Toward the Optimization of an e-Tongue System Using Information

ARTICLE pubs.acs.org/Langmuir

Toward the Optimization of an e-Tongue System Using Information Visualization: A Case Study with Perylene Tetracarboxylic Derivative Films in the Sensing Units Diogo Volpati,† Pedro H. B. Aoki,† Cleber A. R. Dantas,|| Fernando V. Paulovich,‡ Maria Cristina F. de Oliveira,‡ Osvaldo N. Oliveira, Jr.,§ Antonio Riul, Jr.,|| Ricardo F. Aroca,^ and Carlos J. L. Constantino*,† †

Faculdade de Ci^encias e Tecnologia, UNESP Univ Estadual Paulista, Presidente Prudente/SP, 19060-900, Brazil Instituto de Ci^encias Matematicas e Computac-~ao and §Instituto de Física de S~ao Carlos, Universidade de S~ao Paulo, S~ao Carlos/SP, 13560-970, Brazil Universidade Federal de S~ao Carlos, Campus Sorocaba/SP, 18052-780, Brazil ^ Materials and Surface Science Group, University of Windsor, Windsor, Ontario N9B 3P4, Canada

)

‡

bS Supporting Information ABSTRACT: The wide variety of molecular architectures used in sensors and biosensors and the large amount of data generated with some principles of detection have motivated the use of computational methods, such as information visualization techniques, not only to handle the data but also to optimize sensing performance. In this study, we combine projection techniques with microRaman scattering and atomic force microscopy (AFM) to address critical issues related to practical applications of electronic tongues (e-tongues) based on impedance spectroscopy. Experimentally, we used sensing units made with thin films of a perylene derivative (AzoPTCD acronym), coating Pt interdigitated electrodes, to detect CuCl2 (Cu2+), methylene blue (MB), and saccharose in aqueous solutions, which were selected due to their distinct molecular sizes and ionic character in solution. The AzoPTCD films were deposited from monolayers to 120 nm via Langmuir
Blodgett (LB) and physical vapor deposition (PVD) techniques. Because the main aspects investigated were how the interdigitated electrodes are coated by thin films (architecture on e-tongue) and the film thickness, we decided to employ the same material for all sensing units. The capacitance data were projected into a 2D plot using the force scheme method, from which we could infer that at low analyte concentrations the electrical response of the units was determined by the film thickness. Concentrations at 10 μM or higher could be distinguished with thinner films—tens of nanometers at most—which could withstand the impedance measurements, and without causing significant changes in the Raman signal for the AzoPTCD film-forming molecules. The sensitivity to the analytes appears to be related to adsorption on the film surface, as inferred from Raman spectroscopy data using MB as analyte and from the multidimensional projections. The analysis of the results presented may serve as a new route to select materials and molecular architectures for novel sensors and biosensors, in addition to suggesting ways to unravel the mechanisms behind the high sensitivity obtained in various sensors.

1. INTRODUCTION It is now well established that the combination of ultrathin films and impedance spectroscopy leads to highly sensitive electronic tongues (e-tongues),1 which may be used to detect trace amounts of impurities in liquids and discriminate among similar samples of beverages, including coffee, orange juice, wine, milk, coconut water, etc. (for a review on e-tongues, see ref 2). The sensing units in the sensor array comprising the e-tongue may be produced with a variety of materials, mostly in the form of nanostructured films fabricated via the Langmuir
Blodgett (LB),3 the layer-by-layer (LbL),4 and the physical vapor deposition (PVD)5 methods on interdigitated electrodes for impedance spectroscopy measurements. The performance of e-tongues based on global selectivity can be improved with adequate choice of film-forming materials, even without specific interaction between these materials and the analytes.6 In special cases, this r 2011 American Chemical Society

e-tongue concept can be extended upon including sensing units capable of molecular recognition. For instance, LbL films containing enzymes7 and antigens8 were used in sensor arrays proven to be highly selective toward a specific analyte, in addition to having high sensitivity. Apart from impedance spectroscopy, several other methods have been used as principle of detection in e-tongues. The most common involve electrochemical measurements, including potentiometry, amperometry, cyclic voltammetry, chronoamperometry, and electrochemical impedance spectroscopy. Reviews on this subject can be found in Del Valle et al.,9 Ghasemi-Varnamkhasti et al.,10 Ciosek et al.,11 and Zhao et al.12 Received: September 16, 2011 Revised: November 20, 2011 Published: November 21, 2011 1029

dx.doi.org/10.1021/la203641a | Langmuir 2012, 28, 1029–1040

Langmuir There are now three major challenges to reach widespread use for e-tongues in real applications, including clinical diagnosis. The first one is the optimization of the experimental procedures and adequate choice of the sensing units for specific applications, as some materials have better performance for particular analytes. The second challenge is the most relevant from the point of view of technology and practical applications, namely, to solve the intrinsic limitation of fluctuations, when sensing units have to be replaced. This limitation is an undesired side effect of the high sensitivity of e-tongues, which is attributed to the major changes caused by interfacial effects in the electrical properties of the sensing systems. Indeed, the electrical response is not the same even for nominally identical sensing units, since the organic materials or the inorganic
organic hybrids normally used exhibit an intrinsic variability that is detectable in the electrical response. Therefore, every time a sensing unit has to be replaced, the overall response of the array changes. One possible way to solve (or at least minimize) this limitation is via software, using computational methods to map responses from one unit onto another.13 The third challenge is associated with fundamental science; more specifically, one may wish to understand why sensitivity is so high in some applications. To address this problem, one should ideally investigate all the molecular-level interactions governing the electrical response of the sensing units while in contact with the liquid under analysis. This is not feasible presently, from neither a theoretical nor experimental perspective. The electric response depends on too many uncontrollable parameters, such as the double-layer effects at the film/liquid interface, the metal/film interaction, the changes in film morphology, and the electrical properties of the liquid itself.14
18 Some experimental methods can nevertheless be used to probe at least some of the interactions, as is the case of atomic force spectroscopy19 and surface-enhanced Raman scattering (SERS).15 We have devised a study to address some of the issues associated with the three challenges above for e-tongues based on impedance spectroscopy. First, we investigate the effects of the film thickness and type of film (LB and PVD)—used to coat the interdigitated electrode—on the electrical response of the sensing units to determine small concentrations of three analytes, which vary with regard to the amount of charge they generate in aqueous solutions and to their dimensions. Rather than choosing distinct materials for the sensing units to take advantage of the cross sensitivity, which is important in the e-tongue concept, we have used sensing units made with the same material. The rationale behind this choice was to establish a direct relationship among the type of analyte, electrical response, and film characteristics (thickness and type), using only one material (a perylene derivative). While this is already important for the optimization of sensor arrays (first challenge mentioned above), we approach the subject using information visualization methods20 that are essential for the second challenge. We also study the interactions with the analytes, for which spectroscopic and microscopy techniques are used to characterize the organic films in the sensing units, before and after the electrical measurements (third challenge). The paper is organized as follows. In Section II, we provide details of the film fabrication, impedance spectroscopy measurements, and the information visualization methods to treat the data. Section III brings the results with discussion focusing on the optimization procedures and on changes in the film properties caused on the sensing units during data acquisition. The conclusions and perspectives are presented in the final section.

