Cross-reactive, self-encoded polymer film arrays for sensor applications

Abstract

The development of chemical sensors continues to be an active area of research, especially for the development of a practical electronic nose. Here, we present a spectroscopic chemical sensor based on an array of 64 self-encoded polymer films deposited on a microfabricated silicon substrate. The polymer arrays were analyzed by FTIR and Raman spectroscopy before and after exposure to a series of organic volatiles to monitor changes in their vibrational fingerprints. We show here that the spectroscopic changes of self-encoded polymer films can be used to distinguish between volatile organic analytes. Changes induced in the sensor arrays by the analyte vapor were denoted by a spectroscopic response of the self-encoded polymer sensors and transformed into a response pattern by multivariate data analysis using partial least squares regression. The results indicated that the polymer sensors provide a unique and reproducible pattern for each analyte vapor and have the potential to be used in the fabrication of a novel electronic nose device.

---

Introduction

Biomimetic engineering is the application of biological principles to the design of artificial devices or systems.^1–4 For many years, scientists and engineers have recognized the power of naturally occurring systems and their ability to guide technological development. One example of this approach is a device known as the artificial or electronic nose (e-Nose).^5,6 Gardner and Bartlett define an e-Nose as “an instrument which comprises an array of electronic chemical sensors with partial specificity and an appropriate pattern recognition system, capable of recognizing simple or complex odors.”⁵ Applications of e-Noses include food quality control,^7–9 pollution monitoring,¹⁰ medical diagnosis,^11–22 and landmine detection,²³ among others. In this rapidly growing field, arrays of semi-selective chemical sensors are the main component of the devices. Coupled to pattern recognition software, these arrays can parallel the biological olfactory system in which semi-selective olfactory receptors are combined with higher order neural processing.⁶ It is believed that the mammalian olfactory epithelium comprises millions of receptor cells expressing ca. 1000 different receptor types. While none of the receptors are thought to be particularly sensitive to one specific analyte, it is likely that their combined interaction with an analyte leads to the highly sensitive and discriminative sense of smell.⁶ Similarly, in e-Nose system architecture, no individual detector is highly selective toward an individual analyte, as would be the case in the traditional “lock and key” approach to chemical sensing. Instead, each detector responds to many analytes with varying degrees of intensity, creating a unique response pattern for each analyte, a principle known as cross-reactivity.⁶ The resulting odor signature from the array is used to classify, and in some cases quantify, the target analytes.^6,24–31

e-Noses are primarily classified by the sensor-transduction mechanism.²⁹ e-Nose transducers reported to date include semiconducting metal oxide,³² conducting polymer films,^24,33 acoustic wave devices,^34–37 electrochemical systems,^38–40 carbon-black loaded polymer film chemoresistors,^20,25,41 and optical transducers with immobilized dyes such as Nile red^30,42–44 or various metalloporphyrins.^7,45,46 Among these approaches, most array sensors employing polymers have produced high selectivity, taking advantage of properties such as polarity, swelling, conductivity, and sorption.⁴⁷ Regardless of the transduction mechanism of the sensor array, a larger number of unique sensors increases the amount of data gathered, thus producing a more complex and specific pattern for improved analyte identification and classification.^47,48

The research and development of sensor arrays often involves the measurement and analysis of a large number of samples.^49–53 This can be laborious and time-consuming because there are many variables that influence the performance, sensitivity, and precision of the sensors. Therefore, there is currently significant interest in the development of sensor arrays with high-throughput screening capability, as well as high reproducibility, selectivity, and sensitivity for applications in disease diagnosis.^11–19,21

Our group has recently reported on a new class of resins prepared from spectroscopically active styrene monomers, the combination of which produces polymers wherein unique vibrational fingerprints are associated with each polymer.^{49,51,54–59} The spectrum from each polymer can then be converted into a barcode in which the position of each bar matches a peak wavenumber in the spectrum. We have previously demonstrated that large numbers of barcoded resins, (BCRs) can be selectively recognized and classified with full confidence by virtue of their unique vibrational fingerprint resulting from their unique chemical composition.^57,58

In this contribution, we describe both the fabrication of polymer film sensor arrays using BCRs, and the testing of the arrays for the differentiation of volatile organic compounds (VOCs). Polymer sensor arrays were fabricated by depositing an assortment of BCRs into micromachined wells on silicon substrates. The analysis of the vibrational spectroscopic changes by partial least square regression (PLS)^57,60 efficiently captures the variances in the data set and allows discrimination of the VOCs. Contour plots and histogram graphs representing the changes in the composition matrix for each polymer sensor in the array indicated that the array responses to VOCs were sensitive and reproducible.

