Matrix-free analysis of selected benzodiazepines in human serum samples using alternating trilinear decomposition modeling of fast liquid chromatography diode array detection data

Talanta 148 (2016) 454–462

Contents lists available at ScienceDirect

Talanta journal homepage: www.elsevier.com/locate/talanta

Matrix-free analysis of selected benzodiazepines in human serum samples using alternating trilinear decomposition modeling of fast liquid chromatography diode array detection data Maryam Vosough n, Negar J. Iravani Chemistry and Chemical Engineering Research Center of Iran, P.O. Box 14335-186, Tehran, Iran

art ic l e i nf o

a b s t r a c t

Article history: Received 12 September 2015 Received in revised form 29 October 2015 Accepted 31 October 2015 Available online 2 November 2015

This paper describes the development and validation of a simple and efficient bioanalytical procedure for simultaneous determination of alprazolam, clonazepam, diazepam in human serum samples using high performance liquid chromatography with photodiode-array detection, regarding a fast elution methodology in 4 min. Briefly, this method consists of a simple liquid extraction step of serum samples followed by HPLC analysis on a C18 column. After confirming the absence of matrix effect, an external standard methodology has been applied for quantification purposes. Due to the presence of serum endogenous components as uncalibrated components in the sample, the second-order calibration based on alternating trilinear decomposition has been applied on a set of absorbance matrices collected as a function of retention time and wavelengths. Acceptable resolution and quantification results were achieved in the presence of matrix interferences and the second-order advantage was fully exploited. The average recoveries for alprazolam, clonazepam and diazepam were 89.1%, 96.3% and 94.7% and relative standard deviation values for intra- and inter-day precision were equal or lower than 8.1% and 9.4%, respectively. The developed method enabled us to determine the analytes in the various serum samples in the presence of overlapped profiles, while keeping experimental time and extraction step at the minimum. & 2015 Elsevier B.V. All rights reserved.

Keywords: Benzodiazepines Second-order calibration High performance liquid chromatography Alternating trilinear decomposition Human serum

1. Introduction Benzodiazepines (BZDs) are an important class of psychoactive drugs that are frequently used in clinical treatments as antiepileptic hypnotics, tranquilizers, sleep inducers, anticonvulsants and muscle relaxants and amnesic properties [1,2]. Because of their dependency inducing properties in long term consumption, care should be taken in prescribing BZD hypnotics and anxiolytics [3]. Therefore, BZDs are frequently encountered in emergency toxicological screening, drugs-of-abuse testing, and forensic medicine examinations. In fact, because BZDs have a high potential of abusing, their rapid and sensitive quantification is an area of interest [4]. In addition, determination of BZDs in plasma is very effective to optimize chronic dosing, verify compliance and identify changes in pharmacokinetics [5]. So, proposing a rapid, simple and reliable analytical methodology for quantification of BZDs in human serum is required to save time and cost, while keeping the sensitivity and reproducibility of the analytical method at a n

Corresponding author. Fax: þ98 21 44787703. E-mail address: [email protected] (M. Vosough).

http://dx.doi.org/10.1016/j.talanta.2015.10.088 0039-9140/& 2015 Elsevier B.V. All rights reserved.

reasonable level. Several chromatographic methods have been reported for quantification of BZDs and/or their metabolites in biological fluids [6–11]. High performance liquid chromatography (HPLC) is a frequently applied tool for routine quantification of BZDs and the large number of published methods used in this respect are based on using HPLC with MS detection [5,11–18]. In spite of high selectivity of MS detection system, the cost of instrumentation prevents it to be a routinely applicable tool for most of the clinical or toxicological laboratories. So, during the recent years, there have been an extensive interest to develop analytical procedures for determination of this class of drugs in biological fluids using HPLC–UV or diode array detection (DAD) systems, which are easily available in most analytical laboratories [19–26]. Sample preparation is the most important and also vulnerable step in a bioanalytical method for handling of matrix effect and removing sample interferences. So, several extraction approaches such as liquid–liquid extraction [27,28], solid phase extraction (SPE) [8,15,22,24,29], online SPE [30], solid-phase microextraction (SPME) [31], microwave-assisted extraction (MAE) [18,22,32], hollow fiber liquid phase microextraction (HF-LPME) [25],

M. Vosough, N.J. Iravani / Talanta 148 (2016) 454–462

dispersive micro-solid-phase extraction (DMSPE) [26] and dispersive liquid–liquid microextraction (DLLME) [23] have been studied and reported for quantification of BZDs in biological fluids. On the other hand, the presence of background signal, shifted retention times, signal overlapping between analytes and also between analytes and matrix constituents, even at the optimal extraction and separation conditions, may have significant effects on qualitative and quantitative results [33]. Therefore, the validity of the analytical results severely depends on the solutions that the selected technique provides for overcoming such critical problems. Owing to the fact that laborious extraction and sample cleanup are needed before HPLC analysis, there is still a high probability of not-effectiveness of this step especially in case of analyzing new samples. So, sufficient resolutions may not be attained in case of eluting constituents with similar retention factors. However, taking advantage of second-order calibration methods as the assistant mathematical tools for modeling HPLC with DAD or fast scanning fluorescence detection (FSFD) systems, clearly have confirmed

