Optoelectronic nose based on an origami paper sensor for selective detection of pesticide aerosols

This study introduces an applicable colorimetric sensor array for the detection of pesticides in the vapor phase. The array consisted of six metal nanoparticles spotted on the piece of filter paper. 3D-origami pattern was used for the fabrication of a paper-based sensor to decrease the effect of the nanoparticles leaching after exposure to analytes. Exposure to pesticide aerosols caused changes in the color of the array due to the aggregation of nanoparticles. These changes provided selective responses to thion pesticides such as malathion, parathion, chlorpyrifos, and diazinon. The sensing assay could also differentiate between aliphatic and aromatic thions and discriminate amine-containing compounds from the other studied analytes. These finding results are clearly confirmed by both visual detection and multivariate statistical methods. The proposed sensor was successfully developed for the quantitative measurement of pesticide aerosols at a very low concentration. The limit of detection of this method determined for malathion, parathion, chlorpyrifos and diazinon were 58.0, 103.0, 81.0 and 117.0, respectively. Moreover, the array could be employed to simultaneously analyze four studied pesticides. The statistcal results confirmed that the method has high performance for concurrent detection of thions as a major air pollutant without the interference of other species.


Scientific Reports
| (2020) 10:17302 | https://doi.org/10.1038/s41598-020-74509-8 www.nature.com/scientificreports/ the formation of NPs and their immobilization on the paper substrate. The density of the main element in each NP structure was calculated by EDX analysis; the results are represented in Table S1. The repeatability of the procedure for injecting the NPs on the surface of detection spots was examined by the preparation of five individual sensor arrays through the same process. For each sensor, the RGB values of each sensing element were determined, as presented in Table S2. Five individual numerical values were obtained for each color element (red, green, and blue). The relative standard deviation (RSD) of these data was calculated. As observed in Table S2, the amount of RSD % was lower than 2% for each measurement. Therefore, there was no systematic difference between the color intensities obtained for each sensing element. As a result, a high repeatable process was used to make the proposed Opto-E-nose.
To find the tolerance of the sensor against the ambient humidity, the relative humidity of the test box environment changed from 0 to 100%, and the response of the sensor were monitored. As shown in Fig. S4a, no significant changes were observed in the color of NPs when the humidity increased from 0 to 50%. At higher values, NPs aggregated or washed away from the paper surface.
Moreover, the durability of the sensors was checked by other experiment. Here, several sensors were fabricated and packed into a plastic packet for a certain period. The color of each sensing element was determined weekly. As illustrated in Fig. S4b, the color of sensors did not change for three weeks. By increasing the storage time, NPs aggregated due to changes in the physical conditions of the environment.
Optimized condition. In order to experience an analytical assay with good sensitivity and selectivity, parameters with significant effects on the response of the sensor should be optimized. Since this study aims to discrimination and determination of pesticides, optimization should be individually performed for all sensing elements. Each element of the sensor comprises buffer and NPs; hence, it seems that the amount of NPs, type, concentration and pH of the buffer solution, play key roles in achieving suitable results. The optimal value is obtained by discrimination ability function (DAF) 36 defined as the following equation: The numerator of the fraction explains the between-analyte variations, and the denominator of the fraction introduces the within-analyte variations. In this equation, the symbols indicate the total number of analytes (n), the jth is the determination of ith analyte (X ij ), the mean of five repetitive measurements of i th analyte (X i ) , and the total average of assay responses obtained by all studied analytes ( = X). For each parameter, the experiment with high values of DAF shows the optimal condition. The concentration of each analyte for optimization experiments was 450.0 ng mL −1 .
In order to find the appropriate volume of NPs for preparing sensing elements, 0.5 µL of buffer was mixed with different amounts of NPs in the range of 0.1-0.5 µL. As seen in Fig. S5, the response of DAF was improved by increasing the volume of NPs up to 0.4 µL. The higher volumes negatively affected the DAF responses since the color of the NPs masked the considerable changes due to the interactions. Therefore, each sensing element was made by mixing 0.5 µL of the buffer, 0.4 µL of NPs, and 0.1 µL of deionized water.