ARTICLE

Figure 1. Molecular structure of the AzoPTCD, MB, and saccharose used in the sensing experiments.

2. MATERIALS AND METHODS 2.1. Film Fabrication and Characterization. The bis (benzimidazo) perylene, AzoPTCD (536.58 g/mol), was provided by Dr. J. Duff from Xerox Resource Centre of Canada. Phenothiazine methylene blue (MB
319.85 g/mol) and copper(II) chloride dihydrate (CuCl2
170.48 g/mol) were acquired from Sigma-Aldrich Co. Saccharose (342.30 g/mol) was purchased from Synth. The molecular structures of AzoPTCD, MB, and saccharose are shown in Figure 1. The physical vapor deposition (PVD) films were deposited by vacuum thermal evaporation with 5, 10, 15, 20, 30, 40, and 120 nm mass thicknesses monitored by a quartz crystal microbalance. The growth process consists of placing the AzoPTCD powder in a metallic boat (W in this case), where an electric current is passed through. The Pt interdigitated electrodes and the quartz crystal microbalance were placed parallel and positioned 15 cm above the W boat. The evaporation process was performed in a vacuum chamber under 10
7 Torr. The electric current was adjusted slowly up to 2.1 A, with the W boat temperature increased to approximately 410 C, measured with a thermocouple. Then, the AzoPTCD started evaporating, and when a constant rate was reached, the quartz crystal microbalance was brought to zero value and the shutter protecting the substrate was opened allowing the growth of the PVD film until the desirable mass thickness. The films were grown one-by-one from 5 to 40 nm mass thickness in one step, while the 120 nm film was grown with three sequential steps of 40 nm, for the reason that AzoPTCD presents nonlinear growth for a thickness larger than 40 nm if evaporated in one step, as determined in previous work.21 The LB monolayers of AzoPTCD were deposited using a Langmuir trough, KSV 2000 model. A 10
4 mM solution of AzoPTCD (10% in volume of trifluoroacetic acid and dichloromethane) was spread onto ultrapure water with a microsyringe, and after 15 min allowed for solvent evaporation, the floating molecules were symmetrically compressed with a barrier speed of 10 mm/min, thus forming a Langmuir monolayer at the air/water interface, which was further transferred onto solid substrates at a constant surface pressure of 28 mN/m to form the LB film. The substrate dipping speed varied between 0.7 and 1 mm/min, being manually adjusted during the deposition (vertical method), and the LB films were deposited in a Z-type mode (upstroke deposition) with transfer ratio (TR) close to 1. The TR is the relationship between the area of the film removed from the air/water interface—given by the barrier displacement to keep the surface pressure constant—and the area of the substrate covered by the monolayer.22 The PVD and LB films were characterized by various techniques. UV
vis absorption spectra of AzoPTCD films deposited onto quartz 1030

dx.doi.org/10.1021/la203641a |Langmuir 2012, 28, 1029–1040

Langmuir

ARTICLE

Figure 2. (a) UV
vis absorption spectra for AzoPTCD forming PVD films with 10 and 120 nm grown in 3 steps of 40 nm (inset: absorbance at 540 nm for the PVD films), monolayer LB (1-LB, gray line), and solution (dotted line). (b) Variation of surface pressure and cumulative TR vs layer number (fraction of the substrate lifted from the water subphase) during deposition of the monolayer LB. substrates were recorded in a Cary 50 Varian spectrophotometer, in the spectral range between 290 and 900 nm. The growth of PVD films between 10 and 120 nm was monitored using the UV
vis absorption spectroscopy (ex situ), with the mass thickness being measured by a quartz crystal microbalance (in situ). As already mentioned, the PVD films between 10 and 40 nm were grown in a single step, while that with 120 nm was grown in 3 consecutive evaporation steps of 40 nm. For each evaporation process, a UV
vis spectrum was obtained to monitor the growth, as shown in Figure 2a. The inset shows that the absorbance at 540 nm increases almost linearly with the mass thickness up to 120 nm, indicating a constant rate of deposition onto the substrate. The same linear behavior was found for all the PVD films produced here. It is known21 that AzoPTCD molecules in PVD films form H and J aggregates. The molecules are organized in an edge-on fashion, i.e., supported by the longer molecular axis, as determined with FTIR spectroscopy combined with surface selection rules.21 The latter is also consistent with UV
vis absorption and surface-enhanced fluorescence (SEF) data. The transfer of the LB films was controlled by adjusting the dipper speed to keep the cumulative transfer curve (cumulative TR) close to 1. The cumulative TR curve is given by the TR values calculated every 1.1 s during the deposition, which allows for better control of the monolayer transfer onto the solid substrates. Figure 2b shows the layer number (fraction of the substrate already covered by the monolayer) against the cumulative TR and the surface pressure during the deposition onto the Pt interdigitated electrodes. The cumulative TR was kept around 1 during almost the whole deposition process and the surface pressure fluctuation was below 0.4 mN/m. This procedure ensures that the material was transferred homogeneously onto the Pt interdigitated electrodes. A UV
vis spectrum of an LB monolayer is given in Figure 2a in arbitrary units (out of scale) in a gray line. AzoPTCD forms stable Langmuir monolayers at the air/water interface, with the molecules packing through π
π interaction and probably adopting an edge-on organization at the air/water interface, according to surface pressure vs molecular area isotherms (π
A isotherms) and theoretical study using a rigid docking method.23,24 The molecular packing was kept during the LB deposition according to UV
vis absorption spectroscopy, while SEF measurements indicated that the molecules might be organized in a head-on fashion, i.e., supported by the shorter molecular axis in the LB monolayer. The UV
vis absorption spectra of perylene tetracarboxylic derivatives usually consist of one electronic transition with a characteristic vibronic structure.25
28 The spectrum of the AzoPTCD solution in Figure 2a (dotted black line) features a vibronic structure associated with the π
π* transition of the perylene chromophore with a 0
0 band at

