandwidths were set at 5 nm. Scanning speed was at 1200 nm/min with a power at 690 V. Excitation and emission filters were set at the

AC

CE

spectral intervals 250-395 nm and 400-640 nm, respectively.

PARAFAC and Tucker3 models were obtained by means of N-way Toolbox [45], and quantitative computations with regression analysis and graphical illustrations were performed by using in-house m-files in Matlab (MathWorks, USA). 3W-PLS and unfolded PLS calibrations were done by using in-house m-files in Matlab (MathWorks, USA).

9

ACCEPTED MANUSCRIPT 3.2. Chemicals and Reagents

Methanol and hydrochloric acid (HCl) were purchased from Sigma Aldrich (Germany) and of analytical grade. Ultrapure water used for the preparation of solutions was obtained by Millipore Milli-Q system (USA). Standard materials of VAL (99.6 %) and AML (%99.8) were kindly

IP

T

supplied by Sanovel Pharmaceuticals (Turkey). The commercial tablets (Exforge film-coated

CR

tablets, produced by Novartis Inc. (Turkey)), contained 160 mg VAL and 10 mg AML per tablet

US

were analyzed in this study.

AN

3.3. Standard and working solutions

M

Stock solutions of VAL and AML at the concentration level of 10 mg/100 mL were individually

ED

prepared in the solvent system consisting of methanol and 0.05 M HCl (50:50, v/v). The stock solutions were filtered by syringe filters with a pore size of 0.45 µm after sonication for five

PT

minutes. From the stock solutions of the analyzed compounds, a calibration set of 18 binary

CE

mixtures containing VAL and AML was prepared in the working range of 0.25-4.50 μg/mL. In the calibration step of PARAFAC, Tucker 3, 3W-PLS and unfolded PLS models, a 24 full

AC

factorial design with additional two experiment points was used in the above linear concetration range as depicted in Table 1. Table 1. Calibration sample set for VAL and AML in the range of 0.25-4.50 µg/µL µg/µL Sample no. CS1 CS2 CS3 CS4 CS5

VAL 0.25 0.25 0.25 0.25 0.75

µg/µL AML 0.25 0.75 3.00 4.50 0.25

Sample no. CS10 CS11 CS12 CS13 CS14

VAL 3.00 3.00 3.00 4.50 4.50

AML 0.75 3.00 4.50 0.25 0.75 10

ACCEPTED MANUSCRIPT CS6 CS7 CS8 CS9

0.75 0.75 0.75 3.00

0.75 3.00 4.50 0.25

CS15 CS16 CS17 CS18

4.50 4.50 2.00 0.00

3.00 4.50 0.00 4.00

For the method validation a test set of 9 sythetic mixtures, inter-day and intra-day samples (at

IP

T

three concentration levels: 0.40, 3.00 and 4.00 μg/mL) containing VAL and AML within the

CR

mentioned working concentration range was prepared by using the above stock solutions. For the estimation of the specifity and selectivity of the applied methods, standard addition

US

samples were prepared by adding three different quantities of the standard stock solutions of the

AN

analyzed compounds to a portion of the prepared tablet solution within the mentioned working range. In this contex, an extra sample starting from tablet sample solution was prepared without

ED

M

using standard stock solution to calculate the added recovery values.

PT

3.4. Preparation of commercial tablet samples

CE

In order to perform commercial tablet analysis, 10 tablet were weighted and finely powdered in a

AC

mortar. An amount of powder equivalent to half tablet, which contained 80.0 mg VAL and 5.0 mg AML, was weighed and transferred to 50 mL volumetric flask containing 20 mL of the solvent system. The powder was dissolved by sonication for 30 minutes and the flask was made up to 50 mL after cooling down. The content was filtered by a syringe filter with a pore size of 0.45 µm and diluted into the working range of the analytes (theoretical concentration of VAL: 4.5 µg/mL; theoretical concentration of AML: 0.28 µg/mL). This procedure was repeated 10 times.

