Assessment of QCM array schemes for mixture identification: citrus scented odors

Abstract

QCM sensor arrays are promising systems for volatile complex mixture analysis. Such mixtures, sometimes termed odors, can prove to be challenging targets for accurate identification using conventional approaches. As a result, development of novel gas sensing systems and materials for identification of complex mixtures has garnered much interest in recent years. Herein, we present a comparative study between traditional and alternative quartz crystal microbalance (QCM) array sensing schemes for complex mixture identification. In this study, several citrus scented odors were chosen for identification using three different QCM sensing scheme. A traditional multisensor array (MSA) scheme was compared to a recently introduced virtual sensor array (VSA) scheme and identification results were found to be comparable (84–91% to 73–98% accurate). In addition, a new sensing scheme developed by combining complementary MSA and VSA schemes is introduced. In this regard, a virtual multisensor array (V-MSA), with enhanced data density, allowed accurate identification (100%) of complex mixtures (odor samples) over multiple concentrations. While each method employed is promising, the newly presented V-MSA scheme is superior to each of the previously presented array sensing methods for complex mixture analysis. To the best of our knowledge, this is the first report of a QCM V-MSA.

---

1. Introduction

One approach to discriminating complex mixtures of volatile organic compounds (VOCs) that is gaining popularity is use of cross-reactive sensor arrays (CRSAs).^1–6 The advantage of CRSAs lies in their use to identify and discriminate complex samples, without the need for identification of individual VOCs. In fact, this operating principle is similar to that of using the mammalian nose, which is held as the gold standard. As a result, this approach has been applied to fabrication of gas sensing arrays using various combinations of transducers and recognition elements.^7–9 Among possible combinations, quartz-crystal microbalance (QCM) transducers, coupled with ionic liquids as recognition elements, have proven increasingly attractive.^3,5,10–13 As a transducer, the QCM is sensitive, offers potential for miniaturization,¹⁴ and is amenable to fabrication of sensor arrays.^{10–12,14,15} Moreover, ionic liquids (ILs), as recognition elements, have been demonstrated to be promising sensing materials for detection and discrimination of a wide range of organic vapors,^{3,10–13,15,16} ILs, which are defined as organic salts with melting points below 100 °C, are highly tunable, possess low vapor pressure, and allow reversible capture of organic vapors. ILs have also been shown to possess viscoelastic behavior,^17–19 a property which plays a critical role in the present studies. Notably, IL based QCM sensor arrays have great potential to satisfy the requirements of simple, rapid, reproducible and reliable systems for discrimination of complex mixtures.^3,10,15,16

By convention, multisensor arrays (MSA) based on chemical affinity have been employed for applications using QCM based CRSAs. In such schemes, each sensor contains a unique cross-reactive recognition element. Upon exposure to complex mixtures, differential sensor responses are exhibited and subsequently used to generate analyte specific patterns. Such patterns are then analyzed using statistical methods (e.g., principal component analysis (PCA), discriminant analysis (DA), artificial neural networks (ANN), cluster analysis (CA), etc.) to predict the utility of the array for identification/discrimination. However, we have recently introduced an alternative strategy for a virtual sensor array (VSA) based on a single sensor, viscoelasticity, film thickness, and harmonics that exhibits significant advantages as compared to MSAs.¹⁰ In this regard, cost, complexity, and problems associated with sensor drift are minimized in VSAs. However, chemical affinity is lost as a discriminatory factor. Naturally, questions arise as to whether these two schemes are (1) comparable for a given application and (2) can be used as complementary approaches.

Herein, a comparative study between QCM array sensing schemes for complex mixture identification is detailed. Hence, the utility of MSA and VSA schemes for citrus odor recognition is directly compared. Furthermore, we introduce, for the first time, a complementary approach that incorporates the two sensing schemes, i.e. use of chemical affinity and viscoelastic discriminatory factors, to rapidly enhance the information density of QCM based sensor arrays. In this regard, a QCM based virtual multisensor array (V-MSA) for complex mixture discrimination is introduced. Ultimately, the performance of MSAs, VSAs, and V-MSAs are compared, using statistical analysis to reveal the best QCM array strategy for complex mixture identification. As a proof of concept, a set of citrus odors, represented by aroma profiles of five essential oils, were chosen as representative complex mixtures. Essential oils are produced by plants, and are responsible for the characteristic flavor and odors of particular species. As a result, these oils are important stocks for the food, pharmaceutical, and perfume industries.²⁰ Odors emitted from essential oils are highly complex mixtures of VOCs. In fact, a common scent, such as one derived from an orange essential oil may contain more than 300 VOCs although only a small fraction of these are truly responsible for the scent commonly identify as orange.²⁰ Hence, odors are challenging targets for accurate identification and discrimination using conventional approaches.²