455

their important role in resolving real bioanalytical problems [34– 40]. Most of the developed multiway or multiset data analysis strategies in the mentioned papers have been conducted to achieve a short run time and simplified chromatographic conditions regardless of the changes in the sample matrices. The aim of this study was to develop a simple, fast and reliable HPLC–DAD strategy coupled with second-order calibration based on alternating trilinear decomposition (ATLD) algorithm for quantitative determination of alprazolam (ALP), clonazepam (CLZ), diazepam (DZP) in human serum samples. Table 1 displays the chemical structures and chromatographic properties of ALP, CLZ and DZP. Since the isolation of the interfering species of serum is such a costly and time consuming process, the main target in the present study was focused on simplifying the extraction procedure and also providing a rapid HPLC analysis to reduce consumption of time and organic solvents. So, the main concerns in the present work was the overlapped chromatographic peaks of the matrix interferences with those of BZDs and retention time shift between

Table 1 Chemical structures, retention time, time regions considered for ATLD modeling, linear range of direct injection and correlation coefficients of ALP, CLZ and DZP. Analyte

Structure

Chromatographic specifications

H3C

ALP

N

Region

Time region (min)

tR (min)

Linear calibration range (μg mL  1)

R2

1

1.55–2.05

1.63

0.05–11

0.9982

1

1.55–2.05

1.77

0.03–11

0.9948

2

2.33–2.85

2.55

0.05–11

0.9994

N N

N

Cl

H N

CLZ

O

O2N

NH2 Cl

CH3

DZP

O

N

Cl

N

456

M. Vosough, N.J. Iravani / Talanta 148 (2016) 454–462

different chromatographic runs. The multiway modeling using ATLD algorithm and HPLC alignment using second-order chromatographic standardization [41] were used to solve the mentioned problems.

2. Theory The ATLD algorithm [42] which is an improvement of PARAFAC algorithm, is a three way decomposition method, based on an alternating least squares principle and an improved iterative procedure that utilizes the Moore–Penrose generalized inverse obtained by singular value decomposition. The ATLD-based secondorder calibration exploits the second-order advantage and makes the calibration possible, even in the presence of interferences that are not present in the calibration samples, so it can provide satisfactory concentration estimates. For a HPLC–DAD, the three-way array X made of I  J  K elements (xijk), a trilinear model for N number of factors has the form: N

xijk =

∑ ainbjnckn n= 1

+ eijk (i = 1, 2, …, I; j = 1, 2, …, J; k = 1, 2, …, K ),

(1)

where I is the number of elution time points, J is the number of wavelengths and K is the number of calibration samples and prediction samples stacked with each other. eijk is the element of residual array E not involved by the model. Three following objective functions, which are the sum of the squares of the elements of the residual matrices, are alternately minimized in ATLD algorithm: I

σ1(A) =

∑ ∥Xi.. − Bdiag(a(i))CT∥2F i=1

(2)

∑ ∥X . j . − C diag(b(i))AT ∥2F j=1

(3)

K

σ3(C ) =

∑ ∥X .. k − Adiag(c(i))BT∥2F k=1

3.2. Equipment Chromatographic analyses were performed using an Agilent 1200 HPLC system (Agilent Technologies Inc., USA) consisting of a quaternary pump, Rheodyne 7725 manual injector and a 20 μL injection loop, a degasser system, a column oven compartment and a Hewlett-Packard 1200 series photo diode-array detector. Chromatographic separation was carried out on a RP-18 column (70 mm  4.6 mm and 5 mm of particle size) equipped with guard (pre-) column. The mobile phase constituents, (A) ammonium formate buffer (pH ¼8.6) and (B) acetonitrile, were used in a gradient elution program as follows; 50% A as the starting mobile phase composition, descended to 5% in 3 min and hold for 1 min. Afterward, the mobile phase composition returned to the initial condition and the column was allowed to get to the equilibrium for 2 min, before the next run. Mobile phase flow rate was 1.0 mL min  1 and the column oven temperature was set at 25 °C. DAD detector was set to record between 210 and 400 nm with the spectral resolution of 2 nm and integration period of 0.4 s per spectrum. Total run time was 4 min. Data which was collected by Chemstation software (B.03.01), exported as Microsoft Excels file for further processing. Three-way modeling using ATLD was performed in MATLAB environment using the MVC2, a structured MATLAB toolbox for second-order calibration developed by Olivieri [44]. 3.3. Calibration and prediction samples

J

σ2(B) =

Blood Donation Unit of Modarres Hospital (Tehran, Iran) and stored at  20 °C in the freezer. Stock standard solutions of individual drugs (1000 mg L  1) were prepared once per two weeks by dissolving proper weights in 5 mL of ACN: water (50:50, v/v) and stored at  20 °C. Working standard solutions were constructed daily by mixing proper volumes of the stock solutions and further dilution in ACN:water (50:50, v/v) and kept in amber vials at 4 °C.

(4)

During this cyclic minimization process, the qualitative profiles (A and B) and the relative concentrations (C) of analytes can be updated accordingly. ATLD has the advantages of being insensitive to number of components, fast convergence due to the operation based on sliced matrices, and fully exploiting the second-order advantage. More detailed discussions on ATLD can be found in the literature [43].