Experimental

Materials

Sixty four polymers were obtained from a combination of seven styrene monomers (Fig. 1) using a previously reported synthetic strategy.^55–57 The polymer composition and spectra are presented in the ESI.† Each polymer has a unique composition and vibrational fingerprint, which can be easily identified by IR or Raman spectroscopy.

Fig. 1 Structures of seven styrene monomers used for the synthesis of the BCRs.

Five volatile organic analytes (methanol, hexane, methylamine, acetonitrile, and ethylacetate) covering a broad spectrum of physical properties and functional groups (amine, alcohol, nitrile, ester, alkane) were selected to test the effectiveness of the sensor arrays as an e-Nose. We hypothesized that the sensors provide sufficient information (spectral variation) upon exposure to an analyte. Analytical-grade reagents were purchased from Aldrich and used without further purification.

Substrate fabrication

The patterned substrates were fabricated on optical grade silicon (100) wafers (diameter: 3.937′′; thickness: 0.020′′; purity: >99.999%; Lattice Materials Corp., Bozeman, Montana) by a combination of optical lithography and reactive ion etching (RIE).⁶¹ The procedures for the RIE experiment are presented in Fig. 2A. The silicon substrates were piranha-cleaned, HMDS-primed and spin-coated with AZ P4620 photoresists. Areas of the silicon were exposed by irradiating the film (λ = 400 nm) through a lithography mask (built in-house). RIE using a combination of SF₆/C₂F₄, at processing pressures of 125 and 75 scm respectively, was used to etch the exposed silicon on the wafer. In this way, 200 μm-deep wells were obtained. Fig. 2B shows the photograph of the ordered microwells (8 × 8 wells, diameter for each well: ca. 2.5 mm) fabricated using this method. The patterned substrates for Raman spectroscopy were coated with a thin film of copper before polymer sensor array deposition.

Fig. 2 Stepwise procedure for the reactive ion etching (RIE) experiment (A) and photograph (B) of the patterned silicon wafer after RIE.

Polymer sensor array fabrication

Each of the 64 polymers was dissolved in chloroform (ca. 10% W/W). One drop of the homogeneous solution was dispersed into a well on the patterned substrate using a capillary tube. The procedure was repeated for each polymer in the library producing an 8 × 8 array. The coated substrate was heated to 50 °C for ca. 2 h on a hot plate to evaporate residual chloroform from the polymer films. The polymer samples in each well were identified by a number (representing a row) and a letter (representing a column), as shown in Fig. 2B.

Instrumentation

An illustration of the experimental setup is shown in Fig. 3. First the polymer sensor array was placed inside a gas flow cell and screened by FTIR and Raman spectroscopy prior to analyte vapor exposure. Saturated analyte vapor, generated by slowly bubbling nitrogen through a vial containing the liquid of interest, was then passed through the gas flow cell containing the sensor array.⁶² After analyte vapor exposure, the sensing system was allowed to reach equilibrium before spectrum or image acquisition. The system was determined to be at equilibrium when no further changes were recorded in the spectra. Prior to the acquisition of the FTIR spectra specifically, several preprocessing steps were taken. To account for the atmospheric vapor (in air), a background scan was recorded with no VOC and subsequently subtracted from the array spectra after exposure to the VOCs.

Fig. 3 Schematic diagram of analyte delivery setup and analysis system for recording sensor array response to a volatile analyte. Exposure to an analyte vapor induces changes in the FTIR and Raman spectra that are recorded and compared to the unexposed sensor array.

FTIR images were acquired using a Varian Stingray imaging spectrometer, which consists of a UMA 600 microscope coupled to an FTS 7000 rapid scan interferometer. The imaging detector has a focal plane array (FPA) of 64 × 64 mercury cadmium telluride (MCT) elements imaging a spatial area of 352 × 352 μm². One dimension of the image is 350 μm and there are 64 pixels in the same dimension, thus the spatial resolution is ca. 6 μm. The FTIR spectra were recorded from 900–3500 cm⁻¹. The acquisition time of 16 scans with 8 cm⁻¹ spectral resolution for an image was ca. 20 s. Transmission and mosaic mode (8 × 8) were selected to screen the ordered polymer sensor arrays on patterned optical grade silicon.⁶³ The polymer film arrays were placed on the FTIR stage sample holder and imaged automatically and rapidly using a motor-driven x–y stage attachment.