610 nm, followed by subsequent 0
1 and 0
2 transitions, at 563 and 524 nm, respectively. Taking the spectrum in solution as a reference, the UV
vis absorption spectra for 10 and 120 nm PVD films, and LB monolayer onto glass, display an intense blue-shift component with a broad maximum at 540 nm and a red-shift component at 670 nm. The strong red- and blue-shift band components in the UV
vis absorption spectra are indications of an intermediate case of organized head-to-tail molecules described by Kasha’s point dipole model,29 which results in the apparent band splitting, as observed for this and other perylene derivatives.21,28,30
32 Raman scattering spectra were collected using a spectrograph microRaman Renishaw model in-Via, coupled to an optical microscope Leica with a 50 objective lens leading to a spatial resolution at ca. 2 μm2 for the 633 nm laser line. The spectrograph is equipped with a 633 nm laser line and grating with 1800 grooves/mm leading to a spectral resolution of ca. 4 cm
1. Additional notch filters and a computer-controlled threeaxis-encoded (XYZ) motorized stage were also included to take Raman images with a minimum step of 0.1 μm. Different powers of the incoming laser (μW range) were used to improve the signal/noise ratio. The Raman maps were performed using the static mode centered at 1290 cm
1, while the spectra of the materials were carried out using the extended mode from 500 to 2000 cm
1. 2.2. e-Tongue Systems and Impedance Spectroscopy. Two e-tongue systems were produced. One array consisted of 7 sensing units, viz., 1 Pt bare interdigitated electrode and 6 Pt interdigitated electrodes coated with 5, 10, 15, 20, 30, and 40 nm of AzoPTCD PVD films (e-tongue system 1). The other e-tongue system comprised 4 sensing units: a bare Pt interdigitated electrode, 1 Pt interdigitated electrode coated with 1 layer of AzoPTCD LB film, and 2 Pt interdigitated electrodes coated with 10 and 120 nm of AzoPTCD PVD films (e-tongue system 2). The latter e-tongue contains a 120 nm PVD film, which is thicker than the LB films used and the Pt interdigitated electrode geometry (50 pairs of digits 5 mm long, 100 nm high, 10 μm wide, and spaced 10 μm from each other). The bare Pt electrode is used as a reference to check the changes caused by the nanostructured films in the electrical response of the interdigitated electrodes. The electrical measurements were carried out with a Solartron impedance analyzer, model 1260A. The impedance spectra were acquired in the frequency range from 1 Hz to 1 MHz using 50 mV of amplitude, with the electrodes immersed into ultrapure water (18.2 MΩ.cm), used as reference, and in aqueous solutions of MB, Cu2+, and saccharose. Different concentrations were used depending on the e-tongue system. For the e-tongue system 1 (PVD films from 5 to 40 nm), the solutions had concentrations of 586 nM, 58.6 nM, 5.86 nM, and 0.586 nM for Cu2+; and 31.2 μM, 3.12 μM, 312 nM, 31.2 nM, 3.12 nM, and 0.312 nM 1031

dx.doi.org/10.1021/la203641a |Langmuir 2012, 28, 1029–1040

Langmuir

ARTICLE

Figure 3. Impedance spectroscopy results (capacitance vs frequency) for the e-tongue system 2 immersed into Cu2+ aqueous solutions at different concentrations (1 mM, 1 μM, and 10 nM) and into ultrapure water before and after the impedance experiments. for MB. For the e-tongue system 2 (monolayer LB, and 10 and 120 nm PVD films), the solutions were 1 mM, 1 μM, and 10 nM for Cu2+, MB, and saccharose. The measurements were carried out from the lowest to the highest concentrations and the reference spectra were obtained with ultrapure water. One should mention that, before dipping the sensing units into a specific solution, they were rinsed with a stock solution (same analyte and concentration) to avoid interference from the presence of water in the films that might alter the concentration of the measured solution and, consequently, the electrical data. Besides, for the e-tongue system 2, each type of solution (Cu2+, MB, and saccharose) was measured by an e-tongue made for that kind of analyte, i.e., three distinct arrays composed by Pt bare electrode, LB monolayer, and 10 and 120 nm PVD films were produced and applied in the experiments with Cu2+, MB, and saccharose, respectively. As for the e-tongue system 1 comprising the Pt bare electrode and PVD films with 5, 10, 15, 20, 30, and 40 nm, since saccharose was not used, only one array was produced and applied for the experiments with Cu2+ and MB. All bare Pt interdigitated electrodes were carefully tested prior to the film deposition, with all displaying a very similar electrical response over the whole frequency range (results given in the Supporting Information Figure SI 1). Moreover, a sequence of measurements at every 5 min during 1 h was made with the sensing units (120 nm PVD film) dipped into the solution to monitor the temporal stabilization of the electric signal due to the double-layer formation at the electrode/solution interface16 (results given in the Supporting Information Figure SI 2). After stabilization (30 min), the measurements were taken for the e-tongue systems 1 and 2. 2.3. Projection Techniques. In order to compare the sensor responses with regard to film thickness, analyte concentration, and