3. Experimental 3.1. Materials Alprazolam (ALP), clonazepam (CLZ) and diazepam (DZP) (purity higher than 98%) were obtained from the pharmaceutical companies: Tehran Daroo and Dr. Abidi (Tehran, Iran). HPLC-grade methanol (MeOH), acetonitrile (ACN), ammonia and diethyl ether (DEE) were from Merck (Germany). Ammonium formate was of analytical grade from Merck, too. Ultrapure water provided by a Milli-Q purification system (Millipore, Bedford, MA, USA) was used throughout the study. Nylon membranes filters with the pore size of 0.22 mm (Varnian, USA) were used for filtering solvents, calibration and real samples before HPLC analysis. Blood serum samples from the healthy persons were kindly provided by the

A set of 15 calibration samples (C1–C15) containing five concentration levels was prepared. The concentration values of three drugs in each sample were randomly selected from the predetermined linear range of 0.05–11 mg mL  1 for ALP, 0.03– 11 mg mL  1 for CLZ and 0.05–11 mg mL  1 for DZP (Table 1). Chromatographic measurements was performed in random order according to the sample number. Also five samples (P1–P5) was considered as a prediction set with analyte concentrations shown in Table 2. In these samples, no interferences were added and triplicate analysis was performed for each sample. 3.4. Serum analysis Five hundred and fifty microliters of serum samples were mixed with 550 mL of 0.005 M ammonium formate (pH ¼8.6) and 3 mL of diethylether. The mixture was vortexed for 1 min and then centrifuged at 3500 rpm for 10 min (Hettichs 320R, Germany). The supernant layer was transferred into a conical tube and was evaporated to dryness under a stream of nitrogen. Finally, the residue was dissolved in 300 mL of mobile phase in ultrasonic bath (Sonorexs Digital 10P, Germany) and filtered through a 0.22 mm PTFE syringe filter and an aliquot of 20 mL was injected into the HPLC system. 3.5. Validation samples Eight human serum samples from two different subjects were built as validation set. The spiked concentration values for ALP, CLZ and DZP are shown in Table 2. The concentrations of the drugs

M. Vosough, N.J. Iravani / Talanta 148 (2016) 454–462

Table 2 Concentrations of alprazolam (ALP), clonazepam (CLN) and diazepam (DZP) in prediction and validation samples. Sample

Prediction samples P1 P2 P3 P4 P5 Spiked samples S1 S2 S3 S4 S5*a S6* S7* S8*

Analyte concentrations (μg mL  1) ALP

CLN

DZP

0.416 3.333 0.166 0.041 1.666

0.166 2.666 3.333 1.270 0.833

0.211 0.312 0.499 1.995 3.495

2.291 0.792 1.000 3.208 0.025 1.083 1.333 0.792

0.300 2.292 0.016 1.416 0.330 0.080 0.750 2.290

3.333 1.000 1.250 1.910 0.025 0.483 0.300 3.333

a Spiked serum samples indicated by asterisks were provided using the second subject.

were selected considering the levels usually found in serum of different patients, during therapeutic drug monitoring studies. Then, the samples were treated as explained in the previous section. Triplicate analyses were performed for all spiked and blank serum samples during the analytical procedure.

4. Results and discussion 4.1. Optimization of experimental condition In the present work, some preliminary chromatographic tests were performed on C18 columns of different lengths (15 and 7 cm) with various mobile phase conditions. The main target was to find the optimum separation condition (not necessarily completely separated peaks) at the shortest run time while keeping the peak shapes at the best. So, different ratios of acetonitrile/0.005 mol L  1 ammonium formate buffer solution (pH ¼8.6) under various

457

elution programs, were examined. Finally, a simple gradient elution strategy consisting of (A) ammonium formate buffer (pH ¼ 8.6, 0.005 mol L  1) and (B) acetonitrile, were used as follows; 50% A as the starting mobile phase composition, descended to 5% in 3 min and hold for 1 min. Afterward, the mobile phase composition returned to the initial condition and the column was allowed to get to the equilibrium for 2 min, before the next run. With the mentioned defined mobile phase condition, the retention times of ALP, CLZ and DZP were 1.63, 1.77 and 2.55 min, respectively (Table 1). After satisfactory chromatographic separation of pure analytes, a sample of blank serum was fortified with the target analytes for obtaining the chromatographic and background pattern in the presence of the biological matrix. The chromatographic profile of the blank serum with and without added analyte standards was a useful tool to select the calibration strategy for second-order calibration with untreated serum samples. Although in principle, the subsequent chemometric data analysis makes it possible to overcome the spectroscopic interference problem, a simple extraction step was used to preserve the useful lifetime of the chromatographic column. Diethyl ether as an extraction solvent was selected, according to the previous experiments conducted in the laboratory, to extract the drugs ALP, CLZ and DZP with the maximum average recoveries. A three dimensional chromatogram of a typical human serum spiked with the selected benzodiazepines and one of the calibration samples under the described chromatographic condition are illustrated in Fig. 1. The complexity of the data and appearance of a baseline offset in the retention time region of the analytes was clear. In addition, ALP and CLZ showed a moderate co-elution, which could add up further to the co-eluted interefering components from the serum sample and make the quantification difficult with conventional methods. In such cases, the analytes of interest can be quantified through univariate calibration by making the extraction step more selective or adding extra clean-up steps to further remove the uncalibrated components. Combination of the latter procedure with changing the mobile phase condition to longer run times can be an alternative way but not a guarantee for matrix-free quantification for the future samples. So, second-order calibration strategy with the aid of the trilinear algorithm ATLD without the necessity of having the knowledge about the nature of other matrix constituents was exploited in the present work as

Fig. 1. Three-dimensional plots of typical chromatograms of the studied benzodiazepines. (a) A standard mixture at concentration of 2, 4 and 2 mg L  1 of ALP, CLZ and DZP. (b) The spiked serum sample (S8) with 0.791, 2.290 and 3.333 mg L  1 of ALP, CLZ and DZP, respectively. The analytes of interest are indicated.