2. Approach

2.1 Array fabrication

In this study one MSA, four VSAs and eleven V-MSAs were fabricated using ILs immobilized as thin films onto QCM-D transducers. Ionic liquids were chosen as promising chemosensitive materials for odor recognition based on work in the field by Nakamoto and colleagues, who have demonstrated that lipids have favorable aroma/odor sensing properties.^21,22 Since lipids are composed of a polar head group(s) and an aliphatic tail(s), four ionic liquids, 1-nonyl-3-methylimidazolium thiocyanate ([C₉MIm][SCN]), 1-nonyl-3-methylimidazolium bromide ([C₉MIm][Br]), 1-octenyl-3-pyridinium bromide ([C₈Pyr][Br]), and 1-undecenyl-3-pyridinium bromide ([C₁₁Pyr][Br]) were chosen for odor sensing since each IL contains a polar head group represented by the charged imidazolium/pyridinium group and anion and aliphatic tails represented by varying length carbon side chains (saturated and unsaturated). Structures of these four ILs are displayed in Fig. 1. Each of the four sensors acquired by coating four quartz crystal using these different ionic liquid, were installed into the instrument simultaneously. Measurements were taken concurrently for each sensor across multiple harmonics to create one master data set. Subsequently, fabrication of each sensor array scheme was accomplished by considering appropriate measurements from the master data set. The MSA comprises four sensors with coatings [C₉MIm][SCN], [C₉MIm][Br], [C₈Pyr][Br], and [C₁₁Pyr][Br] respectively, where only the response at the first harmonic (i.e. fundamental frequency) was used for data analysis. This is consistent with the standard operation of QCM MSAs in the literature. Four VSAs were employed using the same sensors from the MSA, only individually. In this regard, measurements across all harmonics for an individual sensor were considered for analysis as performed in a previous publication.¹⁰ Finally, to test the utility of these two schemes as complementary methods, multiple V-MSAs were fabricated by considering the aggregate of measurements from all four sensors over all harmonics and these data were analyzed. It should be noted that all measurements considered for each array were taken concurrently. Therefore, any variation in identification accuracy is a result of the scheme as tested and not to differing experimental procedure or conditions. For clarity a scheme of each array type is depicted in Fig. 2.

Fig. 1 Chemical structures of organic salt adlayers.

Fig. 2 QCM array sensing schemes (A) MSA (B) VSA and (C) V-MSA. Where red, green, and blue sensors represent physical sensors. Smaller odd numbered sensors represent harmonics.

2.2 QCM virtual sensor arrays

Development of QCM based virtual sensor arrays has been detailed in a previous publication.¹⁰ For clarification, a brief synopsis is provided here. Briefly, a QCM sensor coated with a viscoelastic material is dynamically operated to obtain analyte specific response patterns, which can be used for identification purposes. Such patterns are obtained by exploiting the effects of film thickness, harmonics, and viscoelasticity on sensor response of the QCM. It is known from the Sauerbrey equation that thin and rigid films exhibit best Sauerbrey like behavior. However, as film thickness increases, deviations from ideal Sauerbrey behavior occur due to increasing viscoelastic effects on sensor response. It is also known that such deviations can be further enhanced depending on the viscoelastic properties of the coating material to obtain all combinations of positive and negative sensor response. In our previous studies, this relationship was exploited to create sensor arrays by performing measurements across multiple harmonics. This strategy is based on the observation that change in sensor response at different harmonics is due to changes in mass and viscoelastic contributions as a result of perceived variations in film thickness. In fact, measurements of a fixed film thickness at multiple harmonics can be viewed as equivalent to measurements of multiple film thicknesses at a single harmonic. This idea is rationalized using the following equation:
where δ is penetration depth, i.e. the distance at which the amplitude of the wave decreases to 1/e of its value at the surface; η is the viscosity of the coating; ρ is the density of the coating; ω is angular frequency, and f is frequency.^10,23,24

