sensors Article Voltammetric Electronic Tongue for the Simultaneous Determination of Three Benzodiazepines Anna Herrera-Chacón 1 , Farzad Torabi 1,2 , Farnoush Faridbod 2 , Jahan B. Ghasemi 3 , Andreu González-Calabuig 1 and Manel del Valle 1, * 1 Sensors and Biosensors Group, Department of Chemistry Universitat Autònoma de Barcelona, Edifici Cn, 08193 Bellaterra, Spain; anna.herrerachacon@gmail.com (A.H.-C.); farzad.torabi@ut.ac.ir (F.T.); andreugc27@gmail.com (A.G.-C.) 2 Center of Excellence in Electrochemistry, Faculty of Chemistry, University of Tehran, Tehran 1417466191, Iran; faridbodf@khayam.ut.ac.ir 3 School of Chemistry, College of Science, University of Tehran, Tehran 1417466191, Iran; jahan.ghasemi@ut.ac.ir * Correspondence: manel.delvalle@uab.es; Tel.: +34-93-581-3235  Received: 16 October 2019; Accepted: 13 November 2019; Published: 16 November 2019  Abstract: The presented manuscript reports the simultaneous detection of a ternary mixture of the benzodiazepines diazepam, lorazepam, and flunitrazepam using an array of voltammetric sensors and the electronic tongue principle. The electrodes used in the array were selected from a set of differently modified graphite epoxy composite electrodes; specifically, six electrodes were used incorporating metallic nanoparticles of Cu and Pt, oxide nanoparticles of CuO and WO3 , plus pristine electrodes of epoxy-graphite and metallic Pt disk. Cyclic voltammetry was the technique used to obtain the voltammetric responses. Multivariate examination using Principal Component Analysis (PCA) justified the choice of sensors in order to get the proper discrimination of the benzodiazepines. Next, a quantitative model to predict the concentrations of mixtures of the three benzodiazepines was built employing the set of voltammograms, and was first processed with the Discrete Wavelet Transform, which fed an artificial neural network response model. The developed model successfully predicted the concentration of the three compounds with a normalized root mean square error (NRMSE) of 0.034 and 0.106 for the training and test subsets, respectively, and coefficient of correlation R ≥ 0.938 in the predicted vs. expected concentrations comparison graph. Keywords: electronic tongue; nanoparticle modifiers; diazepam; lorazepam; flunitrazepam; artificial neural networks 1. Introduction Benzodiazepines (BZs) are a family of drugs massively used worldwide since the 1960s, when they were first introduced. BZs are frequently prescribed as anxiolytics, anticonvulsants, muscle relaxants, and as treatment for alcohol and drug abuse [1]. At the same time, they are also used in clinical anesthesia due to their sedative and relaxant properties in a variety of procedures. The clinical use of these compounds is accomplished through their effects on the central nervous system (CNS) and specifically through the modulation of the inhibitory neurotransmitter gamma-aminobutyric acid (GABA) receptor [2]. BZs enhance GABAA receptor affinity producing sedative, tranquilizing, and sleep-inducing effects. BZs supposedly present low toxicity when administered at clinical doses but may be potentially dangerous when used at high doses. Even though they are well-known drugs, they are difficult to identify in the body after their use. Therefore, the identification of BZs in blood [3], urine [4], hair samples [5], or in post-mortem studies may be interesting in terms of criminology and forensic studies [6]. Sensors 2019, 19, 5002; doi:10.3390/s19225002 www.mdpi.com/journal/sensors Sensors 2019, 19, x FOR PEER REVIEW 2 of 12 the identification of BZs in blood [3], urine [4], hair samples [5], or in post-mortem studies may be interesting Sensors 2019, 19,in terms 5002 of criminology and forensic studies [6]. 2 of 12 A plethora of analytical methods to determine BZs can be found in the literature such as spectrophotometry [7], capillary electrophoresis [8], HPLC/MS [9], LC–MS [10], and immunoassays A plethora [11,12]. However, of analytical these techniquesmethodsrequire to determine high-cost BZsinstrumentation, can be found in arethe literature slow, requiresuch as highly spectrophotometry [7], capillary electrophoresis [8], HPLC/MS [9], trained personnel, and cannot be implemented to perform on-line measurements. Among less LC–MS [10], and immunoassays [11,12]. However, employed these techniqueselectrochemical techniques, require high-costmethods instrumentation, stand out, are slow, require highly specifically, trained personnel, voltammetric methods. and Voltammetric methods have several advantages over classical methods, including goodtechniques, cannot be implemented to perform on-line measurements. Among less employed sensitivity, electrochemical low detection limits, methods stand relative out, specifically, simplicity, portability, voltammetric and low-costmethods.instrumentation Voltammetriccompared methods to other have several They techniques. advantages also have over classical enough methods, versatility to beincluding implemented goodwith sensitivity, portablelow detection limits, instrumentation. The relative study of simplicity, portability, and techniques BZs by electrochemical low-cost instrumentation is centered on compared to other techniques. the determination They also and the quantification have of a enough single BZ versatility at a time. to beAs implemented an examplewith portable instrumentation. of individual BZ studies we The mentionstudydetermining of BZs by electrochemical clonazepam in beverages and serum samples using screen printed electrodes [13] at techniques is centered on the determination and the quantification of a single BZ ora time. using As an example of individual BZ studies we mention determining polyaniline/graphene oxide nanocomposites [14]; detecting diazepam by dendritic silver clonazepam in beverages and serum samples using screen nanostructures printedby supported electrodes graphene [13][15], or using polyaniline/graphene studying its redox behavior oxidein nanocomposites drinks by adsorptive [14]; detecting diazepam by dendritic silver nanostructures supported stripping voltammetry [16], or by the use of fullerene-functionalized carbon nanotubes [17]; by graphene [15], studying its redox behavior determiningin drinks by adsorptive lorazepam stripping voltammetry [16], by polypyrrole@sol-gel@gold or by the use nanoparticles pencil of fullerene-functionalized graphite electrode [18]; carbon nanotubes quantifying [17]; determining olanzapine using quantum lorazepam dots ontoby polypyrrole@sol-gel@gold modified multiwalled carbon nanoparticles nanotubes pencil gold graphite electrode [18]; quantifying olanzapine using quantum dots onto electrode [19]; and, lastly, sensing flunitrazepam in screen-printed electrodes [20], or in alcoholic andmodified multiwalled carbon nanotubes soft drinks gold electrode [19]; and, by screen-printed lastly, sensing drop-volume cellsflunitrazepam in screen-printed [21]. The determination of several electrodes BZs can[20], or in be found alcoholic and soft drinks by screen-printed drop-volume cells [21]. in the recent literature as the determination of diazepam and oxazepam in biological fluids using The determination of several BZs can be found in the recent literature as the determination of diazepam modified carbon paste electrodes [22], or with the use of a multiwall carbon nanotube-ionic liquid and oxazepam in biological fluids pasteusing modified electrode [23];carbon usingpaste electrodes [22], a boron-doped or with the diamond use of a to electrode multiwall determine carbon nanotube-ionic bromazepam and liquid paste electrode [23]; using a boron-doped diamond electrode alprazolam [24]; an attempt to determine