Raman spectra were obtained using a dispersive Almega® visible Raman spectrometer (Thermo Nicolet). A frequency doubled Nd:YY04 DPSS green laser (532 nm) with a beam diameter of 1.9 mm was used as an excitation source for the sample. The experiments were performed in the microscope compartment of the spectrometer under a 10× objective giving a spatial resolution of 14.0 μm. High-resolution gratings were used to give a spectral resolution of 2 cm⁻¹. The spectra were detected on a charge coupled device (CCD) detector. Atlμs® for Almega® software (version 7.1) was used for spectrum processing and for controlling the x–y motorized stage within the microscope compartment. High-resolution spectra of the polymer films were collected using the 532 nm laser line. The spectra were recorded from 448–1598 cm⁻¹. The Raman imaging was performed using the mapping function incorporated in the Atlμs® software. The step size (distance between two Raman spectra) was selected to maintain the same center-to-center distance between adjacent microwells and to complete mapping of the sensor arrays within a reasonable time frame (pixel resolution of 20–50 μm).

Data processing

The preprocessing of the vibrational spectra, extracted from FTIR images and Raman maps, involve background subtraction (FTIR), baseline correction, and normalization to compensate for variations in spectral intensity. This procedure was performed for each spectrum before and after volatile analyte exposure. All spectra were imported into Microsoft Excel in ASCII format to form spectroscopy matrices. The matrices were then imported into Unscrambler® (Camo) software. The weight percent of all seven monomers for each polymer (the reference composition matrix) was added to the file. A partial least squares regression (PLS2) was performed using the Unscrambler® software to correlate the spectral matrix (X) to the reference composition matrix (Y). The PLS2 regression was validated by calculating the prediction composition matrix (Y′) for the polymer array and comparing it to the reference composition matrix (Y). As expected, the model was representative of the spectral data and Y′ was found to be similar to Y. The similarity was represented in a contour plot that compares the angle between the predicted and reference composition vectors (Fig. 5).

Because the Y matrix is a collection of compositional vectors, each polymer had a unique reference composition vector in seven dimensional space, y_i, one dimension for each monomer used in polymer synthesis. The dimensions of the Y matrix are thus 64 rows by 7 columns. The Y′ matrix also consisted of a collection of 7 dimensional vectors, y′_i, and had a dimension of 64 × 7. The angle between every y_i and y′_i was determined by the inner product of the two composition vectors:^60,64,65

This gave 4096 (64 × 64) pairs of vectors or angles. From eqn (1), it can be seen that if y_i = y′_i then cos θ = 1, thus θ = 0. This shows that the closer the predicted vectors are to the reference composition vectors, the smaller the angle between y_i and y′_i. Moreover, by comparing the predicted composition vector for each polymer in the array before (reference) and after (response) the interaction with the analyte vapor, a response pattern could be generated for each analyte in the set. Therefore, the theta (θ) value (obtained from eqn (1)) would reflect the degree of change in the vibrational spectrum of each polymer sensor resulting from exposure to the analyte vapor. The resulting θ values for each sensor in the array are presented as histogram graphs in Fig. 7. Contour plots of all 4096 θ values between sensors are presented in Fig. S3.†

Results and discussion

In this study, polymer film arrays where fabricated by depositing polymers on patterned, etched silicon wafers (Fig. 2B). The support itself provides excellent resistance to aggressive organic solvents and allows the deposition and preparation of BCR polymer films for FTIR and Raman analysis. Because of its excellent stability in organic solvents, the support can be thoroughly cleaned and reused multiple times. The 64 BCR polymer samples in the microwells can be distinguished from one another through their unique Raman and FTIR spectra. Because each polymer sample's vibrational spectrum is also its encoding fingerprint, we refer to these sensor arrays as self-encoded polymer film sensor arrays.

Sensor array preparation and FTIR imaging

Preparation of the sensors results in an uneven polymer film surface topography, which in turn leads to slight variations in peak intensity (i.e. slight differences in color scale for each pixel of the focal plane array, Fig. 4). However, film homogeneity is not necessary for vapor analysis, as the sensing method relies on changes in peak intensity and position as a result of the sensor's interaction with the analyte.