11

ACCEPTED MANUSCRIPT 4. Results and Discussion

In research and development and quality control laboratories, high performance liquid chromatography (HPLC) is the standard method for the quantitative resolution of multicomponent pharmaceutical preparations. As an alternative way, spectroscopic methods,

IP

T

particularly fluorescence spectroscopy can be very suitable for the analysis of samples containing

CR

two or more active ingredients. However, classical spectroscopic approaches may not give desirable analysis results due to overlapping spectral bands in excitation and emission

US

wavelength regions. Hence, chemometric multivariate calibrations with three-way excitationemission dataset are very useful and promising to resolve overlapping excitation-emission spectra

AN

of analyzed drugs in complex commercial samples.

M

In this paper, we mainly focused on the application of fluorescence spectroscopy to the analysis

ED

of VAL and AML in tablets because the analyzed drug substances have native fluorescence. In this approach, the analysis of VAL and AML is not possible by measuring direct emission

PT

intensity due to superposition of emission bands in same wavelength region. After preliminary

CE

experiments, it was concluded that three-way analysis methods and unfolded PLS could be

and AML.

AC

appropriate for the quantitative multicomponent resolution of commercial tablets containing VAL

The fluorescence excitation-emission matrices were recorded as described in the “Experimental Setup” section. In the excitation and emission wavelengths, Rayleigh and Raman scatterings were avoided by using excitation and emission filters. Under the mentioned spectral conditions, the EEM data of calibration set (see Table 1) and other samples of the analyzed compounds were obtained for the calibration and prediction treatments. After calibration modelling procedure,

12

ACCEPTED MANUSCRIPT PARAFAC, Tucker3, 3W-PLS and unfolded PLS were used for the simultaneous quantification of VAL and AML in tablets using EEM data array. Experimental details in the application of the

T

chemometric calibrations to the analysis of binary mixture of analytes were explained below.

IP

4.1. Application of PARAFAC method

CR

In fluorescence analyses, the EEM data of all the samples consisting of calibration set, validation

US

set (test samples, intra-day and inter-day samples and standard addition samples) and commercial tablet samples were recorded. The EEM data array X with a dimension of 121x46x67 (excitation

AN

x emission x samples) was obtained by using the recorded EEM data sets. The fluorescence

M

excitation-emission spectra of calibration set (from CS1 to CS18) and commercial tablet samples

AC

CE

PT

ED

(from CTS1 to CTS10) were illustrated in Figure 1.

13

CE

PT

ED

M

AN

US

CR

IP

T

ACCEPTED MANUSCRIPT

AC

Figure 1. Three-dimensional illustration of EEM data matrices for a) calibration set (from CS1 to CS18) and b) commercial tablet samples (from CTS1 to CTS10)

The PARAFAC decomposition of EEM data array X into three loadings related to excitation spectral profile, emission spectral profile and relative concentration profile was performed as shown in Figure 2a-c.

14

M

AN

US

CR

IP

T

ACCEPTED MANUSCRIPT

PT

ED

Figure 2. PARAFAC decomposition of EEM data array into three loadings: a) excitation profile, b) emission profile, c) relative concentration profile.

CE

In the VAL and AML analyses in samples, PARAFAC model without using any constraint gave

AC

us successful results with two components using ALS algorithm. The model was able to explain 99.94 % of the variations in the dataset. Core consistency diagnostic [46] was computed as 100 % which proved that the applied model with two-component system was confirmed for the EEM dataset and from the decomposition mode, the estimated loadings were in excellent agreement with the original excitation spectra, emission spectra and concentration set. In the quantification of the analyzed compounds, it was concluded that the concentration of VAL and AML in binary mixture system was proportional to the relative concentration profiles of 15

ACCEPTED MANUSCRIPT VAL and AML, obtained by PARAFAC decomposition of three-way EEM data array. Based on the mentioned mathematical relationship, linear regression equations for both compounds were individually obtained by the regression of the actual concentration on the relative concentration profile. Their statistical data results were listed in Table 2. The quantity of VAL and AML in all

IP

T

the analyzed samples were computed using the above linear regression equations. Table 2. Least square regression analysis and statistical results obtained from PARAFAC

M

AN

m n r SD (m) SD (n) SD (r) LOD (µg/mL) LOQ (µg/mL)

