This work explores an alternative vapor sensing mechanism through analyzing dynamic desorption process from butterfly wings for the differentiation of both individual and mixed vapors quantitatively. Morpho butterfly wings have been used in differentiating individual vapors, but it is challenging to use them for the differentiation of mixed vapor quantitatively. This paper demonstrates the use of Morpho butterfly wings for the sensitive and selective detection of closely related vapors in mixtures. Principal components analysis (PCA) is used to process the reflectance spectra of the wing scales during dynamic desorption of different vapors. With the desorption-based detection mechanism, individual vapors with different concentrations and mixed vapors with different mixing ratios can be differentiated using the butterfly wing based sensors. Both the original butterfly wings and butterfly wings with surface modification show the capability in distinguishing vapors in mixtures, which may offer a guideline for further improving selectivity and sensitivity of bioinspired sensors.
Selective vapor sensing is widely used in the areas of biomedical , environmental monitoring , and industrial production . Both individual sensors and sensor arrays are used in vapor sensing. Individual sensors, however, have poor selectivity and are prone to be poisoned by interference, such as water vapor in air , . Sensor arrays combining multiply individual sensors are demonstrated to have improved selectivity, but the complex array structures also lead to challenges including uncorrelated drift from each sensor in the array and additive noises , . In addition, it is difficult for both the individual sensors and sensor arrays to differentiate mixed vapors quantitatively , . To improve the detection capability towards mixed vapors, bioinspired approaches have attracted increased attention recently , , . Among various bioinspired approaches, Morpho butterfly wing-based sensors have been reported to show both high sensitive and selective response to chemical vapors .
The Morpho butterfly wings with multilayered nanostructures have strong interaction with light and show a sharp reflection peak in the visible spectrum , , which can transfer the stimulation of vapors to output optical signals. Moreover, the nanostructures on the wing scale have a polarity gradient along the ridge . Polar molecules can be selectively absorbed at different positions on the ridges, which provides a diverse optical response. Potyrailo et al.  demonstrated the selective differentiation of individual vapors, not in vapor mixtures however, using butterfly wings. In 2015, the same research group further demonstrated the differentiation of individual vapors in vapor mixtures using a nanofabricated artificial sensor, but the quantifications of each component in the mixtures were not reported .
In this work, we demonstrated an alternative detection approach that is based on the dynamic desorption process of vapor molecules from butterfly wings. Figure 1 shows the nanostructured Morpho butterfly wing with the intrinsic polarity gradient along the ridges. The vapor molecules preferentially adsorbed at different positions of ridges due to the differences in their polarity . Such absorbed vapor molecules will reach an adsorption and desorption equilibrium when the vapors flow at a constant rate. When dry nitrogen (N2) flows through the surface of the butterfly wing that reaches the adsorption and desorption equilibrium, such equilibrium is disrupted and more molecules tend to desorb from the surface of the nanostructures. The desorption of the vapor will lead to the alternation of the reflection spectra of the butterfly wings. Such change provides an opportunity for the differentiation of the desorbed vapors in the mixture. Through the use of principal components analysis (PCA) to exam the dynamic changes of the optical spectra during the vapor desorption processes, rapid and sensitive vapor sensing is demonstrated. Such approach also shows the potential in quantification of the relative components of vapors in the mixtures.
Results and discussion
The morphologies and structures of the Morpho sulkowskyi butterfly are shown in Fig. S1. Figure S1a shows an optical image of the butterfly. The stacked scales are arranged on the surface of the wings (Fig. S1b). The scanning electron microscopy (SEM) image and transmission electron microscopy (TEM) image of the butterfly wing scale show the cross-sectional ridge arrays with multilayered lamella structures (Fig. S1c and d). In the experiment, the butterfly wing was fixed on a silicon substrate and placed inside a chamber. An optical fiber was mounted vertically to receive the visible light reflected by the sample (Fig. 2). Two streams of N2 gas, one served as the carrier gas to deliver different vapors into the container and another one served as the diluting gas, were premixed before passing through the chamber. The reflectance spectra were recorded during the process of vapor desorption. The relative reflectance spectra were calculated using Eq. 1 to represent the change of the reflectance spectra :
where Rr(λ) is the relative reflectance at wavelength λ, R0(λ) is the reflectance spectrum collected from butterfly wings upon exposure to dry N2, and R(λ) is the reflectance spectrum collected from the wing during the process of vapor desorption assisted by dry N2. To analyze these spectra quantitatively, principal components analysis (PCA) was used . PCA is a multivariate statistics analysis method and widely used in classifying multivariate data , , , , . In this study, we used PCA to analysis the N2 assisted dynamic desorption process and achieved both selective detection of different concentrations of single vapor and also the vapor mixtures with varied vapor concentration ratios.