2.3 QCM virtual multisensor arrays

Herein, we introduce the first example of a QCM virtual multisensor array scheme. This scheme is based on employing the multisensory array and virtual sensor array schemes in a complementary fashion. Notably, the multisensor array scheme is based on employing multiple sensors, which differ, based on chemical affinity. In contrast, the virtual sensor array scheme is based on dynamic operation of a single sensor, which elicits differential response based on harmonics, film thickness, and viscoelasticity. Thus, the logical conclusion is that the schemes are not required to be mutually exclusive. In fact, since each scheme employs different discriminatory factors, it should be expected that the combination (V-MSA) should yield more discriminatory information. In this regard, the V-MSA scheme should exhibit enhanced data density, which should result in more accurate analyses than either the MSA or VSA alone.

3. Experimental section

3.1 Reagents and materials

Four ILs, i.e. [C₉MIm][SCN], [C₉MIm][Br], [C₈Pyr][Br] and [C₁₁Pyr][Br], were used to prepare quartz crystal coatings in the present studies and were synthesized using previously documented procedures.^10,25 Foods brand 100% pure essential oils were purchased from Whole Foods Inc, i.e. lemon, lime, orange, lemon eucalyptus, and lemongrass. All oils were used as received without further purification. Dichloromethane was purchased from Macron Fine Chemicals (Center Valley, PA, USA) and used without further purification.

3.2 Preparation of stock solutions

Stock solutions of [C₉MIm][SCN], [C₉MIm][Br], [C₈Pyr][Br] and [C₁₁Pyr][Br] (1 mg mL⁻¹) were prepared in dichloromethane (DCM) using 20 mL borosilicate glass scintillation vials.

3.3 Preparation of sensing films

Electrospray was used for deposition of thin films. All films were coated using the same parameters: deposition time of 1.5 min, flow rate of 100 μL min⁻¹, current of 3 A, voltage of 2.9 V, and a working distance of 7 cm. After coating, all films were blown with nitrogen and stored in a desiccator. The change in frequency between coated and uncoated crystal was found to be ∼−1000 Hz at the fundamental frequency. Coatings were characterized using scanning electron microscopy (SEM) and energy dispersive X-ray spectroscopy (EDS) (Fig. S1 and S2†).

3.4 QCM-D data acquisition

The flow type system used to perform these experiments consists of two independent gas flow channels i.e., one channel for sample odors and another channel for carrier gas. As a first step in data collection, ultrapure argon is purged through the system to obtain a stable baseline frequency. Odors are introduced via bubbling of argon gas through the sample reservoir, which was filled with the pure essential oil of interest, in order to generate a sample of saturated vapor pressure. As the sample channel and carrier channel merge, the sample flow is diluted yielding percentages of the respective saturated vapor pressure (SVP) (e.g. 10% 50%, 75%, 100% of saturated vapor pressure). The flow rate was controlled by digital mass flow controllers and adjusted to a total flow rate of 100 sccm. After sufficient mixing over the length of the tubing (1 m), the vapors are passed over the QCM sensor crystal placed inside a flow module. The chamber temperature was precisely regulated (22 °C). Finally, to remove sample vapors, the system was purged with ultrapure argon until recovery of the baseline. A schematic of the experimental system is depicted in Fig. S8.†

3.5 Data analysis

A single data set was generated from these experiments and used to develop statistical models for assessing the identification accuracies of each of the arrays. In this regard, independent predictive models were developed using frequency change (Δf) response values appropriate for a particular array scheme (VSA, MSA, V-MSA). In order to reduce the dimensionality of the data set, principal component analysis (PCA) was performed. Subsequently, discriminant analysis was performed using PCA indices (accounting for 99% of variance in the original data set) as input variables to quantitatively access the ability of each array for identification of selected odors. In this regard, linear discriminant analysis (LDA) and quadratic discriminant analysis (QDA) using cross-validation were used to obtain classification error rates for all models.