458

M. Vosough, N.J. Iravani / Talanta 148 (2016) 454–462

follows. 4.2. Preprocessing steps and second-order data generation

Table 3 Resolved concentrations of prediction samples P1–P5 and serum samples S1–S8 spiked with ALP, CLZ and DZP by using ATLD. Predicted concentration (mg mL  1)a

Before three way modeling, a suitable subset of each data matrix in both ways of measurement was selected. So, according to the retention times of ALP, CLZ and DIA at 1.63, 1.77 and 2.55 min, two time domains from the raw data (Table 1) in the range of 1.46–2.12 min (Ya) and 2.32–3.00 min (Yb) and identical spectral region of 240–400 nm were selected for all matrices. Finally, two matrices were made from each chromatographic run. Because of appearing the baseline offset for each validation sample, implementing background elimination was conducted as the first preprocessing step for each matrix. In the current study, the complexity of the real sample data was reduced as a separate step by background elimination for two-dimensional signals based on asymmetric least squares splines regression approach [45]. So, the matrices Ya and Yb were corrected to Ra and Rb, respectively. The corrected and simplified serum data matrices in different HPLC runs were further corrected due to retention time shift between the calibration and serum samples. Second-order standardization algorithm (introduced by Prazen et al.) [41] was chosen for this purpose, as a multivariate rank alignment strategy. The necessary shift for every candidate test matrix was assigned as to reach a minimum calculated residual variance through singular value decomposition of a row-wise augmentation with reference matrix. Finally, according to the selected retention time and spectral regions, HPLC–DAD three-way data arrays were constructed from every partition with dimensions 101  81  16. The first dimension is the number of retention times, the second is the number of wavelengths and the third is the number of calibration and prediction samples.

Prediction samples P1 P2 P3 P4 P5 Spiked samples Un-spiked S1 S2 S3 S4 S5*d S6* S7* S8*

ALP

CLZ

DZP

0.445 (106.9) 3.327 (99.8) 0.172 (103.4) 0.048 (107.3) 1.012 (102.4)[1.8]b

0.153 (92.1) 2.529 (94.8) 3.550 (106.5) 1.470 (107.6) 0.857 (102.9)[2.6]

0.218 (103.3) 0.290 (92.9) 0.515 (103.3) 2.070 (103.9) 3.120 (89.4)[2.4]

n.dc 1.809 (78.9) 0.756 (95.5) 0.819 (81.9) 2.746 (85.6) [4.6] 0.017 (68.1) 1.152 (106.3) 1.325 (99.4) 0.768 (96.9) [5.7]

n.d 0.235 (78.5) 1.733 (75.6) 0.018 (112.9) 1.305 (92.2) [3.5] 0.302 (91.7) 0.090 (115.1) 0.882 (117.6) 1.987 (86.7) [6.8]

n.d 3.943 (118.3) 0.929 (92.9) 1.282 (102.5) 1.903 (99.6) [3.2] 0.019 (76.8) 0.391 (81.1) 0.280 (93. 3) 3.107 (93.2) [8.1]

a

Recoveries in parenthesis. RSD(%) for four replicates of P5, S4 and S8 in square brackets (intra-day precision). c Not detected. d Spiked serum samples indicated by asterisks were provided using the second subject. b

4.3. ATLD modeling of prediction samples Following the necessary mathematical peak alignment in the present work, because of appearing peak misalignments, three way modeling of prediction samples using ATLD was performed. Results of concentrations of prediction samples through ATLD were shown in Table 3. The recoveries of ALP, CLZ and DZP attained from ATLD were 99.8–107.3%, 92.1–107.6%, 89.4–103.9% respectively. The prediction results were acceptable and satisfactory as shown in Table 2. The results confirmed that the algorithm could give accurate results for prediction samples. 4.4. ATLD modeling of spiked serum data for quantification of ALP, CLZ and DZP The lack of matrix effect was checked by comparing slopes and intercept corresponding to pseudo-univariate calibration curves based on the Milli-Q water and serum samples using a hypothesis tests (p-values o0.1, Statgraphics Centurion XVI, V 16.1.11). So, the second-order external calibration strategy rather than standard addition would be a suitable way for quantifying the analytes in the serum samples. Therefore, three-way analysis of both series of validation samples was accomplished by making 3, 3, 4 and 4 components models for regions r1 and r2 of the first set and r1 and r2 of the second set, respectively. In this work, the optimum number of the components was selected based on core consistency diagnostic test [46]. Also, ADD-ONE-UP test [47] was used to further confirm the proper number of the components. The mentioned analyses were carried out in order to obtain the best resolution and quantification results. Fig. 2A and B shows the total chromatograms of the first and the second pool of the serum samples, spiked with ALP, CLZ and DZP, at multiple wavelengths (240–400 nm). The appearance of