Detection of different concentrations of methanol vapor and ethanol vapor
Figure 3 shows the response of the butterfly wings after exposure to different concentrations of methanol or ethanol. The flow rates of two N2 steams were varied to control the partial pressures of the target vapors. The butterfly wings were exposed to methanol vapor with partial pressures of 0.06 PSM, 0.09 PSM, 0.12 PSM, 0.15 PSM and 0.18 PSM, and also the ethanol vapor with partial pressures of 0.06 PSE, 0.09 PSE, 0.12 PSE, 0.15 PSE and 0.18 PSE. PSM (16.954 kPa) and PSE (7.916 kPa) are the saturation vapor pressures for methanol and ethanol at temperature of 25°C, respectively . When a certain concentration of methanol or ethanol vapor passed through the chamber, the vapor molecules absorbed onto the surface of nanostructures of butterfly wing. After the sample reached to the adsorption-desorption equilibrium, pure dry N2 was introduced to the chamber to break the equilibrium and the vapor molecules desorbed from the nanostructures. Due to the concentration differences of vapors (with different partial pressures), vapor molecules showed different desorption speed, which led to different dynamic responses in the visible reflectance spectra. The spectral change of desorption process was analyzed by PCA as shown in Fig. 3a and c. Each point in the plot corresponds to the reflectance spectrum at the desorption time of 0 min, 1 min, 2 min, 4 min, 6 min, 8 min, and 10 min upon exposure to pure N2. The N2 point in the figure represents the spectra of R0(λ) after the sample was exposed to dry N2 for 20 min. With the vapor molecules desorbed from the wings and replaced by N2 molecules, the curves of methanol and ethanol vapors with different partial pressures all moved towards the N2 point. To compare the similarity of five curves in Fig. 3a for methanol vapor, we calculated the weighted Euclidean distances  between the curves at other four different concentrations and the curve at the concentration of 0.06 PSM by calculating the weighted Euclidean distances between points at the same detection time and then used those distances to calculate the average distance. The weighted Euclidean distance between two corresponding points at the same detection time was calculated using Eq. 2 :
where w1 is the contribution of PC1, 79.1%, w2 is the contribution of PC2, 14.4%. Each curve in this study has seven points, i=1, 2, 3, …, 7. The original PC1 values for each curve are x1, x2, x3,…, and x7 and the original PC2 values for each curve are y1, y2, y3,…, and y7. xi−0.06 and yi−0.06 are the original PC1 and PC2 values of the curve at the concentration of 0.06 PSM. The average weighted Euclidean distance (D) between the curves at other concentrations and the curve at the concentration of 0.06 PSM was calculated using Eq. 3 :
The weighted Euclidean distances between the curves at four other different concentrations and the curve at the concentration of 0.06 PSE in Fig. 3c for ethanol vapors are also calculated by using above process. Figure 3b and d both show an approximate linear relationship between the distance D and the partial pressure of the vapors. Such linear relationship therefore provides the opportunity for the detection of vapors with different concentrations. In addition, we can still detect the nonpolar vapor with different concentrations, such as hexane (Fig. S2a). We used PCA to analysis the combined data of different concentrations of hexane and methanol vapor, and the response of butterfly wings to hexane was smaller than methanol vapor (Fig. S2b). The polarity of molecule plays an important role in the difference of the responses: the polarity of methanol is much larger than that of the hexane, which leads to the larger response for methanol than for hexane.