analyte ionic character, the data were analyzed using an information visualization approach known as Multidimensional Projection Technique, or simply, Projection Technique.33 The sensor responses (capacitance values) were considered as vectors embedded into an m-dimensional space, with m being the number of frequencies (in our tests, m = 61, varying from 1 Hz to 1 MHz, which means that for each frequency there is a corresponding capacitance value, totalizing 61 pairs). A projection technique mapped each vector into a visual element (in this case, a circle), preserving, up to a degree, the original similarity relationships between them. We take as a measure of similarity between two samples the Euclidean distances computed between their corresponding mdimensional vectors. On the resulting layout, similar vectors, i.e., similar sensor responses, are closely positioned and the dissimilar ones are placed far away. Thereby, the human visual ability can be employed to verify and compare the responses from different sensors to check how similar (or dissimilar) they are, by visually inspecting the different projection layouts depicting corresponding measurements for a set of samples. Formally, let δ(xi, xj) be the distance between two m-dimensional vectors xi, xj ∈ Rm (sensor responses), and let d(yi, yj) be the distance between two graphical elements yi, yj ∈ Rp, with p = {1, 2, 3}. A projection technique can be defined as an injective function f that maps f: RmfRp so that |δ(xi, xj)
d(f(xi), f(xj))| ≈ 0, " xi, xj ∈ Rm.34 There are various projection techniques, which differ in their ability to handle nonlinear data, to process large data sets, and so on, and therefore one has to choose the technique that is best suited for the application in hand (see Paulovich et al.35 for a discussion and comparison of different techniques). In this paper, we use a nonlinear, very precise technique, referred to as Force Scheme.34 Given an initial layout of the data vectors, either arbitrary or produced by a dimension 1032

dx.doi.org/10.1021/la203641a |Langmuir 2012, 28, 1029–1040

Langmuir

ARTICLE

reduction technique, Force Scheme applies an iterative approximation approach that, at each step, attempts to reduce the error in the relative positioning of the data points, considering their distance in the original and in the projected spaces. For each projected data sample yi ∈ Y, a vector 6 yj, is computed, then yj is moved on the direction of vij = (yj
yi), " yi ¼ vij. The amount of movement is given by M = δ(xi, xj)
d(yi, yj). By successively applying this process to all samples until a stable layout is reached (i.e., no further improvements are possible), or after a predefined number of iterations, the distances between samples on the visual space approach the corresponding distances in the original mdimensional space, rendering the final projection layout. Besides being very fast, processing hundreds of vectors in seconds, Force Scheme is precise, meaning it preserves on the visual space most of the distance relationships from the original space, an essential feature for the type of analysis conducted here.

3. RESULTS AND DISCUSSION 3.1. Studies for Optimization Using Information Visualization Methods. The ultrathin film coating on the interdigitated

electrodes has been suggested as one of the factors responsible for the high sensitivity of e-tongues.18 This must be associated with the importance of interface effects, but the precise basis for the thickness dependence is not known. Because the optimization of e-tongues may require understanding of how the electrical response depends on the sensor architecture, here we have tested sensing units with organic films possessing different thicknesses to investigate such dependence. We deliberately chose the same material to make the sensing units, so that only the thickness and the processing technique would be varied. For analytes with ionic character, the electric response depends on the analyte concentration, as illustrated in Figure 3 for Cu2+ solutions and e-tongue system 2 (4 sensing units of various thicknesses). Although data were acquired using impedance spectroscopy, all results were analyzed in terms of capacitance.17 The sequence of measurements started from the lowest to the highest concentrations, with a measurement in ultrapure water at the beginning and at the end of each run to check the reversibility of the sensor. There is an upward trend in the measured capacitance at frequencies lower than 105 Hz as the concentration increased, while for very high frequencies, the curves moved down in Figure 3. The latter is attributed to the geometric capacitance of the sensing units,16 being caused by changes in the dielectric constant of the medium or in the cell constant (geometry) of the system.36 If the ionic concentration increases, the resistance of the solution decreases, resulting in higher electrolyte conductivity. In other words, the theoretical expectations about the total impedance of the system, which can be made explicit using equivalent electric circuit models,1,36 were fulfilled. The shifts in capacitance were also observed for MB solutions, which have ionic character, but not for the nonionic saccharose (results given in the Supporting Information
Figure SI 3a,b, respectively). Therefore, the concentration of the analyte appeared to be the most important factor for the electrical response—in the case of ionic solutions— regardless of the film thickness in the sensing units. In order to check the thickness dependence in more detail, another set of experiments were carried out with the e-tongue system 1 comprising 7 sensing units (Pt bare electrode and PVD films with 5, 10, 15, 20, 30, and 40 nm). The results (not shown) were similar to those in Figure 3 for MB and Cu2+ aqueous solutions. The visual inspection of Figure 3 served to indicate the importance of the film thickness in the sensing units, but it does

Figure 4. Projections of data produced by different sensors and analytes with different ionic character and concentrations. For the analytes with ionic character (Cu2+ and MB), the distinguishability is governed by the analyte concentration down to ca. 10 μM. For the analyte with no ionic character (saccharose), separation is governed by the sensor architecture, as the analytes are not distinguishable, or there is no separation based on their concentration. Note that the axes are not labeled, because the plots obtained from projection techniques depict relative distances between data points. The two closer data points represent more similar samples.

not allow for the verification of whether the lowest concentrations will be distinguishable. We have therefore performed a quantitative analysis of the capacitance data for the two e-tongues using the Force Scheme visualization method. This was motivated by our recent work13 where information visualization methods proved useful not only to increase sensitivity and selectivity but also to optimize sensor performance. Figure 4a, b,c shows the plots for the capacitance curves obtained for aqueous solutions containing Cu2+, MB, and saccharose, respectively, using the e-tongue system 2. The color of the circles denotes the sensing unit as indicated in the inset, while the concentration is shown next to each point in the plot. For the analytes with ionic character (Cu2+ and MB), the higher concentration (1 mM) is easily distinguished, meaning that the thickness of the PVD film is not relevant, and the electrical response of the sensing units is basically governed by the 1033

dx.doi.org/10.1021/la203641a |Langmuir 2012, 28, 1029–1040

Langmuir

ARTICLE

Figure 5. Projections of data produced for two analytes removing the samples with high concentration. For low concentrations, the film thickness is the dominant factor, and the similarity between the samples, reflected by the proximity of the circles, is mostly defined by the sensor architectures. Data for saccharose were not obtained because only the film thickness is important, and this has already been shown unequivocally in Figure 4.