Fig. 4 Representative FTIR images of the polymer sensor arrays before (left) and after (right) exposure to hexane vapours recorded at 3024 cm⁻¹, corresponding to C–H stretching mode. Each individual polymer image is 352 × 352 μm².

The wavenumber, intensity, and shape of the FTIR spectrum depend directly on the physical and chemical nature of the polymer and its microenvironment.^66–68 Analyte vapors alter the polymer film microenvironment upon film-analyte exposure due to analyte molecule adsorption, polymer film swelling, or polymer film density change as a result of analyte diffusion into the films. Thus our hypothesis is that FTIR imaging could offer a convenient way to observe the response of a spectroscopically-encoded polymer sensor array to an analyte vapor.

Representative FTIR images of the polymer film arrays (recorded at 3024 cm⁻¹ corresponding to C–H stretching mode) before and after exposure to hexane vapor are shown in Fig. 4 and S2 (ESI†). Comparison of the FTIR images before and after hexane vapor exposure shows that some polymers (e.g. A7, B5, B8, C5 and D7) undergo significant chemometric changes while others (e.g. A6, C8 and D5) underwent minor changes. A complete set of FTIR images of the sensor arrays response to hexane vapor is shown in Fig. S2 (ESI†).

Sensor array response to various analytes

Both FTIR and Raman spectroscopy provide information about the fundamental vibrational modes of a polymer sample and can be used to quantify a multi-component system. Subtle changes in the polymer environment caused by an analyte, such as polarisability and dipole moment, are reflected in the polymer's FTIR and Raman spectra. Using multivariate data analysis (PLS2), the information within vast amounts of spectroscopic data can be extracted, analyzed and correlated to polymer composition.^69–72 For this study, we have shown that each polymer composition (represented as a Y matrix constructed from each polymer composition) can be accurately correlated to its respective FTIR or Raman spectrum (X matrix).⁵⁷ Unscrambler software allows us to perform this correlation by comparing both the position and relative intensity of the spectral information with the monomer mass ratios comprising each composition vector (i.e. polymer). Thus, a library of 64 uniquely composed polymers was converted into a spectral matrix, in which each polymer was assigned a 7-dimensional composition vector. After a prediction calculation was performed, an excellent agreement was found between the predicted and reference composition vector, thus validating the regression model (Fig. 5).

Fig. 5 FTIR (left) and Raman (right) angle maps for the reference, and prediction component vectors obtained from multivariate data analysis. Theta values of a polymer change before and after exposure to an analyte. Notably, comparison of identical polymers (theta values along the diagonal) is diagnostic of the sensor array's response to the analyte: a blue continuous diagonal suggests no effect whereas a discontinuous diagonal suggests an interaction of the analyte with specific elements of the sensor array.

To determine the response of the polymer sensor array to VOCs, we postulated that the analyte vapor would absorb into the polymer matrix and thus alter the polymer's microenvironment. Fig. 6 shows the FTIR and Raman spectra of methanol and polymer C2 before and after exposure to methanol vapor. The difference spectra – before versus after exposure to methanol vapor – show that the analyte spectrum does not appear in the polymer spectra. Thus, the changes in the spectra can be attributed to subtle changes in the polymer vibrational modes in response to methanol rather than additional vibrations unique to methanol. Because the PLS2 approach correlates small changes in the polymer spectrum with its composition, any alteration in the spectrum upon exposure to an analyte would alter this correlation whereas correlating a polymer spectrum exposed to a blank with its composition would establish potential changes due to experimental errors or data processing.

Fig. 6 FTIR (left) and Raman (right) of polymer C2 before and after exposure to methanol vapor. Methanol spectra (black traces) and difference spectra (blue traces) between “before” (red traces) and “after” (green traces) exposure to methanol vapor are shown.