Discrimination of mixed vapors of methanol and ethanol
To demonstrate the capability of discrimination of mixed vapors with different concentration ratios using the butterfly wing based desorption process, we tested the detection performance using the mixed vapors of methanol and ethanol. We prepared five sample vapors with different mixing ratios. In the experiments, the two streams of N2 were set at constant flow rate. According to Raoult’s law, the partial pressure of each vapor is related to mole fraction of this vapor in the mixture. By changing the concentration of methanol and ethanol in the liquid mixture, we can adjust the specific partial pressures of methanol and ethanol. The partial pressures of five sample vapors used for the study were the pure methanol at 0.06 PSM, pure ethanol at 0.06 PSE, the mixture of methanol at 5.1*10−3 PSM and ethanol at 0.055 PSE, the mixture of methanol at 0.020 PSM and ethanol at 0.040 PSE, and the mixture of methanol at 0.042 PSM and ethanol at 0.018 PSE, respectively, where PSM (16.954 kPa) and PSE (7.916 kPa) are the saturation vapor pressure for methanol and ethanol at the temperature of 25°C . Therefore, the ratios of partial pressure between methanol and ethanol in the three vapor mixtures were close to 1:5, 1:1 and 5:1. When the butterfly wing was exposed to the sample vapor, the vapor molecules with different polarity were selectively absorbed at corresponding positions of ridges due to the intrinsic polarity gradient of nanostructures on the ridges. After the wing sample reached the adsorption and desorption equilibrium, pure N2 gas was introduced into the chamber to induce the vapor desorption from the butterfly wing nanostructures. The dynamic responses in the visible reflectance spectra of desorption process were collected and processed (Fig. 4a). In Fig. 4b, we calculated the distance of the curves using the curve of pure ethanol as the baseline. It is shown that vapors with different ratios of methanol to ethanol are clearly separated with the distance decreased as the ratio of ethanol increased. The curves of desorption for all the samples move toward to the same direction where the wing sample was exposed to dry N2 only. The dynamic desorption responses of the butterfly wings at 555 nm, 506 nm, and 461 nm wavelengths are shown in Fig. S3. For each test, we exposed the butterfly wing to the vapor for 10 min and then dry N2 for 10 min alternately and repeat the cycle for multiple times.
Besides the original butterfly wings, we also studied the detection of vapor mixture using surface-modified butterfly wings. The surfaces of butterfly wings were modified by using oxygen plasma and also octyltrichlorosilane. Wing sample 1 was treated under oxygen plasma for 50 s to obtain a hydrophilic surface. The sample was tested right after the oxygen plasma treatment. Wing sample 2 was treated under the oxygen plasma for 50 s and then exposed to octyltrichlorosilane for 20 min to obtain a hydrophobic surface. Figure 4c and e show the change of the reflectance spectra of sample 1 and sample 2 upon N2 assisted desorption. Figure 4d and f show that both two butterfly wing samples can also differentiate the vapors in mixtures. The value of distance in Fig. 4b, however, is larger than the distance in Fig. 4d and f, which means the difference between the curves in Fig. 4a is larger than those for the curves in Fig. 4c and e. The selectivity of original butterfly is therefore showed a better performance than the plasma treated butterfly wing and also octyltrichlorosilane modified butterfly wing . This result should be attributed to the fact that the reduction of intrinsic polarity gradient of surface modified butterfly wing samples led to the decrease of the selectivity to the vapors.
Discrimination of mixed vapors of 1-propanol and 2-propanol
We further tested the vapor sensing performance for the mixture of two more closely related vapors, the isomeric compounds 1-propanol and 2-propanol, using the same butterfly wing-based desorption analysis. The differences of polarity and refractive index between 1-propanol and 2-propanol are smaller than those between methanol and ethanol . The polarities of methanol, ethanol, 1-propanol and 2-propanol are 55.5 kcal mol−1, 51.9 kcal mol−1, 50.7 kcal mol−1 and 48.6 kcal mol−1, respectively. The refractive indices of methanol, ethanol, 1-propanol and 2-propanol are 1.326, 1.359, 1.384 and 1.375, respectively . The partial pressure of five sample vapors are 1-propanol at 0.10 PS1, 2-propanol at 0.10 PS2, the mixture of 1-propanol at 0.028 PS1 and 2-propanol at 0.072 PS2, the mixture of 1-propanol at 0.067 PS1 and 2-propanol at 0.033 PS2, and the mixture of 1-propanol at 0.090 PS1 and 2-propanol at 0.010 PS2. PS1 and PS2 are the saturation vapor pressures for 1-propanol and 2-propanol at the temperature of 25°C, respectively . PS1 equals to 2.880 kPa and PS2 equals to 5.799 kPa. The mixing ratios between 1-propanol and 2-propanol in the three vapor mixtures were therefore ~1:5, 1:1, and 5:1. The reflectance spectra were collected during molecular desorption assisted by N2 and analyzed by PCA, as shown in Fig. 5a and b. Again, the butterfly wing successfully differentiated the mixtures of 1-propanol and 2-propanol with different mixing ratios.
