Trends in Analytical Chemistry 157 (2022) 116765 Contents lists available at ScienceDirect Trends in Analytical Chemistry journal homepage: www.elsevier.com/locate/trac Fundamentals and application of voltammetric electronic tongues in quantitative analysis  a, *, Sandra Pe Xavier Ceto rez b, Beatriz Prieto-Simo  n c, d a Department of Chemistry, Faculty of Sciences, Universitat Autonoma de Barcelona, 08193, Bellaterra, Barcelona, Spain b Institute of Microelectronics of Barcelona (IMB-CNM), CSIC, Campus UAB, 08193, Bellaterra, Barcelona, Spain c Department of Electronic Engineering, Universitat Rovira i Virgili, Tarragona, 43007, Spain d ICREA, Pg. Lluís Companys 23, Barcelona, 08010, Spain a r t i c l e i n f o a b s t r a c t Article history: Electronic tongues (ETs) are bioinspired analytical tools based on the synergies between (bio)sensors and Received 1 May 2022 chemometrics. Through the application of chemometrics, it is possible to infer underlying relationships Accepted 28 August 2022 between the measured analytical signals and the chemical properties of the samples, both for descriptive Available online 5 September 2022 and predictive purposes, otherwise impossible to decipher. Research in voltammetric ETs during the last two decades has demonstrated the benefits derived from the use of sensor arrays with complementary Keywords: response, together with advanced data treatment methods to enhance their overall performance. In this Electronic tongue direction, the different approaches followed when developing voltammetric ETs and some relevant Sensor array Taste sensor applications in quantitative analysis are reviewed herein. Voltammetric sensors © 2022 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND Chemometrics license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Multivariate data analysis Pattern recognition Smart sensors Biosensors 1. Introduction comprehensive information about complex samples by simulta- neously detecting a large spectrum of compounds in a few seconds/ Research in electrochemical sensors during the last two decades minutes, without requiring any pre-treatment step as the data has demonstrated the advantages that can be obtained from the use processing stage may offset any matrix effect from the sample itself of arrays of sensors with complementary response, combined with [1e3,5], and even correct interferences, drifts or non-linearity ef- advanced data treatment methods to enhance their overall per- fects of the sensors [6]. formance [1e3]. Such an approach is known as electronic nose (EN) From the definition, it is clear that ETs have two key compo- when gas samples are analysed, or electronic tongue (ET) when it is nents: the sensor array and the chemometrics part. Those two are applied to the analysis of liquid samples. More specifically, ETs have the ones that define the ET and determine the nature of the infor- been defined by the IUPAC as “an analytical instrument comprising mation obtained. On the one side, the choice of the sensors impacts an array of nonspecific, low-selective, chemical sensors with high on which physicochemical parameters of the sample are measured. stability and cross-sensitivity to different species in solution, and an On the other side, the selection of the multivariate data analysis appropriate method of PARC (pattern recognition) and/or multi- method determines the task (application) carried out by the ET (e.g. variate calibration for data processing” [4]. categorical vs. regression based models). Obviously, the perfor- Thus, ETs are based on multi-sensor devices applied to the mance of the sensors has a direct impact on the performance of the analysis of liquid samples, that harness chemometric methods to system, as their responses have to be related to the parameters overcome the complex nature of some sensors responses. In this being analysed. Nonetheless, the chemometric methods have as regard, ETs main advantages lie in their ability to provide much impact as the sensors, as they are responsible to infer the underlying relationships between the measured analytical signals and the chemical properties of the samples. Among the different types of chemosensors that can constitute * Corresponding author.  ). E-mail address: xavier.ceto@uab.cat (X. Ceto the ET sensor array, electrochemical sensors (including https://doi.org/10.1016/j.trac.2022.116765 0165-9936/© 2022 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). , S. Perez and B. Prieto-Simo X. Ceto n Trends in Analytical Chemistry 157 (2022) 116765 potentiometric, voltammetric and impedimetric) are the most common ones [5]; although other types of sensors such as optical, gravimetric or piezoelectric might also be used. In this regard, it is possible to combine sensors from different types, an approach that is known as data fusion; howbeit this requires further consider- ations on how to integrate these data consistently [7]. More recently, the incorporation of biosensors has opened a new field of applications [3,8]. Such systems are referred to as bioelectronic tongues (BioETs), and do only differ in the incorporation of one or several biosensors into the array. The chemometric methods used can be mainly classified ac- cording to two parameters: (i) the machine learning task (unsu- pervised or supervised methods); and (ii) their application (categorical or regression based models). The latter is related to which is the output of the model, namely divided into models used for qualitative (classification) or quantitative purposes, while the former takes into account whether labelled or unlabelled data is provided to the learning system when building the model. That is, in unsupervised methods only the sensors’ responses (the pre- dictors) are provided to the system to build the model; whereas in supervised methods, both the predictors as well as the target(s) are provided (training set) and used to build the model, which is later applied to estimate those targets for new samples. However, other authors divide the different analysis methods in three different types: exploratory, classification and calibration. That is, dis- tinguishing visualization or clustering methods such as principal component analysis (PCA) or other cluster analysis techniques from the actual unsupervised categorical methods [9]. Up