The performance of the assay depends on the pH of the environment. Thus, the optimal pH influenced the robustness of electrostatic or H-bonding interactions between analyte and sensors 33 . To investigate the pH effect, the sensor response was separately monitored in the media with certain pH values in the range of 3.0-11.0. As observed in Fig. S6, the sensor showed a high potential for detecting the analytes in the alkaline media with a pH value of 9.0. The unfavorable responses were obtained at other pH values. Probably, at lower pH, the reaction sites of capping agents were blocked with hydronium ions 32 . On the other hand, at higher pH, the interaction of NPs and pesticides was disturbed in the presence of alkaline metal or hydroxyl ions as interfering species 32 . As a result, pH 9.0 was used for further studies.
Later, the experiment was performed in two different types of buffer: borate and Tris. As illustrated in Fig. S7, the desirable results were achieved using the borate buffer. For further studies, NPs were mixed with this buffer to create sensing elements.
To evaluate the ionic strength of the media, the concentration of buffer solution was varied from 0.05 to 0.2 M. As seen in Fig. S8, the appropriate response was observed in the buffer solution with the concentration of 0.1 M. The higher concentrations had an unsuitable effect on the desired interaction, and the response of DAF decreased constantly. Therefore, the buffer concentration was adjusted at 0.1 M for further studies.
The incubation between sensing elements and studied pesticides was investigated in a period. In this duration, the sample was sprayed into the test box, penetrated the paper, and interacted completely with NPs. The time required to perform all these steps for each analyte is shown in Fig. S9. As clarified, the response time of the sensor was 15 min for chlorpyrifos and diazinon and 20 min for malathion and parathion. By increasing the time of interaction, the color of the sensor did not change, and the response of the array was constant. Since pesticides should be determined at the same condition, the image of the sensor was recorded after 20 min.
Colorimetric responses. Figure 1 shows the desired pattern for the sensor array fabrication. The paperbased E-nose was individually exposed to aerosols of four pesticides at the optimal conditions. The responses of the assay to each analyte are shown in Fig. 1d. As can be observed, the color of sensing elements changed after exposure to the analyte due to the NPs aggregation. The presence of a pesticide reduced the electron repulsion or created a bridge between two NPs, leading to a decrease in the distance of NPs 37 . Based on this event, the yellow color of AgNPs turned to orange or brown, and the color of AuNPs changed from red to pale or intense www.nature.com/scientificreports/ purple. As seen, all NPs aggregated in the presence of malathion. This pesticide is an aliphatic compound with a range of active sites, including carbonyl, P=S, thiol groups, besides free paired electrons of oxygen. Thus, this material had a strong H-bonding and nucleophilic interactions with NPs. The small size of this chemical allowed it to participate in the interaction with TA-capped NPs. Parathion pesticide had a high tendency to AuNPs due to the high affinity of gold to sulfur atoms. However, it interacted with only Cys-capped AgNPs. The interaction of parathion and NPs generates a complete resonance in the aromatic ring bonded to nitro groups, forming a stable complex between analyte and NPs. Possibly, electrical repulsion and steric hindrance prevented the interaction of this pesticide with AgNPs synthesized by Tyr and TA, respectively. The other aromatic pesticides, diazinon, changed only the color of NPs prepared by cysteamine. Diazinon with some reaction sites, such as P=S, amino, and oxygen groups, had a part in covalent, nucleophilic, and H-bonding interaction with NPs. The positive charges on the surface of NPs caused the electrostatic interaction between NPs and analytes. The NPs synthesized by TA and Tyr could not show any response to diazinon because of the interference methyl groups in its chemical structure. Possibly, Tyr-capped NPs did not accumulate in the presence of diazinon due to both electrostatic and amine-amine repulsion. This happened for Tyr-AgNPs when exposing to chlorpyrifos. Similar to diazinon, chlorpyrifos resulted in aggregation of Cys-modified NPs. It could also affect the Tyr-AuNPs and change the color of these sensing elements. In the chlorpyrifos, chlorine substituents were replaced with methyl groups, and spatial disruptions decreased; thus, it could penetrate the space of AuNPs and create a complex with them. The aggregation of Cys capped AgNPs and Cys modified AuNPs in the presence of malathion were investigated by FE-SEM and the respective images were represented in Fig. S1h and Fig. S1i. Color profiles (Fig. 1e) indicates difference between the color of sensing elements before and after exposure to analytes. Evidently, the visible observations were exactly in line with the results obtained by the image processing method. The color difference map provided a unique response for each analyte. The useful information about chemical structure and intrinsic features of studied pesticides could be extracted from these patterns. The colorimetric profiles showed the additive details for the colorimetric detection. As clarified, the color of each sensing element became brighter from diazinon to malathion, i.e., the aliphatic pesticides had a strong interaction with NPs compared to heterocyclic compounds. Also, the tendency of the studied pesticides (malathion, parathion, and chlorpyrifos) for reaction with Tyr-modified AuNPs was higher than the other AuNPs.