To better understand the mechanism of vapor desorption process, the reflectance spectra for the original butterfly nanostructures were calculated using finite-difference time-domain (FDTD) methods and analyzed by PCA. The polarity gradient of butterfly ridges runs from the polar top to the less polar bottom . In the modeling, we assumed that the 1-propanol absorbed at the top four layers of ridges and 2-propanol absorbed at the layers below the first four layers, since the polarity of 1-propanol is larger than 2-propanol . The simulated PCA results shown in Fig. 5c and d have similar trend to those in Fig. 5a and b. The plots also moved toward the same direction, at where the sample was exposed to pure N2 gas. Both the experimental results and the simulation results show that the butterfly wings can selectively detect the mixtures of 1-propanol and 2-propanol with different mixing ratios by N2 assisted desorption.
Discrimination of mixed vapors in the presence of water vapor background
To test the selectivity in the present of interference, we tried to differentiate the mixed vapors of methanol and ethanol when the butterfly wing was exposed to the air that contained water vapor. The relative humidity of the air was 40% during the study. As shown in Fig. 6a and b, the trend of the curves is similar to the curves in Fig. 4a and b. The distance also decreased as the ratio of ethanol increased. This result demonstrated the butterfly wing-based desorption process maintained the capability of quantitatively differentiating of methanol and ethanol vapors in mixtures even in the presence of water.
In summary, we demonstrated an alternative vapor sensing approach based on dynamic vapor desorption process assisted by N2 using butterfly wing-based sensors. The iridescent Morpho butterfly scales have surface polarity gradient along ridge structures that can help differentiate both individual and mixed vapors quantitatively. Closely related vapor mixtures, including isomeric vapor mixtures such as 1-propanol and 2-propanol, can also be differentiated using this approach. The simulation results are consistent with the experimental results, which further demonstrates that Morpho butterfly can differentiate mixed vapors quantitatively through dynamic process of molecule desorption. The study therefore offers a promising approach for high-performance optical vapor sensing. The findings in this study also provide potential guideline in engineering man-made high performance vapor sensors.
A 20 W tungsten-halogen light source (Ocean Optics, FL, USA) and a bifurcated optical fiber were used to probe the samples. The reflectance spectra during the process of vapor desorption were recorded by an optical spectrometer (Ocean Optics, FL, USA). The samples were placed inside a flow chamber with the probe of the optical fiber mounted vertically to the surface of butterfly wings. The reflection spectrum of BaSO4 was used as the reference.
The flow rates of different vapors were controlled by mass flow controllers (Alicat Scientific, Inc., AZ, USA). N2 gas with constant flow rate of 500 ml min−1 was split into two streams of gas, one served as the carrier gas to deliver different vapors through bubbling the corresponding liquid, another served as diluting gas. The two streams were premixed before passing through the chamber.
In the simulation, the geometry parameters were obtained based on SEM and TEM characterization. The refractive index of the butterfly wing was set as 1.56+0.06i . We hypothesized that the 1-propanol only absorbed and covered the first four layers of the ridges and 2-propanol absorbed at the layers below the first four layers. Based on the differences of the spectra change for 1-propanol and 2-propanol measured experimentally, we made the following assumption for the thickness of adsorbed 1-propanol and 2-paorpanol on the butterfly wing surface. To the plot of 1-propanol, the thicknesses were set to be 8 nm, 7 nm, 6 nm, 5 nm, 4 nm, 3 nm, and 2 nm. To the plot of 2-propanol, the thicknesses were set to be 4 nm, 3.5 nm, 3 nm, 2.5 nm, 2 nm, 1.5 nm, and 1 nm. For the vapor mixtures, the layer thicknesses were calculated according to the ratios of 1-propanol and 2-propanol.