4. Results and discussion

4.1 Sensor response of ILs to odors

Four QCM sensors coated with [C₉MIm][SCN], [C₉MIm][Br], [C₈Pyr][Br], and [C₁₁Pyr][Br] respectively, were installed into a flow type system. Subsequently all sensors were introduced to a set of five citrus scented odors (lemon, lime, orange, lemon eucalyptus, lemongrass) selected for comparative assessment of complex mixture identification ability. This set consists of chemically distinct and closely related odors generated by essential oils. Differences in odor composition were confirmed using GC-MS analysis (Fig. S9†). All sensors were exposed to three different concentrations (10%, 20%, 40% of SVP) of the respective odors for 30 s intervals and Δf at each harmonic was measured. For each analyte, this corresponds to ppm regime concentration ranges depicted in Table 1. Each sensor was found to exhibit stable baseline, and reversible capture (Fig. S3–S7†). Moreover, sensor responses were found to be stable and reproducible. Sensitivity of each sensor at the fundamental frequency is depicted in Table S1.† Fig. 3 depicts sensor response for all sensors across multiple harmonics when exposed to each of the five citrus scented odors. As in our previous study,¹⁰ cross-reactive responses to different analytes were obtained by simply changing harmonics of the same sensor. This suggests that each sensor has potential for not only fabrication of MSAs, but also VSA and V-MSAs. As expected, positive and negative shifts in resonant frequency were obtained. Such shifts can be attributed to relative changes in the mass and viscoelastic contributions of the chemosensitive film to sensor response. In this regard, response variability is primarily governed by film viscoelasticity and penetration depth with changing harmonic. These dependencies have been described in previous publication.¹⁰ Overall, the differential sensing patterns obtained for each sensor seem promising for fabrication of sensor arrays.

Table 1 Analyte concentration ranges

Analyte	Concentration (mg L^−1)
Lemon	1–4.8
Lemon eucalyptus	0.6–4.4
Lemongrass	0.2–1.6
Lime	0.4–3.6
Orange	1.2–2.2

---

Fig. 3 Sensor response of citrus type odors at multiple harmonics at 20% of SVP for (A) [C₉MIm][SCN], (B) [C₉MIm][Br], (C) [C₈Pyr][Br], and (D) [C₁₁Pyr][Br]. Error bars represents standard deviation for three replicate measurements.

4.2 Evaluation of a multisensor array for odor recognition

Fabrication of an MSA was performed in the traditional fashion. In this regard, an MSA consisted of all four chemically distinct sensors, specifically [C₉MIm][SCN], [C₉MIm][Br], [C₈Pyr][Br], and [C₁₁Pyr][Br]. Only measurements taken at the fundamental frequency (first harmonic) for each sensor were considered for MSA performance evaluation. As a result, the data matrix consisted of Δf values, where the columns represented each of the chemically distinct sensors and the rows represented each of the odors tested at three concentrations (10%, 20%, and 40% of SVP). Three replicate measurements were considered for each odor, giving a total of 9 measurements per sample (4 columns × 45 rows). A canonical plot for this MSA is depicted in Fig. 4, wherein ellipses represent 95% confidence. Qualitatively, high level discrimination is typically represented by good clustering within a sample and spatial separation between samples. Therefore, the high degree of overlap exhibited in this plot suggests less than optimal accuracy. This is quantitatively supported, by classification results using LDA with the cross validation method, where an accuracy of 84.5% was achieved. This corresponds to 7 total misclassifications. As suggested by examination of the plot, two major groups of odor confusion were exhibited. The first group contains 5 instances of confusion between lemon, lemongrass, and orange odors, while the second group consists of 2 instances between lime and lemon eucalyptus odors. Typically classification accuracies can be improved by employing quadratic discriminant analysis (QDA). This method allows for better approximation of decision boundaries resulting in better classification. However, it can only be employed when the number of sensors are less than the number of sample measurements for a single analyte. Upon employing QDA on the same data set, an accuracy of 91.2% corresponding to 4 misclassifications, was achieved. Overall, this level of accuracy suggests that the multisensor array as presented is a reasonable method for odor identification. Incidentally, including additional sensors as is done in some studies should further enhance accuracy.^3,11 Yet, when compared to methods presented herein, addition of more sensors could be disadvantageous as it increases experimental time, materials cost, and complexity when using the current system.

Fig. 4 Canonical plot for identification of five citrus type odors with respect to a 4 sensor MSA. Plot considers 45 total measurements consisting of three replicate measurements at three concentrations for each odor (9 measurements per sample).