The colorimetric procedure was repeated in two separate boxes with a length of 25 cm and 50 cm. The distance between spraying and sensing locations increased. As illustrated in Fig. S10, the color intensities of each sensing Discrimination analysis. As mentioned above, the designed PAD can provide unique analytical data for each studied pesticide so that it is possible to differentiate between the thion species. These differences are visible by the naked eye but must be confirmed by some pattern recognition methods. The color difference data were collected in a matrix with a size of 20 × 18 and entered principal component analysis (PCA) or hierarchical clustering analysis (HCA) as input data. The PCA score plot showed the extracted clusters embedded in a numerical data matrix, as in Fig. 2a. This Figure demonstrates that two first PCs included about 95% of total variances of raw data, 86% was distributed on PC 1 and the rest on PC 2. As observed, the thion clusters were well spaced apart and divided into families of aliphatic and aromatic. Further, the aromatic pesticides could be grouped into two individual classes, defined as heterocyclic (diazinon and chlorpyrifos) and non-heterocyclic (parathion). An excellent distinction was obtained between the heterocyclic analyte with a pyridine ring and that with a pyrimidine ring in its structure. The statistical analysis was performed by a full dimensional method of HCA. The dendrograms provided by the HCA analysis (Fig. 2b) indicated the analytes categories in four clusters, belonging to aliphatic, aromatic, and heterocyclic pesticides.
These results were obtained when the concentration of each pesticide was equal to 450.0 ng mL −1 . The qualitative studies of thions were repeated for the other concentrations, including 130.0 ng mL −1 , 170.0 ng mL −1 , 210 ng mL −1 , and 250.0 ng mL −1 . The dendrograms, as shown in Fig. S11, revealed that the analytes were individually separated at lower concentrations while the discrimination of families was achieved at values higher than 210.0 ng mL −1 .

Effect of interferences.
To examine the selectivity of the assay, the PAD was subjected to other materials possibly present in the experiment media. The studied compounds were classified in the different categories, such as oxon organophosphates (paraoxon, dichlorvos, and trichlorfon), carbamate pesticides (carbaryl, pirimicarb, and carbofuran), alcohols (ethanol, methanol, and 1-hexanol), aldehyde (hexanal, heptanal, and benzaldehyde), amines (triethylamine, amylamine, benzylamine, pyridine, aniline, and ammonia), arenes (benzene, toluene, and p-xylene), acids (acetic acid, isobutyric acid, and phosphoric acid), alkanes (heptane, hexane, and heptane), and phosphine (dimethylphenylphosphine). The amount of these materials was similar to analytes concentration (450.0 ng mL −1 ). The colorimetric difference maps are indicated in Fig. S12. As seen, NPs did not aggregate when exposed to these interfering compounds, and no color change was observed. Besides, the studied thions with the concentration of 450.0 ng mL −1 were individually mixed with each foreign material at different www.nature.com/scientificreports/ concentrations. The assay responses to each mixture were determined and compared to that obtained with the only thion aerosols. Probably, The discrimination between thions from the other pesticide compounds is due to the presence of sulfur atoms in the thion structures. This atom has high affinity to not only metal core (gold and silver) but also the fictional groups in the chemical structure of capping agents. As provided in Table S3, all species did not interfere with this measurement; hence, the studied thions could be detected by the proposed assay with high selectivity.