olanzapine and risperidone modifying gold electrodes to determine bromazepam and alprazolam with carbon [24]; an attempt nanotubes to and [25]; determine using aolanzapine and risperidone modified bentonite sonogel modifying carbon electrodegold electrodes with to determine carbon nanotubes [25]; and using a modified bentonite sonogel diazepam chlordiazepoxide determined in urine [26]. Furthermore, the identification or carbon electrode to determine diazepam chlordiazepoxide determination ofdetermined different BZs in urine [26]. Furthermore, by single sensors such theasidentification monitoring or determination five BZs using aofpolymeric different BZs by single interfaces sensors modified electrode such as monitoring in layer by fivelayer BZs manner using a polymeric [27], applying interfaces electrodevoltammetric square-wave modified in layer technique using a hanging mercury drop electrode [28], using silver nanoparticles-carbonmercury by layer manner [27], applying square-wave voltammetric technique using a hanging dots ink drop electrode modified [28], using electrodes [29],silver nanoparticles-carbon or employing a modified dots silverink modified solid amalgam electrodes electrode [29],can or be employing found in athe modified silver literature solid [30]. amalgam However, electrode none of thesecanworks be found in the literature accomplishes [30]. However, the simultaneous none of theseof determination works accomplishes the simultaneous determination of BZs, since they BZs, since they do it individually. Consequently, the simultaneous voltammetric determination of do it individually. Consequently, the simultaneous multiple BZs is still voltammetric a challenge,determination mainly due theofhigh multiple BZs is of overlapping still a challenge, redox peaks when mainly due the multiple BZs high overlapping of redox peaks when multiple BZs are present in the are present in the same solution. As the specific problem concerns finding a sensor-based analytical same solution. As the specific problem method concerns that is able finding a sensor-based to determine and quantify analytical method different that is able to determine BZs simultaneously, the use of and quantify a sensor array different BZs simultaneously, the use of a sensor array becomes an interesting becomes an interesting option. For this reason, this work aims to use an electronic tongue approach option. For this reason, this work aims to solve to usemixture a ternary an electronic of three tongue BZs:approach diazepam, to flunitrazepam, solve a ternary mixture and lorazepam,of three BZs: threediazepam, regulated flunitrazepam, and lorazepam, three regulated substances substances with equivalent functional base, and varying substituents (Figure 1). with equivalent functional base, and varying substituents (Figure 1). Figure 1. Chemical structures of diazepam, lorazepam, and flunitrazepam. Figure 1. Chemical structures of diazepam, lorazepam, and flunitrazepam. Electronic tongues (ETs) have been in use since mid-1990s to overcome limitations of single sensor methodologies, spreading the use of sensor arrays and chemometric tools. An ET consists of a multisensor system formed by low-selective sensors, and uses advanced mathematical procedures for Sensors 2019, 19, x FOR PEER REVIEW 3 of 12 SensorsElectronic tongues 2019, 19, 5002 (ETs) have been in use since mid-1990s to overcome limitations of single 3 of 12 sensor methodologies, spreading the use of sensor arrays and chemometric tools. An ET consists of a multisensor system formed by low-selective sensors, and uses advanced mathematical procedures signal processing for signal basedbased processing on pattern recognition on pattern and/or and/or recognition multivariate analysis, multivariate among these analysis, among Artificial these Neural Network (ANNs) or Principal Component Analysis Artificial Neural Network (ANNs) or Principal Component Analysis (PCA) [31]. These (PCA) [31]. These mathematical tools are widely used to tools mathematical correctareinterference widely used and/or to matrix correct effects in analytical interference and/orprocedures. In recent matrix effects published in analytical literature, procedures. electronic In recenttongues published have been used literature, to differentiate electronic tongues have and quantify been usedanalogue speciesand to differentiate that, quantify thanks analogue to the gatheringspecies that, thanks of multiple sensortodata,the gathering made possibleof multiple sensor resolution the successful data, madeofpossible situations the for successful which resolution traditional of situations techniques functionforwith which traditional difficulty. ETstechniques function can be classified with difficulty. or divided depending ETson canthe be classified techniques or divided employed, depending including on the techniques potentiometric [32,33],employed, amperometric including potentiometric [34], impedance based[32,33], [35–37] amperometric and voltammetric [34], impedance [38]. Some examplesbased in [35–37] recentand voltammetric literature have adopted [38]. Some examples voltammetric ETin recent principles literature to have adopted discriminate red wines voltammetric [39], to assesET wine principles sensorytodescriptors discriminate redtowines [40], detect[39], andtoquantify asses wine nitro sensory and descriptors peroxide [40], [41], explosives to detect and and quantify nitro to distinguish and peroxide aminothiols [42]. explosives An important [41],breakthrough and to distinguish in this aminothiols field has been [42]. theAn important generic approachbreakthrough employing inmolecularly this field hasimprinted been the generic polymers approach to formemploying (bio)sensor molecularly arrays [43]. imprinted polymers to form (bio)sensor arrays [43]. In the In the present present paper, paper, an an efficient efficient ETET approach approach is is suggested suggestedfor forthethesimultaneous simultaneousdetermination determination of ternary mixtures of BZs comprising diazepam, flunitrazepam, and lorazepam usingchemometric of ternary mixtures of BZs comprising diazepam, flunitrazepam, and lorazepam using chemometric tools coupled tools coupled withwith an electrochemical an electrochemical detection detection system. Multidimensional system. Multidimensional voltammetric voltammetric information information is obtained through the use of an array of different modified is obtained through the use of an array of different modified electrodes with a carbon base. electrodes with a carbon For this base. For this purpose voltammogram data were compressed using purpose voltammogram data were compressed using the Discrete Wavelet Transform (DFT), and the the Discrete Wavelet Transform (DFT), and resulting the resulting coefficients werecoefficients used as input were forused as input for the Artificial the Artificial Neural NetworkNeural (ANN)Network model as(ANN) a single model as a single vector formed by all the pre-treated signals. The vector formed by all the pre-treated signals. The resolution of any signal overlapping and quantificationresolution of any signal overlapping and quantification of the individual species considered of the individual species considered is achieved thanks to the extraction of the samples’ fingerprints is achieved thanks to the extraction of the samples’ fingerprints along with the capabilities of along with the capabilities of the ANN predictive model. Satisfactory models for identification and the ANN predictive model. Satisfactory resolution ofmodels for identification their mixtures were thusand resolution developed. of their Figure mixtures 2 shows were thus developed. schematically Figure ET the voltammetric 2 shows schematically the voltammetric ET strategy followed. The approach comprises succeeding strategy followed. The approach comprises succeeding stages: (A) an experimental design is used to stages: A) an experimental design is used to define the standards needed to build the response define the standards needed to build the response model; (B) these are measured using a voltammetric model; B) these are measured using a voltammetric sensor array; C) the data obtained is compressed sensor array; (C) the data obtained is compressed and fingerprints extracted; (D) the model is trained and fingerprints extracted; D) the model is trained with the data extracted from the measurements; with the data extracted from the measurements; and (E) the model is validated with an independent and E) the model is validated with an independent set of samples. set of samples. Figure 2. Scheme depicting the main steps of the ET approach used in the simultaneous determination Figure of BZs. 2. Scheme depicting the main steps of the ET approach used in the simultaneous determination of BZs. 2. Materials and Methods. 