Tucker3

VAL 3.33x10-2 1.12x10-2 0.9946 8.67x10-4 1.10x10-3 6.47x10-3 0.09 0.30

US

PARAFAC VAL AML 3.54x10-2 4.88x10-2 3.32x10-3 1.35x10-3 0.9998 0.9997 -4 1.837x10 3.109x10-4 -4 4.813x10 8.529x10-4 1.297x10-3 1.593x10-3 0.04 0.05 0.14 0.17

CR

and Tucker3 models

AML 4.95x10-2 -2.50x10-4 0.9996 3.54x10-4 9.72x10-4 1.79x10-3 0.06 0.20

CE

PT

ED

m = Slope of the linear regression equation n = Intercept of the linear regression equation r = Correlation coefficient of the linear regression equation SD (m) = Standard deviation of the slope SD (n) = Standard deviation of the intercept SD (r) = Standard deviation of the correlation LOD = Limit of detection LOQ = Limit of detection quantitation

AC

4.2. Application of Tucker3 method

After preliminary implementation of different factor numbers, the 2x2x2 Tucker3 method was an appropriate modelling to decompose three-way array X consisting of EEMs of VAL and AML for the quantitative prediction of their contents in samples. The 2x2x2 Tucker3 model was applied by using non-negativity constraint in each mode. In practice, the EEM data array with dimension 121x46x67(excitation x emission x concentration) obtained from the recorded EEM of

16

ACCEPTED MANUSCRIPT the analyzed samples (calibration set, validation set and real samples) was processed by using the 2x2x2 Tucker3 algorithm to get three loadings consisting of excitation, emission and concentration profiles as displayed in Figure 3. The explained variance for the Tucker3 model was reported as 99.9265 %. Tucker3 decomposition of the EEM array of VAL and AML into

T

three loadings related to the excitation spectra, emission spectra and relative concentration was

AC

CE

PT

ED

M

AN

US

CR

IP

illustrated in Figure 3a-c.

Figure 3. Tucker3 decomposition of EEM data array into three loadings: a) excitation profile, b) emission profile, c) relative concentration profile

17

ACCEPTED MANUSCRIPT The quantification of VAL and AML in samples is based on the use of relative concertation profiles of the analyzed compounds, obtained by application of Tucker3 to the EEM data array (see Figure 3c). For the quantification procedure, the calibration vector of each compound in the calibration set containing VAL and AML was proportional to the relative concentration vector of

T

each compound in the relative concentration profiles. The linear regression functions for VAL

IP

and AML were obtained from the above linear relationship. The statistical results obtained by the

CR

ordinary regression analysis technique were tabulated in Table 2. The amounts of VAL and AML

US

in the analyzed samples were calculated by using the mentioned linear regression functions.

AN

4.3. Application of 3W-PLS method

M

In this study, the matrix (18x5566) obtained by unfolding the three-way array

(121x46x18),

ED

was multiplied by the concentration vector of VAL and AML and tensor Z was obtained. After this, the tensor was decomposed by trilinear PLS algorithm, the weight vectors were used to

PT

calculate the score “T”. 3W-PLS regression was used for the quantitation of analytes in validation

CE

and real samples. Leave-one-out cross validation technique revealed that first four factors were the optimal factor number for the construction of 3W-PLS calibration for both VAL and AML.

AC

Prediction residual error sum of squares (PRESS) values were calculated as 0.0241 and 0.0188 for VAL and AML, respectively.

4.4. Application of unfolded PLS method

18

ACCEPTED MANUSCRIPT Another application of PLS approach, unfolded PLS is one of the latent structured models in unfolded configuration. In order to build the unfolded PLS calibration, firstly raw data matrices of the calibration set were stored as a three-way EEM tensor X of dimensions IxJxK (121x46x18). Then, X was unfolded to a two-way matrix X of Kx(IxJ) (18x5566) dimension as in 3W-PLS

T

model. In the unfolded PLS modelling procedure, the matrix X was mean-centered along K. After

IP

this procedure, ordinary two-way PLS was applied to two-way data matrix of the calibration

CR

samples and then, the constructed unfolded PLS calibration was applied to the quantitative

US

prediction of analytes in two-way data matrix of the validation set and the samples. In the unfolded PLS modelling, the number of factors was estimated by leave-one-out cross-

AN

validation procedure. First three factors were chosen to build the unfolded PLS calibration model

PT

ED

for VAL and AML, respectively.