Fig. 2. Chromatographic profiles, each at a single wavelength (240–400 nm), for spiked samples S4 (serum 1) and S8 (serum 2).

interfering components in the retention time region of the analytes is obvious. Also, the difference between chromatographic patterns of the serum samples is clear. Fig. 3A and B shows the profiles retrieved by ATLD in the temporal mode for the first and the second region of serum-1. As can be appreciated from this figure, the chromatographic profiles are recognized as belonging to the analytes or interferences contained in the serum, although the overlapping is significant. The resolved spectral profiles by the mentioned algorithm is depicted in Fig. 4A and B for both regions. An almost perfect match between the recovered and the normalized pure spectrum of three drugs can be seen in these figures. The quality of the ATLD recovered spectral profiles was further evaluated for both regions using the similarity criterion which considered a comparison through the correlation coefficient (R) between the reference and recovered spectrum. The values of R, found for ALP, CLZ and DZP, were more than 0.998, confirming the

M. Vosough, N.J. Iravani / Talanta 148 (2016) 454–462

459

Fig. 3. ATLD resolved normalized elution time profiles in regions 1 (A) and 2 (B) of the first human serum as well as their corresponding real normalized profiles for three benzodiazepines. (Color figure available online.).

excellent quality of the ATLD resolved profiles. On the other hand, analysis of the second set of serum samples showed different scenario. In fact, different coeluting interferences in this chromatogram (Fig. 2B), further emphasized the need for applying the second-order multivariate calibration methods to avoid extensive sample clean-up procedures following deproteinization which complicated the HPLC assay. Figs. 5 and 6 show the plots of chromatographic and spectral profiles provided by ATLD modeling of the first and second regions of serum 2 validation sample. Again, it is clear that in the presence of serum interferences, resolved spectral profiles nicely matched the spectra of pure standard solutions, which confirms the accuracy and reliability of the proposed strategy. Following decomposition process for all unknown samples, quantification results were collected as shown in Table 3. Acceptable quantitative results for three analytes through modeling of both series of spiked serum samples demonstrates the effectiveness and accuracy of ATLD algorithm for quantifying multiple benzodiazepines in serum samples. The recoveries of ALP, CLZ and DZP were 68.1–106.3%, 75.6–117.6% and 76.8–118.3%, respectively. Also, the relative standard deviations (RSD%) of predicted concentration values for four replicates of S4 and S8 samples can be considered acceptable regarding the fact that no attempt has been performed to remove the interfering compounds before HPLC analysis. The reproducibility of the proposed method was tested by preparing two validation samples for each serum sample and analyzing in three consecutive days. The results for inter-day precision (expressed as RSD) were summarized in Table 4. From the average predicted concentrations and RSD values (lower than 10%), it can be stated that the proposed method can be considered reliable.

4.5. Method validation Analytical performance of the developed method was evaluated through calculating some analytical figures of merit, such as sensitivity (SEN), limit of detection (LOD), limit of quantification (LOQ), according to the following FO equation [40,48]: T T + + SENn = sn{[(Acal (I − A unx A unx )Acal )*(Bcal (I − BunxBunx )Bcal )]−1}nn−1/2 (5)

2 1/2 LODn = 3.3(SEN−n 2σx2 + h0SEN−n 2σx2 + h0SEN−n 2σ ycal )

(6)

2 1/2 LOQ n = 10(SEN−n 2σx2 + h0SEN−n 2σx2 + h0SEN−n 2σ ycal )

(7)

where in Eq. (5), sn is the slope of the ATLD pseudo-univariate plot, the index n corresponds to each analyte, matrices A and B collect the loading values, the subscripts ‘cal’ and ‘unx’ indicates calibration and unexpected components respectively, the symbol ‘*’ is the Hadamard matrix product, the subscript ‘nn’ indicates the (n,n) diagonal element of a matrix, and the symbol ‘þ’ shows the pseudo-inverse of a matrix. In Eqs. (6) and (7), n identifies the analyte; h0 is the un-spiked sample leverage; σx2 is the variance in 2 is the variance in calibration conthe instrumental signal; σycal centrations; and SENn is the analyte sensitivity. Also, statistical parameter of root-mean-square-error of prediction (RMSEP), was estimated as follows:

⎡1 RMSEP(μg L−1) = ⎢ ⎢⎣ n

n

⎤1/2

∑ (cadd. − cpred.)2⎥ n= 1

⎥⎦

(8)

where n is the number of unknown samples, and cadd. and cpred.

Fig. 4. Normalized spectral profiles of benzodiazepines recovered by ATLD in regions 1(A) and 2(B) of the first human serum as well as their corresponding real normalized profiles. The interfering components have been shown for the analytes. (Color figure available online.).

460

M. Vosough, N.J. Iravani / Talanta 148 (2016) 454–462

Fig. 5. ATLD resolved normalized elution time profiles in regions 1 (A) and 2 (B) of the second human serum as well as their corresponding real normalized profiles for three benzodiazepines. (Color figure available online.).