2. Materials and Methods. 2.1. Chemicals and Reagents 2.1. Chemicals and Reagents All chemicals used in this investigation were of analytical reagent grade. Water was purified by a All Milli-Q Ultra used chemicals PureinWater System (Millipore, this investigation were of Bedford, analyticalMA, USA). reagent Diazepam, grade. flunitrazepam, Water was purified by and lorazepam were purchased from Sigma-Aldrich (Madrid, Spain). All the a Milli-Q Ultra Pure Water System (Millipore, Bedford, MA, USA). Diazepam, flunitrazepam,samples were diluted andin lorazepam were purchased Britton-Robinson from Buffer, pH 10 Sigma-Aldrich (0.04 M H3 BO3 , 0.04 (Madrid, M H3 PO Spain). 4 , 0.04 All M CHthe3 COOH, samplesand were0.1diluted M NaOH).in Britton-Robinson Modifiers Buffer, employed pHpreparation in the 10 (0.04 M H of3BO 3, 0.04 Mwere electrodes H3PO 4, 0.04 M CH purchased 3COOH, from and 0.1 M NaOH). Sigma-Aldrich: platinum Modifiers nanopowder, employed in sizethe 50 nm particle (Ref. preparation 6845453); copperof (II) electrodes oxide, 50 nm were purchased particle from size (Ref. 544868); copper nanopowder, 40–60 nm particle size (Ref. 774111); and tungsten (VI) oxide, 100 nm particle size (Ref. 550086). Sensors 2019, 19, x FOR PEER REVIEW 4 of 12 Sigma-Aldrich: platinum nanopowder, 50 nm particle size (Ref. 6845453); copper (II) oxide, 50 nm particle Sensors size 2019, 19, (Ref. 5002 544868); copper nanopowder, 40–60 nm particle size (Ref. 774111); and tungsten 4 of 12 (VI) oxide, 100 nm particle size (Ref. 550086). 2.2. Preparation of the Modified Electrodes The main main electrodes electrodesemployed employedininthis thiswork workare aregraphite graphiteepoxy epoxy composite composite electrodes electrodes modified modified in different in differentvariants. In order variants. to build In order the electrodes to build a copper the electrodes disk wasdisk a copper soldered to an electrical was soldered to an connector. electrical The connector connector. The was then fitted connector into fitted was then a 6 mm intodiameter PVC tube,PVC a 6 mm diameter which provided tube, the mainthe which provided body of main the body electrode. Then a conductive of the electrode. paste was Then a conductive prepared paste using 50-µm was prepared usingparticle 50-µm size graphite particle powder size graphite (BDH powder laboratory Supplies,Supplies, (BDH laboratory Poole, UK) and Epotek Poole, UK) and H77 resin and Epotek H77hardener resin and(both from Epoxy hardener Technology, (both from Epoxy Billerica, Technology, MA,Billerica, USA), which MA, were USA),deposited, which were filling the cavity deposited, in the filling thevoid body. cavity in The the electrodes void body.were The cured in thewere electrodes oven cured for 72 hinand thethen oven repeatedly for 72 hpolished and then with sandpaperpolished repeatedly of decreasingwith grains sandpaperprior of to their readiness for measuring (see Figure 3). decreasing grains prior to their readiness for measuring (see Figure 3). Modified electrodes follow the same building process but instead of using a paste of graphite and epoxy resin, they incorporate a 5% of modifiers acquired from commercial source. In In this this work, work, the modifiers employed were metallic nanoparticles of Cu and Pt, oxide nanoparticles of CuO and WO3, WO3, a Pt disk electrode, and a graphite epoxy electrode (GEC) as a bare electrode. electrode. The use of suchsuch graphite graphite epoxy epoxy electrodes electrodes modified modified withwith nanotechnological nanotechnological components has been reported in our laboratory for its use in the detection of Brett character in wine [44], to asses wine sensory descriptors from a sensory panel [40], to analyze amino acid mixtures [45], to evaluate the antioxidant capacity of red wines [46], to discriminate beers [47], to analysis cava beverages beverages [48], [48], different electronic and for the comparison of different electronic tongues tongues forfor pharmaceutical pharmaceuticalanalysisanalysis[49]. [49]. mentionedbefore, As mentioned before, noble noble metal metal disc disc electrodes electrodes wereemployed were also also employed in the in the array. array. These These electrodes electrodes were were soldering constructed constructed soldering a Pt wire (99.95% a purity, Pt wire (99.95% diameter purity, 1 mm, diameter supplied 1 mm, supplied by Goodfellow, Cambridge, by Goodfellow, UK) Cambridge, to an electrical UK)and connector to then an electrical encasingconnector andin the connector then the encasing PVC tubethe connector (Servei Estació,inBarcelona, the PVC tube (Servei Spain). Estació, The wire Barcelona, was finally coatedSpain). in epoxyThe resin, wire cured was finally coated to and polished in let epoxy exposedresin, onlycured and the wire polished cross to let exposed only the wire cross section. section. Figure 3. Figure 3. Scheme Scheme ofof the the step-by-step step-by-step construction construction of of graphite-epoxy composite electrodes. graphite-epoxy composite electrodes. (I) (I) Copper Copper disk soldered to the connector. (II) Assembly into the PVC tube. (III) Incorporation of disk soldered to the connector. (II) Assembly into the PVC tube. (III) Incorporation of graphite-epoxy graphite-epoxy mixture. (VI) Polishing of the mixture. (VI) Polishing of the hardened surface. hardened surface. 2.3. Electronic Tongue 2.3. Electronic Tongue The The voltammetric voltammetricETET was formed was formedby an byintegrated array of an integrated six modified array graphite epoxy of six modified graphiteelectrodes epoxy as electrodes as working electrodes, described in the previous section. The electrochemicalin cell working electrodes, described in the previous section. The electrochemical cell employed the measurements was formed by 6 working electrodes, an Ag/AgCl reference employed in the measurements was formed by 6 working electrodes, an Ag/AgCl reference electrode, and a platinum electrode electrode,as counter and electrode a platinum (physical electrode area 0.47 as counter cm2 ). (physical area 0.47 cm2). electrode Electrochemical measurementswere Electrochemical measurements were performed performed at at room room temperature temperature (25(25 ◦°C), C), using using aa 6-channel 6-channel AUTOLAB AUTOLAB PGSTAT20 PGSTAT20 (Metrohm (Metrohm Autolab Autolab BV, BV, Utrecht, Utrecht, The The Netherlands) Netherlands) controlled controlled with with the the GPES GPES 4.74.7 Multichannel software package. Multichannel software package. A A complete completevoltammogram voltammogramwas wasrecorded recorded forfor each eachsample sample by by cycling the the cycling potential between potential between−1.5−1.5 V andV +1.5 V vs. Ag/AgCl with a step potential of 9 mV and a scan rate of 100 mV·s −1 and +1.5 V vs. Ag/AgCl with a step potential of 9 mV and a scan rate of 100 mV·s . Acquired . Acquired voltammogram −1 for data processing voltammogram was processing for data the fourth (in wasa the series of 4)(in fourth to aimprove series ofreproducibility of measurements. 