Here we establish that changes in the composition vectors of the 64-polymer sensor array are indeed caused by exposure to analyte vapors. Because the intensity and wavenumber of the spectral peaks change slightly due to exposure to the analyte vapor, we considered the output of composition vectors before analyte vapor exposure as the “reference” and the output of composition vectors after analyte vapor exposure as the “prediction” vectors. From the inner product of the reference and prediction vectors for each polymer, the θ value between each pair of vectors was calculated (Fig. 7). For example, the reference and prediction composition vectors for spectral element (obtained from FTIR spectroscopy for the methanol treatment) of sensor A4 were (41.921, 5.222, −0.543, 43.801, −0.554, −1.332, 11.485) and (36.15, −0.275, 0.830, 49.843, 0.747, 1.675, 11.039) respectively, equivalent to a θ value of 9.77°. This indicated that the composition vector for polymer A4 has changed ca. 10° with respect to the initial composition vector. Using this type of analysis, the responses of the polymer sensor array to the analyte could be converted into an angle change “fingerprint,” when considering the entire array response to a particular analyte. The higher the θ value of a polymer, the larger its response is to a particular analyte. The FTIR images and Raman/FTIR histograms for each of the 5 analyte exposures are unique; this indicates that the polymer sensor array has a unique response to each particular analyte (analyte differentiation). Table 1 shows the average diagonal angle between the “reference” and “prediction” obtained for FTIR and Raman for the five analytes investigated. The results suggest that the θ values depend not only on the type of analyte used but also on the spectroscopic method. Thus, the combined use of Raman and FTIR would allow for a more accurate identification of the target analytes.

Fig. 7 Histogram of the individual FTIR (left) and Raman (right) responses (angle value) of the 64 polymer sensors to (A) hexane; (B) methylamine; (C) acetonitrile; (D) methanol; (E) ethylacetate.

Table 1 Average diagonal θ angle (°) for five different analyte vapor treatments obtained by FTIR and Raman spectroscopy

Analyte	FTIR	Raman
Hexane	12.344	11.904
Methanol	7.800	10.554
Ethylacetate	10.860	8.948
Acetonitrile	13.163	9.556

---

Response specificity of the polymer sensor arrays to the analyte

Central to the success of this methodology in polymer sensor arrays is the specificity of the sensor response to each analyte. In Fig. 7, the θ values for each sensor in the array are displayed as 3D histograms, with the height representing θ. Clearly, each polymer sensor in the array has a unique response (θ) to each analyte, producing a cross-reactive response (fingerprint) representing a single analyte, which can then be used for analyte differentiation. The combination of both histograms and angle maps (Fig. S3†) improves analyte classification. A particularly useful similarity between the natural mammalian olfactory system and the artificial device presented here is that each system employs the simultaneous response of multiple sensors to create a global response pattern for odorant identification. By considering the global response pattern of a large number of individual sensors, we have been able to produce a sensor array with a high degree of selectivity. Similar to scent memory in the mammalian olfactory system, the resulting patterns obtained from the polymer sensor arrays can be stored and later used as reference to identify a specific analyte. This training procedure is a crucial aspect of the sensor array design.

Reproducibility of the response of polymer sensor arrays to analytes

Both biological and artificial chemical sensors have a finite lifetime. In many vertebrate species, for example, olfactory receptor cells are thought to survive for a period of 4–8 weeks before being replaced by new cells.⁷³ Similarly, with the present sensor arrays, replacement will ultimately be required because of polymer degradation or laser exposure over a long period of time. Thus, the ability to reproduce identical results is critical to the success of an e-Nose device.⁷⁴

To test the reproducibility of the sensors, we fabricated several arrays and tested their responses to the same set of analytes. The resulting contour plots and histogram graphs confirmed a high degree of reproducibility of the response (θ value) for the sensor arrays after each analyte vapor exposure, thus indicating that our sensors can be fabricated reproducibly. The average θ values for each sensor are displayed in Fig. 8.

Fig. 8 Reproducibility of the sensor arrays response to different analytes, obtained by FTIR (left) and Raman (right) observation spectroscopies.

Conclusions

Self-encoded polymers from spectroscopically active styrene monomers were used to prepare polymer film sensor arrays and were analyzed using high-throughput FTIR and Raman microspectroscsopy. The arrays were exposed to a set of 5 test vapor analytes to determine the ability of each sensor array to differentiaet them. FTIR images for each polymer sensor showed unique responses to each analyte vapor. Multivariate data analysis was used to process the FTIR and Raman spectra before and after exposure to analyte vapors. The inner product of the “reference” (before analyte vapor exposure) and “prediction” (after analyte vapor exposure) vectors from the multivariate data analysis was converted into an angle, which could be used to quantify the response of the polymer sensor arrays. In this work, we presented three approaches to characterize sensor vapor response: (a) FTIR imaging to reflect a change at a given wavenumber (Fig. 4 and S2†), (b) analysis of the changes in full FTIR and Raman spectra and quantification of sensor response through multivariate data analysis (Fig. 5 and S3†), and angle change calculation (Fig. S7†). The results obtained after data analyses demonstrated that the sensor arrays have unique and reproducible response patterns for each analyte (especially for the FTIR data). Thus, we propose that our sensor arrays can be used as an e-Nose to detect a broad range of volatile analytes after an analyte recognition training process.