to date, a wide variety of applications involving the use of ETs has already been reported [1e3]. Among those, agri-food analysis (including beverages) and environmental monitoring are the two more prominent fields, although significant applications have also been developed in the security, pharmaceutical or chemical in- dustry fields, among others. Particularly, in the agri-food analysis, most of the developed applications deal with the classification of food and beverages according to perceptions of food quality (e.g. taste), varieties, geographical origin, etc. in an effort to provide an automated equivalent to sensory panels; whereas in the environ- mental monitoring side, most of the applications focus on the detection and determination of different contaminants such as heavy metals, pesticides or disinfection-by-products (DBPs), as well as other compounds of agri-environmental interest such as fertil- izers or alkaline ions. This review focuses on the application of ETs based on voltam- Fig. 1. Citation report created from Web of Science through the searches (“electronic tongue” or “taste sensor”) as topic, as well as refining the previous with the topic metric (bio)sensors devoted to quantitative analysis. Instead of “voltam*” (green bars). The percentage of the latter with respect to the former is also providing a comprehensive summary of all of the reported works in plotted (red ,). Search limited to the following document types: review, article and the past years, the aim of this review is to highlight the most book chapter. Accessed on 28/03/2022. relevant advances in the field over different scenarios, being an essential introduction to the key concepts for the non-specialist interested to learn on the topic, and at the same time serving as a both the number of publications, as well as the citations received by springboard for all the researchers working with electrochemical those, have steadily increased over the last 20 years after the term sensors. was coined. Besides the numbers shown, it is important to highlight that there is also a significant number of publications that do not 2. Voltammetric electronic tongues mention the terms ET or taste sensor, and simply describe the combination of a multi-sensor system and some data treatment Among the different types of sensors considered when devel- approaches. Similarly, there is also a significant number of papers in oping ET applications, voltammetric (bio)sensors are the most which ETs’ principles are applied, but making use of only one common ones. To illustrate this, a bibliometric study was per- sensor (e.g., Refs. [10e13]); an approach that is possible thanks to formed making use of the Web of Science database, which results the multi-dimensional response of the sensors. are summarized in Fig. 1. The results from the previous search were refined using the Although the term “electronic tongue” is the most popular one, term “voltam*” to consider only the applications involving vol- and actually the one proposed by the IUPAC, some authors also tammetric sensors, while ensuring to include all the word varia- refer to ETs as “taste sensor”, “multisensor system”, “sensor array” tions such as voltammetric, voltametric or voltammetry. or “multisensor array” [1], and consequently some of those terms Analogously, similar trend can be seen both for the number of were also used when doing the search. It can be clearly seen how manuscripts and citations. If considering the last 10 years, ca. 22% of 2 , S. Perez and B. Prieto-Simo X. Ceto n Trends in Analytical Chemistry 157 (2022) 116765 ETs publications report on voltammetric ETs, whereas in terms of comprehensive information of the sample, rather than using one citations, those receive ca. 26% of total citations; numbers that sensor with high selectivity and sensitivity towards a specific an- confirm the relevance of voltammetric (bio)sensors in the context alyte. Afterwards, through the combination of the sensors’ re- of ETs. sponses and with the aid of chemometric methods, the Over the next sections we will focus on the main parameters classification of samples or the actual quantification of specific that define ETs (Fig. 2): (i) the type of sensors, (ii) the measuring analytes/indexes is achieved. technique, and (iii) the data processing. Finally, an overview of This approach presents two main advantages when compared to different selected applications will be provided to better illustrate classical research in chemo- and bio-sensors. On the one side, the the capabilities of such an approach. difficulties found when developing highly specific sensors are overcome by shifting the complexity from the analytical to the data 2.1. Voltammetric (bio)sensors processing field. On the other side, we are also able to overcome the lack of knowledge of unique characteristic compounds responsible As already stated, the idea behind ETs is to have an array of for certain perceptions or that can be used to achieve samples sensors with cross-response features, capable to provide wide and discrimination. For example, it is even possible to estimate an un- measured variable from other measured variables, what is known as “soft sensor” [14]. This is highly relevant when evaluating complex integral features such as toxicity or perception attributes which depend on numerous chemical components simultaneously. A wide variety of sensors have been considered when devel- oping ET applications (Fig. 3). The first time a voltammetric ET was reported, it was based on the usage of two bare metallic wires as working electrodes, namely Au and Pt [15]. Next, the use of other noble metals was explored (Au, Pt, Ir, Pd, Rh, Ag, etc.) [16e22], while later on, also other metals such as Re, Cu, Ni, Co, W and Ti have been considered [23e25]. The advantages of those lie in their simplicity and durability in terms of material, whereas the use of different metals leads to different electrochemical behaviour and the low-selectivity required to develop ET applications. Despite the development of novel sensing platforms is not the ultimate goal of ETs research, it cannot be neglected that ETs nourish from advances in the sensors field. In this direction, the development of composites based on conductive phases dispersed in polymeric matrices offered attractive