electrolyte. However, the smaller concentrations (1 μM and 10 nM) are not separable, and they are grouped based on the sensor architecture. Therefore, the film thickness is the dominant factor, and there is a tendency to gather the data points of the units with similar thicknesses to remain close. For saccharose (analyte without ionic character), the similarity between the samples is dominated by the sensor architecture. This is shown in the layout by the observable grouping of the sensor responses, based on the type of sensor. There is no clear separation or a tendency towards separation among the concentrations, with larger concentrations being closer to smaller concentrations, and vice versa. When the highest concentrations of Cu2+ are removed from the data in Figure 4, the resulting plot (result not shown) indicates that the sample with 10 μM is now distinguished from those at lower concentrations, thus demonstrating that down to 10 μM the film thickness has only a minor effect on the electrical properties. However, below 10 μM it is crucial to consider the film architecture, especially because the capacitance depends mainly on the film thickness. Figure 5a displays the results only for small concentrations and different film thicknesses for Cu2+ solutions. To get a more accurate dependence on film thickness, we used the e-tongue system 1. There is a trend in the plot for the data points shifting from left to right with increasing film thickness (blue to red). It means that the thickness dominates the sensor responses, consistent with the analysis of Figure 4. It is also worth mentioning that the measurements for pure water before and after a series of experiments give rise to data points close to each other, as expected, but the points do not coincide, showing that the electrical response of the sensing units is not entirely reversible. We shall return to this discussion later on in the paper. When the analyte is MB, which creates less electrical charge in the aqueous solutions than Cu2+, the concentration of the analyte is even less important than the film thickness, as indicated in Figure 5b. There is, nevertheless, some tendency for separation of different concentrations in the vertical direction in the plot. Moreover, thicker films tend to be worse in separating the

samples with distinct analyte concentrations, in agreement with the literature.18 Indeed, at higher concentrations the importance of the film thickness is diminished. From these results, we can infer that sensors with analytes having ionic character can separate the samples down to a certain concentration, below which the sensor architecture dominates the electrical response. For the analyte without ionic character, the responses are mainly dependent on the sensor architecture and, therefore, it is more difficult to distinguish the samples based on their concentrations only, even at higher concentrations. It has recently been shown37 that choosing appropriate projection techniques and treating the whole curves, as it was done to obtain the plots in Figures 4 and 5, it is advantageous compared to the widely used multivariate statistical analysis, including principal component analysis (PCA). Here, we analyzed the capacitance data from the e-tongue system 2 with PCA, upon taking the capacitance value at 6.3 kHz for the 120 nm PVD sensing unit and at 1 kHz for the other units. This choice of frequency was made to yield the best distinction ability. The plots in Figure 6 point to good discrimination for the analytes with ionic character, including PC1 values above 99%, and with a trend from left to right for the placement of solutions with increased concentrations. Furthermore, the sample with the highest concentration (1 mM) is placed apart on the far right of the plots in Figure 6a and b, which is consistent with the projection results. For this concentration, the electrical response is dominated by the solution concentration, in contrast to the case of lower concentrations for which the thin film architecture dominates the electrical response. For saccharose, on the other hand, the results from the PCA analysis were poorer. PC1 accounted for only 51.68% and there was no definite trend in the placement of the samples with increasing concentration. From this direct comparison, one may conclude that while both PCA and the projection technique using Force Scheme provided reasonable discrimination of the samples, with the latter method it was possible to map the electrical response from 1034

dx.doi.org/10.1021/la203641a |Langmuir 2012, 28, 1029–1040

Langmuir

ARTICLE

Figure 6. Principal component analysis of solutions containing (a) Cu2+ ions, (b) MB, and (c) saccharose measured with e-tongue system 2.

sensing units individually, clearly expressing a global measure of similarity between all sensing units. 3.2. Implications for e-Tongue Optimization. The main conclusion to be drawn from analyzing the layouts of the impedance (capacitance) data generated by the force scheme method is that the film thickness plays an important role for the distinguishing ability of e-tongues. For the two e-tongues tested, it was found that for high concentrations of the three analytes (Cu2+, MB, and saccharose) a sensor array with sensing units of the same material (AzoPTCD) is sufficient for a good distinction to be reached, even using thick films, especially for the analytes with ionic character. In contrast, for applications requiring higher sensitivity, with concentrations below 10 μM the films from AzoPTCD need to be very thin (tens of nm at most). Obviously, the sensitivity may be increased still further if different materials with suitable interactions toward the analytes were used, but this was not investigated in the present paper. Of special importance would be the choice of a material that is suitable for nonionic analytes, as indicated by the poor performance in distinguishing saccharose. Furthermore, it was possible to probe the reversibility in the electrical response by measuring the impedance data in pure water, before and after a series of independent measurements with the analytes. As indicated in the plots obtained with the Force Scheme Method in Figures 4 and 5, the response for the final measurements is close but not identical to the first measurements. Therefore, the response is not completely reversible, though the sensing units may be reused in applications where the accuracy required is not

high—then the distance between the data points for the initial and final water measurements is negligible compared to the distance between points representing different analytes. Overall, information visualization methods provide a fast, simple way to identify the film characteristics for sensor arrays to be used in specific applications. 3.3. Interface Effects and the Sensing Ability. We have used microscopy and spectroscopic techniques to try and verify whether the exposure of the sensing units to the three analytes caused changes in film properties that could be correlated with the sensing ability. Prior to this study, we characterized the sensing units with micro-Raman spectroscopy at the micrometer scale by collecting spectra point-by-point along the film surface. Raman maps could then be built by plotting the intensity of a given vibrational band of the analyte or of the AzoPTCD film. Therefore, a distribution of the analyte on the film surface or some structural properties of the film can be followed combining optical microscopy (morphology) and Raman spectra (chemical information). Figure 7a shows the Raman spectra collected with the 633 nm laser line for the 120 nm AzoPTCD film and powders of CuCl2, saccharose, and MB, which are used as references. AzoPTCD displays vibrational bands with a fluorescence background when using the 633 nm laser line. The bands at 1583, 1567, and 1547 cm
1 are assigned to CdC stretching (perylene), ring stretching (chromophore), and CdN stretching + ring stretching (perylene), respectively. The band at 1360 cm
1 is due to C—N stretching + ring stretching + C—H bending (perylene), while the one at 1290 cm
1 is assigned to C—H 1035

dx.doi.org/10.1021/la203641a |Langmuir 2012, 28, 1029–1040

Langmuir

ARTICLE

Figure 7. (a) Raman spectra recorded with the laser at 633 nm for powder samples of AzoPTCD, CuCl2, saccharose, and methylene blue (MB). (b) Optical images of Pt interdigitated electrodes covered by monolayer LB, and 10 and 120 nm PVD film (e-tongue system 2). The optical images are superimposed to Raman maps collected point-by-point with step of 3 μm for an area of 60 μm  60 μm using the 633 nm laser line. The brighter spots refer to higher intensities of the AzoPTCD band at 1290 cm
1. The Raman data were recorded before the impedance experiments.