This approach can be extended for the high-throughput screening and identification of a wide range of analytes. The simple and low-cost fabrication of the present sensor arrays coupled with excellent reproducibility, ensures a facile application in biodiagnostics.

Work to expand the repertoire of analytes and BCR sensors to gain a better handle on the scope and versatility of this system, to test the ability of the sensor arrays in quantitative measurements of mixture components, and to miniaturize the sensor device, are currently underway in our laboratories.

Conflicts of interest

The authors declare no competing financial interest.

Acknowledgements

We thank Northeastern University, the National Research Council of Canada, and the University of Alberta for supporting this work.

Notes and references

1. T. A. Dickinson, J. White, J. S. Kauer and D. R. Walt, Nature, 1996, 382, 697–700.
2. Y.-S. Sohn, A. Goodey, E. V. Anslyn, J. T. McDevitt, J. B. Shear and D. P. Neikirk, Biosens. Bioelectron., 2005, 21, 303–312.
3. J. J. Lavigne, S. Savoy, M. B. Clevenger, J. E. Ritchie, B. McDoniel, S.-J. Yoo, E. V. Anslyn, J. T. McDevitt, J. B. Shear and D. Neikirk, J. Am. Chem. Soc., 1998, 120, 6429–6430.
4. I. Lundström, R. Erlandsson, U. Frykman, E. Hedborg, A. Spetz, H. Sundgren, S. Welin and F. Winquist, Nature, 1991, 352, 47–50.
5. J. W. Gardner and P. N. Bartlett, Sens. Actuators, B, 1994, 18, 210–211.
6. D. R. Walt, S. E. Stitzel and M. J. Aernecke, Am. Sci., 2012, 100, 38–45.
7. C. Di Natale, R. Paolesse, A. Macagnano, A. Mantini, A. D'Amico, A. Legin, L. Lvova, A. Rudnitskaya and Y. Vlasov, Sens. Actuators, B, 2000, 64, 15–21.
8. C. Zhang and K. S. Suslick, J. Agric. Food Chem., 2007, 55, 237–242.
9. A. Jonsson, F. Winquist, J. Schnürer, H. Sundgren and I. Lundström, Int. J. Food Microbiol., 1997, 35, 187–193.
10. S. De Vito, E. Massera, M. Piga, L. Martinotto and G. Di Francia, Sens. Actuators, B, 2008, 129, 750–757.
11. A. D'Amico, G. Pennazza, M. Santonico, E. Martinelli, C. Roscioni, G. Galluccio, R. Paolesse and C. Di Natale, Lung Cancer, 2010, 68, 170–176.
12. M. Phillips, K. Gleeson, J. M. B. Hughes, J. Greenberg, R. N. Cataneo, L. Baker and W. P. McVay, Lancet, 1999, 353, 1930–1933.
13. M. Shirasu and K. Touhara, J. Biochem., 2011, 150, 257–266.
14. H. P. Chan, C. Lewis and P. S. Thomas, Lung Cancer, 2009, 63, 164–168.
15. A. P. Turner and N. Magan, Nat. Rev. Microbiol., 2004, 2, 161–166.
16. W. Miekisch, J. K. Schubert and G. F. E. Noeldge-Schomburg, Clin. Chim. Acta, 2004, 347, 25–39.
17. M. Phillips, G. A. Erickson, M. Sabas, J. P. Smith and J. Greenberg, J. Clin. Pathol., 1995, 48, 466–469.
18. A. Mazzatenta, C. Di Giulio and M. Pokorski, Respir. Physiol. Neurobiol., 2013, 187, 128–134.
19. U. Tisch, M. Nassar, N. Axelrod, F. Azar, A. Marmur, J. Aharon-peretz and H. Haick, Nanomedicine, 2013, 8, 43–56.
20. S. Dragonieri, R. Schot, B. J. A. Mertens, S. Le Cessie, S. A. Gauw, A. Spanevello, O. Resta, N. P. Willard, T. J. Vink, K. F. Rabe, E. H. Bel and P. J. Sterk, J. Allergy Clin. Immunol., 2007, 120, 856–862.