electrochemical, physical, mechanical and economical features compared to the classic metallic conductors [28]; especially considering that the incorpo- ration of modifiers can significantly improve the analytical char- acteristics of the resulting sensors. Among those, carbon paste electrodes (CPEs) and graphite-epoxy composites (GECs) arose as two of the most widely employed electrodes when developing voltammetric ETs [29e36], using as modifiers redox mediators/ electrocatalysts such as metals, conducting polymers, phthalocya- nines or nanoparticles, among others. Nowadays, screen printed electrodes (SPEs) have arisen as the most common sensing platform for the development of electro- chemical sensors due to their simple, rapid and low-cost produc- tion, plus the added advantages of their disposable character, small size suited to work with micro-volumes, and portability as all working, reference and counter electrodes are printed on the same strip. For the fabrication of SPE devices, many configurations have been reported, both in terms of geometry as well as materials [37]. In regards to the former, a three electrode configuration (reference, counter and working electrodes) is the more common one, but several devices including many working electrodes (2,4,8, …,20) in the same miniaturized device have been reported. As for the sensing material, a wide variety of inks are commercially available nowadays; some of the more interesting ones include graphite and different carbon-based materials such as carbon nanotubes (CNTs), graphene or nanofibers, metals such as Au, Pt, Cu, Bi, etc, mediators such as phthalocyanines or Prussian blue, etc. Lastly, additive manufacturing (most commonly known as 3D printing) is emerging as an alternative approach for the fabrication of customized electrochemical sensors, owing to their many unique advantages such as its low-cost (both of the material and equip- Fig. 2. Schematic illustrating (A) the main components that define an ET, and (B) block ment), tuneability and easy prototyping [38,39]. One of the reasons diagram of the ETs approach. that makes 3D printing so appealing is that it combines the ease of 3 , S. Perez and B. Prieto-Simo X. Ceto n Trends in Analytical Chemistry 157 (2022) 116765 fabrication of other composite materials such as CPEs and GECs with the automatized fabrication and miniaturization of SPEs. Concretely, electrodes are fabricated by fused deposition modelling (FDM) from thermoplastics such as poly-lactic acid (PLA) or acrylonitrile-butadiene-styrene (ABS), commonly doped with different carbon-based materials such as carbon black, CNTs or graphene to overcome the insulating nature of PLA and ABS. Additionally, other redox mediators/electrocatalysts, as those above mentioned, can also be incorporated by extrusion of customized filaments. Hitherto the usage of bare materials or their bulk modifications have been discussed, but chemically modified electrodes (CMEs) have also played a key role in the development of voltammetric ETs. Modification of the electrodes is usually performed through the irreversible adsorption of substances with desired functionalities [40,41], covalent binding of components [26], or coating with polymeric or non-polymeric films [13,17,42,43]. The aim of such modifications is to improve the selectivity and the sensitivity of the sensors towards specific analytes. That is, despite is true that ETs are based on the usage of low-selective sensors showing cross- response features, certain degree of selectivity is still required. Lastly, special mention has to be devoted to the incorporation of biosensors into ETs sensor arrays, an approach that has been referred as bioelectronic tongues (BioETs) [3]. In the same line as CMEs, and despite the higher selectivity and specificity associated with the usage of bioreceptors, it has to be taken into account that still those also show some cross-reactivity, although towards a narrower group of compounds. In this line, the coupling of bio- sensors and ETs presents huge potential as it exploits the combi- nation of chemometrics to solve interferences issues from biosensors, and biosensors to improve the lack of selectivity sometimes attributed to ETs when applied to complex samples. Thus, the development of BioETs allows improving ETs perfor- mance and tackling new application fields. Among the different examples that can be found in the literature, two scenarios have been extensively studied: (i) the detection of pesticides through the measurement of enzymatic inhibition, mainly of cholinesterase enzymes (ChE) [44], and (ii) the detection of phenolic compounds using tyrosinase (Tyr) and laccase (Lac) enzymes as bioreceptors [10,31,45]. 2.2. Voltammetric methods In voltammetric methods, both qualitative and quantitative in- formation of the electroactive species present in the sample is obtained from the application of a time-dependent potential (excitation signal), and the measurement of the resulting current as a function of the former [46,47]. Precisely, how the applied po- tential is changed and the current is measured, is what distin- guishes the different voltammetric methods (Fig. 3). In this regard, linear sweep voltammetry (LSV) or cyclic vol- tammetry (CV) are the most common voltammetric techniques (Table 1). This is because they are considered the most classical and simplest voltammetric methods, plus the fact that CV provides, in one step, information of the compounds present in the sample that can be either oxidized or reduced. Fig. 3. Illustration of the different type of sensors used when developing voltammetric With the advance of electrochemistry, new methods appeared ETs, as well as the voltammetric responses obtained with those when using different aiming to overcome some of the limitations found with LSV/CV, voltammetric techniques. Measurement of (A) a wine sample with different composite especially in terms of detection limits. Among them, pulse vol- sensors employing CV, (B) a water sample containing different metal ions with screen- tammetric methods, such as differential pulse voltammetry (DPV) printed electrodes employing DPV, and (C) a vegetable milk with metallic electrodes or square wave voltammetry (SWV), have also been widely used in employing LAPV. Images reproduced with permission from