Fig. 6. Spectral profiles of benzodiazepines recovered by ATLD in regions 1(A) and 2(B) of the second human serum. The interfering components have been shown for the analytes. (Color figure available online.).

are the added and the predicted concentrations, respectively. The average values of the parameters calculated through application of ATLD on all serum samples were shown in Table 5. Linear pseudounivariate calibration curves were obtained with R values of 0.9989, 0.9957 and 0.9994 for ALP, CLZ and DZP, respectively. Linearity of the models for both analytes in their mentioned concentration ranges was confirmed using lack of fit test (Statgraphics Centurion XVI, V 16.1.11). The p-values for lack-of-fits in the ANOVA tables were greater than 0.05, so the linear models appeared to be adequate for the recovered areas at the 95.0% confidence level. The limits of detection (LODs) and limits of quantification (LOQs) values obtained for selected benzodiazepines in the serum samples were acceptable, taking into account the complexity of the studied matrices. Considering the typical values which can be found in serum samples, the proposed method could be directly applied without a pre-concentration step.

5. Conclusion In the present study, we developed a HPLC–DAD method

Table 5 Statistical parameters and figures of merit for quantification of selected benzodiazepines in human serum by ATLD.

RMSEP (μg mL  1) SEN (mL μg  1) R2 LOD (μg mL  1) LOQ (μg mL  1)

ALP

CLZ

DZP

0.2627 39.55 0.9989 0.015 0.044

0.2228 72.54 0.9957 0.008 0.024

0.2351 21.33 0.9994 0.027 0.081

coupled with ATLD modeling for simultaneous determination of alprazolam, clonazepam and diazepam in human serum samples. For this purpose, two chromatographic regions was selected, so that the first region contained alprazolam and clonazepam, and the second region contained diazepam. Before multiway modeling, baseline correction and retention time shift correction steps were implemented to reduce the complexity of the data matrix and to make it compatible with the trilinearity requirement necessary for ATLD method. The good quality of the obtained results suggested that the developed technique was appropriate for the rapid quantification of benzodiazepines in serum samples. The

Table 4 Inter-day precision for determination of selected benzodiazepines by using ATLD (N ¼ 3). Sample

Serum-1 Serum-2

ALP

CLZ

DZP

Added

Found

RSD (%)

Added

Found

RSD (%)

Added

Found

RSD (%)

0.792 3.208 0.025 0.792

0.783 2.864 0.020 0.796

6.8 4.6 9.4 7.5

1.416 2.290 0.080 2.290

1.425 2.235 0.087 1.983

5.9 4.2 7.0 6.9

1.000 1.910 0.025 3.333

0.972 1.995 0.022 3.124

4.6 5.1 9.0 8.3

M. Vosough, N.J. Iravani / Talanta 148 (2016) 454–462

presented method is characterized by its simplicity, very short time of analysis, minimum operator effort and using small amounts of organic solvents, which make this method suitable for rapid analysis of serum samples despite the serious interferences.

Acknowledgments The authors would like to acknowledge the Research Council of Chemistry and Chemical Engineering Research Center (CCERCI) for financial support of this work (Grant No. 1001).