21. C. Wang, C. Ke, X. Wang, C. Chi, L. Guo, S. Luo, Z. Guo, G. Xu, F. Zhang and E. Li, Anal. Bioanal. Chem., 2014, 406, 4757–4763.
22. R. Ionescu, Y. Broza, H. Shaltieli, D. Sadeh, Y. Zilberman, X. Feng, L. Glass-Marmor, I. Lejbkowicz, K. Müllen, A. Mil ler and H. Haick, ACS Chem. Neurosci., 2011, 2, 687–693.
23. Electronic noses & sensors for the detection of explosives, ed. J. W. Gardner and J. Yinon, Kluwer Academic Publishers, Dordrecht, 2004.
24. M. S. Freund and N. S. Lewis, Proc. Natl. Acad. Sci. U. S. A., 1995, 92, 2652–2656.
25. B. C. Sisk and N. S. Lewis, Langmuir, 2006, 22, 7928–7935.
26. J. W. Gardner, H. W. Shin and E. L. Hines, Sens. Actuators, B, 2000, 70, 19–24.
27. R. M. S. Thorn and J. Greenman, J. Breath Res., 2012, 6, 024001.
28. M. J. Aernecke and D. R. Walt, Sens. Actuators, B, 2009, 142, 464–469.
29. J. R. Askim, M. Mahmoudi and K. S. Suslick, Chem. Soc. Rev., 2013, 42, 8649–8682.
30. S. R. Johnson, J. M. Sutter, H. L. Engelhardt, P. C. Jurs, J. White, J. S. Kauer, T. A. Dickinson and D. R. Walt, Differences, 1997, 69, 4641–4648.
31. R. Glatz and K. Bailey-Hill, Prog. Neurobiol., 2011, 93, 270–296.
32. H. Meixner and U. Lampe, Sens. Actuators, B, 1996, 33, 198–202.
33. C. Wei, L. Dai, A. Roy and T. B. Tolle, J. Am. Chem. Soc., 2006, 128, 1412–1413.
34. S. Shiokawa and J. Kondoh, Jpn. J. Appl. Phys., Part 1, 2004, 43, 2799–2802.
35. R. Gabl, H. D. Feucht, H. Zeininger, G. Eckstein, M. Schreiter, R. Primig, D. Pitzer and W. Wersing, Biosens. Bioelectron., 2004, 19, 615–620.
36. S. M. Chang, B. Ebert, E. Tamiya and I. Karube, Biosens. Bioelectron., 1991, 6, 293–298.
37. G. Wingqvist, V. Yantchev and I. Katardjiev, Sens. Actuators, A, 2008, 148, 88–95.
38. S.-Y. Tian, S.-P. Deng and Z.-X. Chen, Sens. Actuators, B, 2007, 123, 1049–1056.
39. J. Gallardo, S. Alegert and M. Del Valle, Sens. Actuators, B, 2004, 101, 72–80.
40. F. Winquist, P. Wide and I. Lundström, Anal. Chim. Acta, 1997, 357, 21–31.
41. A. V. Shevade, M. A. Ryan, M. L. Homer, A. M. Manfreda, H. Zhou and K. Manatt, Sens. Actuators, B, 2003, 93, 84–91.
42. J. White, J. S. Kauer, T. A. Dickinson and D. R. Walt, Anal. Chem., 1996, 68, 2191–2202.
43. H. H. Gorris, T. M. Blicharz and D. R. Walt, FEBS J., 2007, 274, 5462–5470.
44. J. R. Epstein and D. R. Walt, Chem. Soc. Rev., 2003, 32, 203–214.
45. N. A. Rakow and K. S. Suslick, Nature, 2000, 406, 710–713.
46. N. A. Rakow, A. Sen, M. C. Janzen, J. B. Ponder and K. S. Suslick, Angew. Chem., Int. Ed., 2005, 44, 4528–4532.
47. T. A. Dickinson, D. R. Walt, J. White and J. S. Kauer, Anal. Chem., 1997, 69, 3413–3418.
48. K. J. Albert, N. S. Lewis, C. L. Schauer, G. A. Sotzing, S. E. Stitzel, T. P. Vaid and D. R. Walt, Chem. Rev., 2000, 100, 2595–2626.
49. S. Chun, J. Xu, J. Cheng, L. Ding, N. Winograd and H. Fenniri, J. Comb. Chem., 2006, 8, 18–25.
50. D. R. Blais, R. A. Alvarez-Puebla, J. P. Bravo-Vasquez, H. Fenniri and J. P. Pezacki, Biotechnol. J., 2008, 3, 949–953.