Refs. [21,26,27]. ETs (Table 1). The main advantage of those is their high sensitivity and the fact that non-faradaic current contributions are minimized. To achieve the detection of minute concentrations, stripping methods such as anodic (ASV) and cathodic stripping 4 , S. Perez and B. Prieto-Simo X. Ceto n Trends in Analytical Chemistry 157 (2022) 116765 Table 1 Summary of some of the most relevant quantitative applications involving the use of voltammetric ETs. Sample Analyte/Attribute Sensors Technique Data processing Ref. Physicochemical parameters of natural food products Animal feed Amino acids: tryptophan, cysteine and tyrosine Pt sensor DPV DWT þ ANNs [48] Honey Protein, colour, phenolic compounds, acidity, sugars 4 metallic electrodes: Pt, Au, GCE, CV Feature [22] Ag extraction þ PLS Juices and water Nitrate GCE modified biosensor CV SVMs [12] Milk Antibiotics: Chloramphenicol, Erythromycin, Kanamycin sulfate, Neomycin 5 metallic electrodes: Au, Ag, Pt, MLPAV PCR, PLS, LS- [16] sulfate, Streptomycin sulfate and Tetracycline HCl Pd, Ti SVM Oil Carotenoids 7 metallic electrodes: Cu, GCE, CV Feature [17] Au, Ni, Pd, Pt, Ag extraction þ PLS 3 interdigitated electrodes modified with polyelectrolytes Saffron Safranal, Picrocrocin and many esters SPEs of C, Pt, Au and PB CV kernels þ PLS [40] SPEs chemically modified with phtalocyanines Tea Theaflavin and thearubigin 7 metallic electrodes: Au, Ir, Pd, LAPV DWT þ PLS/ [18] Pt, Rh SVMs Wine Total polyphenols in wine 4 graphite-epoxy modified CV FFT þ ANNs [45] enzymatic (bio)sensors Wine Anthocyanins, sugar content, titratable acidity, pH, optical density, colour, etc. PEDOT modified Pt and sonogel DPV PLS [13] carbon Wine (cava) Sugar 6 graphite-epoxy modified CV FFT þ ANNs [29] sensors Contaminants Food and vegetable Organophosphate pesticides: dichlorvos, malaoxon, chlorpyrifos-oxon, 6 AChE-based enzymatic CAmp ANNs [44] extracts, water chlorpyrifosmethyl-oxon,chlorfenvinphos and pir-imiphos-methyl-oxon biosensors Standards Explosive compounds: RDX, TNT and PETN SPE CV DWT þ ANNs [11] Standards Nerve Agent Mimics: diisopropyl fluorophosphates (DFP), diethyl 8 metallic working electrodes: LAPV PLS [19] chlorophosphate (DCP) and diethyl cyanophosphate (DCNP) Au, Pt, Ir, Rh, Cu, Co, Ni and Ag Standards Phenolic compounds Tyrosinase modified graphite- LSV ANNs [10] epoxy composite 2þ 2þ 2þ Water Heavy metals: Cu , Cd , Pb 4 peptide-modified gold SWV PLS, nPLS [41] electrodes 2þ 2þ 2þ Water Heavy metals: Zn , Cd , Pb 3 peptide-modified graphite- AdSV FFT þ ANNs [30] epoxy composite Water, tap Heavy metals: Cd2þ, Pb2þ, Tlþ and Bi3þ in the presence of Zn2þ and In3þ 4 modified SPEs with peptides DPV PLS [26] and Sb Water, tap Haloacetic acids: HAA3 Au sensor CV/SWV FFT þ ANNs [49] Water, wastewater Phenolic compounds 4 graphite-epoxy modified (bio) CV FFT þ ANNs [31] sensors Medical diagnostics Pharmaceutical Diclofenac, paracetamol and naproxen Multiwall carbon nanotubes CPE DPV DWT þ ANNs [32] drugs Urine, serum and Antibiotics: tetracycline and cefixime SPE modified with gold SWV PCA þ ANNs [43] milk nanoparticles and cysteine Urine, human Creatinine 7 metallic electrodes: GCE, Au, Pt, CV PLS [20] Ag, Ni, Pd, Cu Perception attributes Coffee Adulteration CPE DPV MLR, PLS [33] Milk Sensorial attributes 8 metallic electrodes: Au, Pt, Rh, LAPV PLS [21] Ir, Cu, Co, Ag, Ni Oil Bitterness CPE modified with olive oil CV kernels þ PLS [34] Oil Adulteration CPE modified with olive oil SWV kernels þ PLS [35] Pear Firmness and sugar Au, Ag, Pt, Pd, W, Ti MLAPV PCR, PLS, SVMs [24] Rice Flavour attributes: softness, stickiness, sweetness and aroma 6 metallic electrodes: Pt, Au, Pd, MLPAV FFT þ ANNs [25] W, Ti, Ag Wine Scores of a sensory panel 6 graphite-epoxy modified CV FFT þ PLS [36] sensors CPE: carbon paste electrode, GCE: glassy carbon electrode, SPE: screen-printed electrode. AdSV: differential pulse adsorptive stripping voltammetry, CAmp: chronoamperometry; CV: cyclic voltammetry, DPV: differential pulse voltammetry, LAPV: large amplitude pulsed voltammetry, LSV: linear sweep voltammetry, MLPAV: multifrequency LAPV, SWV: square wave voltammetry. ANNs: artificial neural networks, DWT: discrete wavelet transform, FFT: fast Fourier transform, MLR: multiple linear regression, PCA: principal component analysis, PCR: principal component regression, PLS: partial least-squares regression, SVMs: support vector machines. voltammetries (CSV), as well as adsorptive stripping voltammetry surface or through incorporating chemical modifiers. Those (AdSV) have been considered [47]. Stripping methods comprise methods are highly suitable for the detection of metal ions as well three steps: accumulation, rest and measurement. The main dif- as organic molecules that, upon application of a potential, generate ference between ASV/CSV and AdSV lies in the accumulation step: insoluble species adsorbed onto the electrode surface, allowing a in the former the analyte is electroplated through the application of decrease of the detection limits of ca. 2 to 3 orders of magnitude. a negative or positive potential, whereas in the latter, the pre- Finally, not only the use of small amplitude pulsed voltammetric concentration is achieved by direct adsorption onto the electrode (SAPV) methods such as DPV has been considered, but also the use 5 , S. Perez and B. Prieto-Simo X. Ceto n Trends in Analytical Chemistry 157 (2022) 116765 of large amplitude pulsed voltammetry (LAPV), either at constant the focus now is on the reported applications. The potential of ETs is or at different sweep frequencies (multi-frequency LAPV, MLAPV) based on the synergies between electrochemical (bio)sensors and [50]. The application of those has been mainly related to the usage chemometrics, taking advantage of the inherent benefits of both of of different bare metallic electrodes. them to obtain an analytical tool with huge versatility and wide applicability. Precisely, the advances in those two research fields is 2.3. Data analysis what allows to spawn new application areas, offering a wide spectrum of possibilities. As already introduced, data analysis is the other key element of Although traditionally ETs have been mainly related to qualita- ETs in conjunction with the sensor array. Chemometrics is the tive analysis, they have also been used for quantitative