M

for both compounds, corresponding to the minimum PRESS values which are 0.0263 and 0.0132

CE

4.5.Validation of the methods

The validity of proposed methods was evaluated in terms of accuracy, precision, repeatability,

AC

specificity and selectivity. The calibration curves obtained by PARAFAC and Tucker3 were found to be linear for both VAL and AML in the working range of 0.25-4.50 µg/mL. For both drugs, high correlation coefficients were reported for these two methods. The following equations (9) and (10) were used for the computation of the limit of detection (LOD) and limit of quantitation (LOQ) values:

(9)

19

ACCEPTED MANUSCRIPT (10)

where SD denotes the standard deviation of regression equation intercept and m is the slope of regression equation. The values of LOD and LOQ were depicted in Table 2.

T

Evaluation of accuracy and precision was performed by analyzing a test set of 9 solutions using

IP

PARAFAC, Tucker3, 3W-PLS and unfolded PLS methods. Mean percent recovery results,

CR

relative standard deviations and relative standard errors were satisfactory for the precision and accuracy, as seen in Table 3. Accuracy and precision of the methods were also proved by intra-

US

and inter-day studies. Samples with three different concentration levels were analyzed by the

AN

proposed methods for three times in the same day and on three successive days. Numerical results of recovery studies, relative standard deviations and relative standard errors were given in

M

Table 4. The results were obtained from the average of three replications.

PARAFAC Found Recovery (µg/µL) (%)

Found (µg/µL)

PT

Added (µg/µL)

ED

Table 3. Recovery data obtained by analyzing the test set by the proposed methods Tucker3 Recovery (%)

VAL AML VAL AML VAL AML VAL AML

VAL

AML

3W-PLS Found Recovery (µg/µL) (%)

Unfolded PLS Found Recovery (µg/µL) (%)

VAL AML VAL AML VAL AML VAL AML

0.25 0.255 0.253 102.1 101.1 0.255 0.260 102.09 103.97 0.245 0.256 98.1 102.3 0.255 0.260 102.1 104.1

0.75

0.25 0.761 0.253 101.5 101.1 0.755 0.243 100.70 97.10 0.751 0.256 100.1 102.3 0.780 0.266 104.0 106.2

3.00

0.25 2.993 0.244 99.8

97.7 2.873 0.252 95.78 100.71 2.926 0.243 97.5

4.50

0.25 4.494 0.246 99.9

98.5 4.413 0.245 98.06

4.50

CE

0.25

97.4 2.979 0.251 99.3 100.6 98.4

0.25 4.446 0.251 98.8 100.4 4.407 0.247 97.93

98.99 4.380 0.244 97.3

97.5 4.405 0.249 97.9

99.7

4.50

0.75 4.354 0.745 96.8

99.4 4.378 0.745 97.28

99.29 4.331 0.726 96.2

96.7 4.646 0.728 103.2 97.0

4.50

3.00 4.447 2.914 98.8

97.1 4.501 2.877 100.02 95.91 4.473 2.904 99.4

4.50

4.50 4.508 4.424 100.2 98.3 4.677 4.385 103.94 97.45 4.505 4.471 100.1 99.3 4.525 4.483 100.5 99.6

4.00

AC

98.08 4.420 0.250 98.2 100.2 4.451 0.246 98.9

0.25 4.037 0.253 100.9 Mean 99.9 SD 1.60 RSD 1.61 SD = Standard deviation RSD = Relative standard deviation

101.4 3.856 0.256 96.41 102.34 4.097 0.244 102.4 99.4 99.1 99.3 98.8 1.61 2.73 2.61 1.87 1.62 2.76 2.62 1.90