Figure 8. Optical images of Pt interdigitated electrodes covered by monolayer LB, and 10 and 120 nm PVD films (e-tongue system 2). The optical images are superimposed to Raman maps collected point-by-point with step of 3 μm for an area of 60 μm  60 μm using the 633 nm laser line. The brighter spots refer to higher intensities of the AzoPTCD band at 1290 cm
1. The Raman data were recorded after the impedance experiments (before the final washing for taking the “final water” curves).

bending + ring stretching (chromophore).21 CuCl2 has a single, broad band at low frequency within the range investigated, which is related to the heavy metal ion, and saccharose displayed the spectrum in agreement with the literature.38,39 However, Cu2+ and saccharose did not exhibit detectable Raman spectra in solution

even when using a concentration of 1 mM. On the other hand, MB displays a typical Raman spectrum, and because the electronic absorption is in resonance with the laser line, one obtains the resonance Raman scattering (RRS) spectrum. A detailed assignment of the MB Raman bands can be found in Aoki et al.40 1036

dx.doi.org/10.1021/la203641a |Langmuir 2012, 28, 1029–1040

Langmuir

ARTICLE

Figure 9. (a) Raman spectra recorded for AzoPTCD films that cover the Pt interdigitated electrodes (e-tongue system 2) after immersing into MB solutions. (b) Optical images of Pt interdigitated electrode covered by the monolayer LB and 10 and 120 nm PVD film (e-tongue system 2). The optical images are superimposed on Raman maps collected point-by-point with step of 3 μm for an area of 60 μm  60 μm using the 633 nm laser line. The brighter spots refer to higher intensities of the MB band at 1394 cm
1. The Raman data were recorded after the impedance experiments with MB solution (before the final washing for taking the “final water” curves).

Figure 7b shows the Raman maps in situ made for the sensing units before immersion into the analytes’ solutions. Because Cu2+ and saccharose did not present Raman bands for solutions at the concentration used here, we built the Raman maps based on the AzoPTCD films (instead of the analytes). The latter allows one to infer whether the film structure covering the Pt interdigitated electrodes (sensing units) is affected by the impedance experiments. Hence, the band intensity at 1290 cm
1 of the AzoPTCD for LB, 10 and 120 nm films is shown in Figure 7b superimposed onto the optical images of the sensing units. Using a color scale associated with the intensity of the scattered signal, it is possible to establish a pattern and compare the effects of the analyte on the film after immersing the sensing units in the solutions. The Raman maps after immersion in the solutions during the impedance measurements are shown in Figure 8. The film structure covering the interdigitated electrodes was not altered at the micrometer scale by immersion in the saccharose solutions, as shown in Figure 8a through c. A decrease in the AzoPTCD scattered signal intensity was detected with minor changes in the color scale, but the color pattern was kept. Figure 8d through f shows the Raman maps for the sensing units after immersion in the Cu2+ solutions, now featuring major changes for the 120 nm film, while the thinner films presented no notable changes. A similar trend was found for the films immersed into MB solutions in Figure 8g through i. Basically, the changes in color scale with dark regions in the sensing units reveal that electrolytic solutions (Cu2+ and MB) might modify the films leading to irreversible structural modifications, while nonionic solutions (saccharose) do not. The fact that this change is observed mainly for the PVD film with 120 nm is significant since the sensing units widely applied in the e-tongue based on impedance spectroscopy have the interdigitated electrodes covered by films up to (typically) 5 LB layers,41
43 5 LbL bilayers,14 or 10 nm PVD.44 At least for AzoPTCD, using PVD films thicker than 100 nm is not recommended, because the adsorption of the film is not sufficiently strong to withstand the electrical measurements. Indeed, the minor

changes for the thinner films (10 nm PVD and monolayer LB) might be directly related to the stronger adsorption of the films onto the Pt interdigitated electrodes. Taking advantage of the large cross section of the MB RRS effect, its adsorption onto the sensing units could be analyzed by maps based on the RRS spectra. Figure 9a shows the Raman spectra for the AzoPTCD films that cover the Pt interdigitated electrodes after immersing into MB solutions. The appearance of the MB band at 1394 cm
1 is readily noted, and this may be assigned to CdN (lateral and center) symmetric stretching + C—H bending in plane (ring) + C—H bending out of plane (CH3) and CdN stretching + CdC stretching.40 Note that, in the 120 nm PVD film, this band appears as a shoulder in Figure 9a, as in this thicker film there is a larger quantity of AzoPTCD and the band due to MB becomes relatively smaller. Figure 9b shows the Raman maps using the MB band at 1394 cm
1 for the sensing units after immersing into the MB solution, revealing the spatial distribution of MB molecules adsorbed all over the AzoPTCD films. However, the thinner films exhibit a larger number of brighter spots, which is related to a greater number of MB molecules adsorbed onto the AzoPTCD films. The sensing unit made with 120 nm PVD film had the smallest number of brighter spots, which is consistent with the better AzoPTCD adsorption of the thinner films (LB and 10 nm) discussed in the analysis of Figure 9, i.e., there may be desorption of AzoPTCD from the 120 nm PVD film when immersing into the analyte solution, especially the electrolytic ones. This would affect the results for MB adsorption. In conclusion, the sensing units made with thin PVD and LB films are not affected significantly by the electrical impedance measurements, which is consistent with the fact that the electrical response is almost reversible after the units are washed with pure water. Reversibility is not complete though, and this could probably be ascribed to the minor changes noted in the Raman spectra. The high sensitivity of these sensing units may arise from the adsorption of the analytes on the film surface, as indicated with the results for MB (for the other analytes this could not be 1037

dx.doi.org/10.1021/la203641a |Langmuir 2012, 28, 1029–1040

Langmuir

ARTICLE

Figure 10. AFM images for the e-tongue system 2: (a) bare Pt interdigitated electrode, (b) monolayer LB, (c) 10 PVD film, and (d) 120 nm PVD film. The AFM images were recorded after the impedance experiments with saccharose solution (after taking the “final water” curves).