purposes, in discipline that seeks to extract information from chemical systems which herein we will focus. Moreover, despite the ET definition through the application of statistical and mathematical methods. states that a sensor array has to be used, the nature of voltammetric Despite voltammetric signals may provide richer information in responses allows that the same principles can be applied when comparison to other electrochemical techniques given the higher using a single sensor. Such is the case of voltammograms being of order of the obtained data (polarization potentials x sensors x first-order and showing cross-response features. Consequently, sample), such dimensionality can also hinder its analysis, requiring some of those will be also reviewed here. To better illustrate the the use of multi-way processing methods or a pre-processing step potential of voltammetric ETs in quantitative analysis, relevant in order to reduce the dimensionality of the input signal [3,51]. applications in different fields are summarized in Table 1. This pre-processing step usually starts with the unfolding of the Firstly, it is important to note the variety of scenarios studied, not data into a 2D matrix, and might also be followed by a compression only in terms of samples (e.g. honey, juice, oil, milk, water, wine, step devoted to reduce data complexity and redundancies, while drugs, urine, serum, etc.), but also of analytes (e.g. amino acids, preserving the relevant information [51]. The latter, although not phenolic compounds, antibiotics, pesticides, explosives, heavy being always mandatory, it has been demonstrated that it can metals, etc.). In this direction, it is observed that most of the appli- improve the performance of the model and its generalization cations are related to the analysis of physicochemical parameters of ability, as well as reduce the computational requirements. In this food products, in line with initial studies with ETs (i.e. qualitative regard, more common approaches reported in the literature are: analysis of food products), although their usage for the detection of feature selection algorithms (including genetic algorithms) contaminants such as heavy metals is also spreading [41,26]. Sec- [17,22,36], “kernel” functions [34,35,40], fast Fourier transform ondly, ETs success has not only been demonstrated through the (FFT) [25,29e31,36,45,49] or discrete wavelet transform (DWT) determination of specific compounds, but also through the quanti- [11,18,32,48]. fication of subjective indexes such as flavour attributes [25] or the The next step in data analysis is the modelling itself, that is to scores given by a sensory panel [21,36]. Actually, the latter represent infer the underlying relationships between the measured analytical a straightforward solution to the lack of knowledge of the specific signals and the chemical properties of the samples, both for compounds (or concentration levels) responsible for certain sensa- descriptive and predictive purposes. For this purpose, principal tions that do not allow to achieve such tasks with classical methods. components analysis (PCA), partial least squares regression (PLS) Other aspects to be taken into account are related to the oper- and artificial neural networks (ANNs) are the most commonly used ation mode of ETs. In this regard, there are three key factors that methods. In quantitative analysis which is the topic covered in this would help in making ETs more attractive: i) miniaturization, ii) review, methods can be further categorized based on the regression automation and iii) portability. In this direction, electrochemical analysis form, distinguishing between linear and nonlinear ones. In systems present many advantages by themselves as they offer fast the former, linear predictor functions are used to fit the data, and accurate information in a cost-effective manner, while their usually through least-square methods (e.g. PLS), whereas in the instrumentation has the potential for miniaturization and porta- latter, both linear and non-linear functions are combined and bility, minimal power requirement and cost-effectiveness. adjusted through an iterative process (e.g. ANNs or support vector Back on the 2010, Crew et al. actually demonstrated the porta- machines (SVMs)). bility of ETs with a portable BioET prototype for the in situ deter- Despite the unfolding of the data is the most common approach mination of organophosphate pesticides [44]. The automated in ETs, it fails harnessing the so-called “second-order advantage” system consisted of an array of six AChE-based biosensors incor- [6], which allows the prediction of the analytes/indexes of interest porating an ANN program, and powered via the lighter socket from even in the presence of unknown interferents. In this direction, the a car battery. Nowadays, with the significant advances that have alternative is to make use of trilinear methods such as parallel been made since then in the development of portable compact factor analysis (PARAFAC) or Tucker decomposition for qualitative potentiostats, in conjunction with the advance in screen printing approaches, or multiway PLS (nPLS) for quantitative analysis. technology, it is even possible to integrate such a system into a Furthermore, new trends in chemometric analysis focus on the small hand-held device controlled through a smartphone app [12]. development of even more powerful analysis tools to better exploit Automation is key for ETs, because unlike “conventional” the information contained in such complex data sets, and be able to methods (based on univariate regression), multivariate methods cope with the advances in data generation [52]; e.g. when the re- require the preparation and measurement of a high number of sponses of an array are also measured at different pH values [30]. standards and/or samples to build the model. Moreover, the However, despite their advantages, their usage is still very limited preparation of the stocks may become more tedious when tackling when developing ET applications, being rarely used. the multi-determination of several analytes. In this regard, the use For a more in-depth revision of the different chemometric of flow systems such as flow injection analysis (FIA) or sequential methods and the main differences between them, readers may injection analysis (SIA) can provide many benefits as those can consult [51,9,53]. comprise the preparation of the stocks, their measurement and the data collection steps [10,54]. 