96.8 4.452 2.940 98.9

98.0

97.7 4.006 0.248 100.1 99.1 98.9 100.6 100.3 2.22 2.11 2.98 2.25 2.10 2.97

20

IP

T

ACCEPTED MANUSCRIPT

AC

CE

PT

PARAFAC VAL AML -0.28 -0.65 -0.51 1.23 -1.85 -1.15 -3.21 -3.96 -1.11 0.89 -2.51 -2.53

Unfolded PLS VAL AML 0.39 0.42 3.00 3.04 3.94 4.00 0.37 0.42 2.97 3.05 3.91 3.94

CR

US

AN

ED

PARAFAC VAL AML 2.93 2.63 1.56 1.78 0.89 2.04 1.15 3.37 1.71 2.03 0.79 2.12

Intra-day Inter-day

PARAFAC VAL AML 99.7 99.4 99.5 101.2 98.1 98.9 96.8 96.0 98.9 100.9 97.5 97.5

Found (µg/µL) Tucker3 3W-PLS VAL AML VAL AML 0.41 0.42 0.41 0.41 3.09 3.02 2.98 3.05 4.14 3.92 3.96 3.99 0.41 0.41 0.40 0.41 3.07 3.01 2.97 3.05 4.10 3.87 3.93 3.93 Recovery (%) Tucker3 3W-PLS VAL AML VAL AML 101.3 105.6 102.6 102.3 103.0 100.7 99.4 101.7 103.4 98.1 98.9 99.8 101.9 103.2 99.1 103.0 102.4 100.4 98.9 101.7 102.6 96.7 98.2 98.3 Relative Standard Deviation Tucker3 3W-PLS VAL AML VAL AML 3.17 2.43 2.18 2.01 1.70 1.76 1.26 1.92 0.91 2.05 0.88 2.30 2.09 1.94 1.70 2.74 1.88 2.02 2.14 2.06 0.80 2.13 0.69 2.18 Relative Standard Error Tucker3 3W-PLS VAL AML VAL AML 1.33 5.64 2.63 2.27 3.01 0.72 -0.56 1.68 3.39 -1.90 -1.08 -0.17 1.95 3.21 -0.91 3.04 2.37 0.39 -1.07 1.67 2.58 -3.27 -1.79 -1.71

M

PARAFAC VAL AML 0.40 0.40 2.98 3.04 3.93 3.95 0.39 0.38 2.97 3.03 3.90 3.90

Intra-day Inter-day

Added (µg/µL) VAL AML 0.40 0.40 3.00 3.00 4.00 4.00 0.40 0.40 3.00 3.00 4.00 4.00

Intra-day Inter-day

Intra-day Inter-day

Table 4. Analysis results of intra-day and inter-day studies for the proposed methods

Unfolded PLS VAL AML 97.2 104.2 100.1 101.5 98.6 99.9 93.1 103.9 99.1 101.6 97.8 98.4 Unfolded PLS VAL AML 2.68 0.87 1.56 1.83 0.89 2.26 1.78 3.42 2.00 2.10 0.78 2.29 Unfolded PLS VAL AML -2.81 4.19 0.06 1.49 -1.40 -0.07 -6.95 3.94 -0.86 1.60 -2.18 -1.59

21

ACCEPTED MANUSCRIPT Specificity and selectivity studies were conducted by analyzing standard addition samples. Three different solutions which contained different amounts of standard material and a constant amount of tablet solution were analyzed. The estimated concentration of the extra sample which only contained the same amount of tablet solution, was subtracted from the estimated concentration

T

value of three standard addition samples to compute the added concentration of VAL and AML.

IP

Analysis results consisting of recovery percentage, relative standard deviation and relative

CR

standard error were stated in Table 5. In PARAFAC, Tucker3, 3W-PLS and unfolded PLS applications, high recovery results from standard addition studies demonstrated the absence of

US

matrix effects on the analysis of tablets. High core consistency diagnostics obtained from

AN

PARAFAC and Tucker3 models which had two components also proves the absence of excipient

M

effects.