Determination analysis. The efficiency of the proposed assay was examined on different amounts of pesticides with the analyte concentrations varied from 0 to 2.0 µg mL −1 . Fig. S13 indicates the brightened color of each sensing element corresponding with the increment of analyte concentration. At specified concentration, a fingerprint colorimetric difference map was obtained for each thion. Therefore, the assay possible to distinguish the studied pesticides at the desired concentration. For these measurements, the calibration curves are given in Fig. 3 and additive analytical information can be found in Table 1 The responses of all analytes were distributed in a space of three-dimension PCA score plot. The malathion at each concentration was completely discriminated from the other pesticides (Fig. 4). The aromatic compounds with a specified concentration were totally separated from each other. Also, a linear relationship was observed Reproducibility of assay responses. The reproducibility of the developed sensor was investigated by fabricating ten individual sensing array and exposing them to each pesticide with the concentration of 450.0 ng mL −1 . Ten Euclidean norms of the response vector were determined for each analyte. The relative standard error (RSD) of these numerical data was calculated and listed in Table S4. The results show that amount of RSD % is lower than 10%. It means that the assay can detect a certain thion with acceptable reproducibility.

Simultaneous analysis.
In practical, a real sample consists of different pesticides. Therefore, the developed assay need to test for simultaneous determination of diverse analytes. Thus, a set of standard mixtures was prepared using four pesticides with known concentrations. Then, a certain amount of analyte was selected from a credible guideline in each mixture (Table S5). The responses of assay corresponding to each mixture were monitored. The results were collected in a matrix with the size of 20 × 18 cm as a training set which was subjected to partial least square (PLS) as a multivariate statistical method to provide a regression model for each analyte. A validation set containing five mixing solutions of pesticide (Table S5) was used to check the reliability of the PLS model. Several optimal latent variables were detected by leave-one-out cross-validation (LOO-CV) method. The predicted pesticide concentrations obtained by the developed model were compared with corresponding real values by calculating some statistical methods such as root mean square errors (RMSE) and correlation coefficient (R 2 ) ( Table 2). Figure 5 illustrates the relationship between real and predicted concentrations of the analyte. The results show a good correlation coefficient and acceptable error rate for each analyte which prove a high potential to simultaneously analyze the thion pesticides in the mixture of the proposed assay. Finally, the proposed PAD was compared with other E-noses which are fabricated by various sensing elements including enzyme-based sensor, fluorescent probe, commercial gas sensors, and chemical dyes. As illustrated in Table S6, only three works evaluated the discrimination of pesticides in the vapor phase. Among them, only the E-nose prepared by NPs was used for detection of thion pesticides. The fabricated sensor showed excellent accuracy for the classification of pesticides with both pattern recognition methods. Moreover, the NPs based E-nose represented a sensitive response to pesticide vapors. Unlike the previous reports, a filter paper was used as a substrate of the sensor, which was inexpensive and available.

Conclusion
In this paper, an optical E-nose based on nanoparticles fabricating on a filter paper was developed successfully for pesticide aerosols analysis. The paper-based device was fabricated using 3D-origami pattern. In the detection spots, the NPs were formed using a low-cost chemical agent without the requirement of a biomaterial like an enzyme. The experiment results show clearly response for thion organophosphates, discriminated aliphatic, aromatic, and heterocyclic compounds from each other. The responses were not influenced by the other types of pesticides such as oxon organophosphate, carbamate, and different chemical organic compounds. The fabricated  www.nature.com/scientificreports/ sensor showed acceptable performance for both individual and simultaneous quantitative analysis. This device can be embedded in the close, mask and skin of farmer as a diagnosis kit to monitor the pesticide pollution in agriculture environments.