4) to improve reproducibility of A conditioning potential of +1.2 V was applied during 60 s in a pH 10 solution measurements. A conditioning potential of +1.2 V was applied during 60 s in a pH 10 solution after each measurementafter to electrochemically each measurement clean the electrochemical to electrochemically surfaces, clean thereby preventing the electrochemical foulingthereby surfaces, and drifting effects. preventing fouling and drifting effects. 2.4. Model Design and Sample Preparation In order to build an acceptable model with good generalization capabilities a considerable number of data points are usually needed for the training of the model. As stated in the European 2.4. Model Design and Sample Preparation In order to build an acceptable model with good generalization capabilities a considerable number Sensorsof data 2019, points are usually needed for the training of the model. As stated in the 5European 19, 5002 of 12 Pharmacopoeia guidelines for multivariate calibrations [50], the size of the dataset needed for building the calibration is dependent on interfering properties and the number of analytes that the Pharmacopoeia guidelines for multivariate calibrations [50], the size of the dataset needed for building model is expected to handle. Thus, this leads to two options: increase the size of data set to overcome the calibration is dependent on interfering properties and the number of analytes that the model is the variability of the measurements [51], or employ a statistically defined data set [41]. expected to handle. Thus, this leads to two options: increase the size of data set to overcome the In this case, the second option was preferred. The model employed was based on a factorial variability of the measurements [51], or employ a statistically defined data set [41]. design [52] with In this 3 levels case, and 3option the second factorswas (27preferred. samples).TheThemodel final model employedwaswas obtained after based on the factorial a factorial wasdesign tilted [52] 45° in each axis for a better final representation. The training subset comprised with 3 levels and 3 factors (27 samples). The final model was obtained after the factorial75% of the totalwas samples. ◦ tilted 45 in each axis for a better final representation. The training subset comprised 75% of the Once total the model was trained, an external test subset of 9 samples (25%) was employed to samples. validateOnce the model model. the response was trained, The an external test subset concentrations of 9 sample of each samples (25%) of thewas testemployed to validate subset were generated the response model. The concentrations of each sample of the randomly within the limits of the experimental domain (0–30 ppm) (Figure 4).test subset were generated randomly within the limits of the experimental domain (0–30 ppm) (Figure 4). Figure 4. Scheme of the modified full factorial design used as train subset (blue) and randomly generated Figure external 4. Scheme of test thesubset (green). modified full factorial design used as train subset (blue) and randomly generated external test subset (green). Both sample subsets were prepared in the same way and the samples were prepared in a Britton–Robinson buffer, pH = 10.0. In order to minimize the degradation of the chemical species Both sample subsets were prepared in the same way and the samples were prepared in a in solution and reduce a possible source of data variability, fresh stock solutions of diazepam, Britton–Robinson buffer, pH = 10.0. In order to minimize the degradation of the chemical species in flunitrazepam, and lorazepam were prepared on the same day the measurements were made. solution and reduce a possible source of data variability, fresh stock solutions of diazepam, 2.5. Data Processing flunitrazepam, and lorazepam were prepared on the same day the measurements were made. Statistical treatment and data analysis were performed using routines written by the authors using 2.5. MATLAB Data Processing R2017a (MathWorks, Natick, MA, USA), together with its Neural Network and Wavelet toolboxes. StatisticalSigmaplot (Systat treatment and Software Inc., San data analysis Jose,performed were CA, USA) was usedroutines using to graphically writtenrepresent by theand authors analyze the results. using MATLAB R2017a (MathWorks, Natick, MA, USA), together with its Neural Network and Wavelet toolboxes. Sigmaplot (Systat Software Inc., San Jose, CA, USA) was used to graphically 3. Results represent and analyze the results. 3.1. Voltammetric Array Response 3. Results Electrochemical response of BZs is largely dictated by its common core element, i.e., the 4,5- azomethine group. The electrochemical behavior has been explained by the 2e− , 2H+ reduction of the 3.1. 4,5-azomethine Voltammetric Array groupResponse at the ring to give the corresponding dihydro species (see Figure 5), plus extra reduction of additionally present active moieties, e.g., the nitro group in the case of flunitrazepam. Electrochemical response of BZs is largely dictated by its common core element, i.e., the 4,5-azomethine group. The electrochemical behavior has been explained by the 2e−, 2H+ reduction of the 4,5-azomethine group at the ring to give the corresponding dihydro species (see Figure 5), plus extra reduction of additionally present active moieties, e.g., the nitro group in the case of flunitrazepam. Sensors 2019, 19, x FOR PEER REVIEW 6 of 12 Sensors 2019, 19, 5002 6 of 12 Sensors 2019, 19, x FOR PEER REVIEW 6 of 12 Figure 5. The oxidation of 4,5-azomethine redox mechanism. In Figure 6 we can see 5. an example of of the different voltammograms obtained for 25 ppm of 5. TheThe Figure Figure oxidation oxidation 4,5-azomethine of 4,5-azomethine redox redoxmechanism. mechanism. diazepam, flunitrazepam, and lorazepam in a Britton–Robinson buffer at pH 10.0 (conditions In Figure according 6 we can to similar works seeinanthe example literatureof [22]), the different voltammograms for: (A) bare epoxy-graphite obtained for electrode, composite 25 ppm of In Figure nanoparticles diazepam, 6 we can seeand (B) CuO flunitrazepam, anlorazepam example modified ofathe in graphite different voltammograms Britton–Robinson electrode, (C) Cu buffer at pH 10.0obtained nanoparticles (conditionsfor 25 ppm of modified according graphite diazepam, to similarflunitrazepam, electrode, works and lorazepam (D) Ptin nanoparticles the literature [22]), for: in modified (A)a bare graphiteBritton–Robinson epoxy-graphite electrode, (E) WO buffer composite at electrode, pH 10.0 3 nanoparticles (conditions (B) CuO modified according to similar graphite electrode, nanoparticles works in the and graphite modified literature (F) Pt metal [22]), electrode. electrode, (C) Cu for: (A) bare As nanoparticles epoxy-graphite can be seen inmodified Figure 6,graphite composite each sensor electrode, provides electrode, (D) aPt (B) CuO nanoparticles singular nanoparticles modified modified voltammetric response. graphite The graphite electrode, (E) electrode, employed WO sensors 3 (C) presented nanoparticles Cu nanoparticles different modified modified sensitivities graphite towards electrode, graphite andthe(F) different electrode, Pt metal BZs, displaying (D)electrode. a Pt