2.4. Applications Even more interesting is the use of microfluidic devices to develop ET applications; an approach that has been already re- After discussing the sensors, the measuring techniques and the ported carrying out capacitive measurements [55], but not yet with data processing approaches used when developing ET applications, voltammetric sensors to the best of authors knowledge. 6 , S. Perez and B. Prieto-Simo X. Ceto n Trends in Analytical Chemistry 157 (2022) 116765 3. Conclusions and future trends the arrays and the effect that such selection has in the overall performance have been rarely reported. Lastly, as already pointed This review provides an overview of the approaches followed out, the combination of different types of sensors can actually boost when developing voltammetric ETs as well as their capabilities for the performance of such biomimetic systems as it clearly maxi- quantitative analysis. The publications reviewed herein were mizes the information extracted from the samples as some types of selected to illustrate the huge versatility and wide applicability of sensors might be more suitable than others depending on the ETs. analytes considered. ETs take advantages of the synergies between (bio)sensors and chemometrics to provide comprehensive information of complex Declaration of competing interest samples by simultaneously detecting a large spectrum of com- pounds in a rapid manner, without requiring any pre-treatment The authors declare that they have no known competing step as the data processing stage may offset potential matrix ef- financial interests or personal relationships that could have fects from the sample. This is especially significant when tackling appeared to influence the work reported in this paper. the classification of samples or the quantification of subjective perceptions where no specific compounds (or concentration levels) Acknowledgments can be associated with such characteristics, focusing in this way on the overall profile rather than on specific compounds. Conse- X. Ceto thanks the support of the Beatriu de Pino s postdoctoral quently, ETs represent an attractive alternative to conventional or programme of the Government of Catalonia's Secretariat for Uni- sophisticated methods, with many advantages over those such as versities and Research of the Ministry of Economy and Knowledge their low-cost, portability, or fast-responses; or even allowing to (Catalonia, Spain). Spanish Ministry of Science and Innovation meet new challenges in environmental monitoring, food safety and  n acknowledges financial support from grant (Spain) B. Prieto-Simo public health. PID2021-124867OB-I00 funded by MCIN/AEI/ 10.13039/ However, despite the huge capabilities of ETs, their imple- 501100011033 and by “ERDF A way of making Europe”. mentation is still quite meagre and mainly limited to research purposes. One of the reasons is that despite ETs are sometimes References defined as an instrument, it has to be reckoned that the number of commercial devices is still rather low, being mainly used in [1] P. Ciosek, W. Wroblewski, Sensor arrays for liquid sensing - electronic tongue systems, Analyst 132 (2007) 963. https://doi.org/10.1039/B705107G. research laboratories working on their development. Another [2] A. Riul Jr., C.A.R. Dantas, C.M. Miyazaki, O.N. Oliveira Jr., Recent advances in reason might be the mathematical background behind these electronic tongues, Analyst 135 (2010) 2481. https://doi.org/10.1039/ methods, which were usually coded in e.g. Matlab or R, and that can c0an00292e. [3] X. Ceto, N.H. Voelcker, B. Prieto-Simo n, Bioelectronic tongues: new trends and be quite complex and bewildering for the non-expert. However, applications in water and food analysis, Biosens. Bioelectron. 79 (2016) 608. this could change as nowadays many software packages with user- https://doi.org/10.1016/j.bios.2015.12.075. friendly interfaces are available. Besides, the large number of [4] Y. Vlasov, A. Legin, A. Rudnitskaya, C. Di Natale, A. D'Amico, Nonspecific sensor samples required to build the model and the recalibration of the arrays ("electronic tongue") for chemical analysis of liquids (IUPAC Technical Report), Pure Appl. Chem. 77 (2005) 1965. https://doi.org/10.1351/ system might also be a disincentive. To this aim, usage of model pac200577111965. transfer methods represents a very promising approach to tackle [5] M. del Valle, Electronic tongues employing electrochemical sensors, Electro- this problem [56]. analysis 22 (2010) 1539. https://doi.org/10.1002/elan.201000013. [6] G.M. Escandar, A.C. Olivieri, N.M. Faber, H.C. Goicoechea, A. Mun ~ oz de la Pen ~ a, Finally, future trends in ETs will be obviously based on the R.J. Poppi, Second- and third-order multivariate calibration: data, algorithms separate advances achieved both in the (bio)sensors and the che- and applications, TrAC, Trends Anal. Chem. 26 (2007) 752. https://doi.org/ mometrics fields, and more interestingly in the new concepts that 10.1016/j.trac.2007.04.006. [7] F. Castanedo, A review of data fusion techniques, Sci. World J. (2013) 19. arise from their combination such as data fusion approaches, or the https://doi.org/10.1155/2013/704504, 2013. development of systems that lead to and actually exploit high [8] J. Zeravik, A. Hlavacek, K. Lacina, P. Skladal, State of the art in the field of multi-dimensional data. On the (bio)sensors side, advances in electronic and bioelectronic tongues - towards the analysis of wines, Elec- troanalysis 21 (2009) 2509. https://doi.org/10.1002/elan.200900285. materials science and nanotechnology have underpinned the evo- [9] E. Richards, C. Bessant, S. Saini, Multivariate data analysis in electroanalytical lution of voltammetric sensors and spurred their potential. How- chemistry, Electroanalysis 14 (2002) 1533. https://doi.org/10.1002/1521- ever, ETs are one step behind “conventional” research in (bio) 4109(200211)14:22<1533::aid-elan1533>3.0.co;2-t. [10] A. Gute s, F. Cespedes, S. Alegret, M. del Valle, Determination of phenolic sensors, thus the incorporation of more recently developed sensing compounds by a polyphenol oxidase amperometric biosensor and artificial devices will foster ETs performance. For example, sensors based on neural network analysis, Biosens. Bioelectron. 