References [1] D. Charney, J. Mihic, R. Harris, The pharmacological basis of therapeutics, in: J. Hardman, L. Limbird (Eds.), Goodman & Gilman's The Pharmacological Basis of Therapeutics, McGraw-Hill, NewYork, 2001, pp. 399–412. [2] K.N. van Dijk, C.S. de Vries, K. ter Huurne, P.B. van den Berg, J.R.B.J. Brouwers, L. T.W. de Jong-van den Berg, Concomitant prescribing of benzodiazepines during antidepressant therapy in the elderly, J. Clin. Epidemiol. 55 (2002) 1049–1053. [3] A. Heather, Toxicity and adverse consequences of benzodiazepine use, Psychiatr. Ann. 25 (1995) 158–165. [4] A. El Mahjoub, C. Staub, High-performance liquid chromatographic method for the determination of benzodiazepines in plasma or serum using the columnswitching technique, J. Chromatogr. B 742 (2000) 381–390. [5] S.J. Marin, R. Coles, M. Merrell, G.A. McMillin, Quantitation of benzodiazepines in urine, serum, plasma, and meconium by LC–MS–MS, J. Anal. Toxicol. 32 (2008) 491–498. [6] S. Pirnay, I. Ricordel, D. Libong, S. Bouchonnet, Sensitive method for the detection of 22 benzodiazepines by gas chromatography–ion trap tandem mass spectrometry, J. Chromatogr. A 954 (2002) 235–245. [7] M. Wilhelm, H.-J. Battista, D. Obendorf, HPLC with Simultaneous UV and reductive electrochemical detection at the hanging mercury drop electrode: a highly sensitive and selective tool for the determination of benzodiazepines in forensic samples, J. Anal. Toxicol. 25 (2001) 250–257. [8] M.N. Uddin, V.F. Samanidou, I.N. Papadoyannis, Development and validation of an HPLC method for the determination of six 1, 4‐benzodiazepines in pharmaceuticals and human biological fluids, J. Liq. Chromatogr. Relat. Technol. 31 (2008) 1258–1282. [9] S. Pirnay, S. Bouchonnet, F. Hervé, D. Libong, N. Milan, P. d'Athis, F. Baud, I. Ricordel, Development and validation of a gas chromatography–mass spectrometry method for the simultaneous determination of buprenorphine, flunitrazepam and their metabolites in rat plasma: application to the pharmacokinetic study, J. Chromatogr. B 807 (2004) 335–342. [10] T. Gunnar, K. Ariniemi, P. Lillsunde, Fast gas chromatography–negative-ion chemical ionization mass spectrometry with microscale volume sample preparation for the determination of benzodiazepines and α-hydroxy metabolites, zaleplon and zopiclone in whole blood, J. Mass Spectrom. 41 (2006) 741–754. [11] T.L. Laurito, G.D. Mendes, V. Santagada, G. Caliendo, M.E. de Moraes, G. De Nucci, Bromazepam determination in human plasma by high-performance liquid chromatography coupled to tandem mass spectrometry: a highly sensitive and specific tool for bioequivalence studies, J. Mass Spectrom. 39 (2004) 168–176. [12] A. Bugey, C. Staub, Application of monolithic supports to online extraction and LC–MS analysis of benzodiazepines in whole blood samples, J. Sep. Sci. 30 (2007) 2967–2978. [13] B.E. Smink, J.E. Brandsma, A. Dijkhuizen, K.J. Lusthof, J.J. de Gier, A.C. Egberts, D.R. Uges, Quantitative analysis of 33 benzodiazepines, metabolites and benzodiazepine-like substances in whole blood by liquid chromatography-(tandem) mass spectrometry, J. Chromatogr. B 811 (2004) 13–20. [14] W.-C. Cheng, T.-S. Yau, M.-K. Wong, L.-P. Chan, V.K.-K. Mok, A high-throughput urinalysis of abused drugs based on a SPE-LC–MS/MS method coupled with an in-house developed post-analysis data treatment system, Forensic Sci. Int. 162 (2006) 95–107. [15] C. Liang, H. Ye, R. Wang, C. Ni, Y. Rao, Y. Zhang, Identification and quantification of 34 drugs and toxic compounds in blood, urine, and gastric content using liquid chromatography with tandem mass spectrometry, J. Sep. Sci. 38 (2015) 1680–1690. [16] D. Montenarh, M. Hopf, H. Maurer, P. Schmidt, A. Ewald, Detection and quantification of benzodiazepines and Z-drugs in human whole blood, plasma, and serum samples as part of a comprehensive multi-analyte LC–MS/MS approach, Anal. Bioanal. Chem. 406 (2014) 803–818. [17] M. Nakamura, Analyses of benzodiazepines and their metabolites in various biological matrices by LC–MS(/MS), Biomed. Chromatogr. 25 (2011) 1283–1307. [18] A. Woźniakiewicz, R. Wietecha-Posłuszny, M. Woźniakiewicz, J. Nowak, P. Kościelniak, Development of the MAE/UHPLC-MS-TOF method for determination of benzodiazepines in human bio-fluids for toxicological analysis,