4.3 Evaluation of virtual sensor arrays for odor recognition

To examine the utility of VSAs for applications in complex mixture identification, each sensor was analyzed as an independent system. In this regard, an individual sensor, coated with a single ionic liquid and its respective multiple harmonic data (frequency change at different harmonics) would constitute an array. The analyte-selective sensing patterns generated by measurements at multiple harmonics are subsequently used for data analysis. Thus, the data matrix would consist of Δf values where the columns represented each of the harmonics and the rows represented each of the odors tested (7 columns × 45 rows). Three replicate measurements were considered for each odor. Each ionic liquid was tested as a separate VSA. Fig. 5 depicts canonical plots for each of the 4 VSAs tested, where ellipses represent 95% confidence. Al-though the discriminating factors for the two array sensing schemes (MSA and VSA) are different, it is interesting to see that the same overlaps between certain odors are present. This would suggest that both array types yield comparable information. This is further supported when considering the identification accuracies obtained via LDA. In this regard, accuracies of 88.9%, 73.4%, 80%, 73.4% were acquired for [C₉MIm][SCN], [C₉MIm][Br], [C₈Pyr][Br], and [C₁₁Pyr][Br] based VSAs, respectively. Furthermore, these accuracies can be enhanced by employing QDA. In this regard, accuracies of 97.8%, 91.2%, 97.8%, and 97.8% were obtained for [C₉MIm][SCN], [C₉MIm][Br], [C₈Pyr][Br], and [C₁₁Pyr][Br] based VSAs, respectively. Thus, depending on sensor composition, the accuracies obtained are comparable or better than the presented MSA. However, when considering the ease of implementation and significant advantages that arise when using a VSA, this method may have better utility for certain applications. Although, these accuracies represent a vast improvement compared to earlier results (MSA), further optimization would be needed to obtain 100% accuracy.

Fig. 5 Canonical plots for identification of five citrus type odors with respect to four VSAs based on (A) [C₉MIm][SCN] (B) [C₉MIm][Br] (C) [C₈Pyr][Br], and (D) [C₁₁Pyr][Br] respectively. Each plot considers 45 total measurements consisting of three replicate measurements at three concentrations for each odor (9 measurements per sample).