checked because they are Raman undetectable in the concentration range investigated). Adsorption is reversible as the electrical response of the sensing units for pure water almost coincides with that for pure water before the measurements with MB. For the thicker PVD film, the adsorption of MB does cause changes, but no straightforward conclusion can be drawn since film desorption appears to occur as well. The morphology of the sensing units at the nanometer scale was investigated through AFM. Here, an important issue is to determine how the interdigitated electrodes are covered by the monolayer LB and the PVD films with different thicknesses. This study was carried out after immersing the e-tongue system 2 into the saccharose solutions in the impedance spectroscopy experiments, and again washing the units with pure water. Figure 10a shows the AFM images for the Pt bare electrode in two and three dimensions, while the images for the sensing units coated by monolayer LB, and 10 and 120 nm PVD films, are shown in Figure 10b,c,d, respectively. Also shown is a profile of the sensing unit surfaces with morphological details along the Pt interdigitated electrodes (coated and uncoated). Some relevant information, which is not obvious at the nanometer level, can be extracted: (i) it seems that even the internal wall of the digits was covered by both LB and PVD films; (ii) a LB monolayer is already sufficient to modify the morphology of the interdigitated electrode surface; (iii) a

difference in height within 100 and 120 nm between the top and the bottom was kept for all the films, even in the case of the 120 nm PVD film. Probably, the films grow uniformly in and out of the digit keeping the difference of ca. 100 nm (height of the digit); (iv) in agreement with the micro-Raman data, the thicker film (120 nm PVD) exhibits some desorption of the AzoPTCD PVD film during the impedance spectroscopy experiments. On the other hand, only small changes could be noted on the morphology of the other sensing units, again consistent with an almost complete reversibility of the electrical response.

’ CONCLUSIONS AND FURTHER WORK The main aim in this work was to determine the role played by film thickness and architecture on e-tongue performance for solutions varying in concentration and ionic character. The combination of computational methods and surface characterization techniques has been proven suitable to address some issues related to optimization of sensing performance. With a direct comparison of sensing units of different thicknesses of the same material, AzoPTCD, we showed through multidimensional projections that the film thickness dominates the electrical response of sensing units in e-tongues when the analyte concentration is very low. For higher concentrations, good distinction ability could only be obtained with thin AzoPTCD films, produced 1038

dx.doi.org/10.1021/la203641a |Langmuir 2012, 28, 1029–1040

Langmuir either with the Langmuir
Blodgett methodology or via physical vapor deposition, especially in the cases where the analyte has an ionic character. Moreover, if the sensitivity required is much higher, i.e., with limit of detection at the μM level (or below), or if nonionic analytes such as saccharose are to be used, one has to resort to other materials rather than AzoPTCD. All these conclusions could be drawn by processing the electrical impedance data with a visualization method, available within a suite of tools referred to as PEx-Sensors,13 which also allowed us to verify that the sensing units are not entirely reversible in terms of their electrical response. The possible reversibility and the reasons for the sensitivity of the thin films could be studied with the help of Raman scattering maps and AFM images. First, the structure and morphology of the sensing units were not affected to any great extent by the impedance measurements, provided that the films coating the interdigitated electrodes were sufficiently thin. Interestingly, there is some adsorption of MB on the sensing units, as probed with the Raman scattering measurements, and this may be at the root of the high sensitivity observed in e-tongues. The adsorption, however, must be relatively weak and reversible, for upon washing the sensing units the original film properties are almost entirely recovered. The latter was shown with electrical impedance measurements and Raman scattering. The small changes observed in the Raman maps, nevertheless, appear to show that alterations in film structure and irreversible adsorption of analytes (even if in small quantities) could be the reason for the lack of reproducibility in impedance spectroscopy measurements. Indeed, the electrical response is so sensitive to minor changes in the film and/or liquid under analysis that it is very hard to obtain completely reproducible data even for nominally identical sensing units. Here, it is important to emphasize that our choice of sensing units made of a single material was useful for the comparative study, but other, equally efficient types of e-tongues may be built with sensing units containing distinct materials. For practical applications of e-tongues, one must circumvent these difficulties associated with fluctuations, and this can only be done by applying computational methods—including those related to information visualization—that allow for mapping of data from one sensing unit onto another. This is an ongoing project in our groups.

’ ASSOCIATED CONTENT Supporting Information. Additional figures as described in the text. This material is available free of charge via the Internet at http://pubs.acs.org.

bS

’ AUTHOR INFORMATION Corresponding Author

*E-mail: [email protected].

’ ACKNOWLEDGMENT This work was supported by FAPESP, CNPq/INEO, CAPES, and rede nBioNet (Brazil). ’ REFERENCES (1) Riul, A., Jr.; dos Santos, D. S.; Wohnrath, K.; Di Tommazo, R.; Carvalho, A.; Fonseca, F. J.; Oliveira, O. N., Jr.; Taylor, D. M.; Mattoso, L. H. C. Langmuir 2002, 18, 239–245.