nanoparticles complementary As can be modified redox seen in Figure profile. graphite 6, each This being electrode, one of the necessary (E) WOa3 singular sensor provides conditions nanoparticles for modified voltammetric obtaining response. graphite a properly The electrode, andfunctioning employed(F)sensors Pt metal ET, obtaining presented differentiated different electrode. As can besignals sensitivities forFigure seentowards in eachthe considered 6, each substance different BZs, sensor with displaying provides a aeach sensor. complementary redox profile. This being one of the necessary conditions singular voltammetric response. The employed sensors presented different sensitivities towards the for obtaining a properly functioning different ET, obtaining BZs, displaying differentiated signals a complementary redoxfor each considered profile. This beingsubstance one of thewith each sensor. necessary conditions for obtaining a properly functioning ET, obtaining differentiated signals for each considered substance with each sensor. Figure 6. Voltammograms obtained with the three benzodiazepines used in this study (25 ppm) using a (A) bare epoxy-graphite composite electrode, and graphite electrodes modified with (B) Figure 6. Voltammograms obtained with the three benzodiazepines used in this study (25 ppm) using a CuO nanoparticles, (C) Cu nanoparticles, (D) Pt Nanoparticles, (E) WO3 nanoparticles, and a (F) Pt (A) bare epoxy-graphite composite electrode, and graphite electrodes modified with (B) CuO metal electrode. nanoparticles, (C) Cu nanoparticles, (D) Pt Nanoparticles, (E) WO3 nanoparticles, and a (F) Pt metal In order toelectrode. mathematically assess whether the selected sensor array was able to provide the necessary signal complementarity when the different BZs are measured, a Principal Component In order to mathematically assess whether the selected sensor array was able to provide the Analysis (PCA) treatment was performed on a series of measurements using the three substances necessary signal complementarity when the different BZs are measured, a Principal Component studied. Figure 6.Specifically, Voltammograms this was treatedwith obtained withthe separate voltammogramsused three benzodiazepines for each electrode in this andppm) study (25 eachusing samplea Analysis (PCA) treatment was performed on a series of measurements using the three substances employed (A) bare in this study. With epoxy-graphite this multivariate composite visualization, electrode, and each sensor graphite appears electrodes differentiated modified with (showing (B) CuO studied. Specifically, this was treated with separate voltammograms for each electrode and each absence of collinearity) nanoparticles, (C) Cu and even the different nanoparticles, (D)this chemical Pt compounds Nanoparticles, (E) group WO3 together, suggesting nanoparticles, and the a sample employed in this study. With multivariate visualization, each sensor appears proper (F) clustering Pt metal electrode.for the intended application (see Figure 7). Moreover, and as can be seen in the score differentiated (showing absence of collinearity) and even the different chemical compounds group plot, the sensor’s scores are located in different quadrants of the plot, demonstrating complementarity together, suggesting the proper clustering for the intended application (see Figure 7). Moreover, and of the sensors employed. These appear also away from the 0,0 origin, which suggests they provide In as order can be to mathematically seen in the score plot, assess whether the sensor’s the are scores selected locatedsensor arrayquadrants in different was ableoftothe provide plot, the useful information for the discrimination of samples. necessary signal complementarity when the different BZs are measured, a Principal Component Analysis (PCA) treatment was performed on a series of measurements using the three substances studied. Specifically, this was treated with separate voltammograms for each electrode and each sample employed in this study. With this multivariate visualization, each sensor appears differentiated (showing absence of collinearity) and even the different chemical compounds group together, suggesting the proper clustering for the intended application (see Figure 7). Moreover, and Sensors 2019, 19, x FOR PEER REVIEW 7 of 12 demonstrating complementarity of the sensors employed. These appear also away from the 0,0 origin, which Sensors suggests 2019, 19, 5002 they provide useful information for the discrimination of samples. 7 of 12 Figure 7. Score plot of the first two principal components obtained after PCA. Five replicates for each Figure 7. Score plot of the first two principal components obtained after PCA. Five replicates for each compound (concentration of 25 ppm) were measured with the six sensors in the ET array. Diazepam, compound (concentration (red circle); of 25 flunitrazepam ppm) (pink wereand triangle), measured with lorazepam (bluethe six sensors in the ET array. Diazepam, cross). (red circle); flunitrazepam (pink triangle), and lorazepam (blue cross). 3.2. Quantitative Analysis: Artificial Neural Network Modeling An ANN 3.2. Quantitative response Analysis: model was Artificial built from Neural the setModeling Network of compressed voltammograms in order to predict the three BZs concentrations in a given sample. For this a training subset of standards (n = 27) was AnusedANN response as defined by amodel was built tilted factorial design.fromNext,thetheset of compressed prediction voltammograms and generalization capabilities in of order to predict the themodel threewereBZs tested concentrations by employing in aa subset givenofsample. unrelated Forand this a training unknown subset samples to ourofmodel, standards i.e., (n = 27) the test subset (n = 9), defined by random concentration values of the was used as defined by a tilted factorial design. Next, the prediction and generalization capabilitiesthree BZs. As commented, given the complexity of the electrochemical information obtained from each of the model were tested by employing a subset of unrelated and unknown samples to our model, sample, it is necessary to perform a pre-processing compression step. This step is necessary in order i.e., the to test subset avoid (n = 9),ofdefined over-fitting by random the developed models concentration and to reduce thevalues of thetime computation three BZs. spent during the Astraining commented, given the complexity of the electrochemical information of the different models [31]. In this particular case, to reduce the data complexity obtained obtained from each sample,from it is the necessary samples, totheperform voltammetrica pre-processing data, the Discrete compression Wavelet Transform step. This (DWT) step wasis necessary employed to in order reduce the 4512 current values obtained per sample (i.e., the six to avoid over-fitting of the developed models and to reduce the computation time spent during the voltammograms) to a set of 144 wavelet coefficients per sample, in this case the approximation coefficients. The mother wavelet used training of the different models [31]. In this particular case, to reduce the data complexity obtained was Daubechies 3, and the compression level used was 3. Values of compression achieved in the from the samples, the preprocessing stepvoltammetric for the different data, the Discrete compression Wavelet levels assayed wereTransform 88.9% for the(DWT) was first, 94.1% foremployed the to reduce second, the 4512 andcurrent 96.7% forvalues the thirdobtained variant, theper sample latter of which (i.e., was the six voltammograms) to a set of 144 chosen. wavelet coefficients The obtained per sample, model in this