4.4 Evaluation of virtual multisensor sensor arrays for odor recognition

Thus far, evaluation of data presented herein, supports the assertion that MSAs and VSAs can be used interchangeably. Notably, such array schemes are not mutually exclusive. In addition, the discriminating factor for each scheme is quite different. Thus, it is reasonable to expect that the two schemes could be complementary. Since a VSA is based on dynamic operation of a single sensor and an MSA is based on utility of multiple chemically distinct sensors, a complementary system could consist of several dynamically operated sensors, which are chemically distinct. In order to assess the utility of using the MSA and VSA methods as complementary methods, several V-MSAs were fabricated using previously collected data. First, a set of V-MSAs, consisting of two sensors, was fabricated. In this regard, response data for two chemically distinct sensors at all harmonics were considered, effectively combining each of the array schemes. By combining the two schemes, the response output has been augmented 2 fold in comparison to a VSA and 7 fold when compared to 2-sensor MSA. In terms of dimensions, the resultant data matrix consisted of 14 columns representing the harmonics (i.e. 2 VSAs or 7 harmonics per sensor) and 45 rows representing three replicate tests at three concentrations of 5 odors. Fig. 6 is a depiction of canonical plots for the six possible 2-sensor V-MSA combinations derived from the four chemically distinct sensors (1 – [C₉MIm][SCN], 2 – [C₉MIm][Br], 3 – [C₈Pyr][Br], and 4 – [C₁₁Pyr][Br]). While it is challenging to discern qualitative differences between these plots, quantitative accuracies showed a marked increase in identification capability for each of the two sensor V-MSAs. In this regard, LDA accuracies of 97.8%, 100%, 97.8%, 93.4%, 84.5% and 93.4% were obtained for sensor combination (A) 1–2 (B) 1–3 (C) 1–4 (D) 2–3 (E) 2–4 and (F) 3–4, respectively. It should be noted that these accuracies are obtained through LDA since QDA is not applicable for this system given that the number of sensors are more than the number of measurements per sample. Therefore, performance comparisons should be made by considering the LDA results for the previous array sensing schemes, which indicates that the V-MSA method is very promising. Moreover, there is a trend in accuracies for the V-MSAs comprised of 2 chemical sensors that can be predicted by considering their constituent VSA accuracies. In this regard, V-MSAs comprising more accurate VSA sensors yield higher identification accuracies, while V-MSAs comprising less accurate VSA sensors yield lower accuracies. This can be clearly seen when examining V-MSA B, which is comprised of a [C₉MIm][SCN] VSA (88.9%) and a [C₈Pyr][Br] VSA (80%) to V-MSA E, which consists of a [C₉MIm][Br] VSA (73.4%) and a [C₁₁Pyr][Br] VSA (73.4%). The former combination comprises the most accurate VSAs and exhibits the highest accuracy V-MSA (100%) as compared to the latter combination, which comprises the least accurate VSAs and exhibits the lowest accuracy V-MSA (84.5%). While these results are quite reasonable, it is still possible to further enhance the discriminatory power and identification accuracies of V-MSAs. In this regard, a set of V-MSAs consisting of three chemical sensors was fabricated. Fig. 7 is a depiction of LDA canonical plots for 4 possible combinations of V-MSAs (comprised of three chemical sensors) derived from the four chemically distinct sensors (1 – [C₉MIm][SCN], 2 – [C₉MIm][Br], 3 – [C₈Pyr][Br], and 4 – [C₁₁Pyr][Br]). For each array, the resultant data matrix consisted of 21 columns representing the harmonics and 45 rows representing three replicate tests at three concentrations of 5 odors. This corresponds to a 3-fold enhancement of data output as compared to a VSA and a 7-fold enhancement as compared to a 3 sensor MSA. Again, LDA was employed using the cross validation method, and accuracies of 100%, 97.8%, 100%, 97.8% were obtained for sensor combination (A) 1–2–3 (B) 1–2–4 (C) 1–3–4 (D) 2–3–4, respectively. Interestingly, two observations can be made when examining the resultant plots and accuracies. When considering the plots, it is clear that the addition of a 3^rd sensor to the V-MSA, significantly enhances spatial separation and clustering of individual samples. In fact, the degree of overlap displayed in these plots is by far the least when compared to any of the previous sensing schemes. Qualitatively, this agrees with the excellent identification accuracies obtained. When considering accuracies, it is clear that addition of an extra sensor further enhances V-MSA results as would be expected from comparing the V-MSA comprised of 2 sensors to the single sensor VSA. Notably, sensor combinations, which were 100% accurate in the two-sensor iteration, retained their level of accuracy upon addition of a third sensor as would be expected. Moreover, two sensor combinations which were less accurate exhibited an increase in accuracy upon addition of a third sensor. Such results suggest that even the least accurate systems can be suitably optimized by addition of a single sensor. Finally, a V-MSA was fabricated by using all four ionic liquid sensors, and statistical analysis was used to measure identification accuracy. The resultant data matrix of this array consisted of 28 columns, representing the harmonics and 45 rows, representing three replicate tests at three concentrations of 5 odors. Apparently, this corresponds to a 4-fold enhancement of data output as compared to a VSA and a 7-fold enhancement as compared to a 4 sensor MSA. Fig. 8 is a canonical plot for the 4 sensor V-MSA comprised of four chemically distinct sensors (1 – [C₉MIm][SCN], 2 – [C₉MIm][Br], 3 – [C₈Pyr][Br], and 4 – [C₁₁Pyr][Br]). In comparison to plots generated from previous sensing schemes, it is easy to see that clustering within a sample and spatial separation between samples is significantly improved. Qualitatively, this plot should represent a significant increase in odor identification accuracy, as sample overlap is almost nonexistent. This supposition is supported quantitatively, where an odor identification accuracy of 100% was obtained using the cross validation method. This result is truly logical based on results from the three sensor V-MSA. Taken in aggregate, the results herein extoll the effectiveness of the new V-MSA scheme. Hence, we have proven that the V-MSA is an excellent approach for complex mixtures analysis. High discrimination accuracy was acquired as the number of sensors increased within a given V-MSA scheme.

Fig. 6 Canonical plots for identification of five citrus type odors with respect to several two sensor V-MSAs based on combinations of 1 – [C₉MIm][SCN], 2 – [C₉MIm][Br], 3 – [C₈Pyr][Br], and 4 – [C₁₁Pyr][Br] where arrays consist of sensors (A) 1–2 (B) 1–3 (C) 1–4 (D) 2–3 (E) 2–4 and (F) 3–4 respectively. Each plot considers 45 total measurements consisting of three replicate measurements at three concentrations for each odor (9 measurements per sample).