ARTICLE

(2) Riul, A., Jr.; Dantas, C. A. R.; Miyazaki, C. M.; Oliveira, O. N., Jr. Analyst 2010, 135 (10), 2481–2495. (3) Blodgett, K. B. J. Am. Chem. Soc. 1935, 57, 1007–1022. (4) Decher, G.; Hong, J. D.; Schmitt, J. Thin Solid Films 1992, 210, 831–835. (5) Reinders, W.; Hamburger, L. Proceedings of the Koninklijke Akademie Van Wetenschappen Te Amsterdam 1917, 19, 958–968. (6) Borato, C. E.; Riul, A., Jr.; Ferreira, M.; Oliveira, O. N., Jr.; Mattoso, L. H. C. Instrum. Sci. Technol. 2004, 32, 21–30. (7) Zucolotto, V.; Pinto, A. P. A; Tumolo, T.; Moraes, M. L.; Baptista, M. S.; Riul, A., Jr.; Araujo, A. P. U; Oliveira, O. N., Jr. Biosens. Bioelectron. 2006, 21, 1320–1326. (8) Perinoto, A. C.; Maki, R. M.; Colhone, M. C.; Santos, F. R.; Migliaccio, V.; Daghastanli, K. R.; Stabeli, R. G.; Ciancaglini, P.; Paulovich, F. V.; Oliveira, M. C. F.; Oliveira, O. N., Jr.; Zucolotto, V. Anal. Chem. 2010, 82, 9763–9768. (9) Del Valle, M. Electroanalysis 2010, 22, 1539–1555. (10) Ghasemi-Varnamkhasti, M.; Mohtasebi, S. S.; Siadat, M. J. Food Eng. 2010, 100, 377–387. (11) Ciosek, P.; Wroblewski, W. Analyst 2007, 132, 963–978. (12) Zhao, W.; Xu, J.-J.; Chen, H.-Y. Electroanalysis 2006, 18, 1737–1748. (13) Paulovich, F. V.; Moraes, M. L.; Maki, R.; Oliveira, M. C. F.; Oliveira, O. N., Jr. Analyst 2011, 136, 1344–1350. (14) Aoki, P. H. B.; Volpati, D.; Riul, A., Jr.; Caetano, W.; Constantino, C. J. L. Langmuir 2009, 25, 2331–2338. (15) Aoki, P. H. B.; Alessio, P.; Riul, A., Jr.; Saez, J. A. D.; Constantino, C. J. L. Anal. Chem. 2010, 82, 3537–3546. (16) Taylor, D. M.; Macdonald, A. G. J. Phys. D: Appl. Phys. 1987, 20, 1277–1283. (17) Riul, A., Jr.; Malmegrim, R. R.; Fonseca, F. J.; Mattoso, L. H. C. Biosens. Bioelectron. 2003, 18, 1365–1369. (18) Ferreira, M.; Riul, A., Jr.; Wohnrath, K.; Fonseca, F. J.; Oliveira, O. N., Jr.; Mattoso, L. H. C. Anal. Chem. 2003, 75, 953–955. (19) Leite, F. L.; Firmino, A.; Borato, C. E.; Mattoso, L. H. C.; da Silva, W. T. L.; Oliveira, O. N., Jr. Synth. Met. 2009, 159 (21
22), 2333–2337. (20) Oliveira, M. C. F.; Levkowitz, H. IEEE Transactions on Visualization and Computer Graphics 2003, 9, 378–394. (21) Volpati, D.; Job, A. E.; Aroca, R. F.; Constantino, C. J. L. J. Phys. Chem. B 2008, 112, 3894–3902. (22) Petty, M. C. Langmuir-Blodgett films - An introduction; Cambridge University Press: Cambridge, 1996. (23) Constantino, C. J. L.; Aroca, R. F. J. Raman Spectrosc. 2000, 31, 887–U3. (24) Constantino, C. J. L.; Antunes, P. A.; Oliveira, C. B.; Trsic, M.; Caracelli, I.; Aroca, R. F. Canadian Journal of Analytical Sciences and Spectroscopy 2004, 49, 64–72. (25) Adachi, M.; Murata, Y.; Nakamura, S. J. Phys. Chem. 1995, 99, 14240–14246. (26) Mercadante, R.; Trsic, M.; Duff, J.; Aroca, R. THEOCHEM 1997, 394, 215–226. (27) Aroca, R. F.; Constantino, C. J. L. Langmuir 2000, 16, 5425– 5429. (28) Antunes, P. A.; Constantino, C. J. L.; Aroca, R.; Duff, J. Appl. Spectrosc. 2001, 55, 1341–1346. (29) Ramamurthy, V. Photochemistry in Organized and Constrained Media; VCH: New York, 1991. (30) Aroca, R. F.; Constantino, C. J. L.; Duff, J. Appl. Spectrosc. 2000, 54, 1120–1125. (31) Antunes, P. A.; Constantino, C. J. L.; Aroca, R. F.; Duff, J. Langmuir 2001, 17, 2958–2964. (32) Constantino, C. J. L.; Duff, J.; Aroca, R. Spectrochim. Acta, Part A 2001, 57, 1249–1259. (33) Paulovich, F. V.; Minghim, R. IEEE Transactions on Visualization and Computer Graphics 2008, 14, 1229–1236. (34) Tejada, E.; Minghim, R.; Nonato, L. G. Information Visualization Journal 2003, 2, 218–231. 1039

dx.doi.org/10.1021/la203641a |Langmuir 2012, 28, 1029–1040

Langmuir

ARTICLE

(35) Paulovich, F. V.; Silva, C. T.; Nonato, L. G. IEEE Transactions on Visualization and Computer Graphics 2010, 16, 1281–1290. (36) Hong, J.; Yoon, D. S.; Kim, S. K.; Kim, T. S.; Kim, S.; Pak, E. Y.; No, K. Lab Chip 2005, 5, 270–279. (37) Paulovich, F.; Maki, R.; de Oliveira, M. C. F.; Colhone, M.; Santos, F.; Migliaccio, V.; Ciancaglini, P.; Perez, K.; Stabeli, R.; Perinoto, A.; Oliveira, O. N., Jr.; Zucolotto, V. Anal. Bioanal. Chem. 2011, 400, 1153–1159. (38) Gift, A. D.; Taylor, L. S. J. Pharm. Biomed. Anal. 2007, 43, 14–23. (39) Kaminskii, A. A. Crystallogr. Rep. 2003, 48, 295–299. (40) Aoki, P. H. B.; Volpati, D.; Caetano, W.; Constantino, C. J. L. Vibr. Spectrosc. 2010, 54, 93–102. (41) Silva, B. A.; Antunes, P. A.; Pasquini, D.; Curvelo, A. A. S.; Aroca, R. F.; Riul, A., Jr.; Constantino, C. J. L. J. Nanosci. Nanotechnol. 2007, 7, 510–514. (42) Constantino, C. J. L.; Antunes, P. A.; Venancio, E. C.; Consolin, N.; Fonseca, F. J.; Mattoso, L. H. C.; Aroca, R. F.; Oliveira, O. N., Jr.; Riul, A., Jr. Sens. Lett. 2004, 2, 95–101. (43) Aoki, P. H. B.; Alessio, P.; Rodriguez-Mendez, M. L.; Saez, J. A. D.; Constantino, C. J. L. Langmuir 2009, 25, 13062–13070. (44) Aoki, P. H. B.; Caetano, W.; Volpati, D.; Riul, A., Jr.; Constantino, C. J. L. J. Nanosci. Nanotechnol. 2008, 8, 4341–4348.

1040

dx.doi.org/10.1021/la203641a |Langmuir 2012, 28, 1029–1040