ofcase was composed the approximation 144 neurons in the input layercoefficients. (24 per sensor),Thefive mother neurons wavelet used was in the hidden layer, Daubechies 3, and andthree the neurons compression in the output layer. The level used wastransfer 3. Values functions employed in the of compression achieved in hidden layer and output layer were satlins and tansig, respectively. These configuration details were the preprocessing step for the different compression levels assayed were 88.9% for the first, 94.1% chosen after exhaustive evaluation of roughly 200 different possibilities for the ANN, originating in a for the second, and 96.7% test of a number betweenfortwo theand thirdninevariant, neuronsthe latter in the hidden of which layers, andwasthe chosen. use of the five transfer The obtained functions model hardlims, was purelin satlins, composed(linearlyof shaped), 144 neurons and satlins in the input(S-shaped) and tansig layer (24both perinsensor), the five neurons in the hidden layer, and three neurons in the output layer. The transfer functions employed hidden and output layers. Choice was performed looking for the minimum normalized Root Mean Squared Error in the hidden layerasand the measure outputused layerto quantify the fitting were satlins degree. and tansig, respectively. These configuration Performance of the trained ET is summarized in Figure 8. Comparison graphs of obtained details were chosen after exhaustive evaluation of roughly 200 different possibilities for the ANN, vs. expected BZs concentrations were built in order to visualize the prediction ability of the ANN originating model. in Moreover a test of their a number between best linear regressiontwolineand wasnine neurons calculated in the hidden (separately layers,and for the training andthethe use of the fivetesting transfer subsets)functions for the threehardlims, satlins, different BZs. Idealpurelin values for (linearly shaped), the comparison lines and should satlins and tansig be 1.0 for (S-shaped) both 0.0 the slope, in the theintercept, hidden and and1.0output for thelayers. correlationChoice was performed coefficient, respectively. looking The figurefor the aminimum shows normalized Root Mean Squared Error as the measure used to quantify the fitting degree. Performance of the trained ET is summarized in Figure 8. Comparison graphs of obtained vs. expected BZs concentrations were built in order to visualize the prediction ability of the ANN model. Moreover their best linear regression line was calculated (separately for the training and the Sensors 2019, 19, 5002 8 of 12 Sensors 2019, 19, x FOR PEER REVIEW 8 of 12 satisfactory fit for the training data and also the correct trend for the external test subset, with perhaps aperhaps somewhat a somewhat higher for higher dispersion dispersion forThis the latter. theislatter. Thisresult the usual is thewhen usualtheresult properwhen the proper modelling has been attained—almost perfect equivalence for the training subset, given the model is built from thisis modelling has been attained—almost perfect equivalence for the training subset, given the model builtand data, from this data, a correct trendandforathe correct trend external testfor the external subset, perhapstestwithsubset, perhapsHowever less accuracy. with lessit accuracy. must be However said it must these data werebenot saidinvolved these data were in the not involved development of in thethe development model. of the Correlation model. Correlation coefficients indicated coefficients on the graphindicated on the1.0 are practically graph for are the practically 1.0 forand training subset thebetween training subset and between 0.938–0.961 0.938–0.961 for the test subset, for the test subset, values values also close to ideality. also close to ideality. Figure8.8.Obtained Figure Obtainedvs. vs.expected expectedconcentrations concentrationsresults resultsplots plotsfor forthe thetraining trainingset set(black (blackdots) dots)and andthe the testing set (white dots) for (A) diazepam, (B) flunitrazepam, and (C) lorazepam. testing set (white dots) for (A) diazepam, (B) flunitrazepam, and (C) lorazepam. The detailed results of regression lines for each compound and both train and test sub-sets are The detailed results of regression lines for each compound and both train and test sub-sets are presented in Table 1. Acceptable values close to the ideal values were obtained for all the compounds presented in Table 1. Acceptable values close to the ideal values were obtained for all the with better values for the training set than for the testing set as it is usually obtained. Additionally, compounds with better values for the training set than for the testing set as it is usually obtained. the table also includes NRMSE values as a measure of the goodness of fit with figures close to zero for Additionally, the table also includes NRMSE values as a measure of the goodness of fit with figures both training and testing sets. close to zero for both training and testing sets. Table 1. Results of the fitted regression lines for obtained vs. expected values for the training and Table 1. Results of the fitted regression lines for obtained vs. expected values for the training and testing subsets of samples and the three considered compounds. testing subsets of samples and the three considered compounds. Set Analyte R2 Slope Intercept (ppm) NRMSE Combined NRMSE Intercept Combined Set Analyte Diazepam R2 0.996 0.932 Slope1.358 (ppm) 0.035 NRMSE Training Set NRMSE Flunitrazepam 0.996 0.939 1.196 0.034 0.034 (n Training = 27) Diazepam 0.996 Lorazepam 0.997 0.923 0.932 1.498 1.358 0.034 0.035 Set Flunitrazepam 0.996 0.939 1.196 0.034 0.034 Diazepam 0.961 0.977 0.075 0.105 (n =Testing 27) Set Lorazepam Flunitrazepam 0.997 0.943 0.909 0.923 2.714 1.498 0.113 0.034 0.106 (n = 9) Testing Lorazepam Diazepam 0.938 0.961 0.884 0.977 2.044 0.075 0.114 0.105 Set Flunitrazepam 0.943 0.909 2.714 0.113 0.106 (n =Finally, 9) toLorazepam give a contrast to 0.938 0.884Table 2 was the results herein, 2.044 prepared for0.114 the examination and comparison of the performance of the present work and those of related papers in the literature. As can be seenFinally, to give from Table a contrast 2 linear rangestoand thelimits results of herein, Table detections are 2similar was prepared forworks to those of the examination and in the literature comparison but of the the presented performance work of the has the clear present work advantage and those of quantifying of simultaneously related papers in BZs three the literature. species. As can be seen from Table 2 linear ranges and limits of detections are similar to those of works in the literature but the presented work has the clear advantage of simultaneously quantifying three BZs species. Sensors 2019, 19, 5002 9 of 12 Table 2. Analytical results of the work presented herein and most relevant references, This table compares the substances detected, the sensor and employed electrochemical technique, the linear range, and the limit of detection. Lineal Range Limit of Substances. Electrode Technique Reference (M) Detection (µM) Ag nanodendrimers supported by Differential pulse Diazepam graphene nanosheets modified 1.0 × 10−7 –1.0 × 10−6 0.086 [15] voltammetry glassy carbon electrode Differential pulse adsorptive Diazepam Unmodified SPCE 2.5 × 10−5 –1 × 10−3 6.3 [16] stripping voltammetry Fullerene functionalized carbon Differential pulse Diazepam nanotubes/ionic 0.3 × 10−6 –7 × 10−4 0.0087 [17] voltammetry liquidnanocomposite GCE Multiwall Carbon Nanotube-Ionic Diazepam Square wave voltammetry 7 × 10−8 –2.5 × 10−6 0.014 [23] Liquid Modified Paste Electrode Modified polypyrrole@sol-gel@gold Lorazepam