Fig. 7 Canonical plots for identification of five citrus type odors with respect to several three sensor V-MSAs based on combinations of 1 – [C₉MIm][SCN], 2 – [C₉MIm][Br], 3 – [C₈Pyr][Br], and 4 – [C₁₁Pyr][Br] where arrays consist of sensors (A) 1–2–3 (B) 1–2–4 (C) 1–3–4 (D) 2–3–4 respectively. Each plot considers 45 total measurements consisting of three replicate measurements at three concentrations for each odor (9 measurements per sample).

Fig. 8 Canonical plot for identification of five citrus type odors with respect to a four sensor V-MSA comprised of sensors 1–2–3–4 where 1 – [C₉MIm][SCN], 2 – [C₉MIm][Br], 3 – [C₈Pyr][Br], and 4 – [C₁₁Pyr][Br]. Plot considers 45 total measurements consisting of three replicate measurements at three concentrations for each odor (9 measurements per sample).

5. Conclusion

In conclusion, a comparative study to assess the utility of QCM array sensing schemes for complex mixture identification was performed by employing four chemically distinct ionic liquids. It was observed that MSA and VSA schemes could be used interchangeably since accuracy levels were comparable. However, the VSA scheme is potentially more promising than MSA in terms of cost, labor, and time expenditure. Furthermore, a new sensing scheme based on complementary use of MSA and VSA schemes was introduced. In this regard, multiple V-MSAs were systematically developed and analyzed using two, three, and four ionic liquid based sensors. After comparing results, it is clear that the V-MSA scheme is extremely promising as compared to existing QCM sensing schemes. It is also apparent from this scheme that increased data density is important for achieving highly accurate identification of complex mixtures. Overall, these studies are particularly promising for use of QCM sensor arrays in applications involving odor recognition. Potential examples of such applications would include: quality control of perfume/scents and vapor phase assessment of food quality. Currently, additional studies are underway to ascertain the full potential of this method with regard to complex mixture analysis.

Acknowledgements

This material is based upon work supported by the National Science Foundation Graduate Research Fellowship Program under grant number DGE-1247192 to NCS; National Science Foundation under grant numbers CHE-1307611; and funds form the Phillip W. West Endowment to IMW.