Added (µg/mL) VAL AML 0.75 1.75

1.51 2.05

1.52 2.06

3.25 3.98

Unfolded PLS VAL AML 0.75 1.79

3.23 1.55 3.29 1.53 3.29 3.95 2.08 4.03 2.07 4.04 Recovery (%) PARAFAC Tucker3 3W-PLS Unfolded PLS VAL AML VAL AML VAL AML VAL AML 99.4 103.7 103.1 103.0 101.7 102.8 99.4 102.4 100.5 100.1 101.6 99.3 103.7 101.3 102.0 101.2 102.3 99.5 103.1 98.7 104.0 100.8 103.4 101.1 Relative Standard Deviation PARAFAC Tucker3 3W-PLS Unfolded PLS VAL AML VAL AML VAL AML VAL AML 0.01 0.03 0.01 0.03 0.02 0.02 0.01 0.02 0.02 0.06 0.02 0.06 0.03 0.08 0.04 0.06 0.01 0.08 0.02 0.08 0.02 0.11 0.03 0.10 *Theoretical concentrations of the drugs in tablet are 2.25 µg/mL VAL and 0.14 µg/mL in standard addition samples.

PT

3.25 4

Found (µg/mL) Tucker3 3W-PLS VAL AML VAL AML 0.77 1.80 0.76 1.80

AC

CE

1.5 2

PARAFAC VAL AML 0.75 1.82

ED

*Tablet

Table 5. Numerical results obtained from the standard addition samples by the proposed methods

4.6. Analysis of commercial tablet samples 22

ACCEPTED MANUSCRIPT Using the multivariate calibration methods, PARAFAC, Tucker3, 3W-PLS and unfolded PLS, we have analyzed the commercial tablets containing VAL and AML substances. Commercial tablet samples were prepared as we have described in the preparation of commercial tablet samples. The EEM data of commercial tablet samples were recorded and arranged as three-way EEM

T

array as illustrated in Figure 1b. According to the proposed methods after EEM dataset was

IP

arranged, the calibrations of PARAFAC, Tucker3, 3W-PLS and unfolded PLS models were

CR

applied to the EEM data to simultaneously predict the amount of VAL and AML in tablets. The

US

assay results showed a good agreement with the label claim as listed in Table 6.

AC

CE

PT

ED

M

AN

Table 6. Determination results of VAL and AML in commercial tablet samples by the proposed PARAFAC, Tucker3, 3W-PLS and unfolded PLS methods. mg/tablet PARAFAC Tucker3 3W-PLS Unfolded PLS VAL AML VAL AML VAL AML VAL AML CTS1 160.5 10.2 162.3 10.0 161.1 10.1 160.4 10.3 CTS2 160.2 10.4 159.1 10.2 160.4 9.9 161.3 10.3 CTS3 160.3 9.9 164.0 9.7 161.0 9.8 161.3 10.5 CTS4 158.4 9.8 160.0 10.0 158.8 10.2 157.3 10.0 CTS5 160.9 10.2 162.8 10.0 160.7 10.1 161.3 9.8 CTS6 162.1 10.0 161.1 9.8 163.7 9.6 161.0 10.1 CTS7 162.0 10.2 163.0 10.0 162.7 10.1 162.4 9.9 CTS8 160.9 9.9 161.7 9.6 167.3 9.7 161.8 9.9 CTS9 157.9 9.9 163.4 9.7 161.3 9.9 164.0 9.8 CTS10 158.5 9.9 165.1 9.7 161.5 9.7 162.9 10.2 Mean 160.2 10.0 162.3 9.9 161.9 9.9 161.4 10.1 SD 1.47 0.20 1.84 0.20 2.32 0.21 1.77 0.24 RSD 0.92 1.95 1.13 2.02 1.43 2.09 1.10 2.36 SD = Standard deviation RSD = Relative standard deviation

4.7. Statistical comparison of results

23

ACCEPTED MANUSCRIPT In this work, one-way analysis of variance (ANOVA) was used for the comparison of assay results obtained by applying four different chemometric approaches, PARAFAC, Tucker3, 3WPLS and unfolded PLS to the quantification of VAL and AML in marketed tablets. From the statistical results, it was reported that calculated p-values, p=0.0894 for VAL and p=0.0909 for

T

AML, were greater than the value of p=0.05 at confidence level of 95 % for both VAL and AML.