Cyclic voltammetry 0.2 × 10−9 –2 × 10−9 0.0009 [18] nanoparticles/pencil graphite electrode Flunitrazepam SPGE Cyclic voltammetry 3.2 × 10−8 –6.4 × 10−7 0.019 [20] Diazepam Cyclic 3.5 × 10−5 –1.1 × 10−4 6.0 Metal nanoparticle modified Flunitrazepam voltammetry–Electronic 3.2 × 10−5 –9.6 × 10−4 5.6 This work graphite epoxy sensor array Lorazepam tongue 3.1 × 10−5 –9.3 × 10−5 4.6 4. Conclusions The proposed approach reports the use of a voltammetric sensor array that has been developed and validated as a rapid, reliable, and accurate method for simultaneous determination of ternary mixtures of BZs products diazepam, flunitrazepam, and lorazepam. A set of standards based on modified full factorial design was prepared to build the model. After validation with an external test set the model successfully predicted the concentration of three BZs despite accused overlapping of the voltammetric signals. The work presented herein shows a promising tool for simultaneously determining mixtures of these regulated substances in a fast, cost efficient, and reliable manner with performance close to that of sophisticate equipment such as HPLC. The ET developed was able to quantify the individual components of ternary mixtures of diazepam, flunitrazepam, and lorazepam with a NRMSE as low as 0.105 for the independent sample subset. However, the drawback of the reported procedure might be the need for a specifically defined buffer and measurement media, i.e., the Britton–Robinson buffer, although the involved effort for a given sample is minimum. Nevertheless, future works should consider compatibility with different biologic media, such as urine or blood serum. These conditions need to be further explored in order to maximize the strong points of the ET approach. The presented work nevertheless opens a promising forensic/illicit drug analysis as an alternative to traditional methodologies. Author Contributions: F.T. and A.H.-C. did the experimental measurements. A.G.-C. performed the calculations and the ANN modelling. A.H.-C.; M.d.V. and A.G. wrote the manuscript. F.F. and J.B.G. supervised work of F.T. M.d.V. planned and supervised the research. Funding: Financial support for this work was provided by Spanish Ministry, Ministerio de Economía y Competitividad, through the project CTQ2016-80170-P. Acknowledgments: Anna Herrera-Chacon thanks Universtitat Autònoma de Barcelona (UAB) for the PIF fellowship. Manel del Valle is thankful for support from the program ICREA Academia from Generalitat de Catalunya. The authors want to acknowledge the enlightening discussions with Adrià Aguilar-Catalan from Veterinary Faculty, Universitat Autònoma de Barcelona. Conflicts of Interest: The authors declare no conflict of interest. References 1. Persona, K.; Madej, K.; Knihnicki, P.; Piekoszewski, W. Analytical methodologies for the determination of benzodiazepines in biological samples. J. Pharm. Biomed. Anal. 2015, 113, 239–264. [CrossRef] Sensors 2019, 19, 5002 10 of 12 2. Lévi, S.; Le Roux, N.; Eugène, E.; Poncer, J.C. Benzodiazepine ligands rapidly influence GABAA receptor diffusion and clustering at hippocampal inhibitory synapses. Neuropharmacology 2015, 88, 199–208. [CrossRef] [PubMed] 3. Ariffin, M.M.; Miller, E.I.; Cormack, P.A.G.; Anderson, R.A. Molecularly imprinted solid-phase extraction of diazepam and its metabolites from hair samples. Anal. Chem. 2007, 79, 256–262. [CrossRef] [PubMed] 4. Uddin, M.N.; Samanidou, V.F.; Papadoyannis, I.N. 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. 2008, 31, 1258–1282. [CrossRef] 5. Anderson, R.A.; Ariffin, M.M.; Cormack, P.A.G.; Miller, E.I. Comparison of molecularly imprinted solid-phase extraction (MISPE) with classical solid-phase extraction (SPE) for the detection of benzodiazepines in post-mortem hair samples. Forensic Sci. Int. 2008, 174, 40–46. [CrossRef] [PubMed] 6. Dziadosz, M.; Teske, J.; Henning, K.; Klintschar, M.; Nordmeier, F. LC–MS/MS screening strategy for cannabinoids, opiates, amphetamines, cocaine, benzodiazepines and methadone in human serum, urine and post-mortem blood as an effective alternative to immunoassay based methods applied in forensic toxicology for preliminary. Forensic Chem. 2018, 7, 33–37. [CrossRef] 7. Riahi, S.; Bagherzadeh, K.; Davarkhah, N.; Ganjali, M.R.; Norouzi, P. Spectrophotometric and chemometric studies on the simultaneous determination of two benzodiazepines in human plasma. Mater. Sci. Eng. C 2011, 31, 992–996. [CrossRef] 8. Furlanetto, S.; Orlandini, S.; Massolini, G.; Faucci, M.T.; La Porta, E.; Pinzauti, S. Optimisation and validation of a capillary electrophoresis method for the simultaneous determination of diazepam and otilonium bromide. Analyst 2001, 126, 1700–1706. [CrossRef] 9. Piergiovanni, M.; Cappiello, A.; Famiglini, G.; Termopoli, V.; Palma, P. Determination of benzodiazepines in beverages using green extraction methods and capillary HPLC-UV detection. J. Pharm. Biomed. Anal. 2018, 154, 492–500. [CrossRef] 10. Pettersson Bergstrand, M.; Helander, A.; Beck, O. Development and application of a multi-component LC–MS/MS method for determination of designer benzodiazepines in urine. J. Chromatogr. B Anal. Technol. Biomed. Life Sci. 2016, 1035, 104–110. [CrossRef] 11. Dickson, S.; Park, A.; Nolan, S.; Kenworthy, S.; Nicholson, C.; Midgley, J.; Pinfold, R.; Hampton, S. The recovery of illicit drugs from oral fluid sampling devices. Forensic Sci. Int. 2007, 165, 78–84. [CrossRef] [PubMed] 12. Bertol, E.; Vaiano, F.; Furlanetto, S.; Mari, F. Cross-reactivities and structure-reactivity relationships of six benzodiazepines to EMIT® immunoassay. J. Pharm. Biomed. Anal. 2013, 84, 168–172. [CrossRef] [PubMed] 13. Honeychurch, K.C.; Brooks, J.; Hart, J.P. Development of a voltammetric assay, using screen-printed electrodes, for clonazepam and its application to beverage and serum samples. Talanta 2016, 147, 510–515. [CrossRef] [PubMed] 14. Jain, R.; Sinha, A.; Kumari, N.; Khan, A.L. A polyaniline/graphene oxide nanocomposite as a voltammetric sensor for electroanalytical detection of clonazepam. Anal. Methods 2016, 8, 3034–3045. [CrossRef] 15. Majidi, M.R.; Ghaderi, S.; Asadpour-Zeynali, K.; Dastangoo, H. Synthesis of dendritic silver nanostructures supported by graphene nanosheets and its application for highly sensitive detection of diazepam. Mater. Sci. Eng. C 2015, 57, 257–264. [CrossRef] 16. Honeychurch, K.C.; Crew, A.; Northall, H.; Radbourne, S.; Davies, O.; Newman, S.; Hart, J.P. The redox behaviour of diazepam (Valium® ) using a disposable screen-printed sensor and its determination in drinks using a novel adsorptive stripping voltammetric assay. Talanta 2013, 116, 300–307. [CrossRef] 17. Rahimi-Nasrabadi, M.; Khoshroo, A.; Mazloum-Ardakani, M. Electrochemical determination of diazepam in real samples based on fullerene-functionalized carbon nanotubes/ionic liquid nanocomposite. Sens. Actuators B Chem. 2017, 240, 125–131. [CrossRef] 18. Rezaei, B.; Boroujeni, M.K.; Ensafi, A.A. A novel electrochemical nanocomposite imprinted sensor for the determination of lorazepam based on modified polypyrrole@sol-gel@gold nanoparticles/pencil graphite electrode. Electrochim. Acta 2014, 123, 332–339. [CrossRef] 19. Mohammadi-Behzad, L.; Gholivand, M.B.; Shamsipur, M.; Gholivand, K.; Barati, A.; Gholami, A. Highly sensitive voltammetric sensor based on immobilization of bisphosphoramidate-derivative and quantum dots onto multi-walled carbon nanotubes modified gold electrode for the electrocatalytic determination of olanzapine. Mater. Sci. Eng. C 2016, 60, 67–77. [CrossRef] Sensors 2019, 19, 5002 11 of 12 20. Garcia-Gutierrez, E.; Lledo-Fernandez, C. Electroanalytical sensing of Flunitrazepam based on screen printed graphene electrodes. Chemosensors 2013, 1, 68–77. [CrossRef] 21. Tseliou, F.; Pappas, P.; Spyrou, K.; Hrbac, J.; Prodromidis, M.I. Lab-On-A-Screen-Printed electrochemical cell for drop-volume voltammetric screening of flunitrazepam in untreated, undiluted alcoholic and soft drinks. Biosens. Bioelectron. 