References

1. J. R. Carey, K. S. Suslick, K. I. Hulkower, J. A. Imlay, K. R. Imlay, C. K. Ingison, J. B. Ponder, A. Sen and A. E. Wittrig, Rapid identification of bacteria with a disposable colorimetric sensing array, J. Am. Chem. Soc., 2011, 133(19), 7571–7576.
2. B. A. Suslick, L. Feng and K. S. Suslick, Discrimination of complex mixtures by a colorimetric sensor array: coffee aromas, Anal. Chem., 2010, 82(5), 2067–2073.
3. R. Toniolo, A. Pizzariello, N. Dossi, S. Lorenzon, O. Abollino and G. Bontempelli, Room temperature ionic liquids as useful overlayers for estimating food quality from their odor analysis by quartz crystal microbalance measurements, Anal. Chem., 2013, 85(15), 7241–7247.
4. S. Capone, M. Epifani, F. Quaranta, P. Siciliano, A. Taurino and L. Vasanelli, Monitoring of rancidity of milk by means of an electronic nose and a dynamic PCA analysis, Sens. Actuators, B, 2001, 78(1), 174–179.
5. W. I. Galpothdeniya, K. S. McCarter, S. L. De Rooy, B. P. Regmi, S. Das, F. Hasan, A. Tagge and I. M. Warner, Ionic liquid-based optoelectronic sensor arrays for chemical detection, RSC Adv., 2014, 4(14), 7225–7234.
6. H. Kwon, F. Samain and E. T. Kool, Fluorescent DNAs printed on paper: Sensing food spoilage and ripening in the vapor phase, Chem. Sci., 2012, 3(8), 2542–2549.
7. K. J. Albert, N. S. Lewis, C. L. Schauer, G. A. Sotzing, S. E. Stitzel, T. P. Vaid and D. R. Walt, Cross-reactive chemical sensor arrays, Chem. Rev., 2000, 100(7), 2595–2626.
8. J. R. Askim, M. Mahmoudi and K. S. Suslick, Optical sensor arrays for chemical sensing: the optoelectronic nose, Chem. Soc. Rev., 2013, 42(22), 8649–8682.
9. J. R. Stetter and W. R. Penrose, Understanding Chemical Sensors and Chemical Sensor Arrays (Electronic Noses): Past, Present, and Future, Sens. Update, 2002, 10(1), 189.
10. N. C. Speller, N. Siraj, B. P. Regmi, H. Marzoughi, C. Neal and I. M. Warner, Rational design of QCM-D virtual sensor arrays based on film thickness, viscoelasticity, and harmonics for vapor discrimination, Anal. Chem., 2015, 87(10), 5156–5166.
11. X. Jin, L. Yu, D. Garcia, R. X. Ren and X. Zeng, Ionic liquid high-temperature gas sensor array, Anal. Chem., 2006, 78(19), 6980–6989.
12. A. Rehman, A. Hamilton, A. Chung, G. A. Baker, Z. Wang and X. Zeng, Differential solute gas response in ionic-liquid-based QCM arrays: elucidating design factors responsible for discriminative explosive gas sensing, Anal. Chem., 2011, 83(20), 7823–7833.
13. X. Xu, C. Li, K. Pei, K. Zhao, Z. K. Zhao and H. Li, Ionic liquids used as QCM coating materials for the detection of alcohols, Sens. Actuators, Bl, 2008, 134(1), 258–265.
14. X. Jin, Y. Huang, A. Mason and X. Zeng, Multichannel monolithic quartz crystal microbalance gas sensor array, Anal. Chem., 2008, 81(2), 595–603.
15. X. Xu, H. Cang, C. Li, Z. K. Zhao and H. Li, Quartz crystal microbalance sensor array for the detection of volatile organic compounds, Talanta, 2009, 78(3), 711–716.
16. C. Liang, C. Y. Yuan, R. J. Warmack, C. E. Barnes and S. Dai, Ionic liquids: a new class of sensing materials for detection of organic vapors based on the use of a quartz crystal microbalance, Anal. Chem., 2002, 74(9), 2172–2176.
17. B. P. Regmi, N. C. Speller, M. J. Anderson, J. O. Brutus, Y. Merid, S. Das, B. El-Zahab, D. J. Hayes, K. K. Murray and I. M. Warner, Molecular weight sensing properties of ionic liquid-polymer composite films: theory and experiment, J. Mater. Chem. C, 2014, 2(24), 4867–4878.
18. W. Makino, R. Kishikawa, M. Mizoshiri, S. Takeda and M. Yao, Viscoelastic properties of room temperature ionic liquids, J. Chem. Phys., 2008, 129(10), 104510.
19. T. Yamaguchi, S. Miyake and S. Koda, Shear Relaxation of Imidazolium-Based Room-Temperature Ionic Liquids, J. Phys. Chem. B, 2010, 114(24), 8126–8133.
20. Flavours and fragrances: chemistry, bioprocessing and sustainability, ed. R. G. Berger, Springer Science & Business Media, 2007.
21. T. Nakamoto, A. Fukuda and T. Moriizumi, Perfume and flavour identification by odour-sensing system using quartz-resonator sensor array and neural-network pattern recognition, Sens. Actuators, B, 1993, 10(2), 85–90.
22. T. Nakamoto, Y. Nakahira, H. Hiramatsu and T. Moriizumi, Odor recorder using active odor sensing system, Sens. Actuators, B, 2001, 76(1), 465–469.
23. B. D. Vogt, E. K. Lin, W. L. Wu and C. C. White, Effect of film thickness on the validity of the Sauerbrey equation for hydrated polyelectrolyte films, J. Phys. Chem. B, 2004, 108(34), 12685–12690.
24. G. McHale, R. Lücklum, M. I. Newton and J. A. Cowen, Influence of viscoelasticity and interfacial slip on acoustic wave sensors, J. Appl. Phys., 2000, 88(12), 7304–7312.
25. F. Hasan, P. Vidanapathirana, S. Das, V. E. Fernand, N. Siraj, J. N. Losso and I. M. Warner, Ionic liquids as buffer additives in ionic liquid-polyacrylamide gel electrophoresis separation of mixtures of low and high molecular weight proteins, RSC Adv., 2015, 5(85), 69229–69237.

---

Footnote

† Electronic supplementary information (ESI) available. See DOI: 10.1039/c6ra16988k

---