20 (2005) 1668. https://doi.org/ novel nanomaterials such as porous silicon (pSi) or mesoporous 10.1016/j.bios.2004.07.026. carbon surpass the performance of the above-mentioned elec- [11] X. Ceto, A.M. O'Mahony, J. Wang, M. del Valle, Simultaneous identification and quantification of nitro-containing explosives by advanced chemometric data trodes. Similarly, 3D printing offers enormous potential not only for treatment of cyclic voltammetry at screen-printed electrodes, Talanta 107 the development of novel sensing devices, but also due to its pro- (2013) 270. https://doi.org/10.1016/j.talanta.2012.12.042. totyping capabilities. Lastly, the incorporation of different bio- [12] J. Massah, K. Asefpour Vakilian, An intelligent portable biosensor for fast and accurate nitrate determination using cyclic voltammetry, Biosyst. Eng. (2018). recognition elements such as antibodies or aptamers is a topic https://doi.org/10.1016/j.biosystemseng.2018.09.007. that has still not been exploited, but also the usage of artificial re- [13] L. Pigani, G. Vasile Simone, G. Foca, A. Ulrici, F. Masino, L. Cubillana-Aguilera, ceptors such as molecularly imprinted polymers (MIPs), which R. Calvini, R. Seeber, Prediction of parameters related to grape ripening by multivariate calibration of voltammetric signals acquired by an electronic opens the door for the development of many compound-tailored tongue, Talanta 178 (2018) 178. https://doi.org/10.1016/j.talanta.2017.09.027. quantitative applications. On the chemometrics part, one of the [14] S. James, R. Legge, H. Budman, Comparative study of black-box and hybrid biggest challenges is tackling the absence of an established meth- estimation methods in fed-batch fermentation, J. Process Control 12 (2002) 113. https://doi.org/10.1016/S0959-1524(00)00065-2. odology in the field of ETs. For example, the lack of classical €m, An electronic tongue based on voltam- [15] F. Winquist, P. Wide, I. Lundstro analytical figures of merit such as sensitivity, selectivity or limit of metry, Anal. Chim. Acta 357 (1997) 21. https://doi.org/10.1016/S0003- detection (LOD). Another aspect that has to be further investigated 2670(97)00498-4. is the selection of the sensors that constitute the ET array. That is, [16] Z. Wei, J. Wang, Detection of antibiotic residues in bovine milk by a voltam- metric electronic tongue system, Anal. Chim. Acta 694 (2011) 46. https:// while the benefits of sensor arrays have been widely demonstrated, doi.org/10.1016/j.aca.2011.02.053. the criteria employed for the selection of the sensors that constitute [17] K. Tahri, A.A. Duarte, G. Carvalho, P.A. Ribeiro, M.G. da Silva, D. Mendes, N. El 7 , S. Perez and B. Prieto-Simo X. Ceto n Trends in Analytical Chemistry 157 (2022) 116765 Bari, M. Raposo, B. Bouchikhi, Distinguishment, Identification and aroma 162 (2017) 218. https://doi.org/10.1016/j.talanta.2016.09.055. compound quantification of Portuguese olive oils based on physicochemical [37] M. Trojanowicz, Impact of nanotechnology on design of advanced screen- attributes, HS-GC/MS analysis and voltammetric electronic tongue, J. Sci. Food printed electrodes for different analytical applications, TrAC, Trends Anal. Agric. 98 (2018) 681. https://doi.org/10.1002/jsfa.8515. Chem. 84 (2016) 22. https://doi.org/10.1016/j.trac.2016.03.027. [18] A. Ghosh, B. Tudu, P. Tamuly, N. Bhattacharyya, R. Bandyopadhyay, Prediction [38] G. Gaal, T.A. da Silva, V. Gaal, R.C. Hensel, L.R. Amaral, V. Rodrigues, A. Riul, 3D of theaflavin and thearubigin content in black tea using a voltammetric Printed e-Tongue, Front. Chem. 6 (2018). https://doi.org/10.3389/ electronic tongue, Chemometr. Intell. Lab. Syst. 116 (2012) 57. https://doi.org/ fchem.2018.00151. 10.1016/j.chemolab.2012.04.010. [39] R.M. Cardoso, C. Kalinke, R.G. Rocha, P.L. dos Santos, D.P. Rocha, P.R. Oliveira, [19] I. Campos, L. Gil, R. Martínez-Man ~ ez, J. Soto, J.-L. Vivancos, Use of a voltam- B.C. Janegitz, J.A. Bonacin, E.M. Richter, R.A.A. Munoz, Additive-manufactured metric electronic tongue for detection and classification of nerve agent (3D-printed) electrochemical sensors: a critical review, Anal. Chim. Acta 1118 mimics, Electroanalysis 22 (2010) 1643. https://doi.org/10.1002/ (2020) 73. https://doi.org/10.1016/j.aca.2020.03.028. elan.200900625. [40] K. Heidarbeigi, S.S. Mohtasebi, J. Serrano-Diaz, C. Medina-Plaza, M. Ghasemi- [20] T. Saidi, M. Moufid, O. Zaim, N. El Bari, B. Bouchikhi, Voltammetric electronic Varnamkhasti, G.L. Alonso, M.V. Garcia-Rodriguez, S. Rafiee, K. Rezaei, tongue combined with chemometric techniques for direct identification of C. Garcia-Hernandez, J.A.d. Saja, M.L. Rodriguez-Mendez, Flavour character- creatinine level in human urine, Measurement 115 (2018) 178. https:// istics of Spanish and Iranian saffron analysed by electronic tongue, Qual. doi.org/10.1016/j.measurement.2017.10.044. Assur. Saf. Crop Foods 8 (2016) 359. https://doi.org/10.3920/qas2015.0591. [21] L. Pascual, M. Gras, D. Vidal-Broto  ns, M. Alcan ~ iz, R. Martínez-M an~ ez, J.V. Ros- [41] D. Ebrahimi, E. Chow, J.J. Gooding, D.B. Hibbert, Multi-analyte sensing: a Lis, A voltammetric e-tongue tool for the emulation of the sensorial analysis chemometrics approach to understanding the merits of electrode arrays and the discrimination of vegetal milks, Sensor. Actuator. B Chem. 270 (2018) versus single electrodes, Analyst 133 (2008) 1090. https://doi.org/10.1039/ 231. https://doi.org/10.1016/j.snb.2018.04.151. B804811H. [22] N. El Alami El Hassani, K. Tahri, E. Llobet, B. Bouchikhi, A. Errachid, N. Zine, [42] C. Apetrei, M.L. Rodriguez-Mendez, V. Parra, F. Gutierrez, J.A. de Saja, Array of N. El Bari, Emerging approach for analytical characterization and geographical voltammetric sensors for the discrimination of bitter solutions, Sens. Actua- classification of Moroccan and French honeys by means of a voltammetric tors B-Chem. 103 (2004) 145. https://doi.org/10.1016/j.snb.2004.04.047. electronic tongue, Food Chem. 243 (2018) 36. https://doi.org/10.1016/ [43] M. Asadollahi-Baboli, A. Mani-Varnosfaderani, Rapid and simultaneous j.foodchem.2017.09.067. determination of tetracycline and cefixime antibiotics by mean of gold [23] F. Winquist, S. Holmin, C. Krantz-Rülcker, P. Wide, I. Lundstro € m, A hybrid nanoparticles-screen printed gold electrode and chemometrics tools, Mea- electronic tongue, Anal. Chim. Acta 406 (2000) 147. https://doi.org/10.1016/ surement 47 (2014) 145. https://doi.org/10.1016/j.measurement.2013.08.029. S0003-2670(99)00767-9. [44] A. Crew, D. Lonsdale, N. Byrd, R. Pittson, J.P. Hart, A screen-printed, ampero- [24] Z. Wei, J. Wang, The evaluation of sugar content and firmness of non- metric biosensor array incorporated into a novel automated system for the climacteric pears based on voltammetric electronic tongue, J. Food Eng. 117 simultaneous determination of organophosphate pesticides, Biosens. Bio- (2013) 158. https://doi.org/10.1016/j.jfoodeng.2013.02.007. electron. 