2019, 132, 136–142. [CrossRef] [PubMed] 22. Lozano-Chaves, M.E.; Palacios-Santander, J.M.; Cubillana-Aguilera, L.M.; Naranjo-Rodríguez, I.; Hidalgo-Hidalgo-de-Cisneros, J.L. Modified carbon-paste electrodes as sensors for the determination of 1,4-benzodiazepines: Application to the determination of diazepam and oxazepam in biological fluids. Sens. Actuators B Chem. 2006, 115, 575–583. [CrossRef] 23. Zare, M.A.; Tehrani, M.S.; Husain, S.W.; Azar, P.A. Multiwall carbon nanotube-ionic liquid modified paste electrode as an efficient sensor for the determination of diazepam and oxazepam in real samples. Electroanalysis 2014, 26, 2599–2606. [CrossRef] 24. Samiec, P.; Švorc, L.; Stanković, D.M.; Vojs, M.; Marton, M.; Navrátilová, Z. Mercury-Free and modification-free electroanalytical approach towards bromazepam and alprazolam sensing: A facile and efficient assay for their quantification in pharmaceuticals using boron-doped diamond electrodes. Sens. Actuators B Chem. 2017, 245, 963–971. [CrossRef] 25. Merli, D.; Dondi, D.; Pesavento, M.; Profumo, A. Electrochemistry of olanzapine and risperidone at carbon nanotubes modified gold electrode through classical and DFT approaches. J. Electroanal. Chem. 2012, 683, 103–111. [CrossRef] 26. Naggar, A.H.; Elkaoutit, M.; Naranjo-Rodriguez, I.; El-Sayed, A.E.A.Y.; De Cisneros, J.L.H.H. Use of a Sonogel-Carbon electrode modified with bentonite for the determination of diazepam and chlordiazepoxide hydrochloride in tablets and their metabolite oxazepam in urine. Talanta 2012, 89, 448–454. [CrossRef] 27. Ashrafi, H.; Mobed, A.; Hasanzadeh, M.; Babaie, P.; Ansarin, K.; Jouyban, A. Monitoring of five benzodiazepines using a novel polymeric interface prepared by layer by layer strategy. Microchem. J. 2019, 146, 121–125. [CrossRef] 28. Dos Santos, M.M.C.; Famila, V.; Gonçalves, M.L.S. Square-Wave voltammetric techniques for determination of psychoactive 1,4-benzodiazepine drugs. Anal. Bioanal. Chem. 2002, 374, 1074–1081. [CrossRef] 29. Ashrafi, H.; Hassanpour, S.; Saadati, A.; Hasanzadeh, M.; Ansarin, K.; Ozkan, S.A.; Shadjou, N.; Jouyban, A. Sensitive detection and determination of benzodiazepines using silver nanoparticles-N-GQDs ink modified electrode: A new platform for modern pharmaceutical analysis. Microchem. J. 2019, 145, 1050–1057. [CrossRef] 30. Samiec, P.; Navrátilová, Z.; Fischer, J. Voltammetry of benzodiazepines on meniscus-modified silver solid amalgam electrode. Mon. Chem. 2016, 147, 127–134. [CrossRef] 31. Moreno-Barón, L.; Cartas, R.; Merkoçi, A.; Alegret, S.; Gutiérrez, J.M.; Leija, L.; Hernandez, P.R.; Muñoz, R.; del Valle, M. Data compression for a voltammetric electronic tongue modelled with artificial neural networks. Anal. Lett. 2005, 38, 2189–2206. [CrossRef] 32. Dias, L.G.; Fernandes, A.; Veloso, A.C.A.; Machado, A.A.S.C.; Pereira, J.A.; Peres, A.M. Single-cultivar extra virgin olive oil classification using a potentiometric electronic tongue. Food Chem. 2014, 160, 321–329. [CrossRef] [PubMed] 33. Ciosek, P.; Wróblewski, W. Potentiometric electronic tongues for foodstuff and biosample recognition-an overview. Sensors 2011, 11, 4688–4701. [CrossRef] [PubMed] 34. Scampicchio, M.; Benedetti, S.; Brunetti, B.; Mannino, S. Amperometric electronic tongue for the evaluation of the tea astringency. Electroanalysis 2006, 18, 1643–1648. [CrossRef] 35. Pioggia, G.; Di Francesco, F.; Ferro, M.; Sorrentino, F.; Salvo, P.; Ahluwalia, A. Characterization of a carbon nanotube polymer composite sensor for an impedimetric electronic tongue. Microchim. Acta 2008, 163, 57–62. [CrossRef] 36. Cortina-Puig, M.; Muñoz-Berbel, X.; Alonso-Lomillo, M.A.; Muñoz-Pascual, F.J.; del Valle, M. EIS multianalyte sensing with an automated SIA system—An electronic tongue employing the impedimetric signal. Talanta 2007, 72, 774–779. [CrossRef] [PubMed] 37. Braga, G.S.; Paterno, L.G.; Fonseca, F.J. Performance of an electronic tongue during monitoring 2-methylisoborneol and geosmin in water samples. Sens. Actuators B Chem. 2012, 171, 181–189. [CrossRef] 38. Arrieta, Á.A.; Rodríguez-Méndez, M.L.; de Saja, J.A.; Blanco, C.A.; Nimubona, D. Prediction of bitterness and alcoholic strength in beer using an electronic tongue. Food Chem. 2010, 123, 642–646. [CrossRef] Sensors 2019, 19, 5002 12 of 12 39. Lvova, L.; Yaroshenko, I.; Kirsanov, D.; Di Natale, C.; Paolesse, R.; Legin, A. Electronic tongue for brand uniformity control: A case study of apulian red wines recognition and defects evaluation. Sensors 2018, 18, 2584. [CrossRef] 40. Cetó, X.; González-Calabuig, A.; Crespo, N.; Pérez, S.; Capdevila, J.; Puig-Pujol, A.; del Valle, M. Electronic tongues to assess wine sensory descriptors. Talanta 2017, 162, 218–224. [CrossRef] 41. González-Calabuig, A.; Cetó, X.; del Valle, M. Electronic tongue for nitro and peroxide explosive sensing. Talanta 2016, 153, 340–346. [CrossRef] [PubMed] 42. Pérez-Ràfols, C.; Gómez, A.; Serrano, N.; Díaz-Cruz, J.M.; Ariño, C.; Esteban, M. A voltammetric electronic tongue based on commercial screen-printed electrodes for the analysis of aminothiols by differential pulse voltammetry. Electroanalysis 2017, 29, 1559–1565. [CrossRef] 43. Herrera-Chacon, A.; González-Calabuig, A.; Campos, I.; del Valle, M. Bioelectronic tongue using MIP sensors for the resolution of volatile phenolic compounds. Sens. Actuators B Chem. 2018, 258, 665–671. [CrossRef] 44. González-Calabuig, A.; del Valle, M. Voltammetric electronic tongue to identify Brett character in wines: On-Site quantification of its ethylphenol metabolites. Talanta 2018, 179, 70–74. [CrossRef] 45. Faura, G.; González-Calabuig, A.; del Valle, M. Analysis of amino acid mixtures by voltammetric electronic tongues and artificial neural networks. Electroanalysis 2016, 28, 1894–1900. [CrossRef] 46. Cetó, X.; Apetrei, C.; del Valle, M.; Rodríguez-Méndez, M.L. Evaluation of red wines antioxidant capacity by means of a voltammetric e-tongue with an optimized sensor array. Electrochim. Acta 2014, 120, 180–186. [CrossRef] 47. Gutiérrez, J.M.; Haddi, Z.; Amari, A.; Bouchikhi, B.; Mimendia, A.; Cetó, X.; del Valle, M. Hybrid electronic tongue based on multisensor data fusion for discrimination of beers. Sens. Actuators B Chem. 2013, 177, 989–996. [CrossRef] 48. Cetó, X.; Gutiérrez, J.M.; Moreno-Barón, L.; Alegret, S.; del Valle, M. Voltammetric electronic tongue in the analysis of cava wines. Electroanalysis 2011, 23, 72–78. [CrossRef] 49. Pein, M.; Kirsanov, D.; Ciosek, P.; del Valle, M.; Yaroshenko, I.; Wesoły, M.; Zabadaj, M.; Gonzalez-Calabuig, A.; Wróblewski, W.; Legin, A. Independent comparison study of six different electronic tongues applied for pharmaceutical analysis. J. Pharm. Biomed. Anal. 2015, 114, 321–329. [CrossRef] 50. European Directorate for the Quality of Medicines & Healthcare. Ch. 8.7 Chemometric Methods Applied to Analytical Data. In European Pharmacopoeia, 9th ed.; Council of Europe: Strasbourg, France, 2016. 51. Calvo, D.; Durán, A.; del Valle, M. Use of sequential injection analysis to construct an electronic-tongue. Application to multidetermination employing the transient response of a potentiometric sensor array. Anal. Chim. Acta 2007, 600, 97–104. [CrossRef] 52. Cetó, X.; Céspedes, F.; Pividori, M.I.; Gutiérrez, J.M.; del Valle, M. Resolution of phenolic antioxidant mixtures employing a voltammetric bio-electronic tongue. Analyst 2012, 137, 349–356. [CrossRef] [PubMed] © 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).