51. H. Fenniri, O. Terreau, S. Chun, S. J. Oh, W. F. Finney and M. D. Morris, J. Comb. Chem., 2006, 8, 192–198.
52. E. H. Oh, H. S. Song and T. H. Park, Enzyme Microb. Technol., 2011, 48, 427–437.
53. A. A. Farah, R. A. Alvarez-Puebla and H. Fenniri, J. Colloid Interface Sci., 2008, 319, 572–576.
54. J. P. Bravo-Vasquez, R. A. Alvarez-Puebla and H. Fenniri, Sens. Actuators, B, 2007, 125, 357–359.
55. H. Fenniri, L. Ding, A. E. Ribbe and Y. Zyrianov, J. Am. Chem. Soc., 2001, 123, 8151–8152.
56. H. Fenniri, S. Chun, L. Ding, Y. Zyrianov and K. Hallenga, J. Am. Chem. Soc., 2003, 125, 10546–10560.
57. H. Fenniri, S. Chun, O. Terreau and J. P. Bravo-Vasquez, J. Comb. Chem., 2008, 10, 31–36.
58. H. Fenniri, H. G. Hedderich, K. S. Haber, J. Achkar and B. Taylor, Angew. Chem., Int. Ed., 2000, 39, 4483–4485.
59. J. Raez, D. R. Biais, Y. Zhang, R. A. Alvarez-Puebla, J. P. Bravo-Vasquez, J. P. Pezacki and H. Fenniri, Langmuir, 2007, 23, 6482–6485.
60. K. H. Esbensen, Multivariate Data Analysis - in Practice, Camo Technologies, Woodbridge, NJ, 5th edn, 2002.
61. S. Grego, T. W. Jarvis, B. R. Stoner and J. S. Lewis, Langmuir, 2005, 21, 4971–4975.
62. S. M. Drew, D. E. Janzen, C. E. Buss, D. I. MacEwan, K. M. Dublin and K. R. Mann, J. Am. Chem. Soc., 2001, 123, 8414–8415.
63. A. Leugers, D. R. Neithamer, L. S. Sun, J. E. Hetzner, S. Hilty, S. Hong, M. Krause and K. Beyerlein, J. Comb. Chem., 2003, 5, 238–244.
64. S. J. Hasweii, Practical Guide to Chemometrics, Marcel Dekker, Inc., New York, 1992.
65. M. J. Adams, Chemometrics in Analytical Spectroscopy, Royal Society of Chemistry, Cambridge, 1995.
66. K. L. A. Chan and S. G. Kazarian, J. Comb. Chem., 2006, 8, 26–31.
67. R. A. Potyrailo, W. G. Morris, R. J. Wroczynski and P. J. McCloskey, J. Comb. Chem., 2004, 6, 869–873.
68. B. A. Miller-Chou and J. L. Koenig, Macromolecules, 2002, 35, 440–445.
69. T. Hasegawa, TrAC, Trends Anal. Chem., 2001, 20, 53–64.
70. J. F. Aust, K. S. Booksh, C. M. Stellman, R. S. Parnas and M. L. Myrick, Appl. Spectrosc., 1997, 51, 247–252.
71. R. A. Ozzetti, A. P. De Oliveira Filho, U. Schuchardt and D. Mandelli, J. Appl. Polym. Sci., 2002, 85, 734–745.
72. L. Rintoul, H. Panayiotou, S. Kokot, G. George, G. Cash, R. Frost, T. Bui and P. Fredericks, Analyst, 1998, 123, 571–577.
73. R. M. Costanzo and P. P. C. Graziadei, in Neurobiology of taste and smell, Wiley, New York, NY, 1987.
74. T. A. Dickinson, K. L. Michael, J. S. Kauer and D. R. Walt, Anal. Chem., 1999, 71, 2192–2198.

---

Footnotes

† Electronic supplementary information (ESI) available: Composition of each copolymer used in this paper (S1); additional FTIR images (S2). See DOI: 10.1039/c6ra13874h

‡ Current address: Professor Jintao Zhu, School of Chemistry and Chemical Engineering, Huazhong University of Science and Technology, 1037 Luoyu Road, Wuhan, China, E-mail: jtzhu@mail.hust.edu.cn

§ Current address: Dr Juan-Pablo Bravo-Vasquez, Pureleau Inc., 101-9865 West Saanich Road, North Saanich, British Columbia V8L 5Y8, Canada, E-mail: juanb@pureleau.ca

---