26 (2011) 2847. https://doi.org/10.1016/j.bios.2010.11.018. [25] L. Lu, X. Hu, S. Tian, S. Deng, Z. Zhu, Visualized attribute analysis approach for [45] X. Ceto, F. Ce spedes, M. del Valle, BioElectronic Tongue for the quantification characterization and quantification of rice taste flavor using electronic tongue, of total polyphenol content in wine, Talanta 99 (2012) 544. https://doi.org/ Anal. Chim. Acta 919 (2016) 11. https://doi.org/10.1016/j.aca.2016.03.019. 10.1016/j.talanta.2012.06.031. [26] C. Perez-Ra fols, N. Serrano, J.M. Díaz-Cruz, C. Arin ~ o, M. Esteban, A screen- [46] D.A. Skoog, F.J. Holler, T. Nieman, Principles of Instrumental Analysis, 5a ed., printed voltammetric electronic tongue for the analysis of complex mixtures McGraw-Hill, Madrid, Spain, 2001. of metal ions, Sensor. Actuator. B Chem. 250 (2017) 393. https://doi.org/ [47] F. Scholz, Voltammetric techniques of analysis: the essentials, ChemTexts 1 10.1016/j.snb.2017.04.165. (2015) 17. https://doi.org/10.1007/s40828-015-0016-y. [27] X. Ceto , A. Gonz alez-Calabuig, J. Capdevila, A. Puig-Pujol, M. del Valle, [48] L. Moreno-Baro n, R. Cartas, A. Merkoçi, S. Alegret, M. del Valle, L. Leija, Instrumental measurement of wine sensory descriptors using a voltammetric P.R. Hernandez, R. Mun ~ oz, Application of the wavelet transform coupled with electronic tongue, Sensor. Actuator. B Chem. 207 (2015) 1053. Part B, https:// artificial neural networks for quantification purposes in a voltammetric doi.org/10.1016/j.snb.2014.09.081. electronic tongue, Sens. Actuators B-Chem. 113 (2006) 487. https://doi.org/ [28] F. Ce spedes, E. Martínez-Fa bregas, S. Alegret, New materials for electro- 10.1016/j.snb.2005.03.063. chemical sensing I. Rigid conducting composites, TrAC, Trends Anal. Chem. 15 [49] X. Ceto, C. Saint, C.W.K. Chow, N.H. Voelcker, B. Prieto-Simo  n, Electrochemical (1996) 296. https://doi.org/10.1016/0165-9936(96)00042-8. fingerprints of brominated trihaloacetic acids (HAA3) mixtures in water, [29] X. Ceto , J. Capdevila, A. Puig-Pujol, M. del Valle, Cava wine authentication Sensor. Actuator. B Chem. 247 (2017) 70. https://doi.org/10.1016/ employing a voltammetric electronic tongue, Electroanalysis 26 (2014) 1504. j.snb.2017.02.179. https://doi.org/10.1002/elan.201400057. [50] F. Winquist, Voltammetric electronic tongues e basic principles and appli- [30] N. Serrano, B. Prieto-Simo n, X. Ceto, M. del Valle, Array of peptide-modified cations, Microchim. Acta 163 (2008) 3. https://doi.org/10.1007/s00604-007- electrodes for the simultaneous determination of Pb(II), Cd(II) and Zn(II), 0929-2. Talanta 125 (2014) 159. https://doi.org/10.1016/j.talanta.2014.02.052. [51] X. Ceto, F. Ce spedes, M. del Valle, Comparison of methods for the processing of [31] X. Ceto , A. Gonza lez-Calabuig, M. del Valle, Use of a bioelectronic tongue for voltammetric electronic tongues data, Microchim. Acta 180 (2013) 319. the monitoring of the photodegradation of phenolic compounds, Electro- https://doi.org/10.1007/s00604-012-0938-7. analysis 27 (2015) 225. https://doi.org/10.1002/elan.201400394. [52] R.M. Maggio, A. Mun ~ oz de la Pen ~ a, A.C. Olivieri, Unfolded partial least-squares [32] G.Y. Aguilar-Lira, J.M. Gutie rrez-Salgado, A. Rojas-Hern andez, J.A. Rodríguez- with residual quadrilinearization: a new multivariate algorithm for process-  Avila, M.E. P aez-Herna  ndez, G.A. Alvarez-Romero, Artificial neural network ing five-way data achieving the second-order advantage. Application to for the voltamperometric quantification of diclofenac in presence of other fourth-order excitation-emission-kinetic-pH fluorescence analytical data, nonsteroidal anti-inflammatory drugs and some commercial excipients, Chemometr. Intell. Lab. Syst. 109 (2011) 178. https://doi.org/10.1016/ J. Electroanal. Chem. 801 (2017) 527. https://doi.org/10.1016/ j.chemolab.2011.09.002. j.jelechem.2017.08.029. [53] R. Bro, Multi-way Analysis in the Food Industry. Models, Algorithms and [33] T.C.B. de Morais, D.R. Rodrigues, U.T. de Carvalho Polari Souto, S.G. Lemos, Applications, University of Amsterdam, 1998. A simple voltammetric electronic tongue for the analysis of coffee adultera- [54] F. Winquist, E. Rydberg, S. Holmin, C. Krantz-Rülcker, I. Lundstro € m, Flow in- tions, Food Chem. (2018). https://doi.org/10.1016/j.foodchem.2018.04.136. jection analysis applied to a voltammetric electronic tongue, Anal. Chim. Acta [34] C. Apetrei, F. Gutierez, M.L. Rodriguez-Mendez, J.A. de Saja, Novel method 471 (2002) 159. https://doi.org/10.1016/S0003-2670(02)00923-6. based on carbon paste electrodes for the evaluation of bitterness in extra [55] C.M. Daikuzono, C.A.R. Dantas, D. Volpati, C.J.L. Constantino, M.H.O. Piazzetta, virgin olive oils, Sens. Actuators B-Chem. 121 (2007) 567. https://doi.org/ A.L. Gobbi, D.M. Taylor, O.N. Oliveira, A. Riul, Microfluidic electronic tongue, 10.1016/j.snb.2006.04.091. Sensor. Actuator. B Chem. 207 (2015) 1129. https://doi.org/10.1016/ [35] I.M. Apetrei, C. Apetrei, Detection of virgin olive oil adulteration using a vol- j.snb.2014.09.112. tammetric e-tongue, Comput. Electron. Agric. 108 (2014) 148. https://doi.org/ [56] M. Khaydukova, V. Panchuk, D. Kirsanov, A. Legin, Multivariate calibration 10.1016/j.compag.2014.08.002. transfer between two potentiometric multisensor systems, Electroanalysis 29 [36] X. Ceto , A. Gonza lez-Calabuig, N. Crespo, S. Pe rez, J. Capdevila, A. Puig-Pujol, (2017) 2161. https://doi.org/10.1002/elan.201700190. M. del Valle, Electronic tongues to assess wine sensory descriptors, Talanta 8