IP

This indicated that there was not a significant difference between the analysis results provided by

CR

the PARAFAC, Tucker3, 3W-PLS and unfolded PLS models. We observed that all the applied

AN

US

chemometric approaches gave us comparable determination results for the analysis of tablets.

5. Conclusions

M

In this research, four different chemometric calibration models based on three-way and two-way

ED

analyses were implemented for the quantitative resolution of overlapping excitation-emission fluorescence spectra for the analysis of two-component mixture systems containing VAL and

PT

AML. In fluorescence analyses of the related drugs, the proposed chemometric calibration

CE

approaches did not require preliminary separation step. The analytical performances of the implemented PARAFAC, Tucker3, 3W-PLS and unfolded PLS techniques were validated by

AC

analysing validation samples. Then successful results were reached for the analysis of commercial tablet samples containing VAL and AML. In practice, we obtained comparable tablet analysis results from four different methods. We observed that three-way and two-way data analysis methods presented in this study were very promising for the spectrofluorimetric quantification and routine analysis of commercial tablets containing the analyzed drug substances.

24

AC

CE

PT

ED

M

AN

US

CR

IP

T

ACCEPTED MANUSCRIPT

25

ACCEPTED MANUSCRIPT References R. Tauler, Chemometr. Intell. Lab. 1995, 30, 133.

[2]

J. Jaumot, R. Gargallo, A.de Juan, R. Tauler, Chemometr. Intell. Lab. 2005, 76, 101.

[3]

R. Henrion, Chemometr. Intell. Lab. 1994, 25, 1.

[4]

P. Geladi, Chemometr. Intell. Lab. 1989, 7, 11.

[5]

R. Bro, Chemometr. Intell. Lab. 1997, 38, 149.

[6]

E. Sanchez, B. Kowalski. Anal. Chem. 1986, 58, 496.

[7]

R. Bro, J. Chemometr. 1996, 10, 47.

[8]

L.R. Tucker, Psychometrika. 1966. 31, 279.

[9]

R. M. Maggio, A. Muñoz De La Peña, A.C. Olivieri, Chemometr. Intell. Lab. 2011, 109,

AN

US

CR

IP

T

[1]

M

178.

[10] M. Vosough, S.N. Eshlaghi, R. Zasmard, Spectrochim. Acta A. 2015, 136, 618.

ED

[11] F. Samari, B. Hemmateenejad, M. Shamsipur, Analytica Chimica Acta, 2010, 667, 49.

PT

[12] M. A. E. M. Hegazy, M. S. Eissa, O. I. Abd El-Sattar, M. Abd El-Kawy, Spectrochim. Acta A. 2014, 128, 231.

CE

[13] M. Cocchi, C. Durante, C. Grandi, M. Manzini, & A. Marchetti, Talanta, 2008, 74, 547.

AC

[14] G. Tomasi, J. H. Christensen, 2009, J. Chromatogr. A. 1216, 7865. [15] E. Dinç, Z. C. Ertekin, E. Büker, J. Sep. Sci. 2016, 39, 3488. [16] J. E. Frampton, L. J. Scott, Am. J. Cardiovasc. Drug. 2009, 9, 309. [17] M. Destro, A. Luckow, M. Samson, A. Kandra, P. Brunel, J. Am. Soc. Hypertens. 2008, 2, 294. [18] E. Dinç, D. Baleanu, Revista de Chimie. 2010, 61, 290. [19] N. G. Mohamed, Anal. Chem. Insights. 2011,6, 53.

26

ACCEPTED MANUSCRIPT [20] N. K Ramadan, H. M. Mohamed, A. A. Moustafa, Anal. Lett. 2010, 43, 570. [21] M. Çelebier, M. S. Kaynak, S. Altınoz, S. Sahin, Brazilian Journal of Pharmaceutical Sciences. 2010, 46, 761. [22] E. S. Abu-Nameh, K. Abu-Shandi, M. Saket, M. Salim, O. M. Othman, Y. Mohammad,

T

Jordan Journal of Pharmaceutical Sciences. 2011, 4, 105.