Principal components analysis
PCA is a multivariate statistics analysis method and widely used in solving multi-variable problems. This orthogonal linear combination method can convert the data to a new coordinate system and all the variance of the data can be projected to a minimum number of dimensions. Here, we use PCA to analyze the relative reflectance spectra. We can obtain an n-dimensional matrix Xm×n from the relative reflectance Rr after mean centering. The covariance matrix Cn×n can be calculated from the equation Cn×n=XTn×mXm×n/(m−1), and the eigenvalues and eigenvectors of Cn×n can also be calculated. The eigenvector matrix of Cn×n, defined as Wn×n, can thus be obtained, which is also the orthogonal transformation matrix of Xm×n. As a result, the k-dimensional principal components matrix are calculated by Tm×k=Xm×nWn×k, where Wn×k is the first k columns of Wn×n, and the columns of it are eigenvectors corresponding to the largest k eigenvalues. In our experiment, the contributions of first and second principles are high enough so that we only need to use them to represent the whole wavelengths. The first and second principal component scores are corresponding to the first and second columns of Tm×k. The variances in the spectra is transferred to the first and second principles (PC1 and PC2). Each point in PCA plot corresponds to a reflectance spectrum. Each plot corresponds to the desorption process of a vapor. Detailed calculation process can be referred to the work of Johnson and Wichern .
A collection of invited papers based on presentations at the 4th International Conference on Bioinspired and Biobased Chemistry & Materials (NICE-2018), Nice, France, 14–17 October 2018.
Funding source: National Natural Science Foundation of China
Award Identifier / Grant number: 51420105009, 51521004
Funding statement: The authors thank the financial support from National Key R&D Program of China (No. 2016YFB0402100), National Natural Science Foundation of China (Funder Id: http://dx.doi.org/10.13039/501100001809, 51420105009, 51521004), and the Zhi-Yuan Endowed fund from Shanghai Jiao Tong University. The authors thank Instrumental Analysis Center of Shanghai Jiao Tong University for access to SEM and TEM.
 A. Lichtenstein, E. Havivi, R. Shacham, E. Hahamy, R. Leibovich, A. Pevzner, V. Krivitsky, G. Davivi, I. Presman, R. Elnathan, Y. Engel, E. Flaxer, F. Patolsky. Nat. Commun. 5, 4195 (2014).10.1038/ncomms5195Search in Google Scholar PubMed
 R. A. Potyrailo, T. A. Starkey, P. Vukusic, H. Ghiradella, M. Vasudev, T. Bunning, R. R. Naik, Z. Tang, M. Larsen, T. Deng, S. Zhong, M. Palacios, J. C. Grande, G. Zorn, G. Goddard, S. Zalubovsky. Proc. Natl. Acad. Sci. USA 110, 15567 (2013).10.1073/pnas.1311196110Search in Google Scholar PubMed PubMed Central
 R. A. Potyrailo, R. K. Bonam, J. G. Hartley, T. A. Starkey, P. Vukusic, M. Vasudev, T. Bunning, R. R. Naik, Z. Tang, M. A. Palacios, M. Larsen, L. A. Le Tarte, J. C. Grande, S. Zhong, T. Deng. Nat. Commun. 6, 7959 (2015).10.1038/ncomms8959Search in Google Scholar PubMed PubMed Central
 J. W. Oh, W. J. Chung, K. Heo, H. E. Jin, B. Y. Lee, E. Wang, C. Zueger, W. Wong, J. Meyer, C. Kim, S. Y. Lee, W. G. Kim, M. Zemla, M. Auer, A. Hexemer, S. W. Lee. Nat. Commun. 5, 3043 (2014).10.1038/ncomms4043Search in Google Scholar PubMed
 G. Peng, U. Tisch, O. Adams, M. Hakim, N. Shehada, Y. Y. Broza, S. Billan, R. Abdah-Bortnyak, A. Kuten, H. Haick. Nat. Nanotechnol. 4, 669 (2009).10.1038/nnano.2009.235Search in Google Scholar PubMed
 D. S. Wilks. Statistical Methods in the Atmospheric Sciences, Elsevier, Amsterdam (2011).Search in Google Scholar
The online version of this article offers supplementary material (https://doi.org/10.1515/pac-2019-0118).
©2020 IUPAC & De Gruyter. This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License. For more information, please visit: http://creativecommons.org/licenses/by-nc-nd/4.0/