461

J. Pharm. Biomed. Anal. 108 (2015) 97–101. [19] L. Mercolini, R. Mandrioli, M. Amore, M.A. Raggi, Separation and HPLC analysis of 15 benzodiazepines in human plasma, J. Sep. Sci. 31 (2008) 2619–2626. [20] M.N. Uddin, V.F. Samanidou, I.N. Papadoyannis, Validation of SPE-HPLC determination of 1,4-benzodiazepines and metabolites in blood plasma, urine, and saliva, J. Sep. Sci. 31 (2008) 3704–3717. [21] J. Adrian, M. Gratacós-Cubarsí, F. Sánchez-Baeza, J.-A. Garcia Regueiro, M. Castellari, M.P. Marco, Traceability of sulfonamide antibiotic treatment by immunochemical analysis of farm animal hair samples, Anal. Bioanal. Chem. 395 (2009) 1009–1016. [22] P. Fernandez, C. Vazquez, R.A. Lorenzo, A.M. Carro, I. Alvarez, P. Cabarcos, Experimental design for optimization of microwave-assisted extraction of benzodiazepines in human plasma, Anal. Bioanal. Chem. 397 (2010) 677–685. [23] P. Fernandez, C. Gonzalez, M.T. Pena, A.M. Carro, R.A. Lorenzo, A rapid ultrasound-assisted dispersive liquid-liquid microextraction followed by ultraperformance liquid chromatography for the simultaneous determination of seven benzodiazepines in human plasma samples, Anal. Chim. Acta 767 (2013) 88–96. [24] M.N. Uddin, V.F. Samanidou, I.N. Papadoyannis, Development and validation of an HPLC method for the determination of benzodiazepines and tricyclic antidepressants in biological fluids after sequential SPE, J. Sep. Sci. 31 (2008) 2358–2370. [25] F. Rezaei, Y. Yamini, M. Moradi, B. Daraei, Supramolecular solvent-based hollow fiber liquid phase microextraction of benzodiazepines, Anal. Chim. Acta 804 (2013) 135–142. [26] A.A. Asgharinezhad, H. Ebrahimzadeh, F. Mirbabaei, N. Mollazadeh, N. Shekari, Dispersive micro-solid-phase extraction of benzodiazepines from biological fluids based on polyaniline/magnetic nanoparticles composite, Anal. Chim. Acta 844 (2014) 80–89. [27] D.S. Ming, J. Heathcote, A rapid and accurate UPLC/MS/MS method for the determination of benzodiazepines in human urine, J. Chromatogr. B 879 (2011) 421–428. [28] A. Bugey, C. Staub, Rapid analysis of benzodiazepines in whole blood by highperformance liquid chromatography: use of a monolithic column, J. Pharm. Biomed. Anal. 35 (2004) 555–562. [29] W. He, N. Parissis, Simultaneous determination of flunitrazepam and its metabolites in plasma and urine by HPLC/DAD after solid phase extraction, J. Pharm. Biomed. Anal. 16 (1997) 707–715. [30] M.-R. Fuh, S.-W. Lin, L.-L. Chen, T.-Y. Lin, Determination of flunitrazepam and 7-aminoflunitrazepam in human urine by on-line solid phase extraction liquid chromatography–electrospray-tandem mass spectrometry, Talanta 72 (2007) 1329–1335. [31] Y. Luo, L. Pan, J. Pawliszyn, Determination of five benzodiazepines in aqueous solution and biological fluids using solid-phase microextraction with carbowaxTM/DVB fiber coating, J. Microcolumn Sep. 10 (1998) 193–201. [32] A. Wozniakiewicz, R. Wietecha-Posłuszny, M. Wozniakiewicz, E. Bryczek, P. Kościelniak, A quick method for determination of psychoactive agents in serum and hair by using capillary electrophoresis and mass spectrometry, J. Pharm. Biomed. Anal. 111 (2015) 177–185. [33] J.M. Amigo, T. Skov, R. Bro, ChroMATHography: solving chromatographic issues with mathematical models and intuitive graphics, Chem. Rev. 110 (2010) 4582–4605. [34] L. Vera-Candioti, M.J. Culzoni, A.C. Olivieri, H.C. Goicoechea, Chemometric resolution of fully overlapped CE peaks: quantitation of carbamazepine in human serum in the presence of several interferences, Electrophoresis 29 (2008) 4527–4537. [35] J.A. Arancibia, P.C. Damiani, G.M. Escandar, G.A. Ibañez, A.C. Olivieri, A review on second- and third-order multivariate calibration applied to chromatographic data, J. Chromatogr. B 910 (2012) 22–30. [36] H.C. Goicoechea, M.J. Culzoni, M.D. Garcia, M.M. Galera, Chemometric strategies for enhancing the chromatographic methodologies with second-order data analysis of compounds when peaks are overlapped, Talanta 83 (2011) 1098–1107. [37] M.J. Culzoni, R.Q. Aucelio, G.M. Escandar, High-performance liquid chromatography with fast-scanning fluorescence detection and multivariate curve resolution for the efficient determination of galantamine and its main metabolites in serum, Anal. Chim. Acta 740 (2012) 27–35. [38] M. Vosough, S. Ghafghazi, M. Sabetkasaei, Chemometrics enhanced HPLC– DAD performance for rapid quantification of carbamazepine and phenobarbital in human serum samples, Talanta 119 (2014) 17–23. [39] J. Zhao, H.-L. Wu, J.-F. Niu, Y.-J. Yu, L.-L. Yu, C. Kang, Q. Li, X.-H. Zhang, R.-Q. Yu, Chemometric resolution of coeluting peaks of eleven antihypertensives from multiple classes in high performance liquid chromatography: a comprehensive research in human serum, health product and Chinese patent medicine samples, J. Chromatogr. B 902 (2012) 96–107. [40] G.M. Escandar, H.C. Goicoechea, A. Muñoz de la Peña, A.C. Olivieri, Secondand higher-order data generation and calibration: a tutorial, Anal. Chim. Acta 806 (2014) 8–26. [41] B.J. Prazen, R.E. Synovec, B.R. Kowalski, Standardization of second-order chromatographic/spectroscopic data for optimum chemical analysis, Anal. Chem. 70 (1998) 218–225. [42] H.L. Wu, M. Shibukawa, K. Oguma, An alternating trilinear decomposition algorithm with application to calibration of HPLC–DAD for simultaneous determination of overlapped chlorinated aromatic hydrocarbons, J. Chemom. 12 (1998) 1–26. [43] H.L. Wu, J.F. Nie, Y.J. Yu, R.Q. Yu, Multi-way chemometric methodologies and

462

M. Vosough, N.J. Iravani / Talanta 148 (2016) 454–462

applications: a central summary of our research work, Anal. Chim. Acta 650 (2009) 131–142. [44] A.C. Olivieri, H.-L. Wu, R.-Q. Yu, MVC2: a MATLAB graphical interface toolbox for second-order multivariate calibration, Chemom. Intell. Lab. Syst. 96 (2009) 246–251. [45] P.H. Eilers, I.D. Currie, M. Durbán, Fast and compact smoothing on large multidimensional grids, Comput. Stat. Data Anal. 50 (2006) 61–76.

[46] R. Bro, H.A. Kiers, A new efficient method for determining the number of components in PARAFAC models, J. Chemom. 17 (2003) 274–286. [47] Z.-P. Chen, Z. Liu, Y.-Z. Cao, R.-Q. Yu, Efficient way to estimate the optimum number of factors for trilinear decomposition, Anal. Chim. Acta 444 (2001) 295–307. [48] A.C. Olivieri, Analytical figure of merit: from univariate to multiway calibration, Chem. Rev. 114 (2014) 5358–5378.