IP

[23] O. A. R. Amin, F. H. Bamane, A. Hanafy, J. Liq. Chromatogr. R. T. 2013, 36:16, 2220. doi:

CR

10.1080/10826076.2012.717060

[24] N. Y. Khalil, T. A. Wani, M. A. Abunassif, I. A. Darwish, J. Liq. Chromatogr. R. T. 2011.

US

34, 2583.

AN

[25] T. Inglot, A. Gumieniczek, P. Maczka, E. Rutkowska, American Journal of Analytical Chemistry. 2013, 4, 17.

M

[26] S. G. Gadepalli, P. Deme, M. Kuncha, R. Sistla, Journal of Pharmaceutical Analysis, 2014,

ED

4, 399.

[27] R. A. Shaalan, T. S. Belal, Drug Testing and Analysis, 2010, 2, 489.

PT

[28] S. Shalan, N. El-Enany, F. Belal, Anal. Methods. 2015, 7, 8060.

Journal

of

CE

[29] H. Y. Fu, H. D. Li, C. Ni, T. M. Yang, Y. Fan, H. Zhang, J. Yang, L. Chen, She, Y. B. spectroscopy,

2015,

Article

ID

681320,

11

pages,

2015.

AC

doi:10.1155/2015/681320 [30] A. M. El-Kosasy, S. M. Tawakkol, M. F. Ayad, A. I. Sheta, Talanta. 2015, 143, 402. [31] R.A. Harshman, UCLA Working Papers Phonet. 1970, 16, 1. [32] J.D. Carroll, J.J. Chang, Psychometrika, 1970, 35, 283. [33] R. B. Cattell, Psychometrika, 1944, 9, 267. [34] C. M. Andersen, R. Bro, J. Chemometr. 2003, 17, 200.

27

ACCEPTED MANUSCRIPT [35] A. M. de la Peña, N. M. Díez, D. B. Gil, A. C. Olivieri, G. M. Escandar, Anal. Chim. Acta. 2006, 569, 250. [36] F. Alarcón, M. E. Báez, M. Bravo, P. Richter, G. M. Escandar, A. C. Olivieri, E. Fuentes, Talanta. 2013, 103, 361.

T

[37] M. Vosough, S. N. Eshlaghi, R. Zadmard, Spectrochim. Acta A. 2015, 136, 618-624.

IP

[38] S. M. Sajjadi, H. Abdollahi, R. Rahmanian, L. Bagheri, Spectrochim. Acta A. 2016, 156,

CR

63-69.

[39] R. Bro, (1998) Multi-way analysis in the food industry: models, algorithms, and

AN

[40] L.R. Tucker, Psychometrika 1966, 31, 279.

US

applications, Ph.D. Thesis, University of Amsterdam, The Netherlands.

[41] E. Acar, B. Yener, IEEE. T. Knowl. Data. En. 2009, 21, 6.

M

[42] A.K. Smilde, J. Chemometr. 1997, 11, 367.

ED

[43] A. Smilde, R. Bro, P. Geladi, Wiley, Chichester, 2004. [44] S. Wold, P. Geladi, K. Esbensen, J. Öhman, J. Chemometr. 1987, 1, 41.

PT

[45] C. A. Andersson, R. Bro, Chemometr. Intell. Lab. 2000, 52, 1.

AC

CE

[46] R. Bro, H. A. Kiers, J. Chemometr. 2003, 17, 274.

28

ED

M

AN

US

CR

IP

T

ACCEPTED MANUSCRIPT

AC

CE

PT

Graphical abstract

29

ACCEPTED MANUSCRIPT

T IP CR US AN M ED PT CE AC

    

Highlights Excitation-emission data was used to quantify valsartan and amlodipine in tablets. Resolution of binary mixture of drugs was achieved by three-way analysis methods. The implemented methods were carefully checked by using validation parameters. Three-way methods were compared with unfolded PLS in terms of analysis results. Three-way calibration methods were very useful for routine analysis of the drugs.

30

Multiway analysis methods applied to the fluorescence excitation-emission dataset for the simultaneous quantification of valsartan and amlodipine in tablets.

In this study, excitation-emission matrix datasets, which have strong overlapping bands, were processed by using four different chemometric calibratio...
2MB Sizes 0 Downloads 12 Views