Method
Materials. All compounds used in this study were in the analytical grades. The studied pesticides were malathion, parathion, paraoxon, dichlorvos, trichlorfon, carbaryl, pirimicarb, carbofuran chlorpyrifos, and diazinon. These materials and other chemicals, including cysteamine (Cys), tyrosine (Tyr), and tannic acid (TA), were purchased from Sigma Aldrich. Ethanol, methanol, 1-hexanol, hexanal, heptanal, benzaldehyde, triethylamine, amylamine, benzylamine, pyridine, aniline, ammonia, benzene, toluene, p-xylene, acetic acid, isobutyric acid, phosphoric acid, pentane, hexane, heptane, dimethylphenylphosphine, silver nitrate (AgNO 3 ), gold (III) chloride trihydrate (HAuCl 4 ·3H 2 O), sodium borohydride (NaBH 4 ), boric acid, tris-hydroxymethyl methane (Tris), sodium hydroxide (NaOH), and hydrochloric acid (HCl) were obtained from Merck Chemical Company. Whatman Grade No. 2 filter paper was used as a sensor array substrate. The standard solution of pesticide with a concentration of 30.0 µg mL −1 was made in ethanol. This solution was diluted by deionized water to prepare the pesticide solution with lower concentrations. The buffer was provided by dissolving a desirable amount of Tris or boric acid in a certain volume of deionized water. The pH of the buffer was adjusted at a specified value by adding drop by drop of NaOH and HCl solution (1.0 M). Instrument and software. UV/Vis spectrophotometer (P, Model V-570) was used to record the SPR spectra of synthesized nanoparticles. The modification of nanoparticles with a capping agent was evaluated by FT-IR spectroscopy [Thermo Scientific Nicolet IR100 (Madison, WI)]. The hydrodynamic size of nanoparticles and the electrical charge of their surface were determined by Zetasizer Nano ZS90 (Malvern, UK). The immobilization process of nanoparticles on the surface of the paper was investigated using field emission scanning electron microscopy (FE-SEM; MIRA3 TESCAN) and SEM-attached energy dispersed spectroscopy (EDX). The desired pattern of the sensor was designed in AutoCAD 2016 software (https ://www.autod esk.com/produ cts/autoc ad) and printed on the piece of paper by an HP LaserJet printer 1320. Changes in the color of nanoparticles were captured by a Canon EOS 750D digital camera. Image processing was performed by Image J software (1.51n, National Institutes of Health, USA) (https ://image j.nih.gov/ij/downl oad.html). All discrimination and simultaneous quantitative analyses were done in the MATLAB R2015 scientific software (https ://www.mathw orks.com). gold NPs (AuNPs) and 3 silver NPs (AgNPs). The synthesized NPs were coated with 3 different stabilizing agents consisting of biogenic amine (cysteamine), amino acid (tyrosine), and polyphenol antioxidant (tannic acid). The procedures for the synthesis of NPs are described in the supplemental document ("Introduction"). The prepared NPs were characterized by spectrophotometric methods, and the results are represented in supporting information ("Results and discussion").

Design of paper-based Opto-E-nose.
The desired pattern for the fabricated sensor array is shown in Fig. 1a. The device was made up of two zones, which are separated by a narrow line. Each zone had a hydrophilic site surrounded by hydrophobic barriers. The proposed pattern was depicted in the AutoCAD environment. The drawn pattern was projected on the Whatman filter paper by the ink-printing method. The paper was transferred into an oven with a temperature of 200 °C for 45 min 38 . During this time, the printer ink was melt and flowed in the layer of paper, filling the holes on that 38 . This process led to an increase in the hydrophobicity of barrier parts 38 . The sampling zone was a rectangular hydrophilic space. The detection zone included six circleshaped hydrophilic sites. 1.0 µL aqueous solutions of NPs were injected into detection zones. The injection was performed by a digital micropipette. The tip of the micropipette was adjusted to the center of the zone. The solution was spread radially on the surface of the paper and took up the total space of zone. The elements of the sensor were provided by mixing the synthesized NPs with a buffer solution. The volume of NPs in the mixture, as well as type, pH, and concentration of buffer solution, was optimized. The image of the fabricated detection zone is shown in Fig. 1b. Sensing procedure. A general procedure for the colorimetric detection of pesticides is schematically shown in Scheme 1. The experiment was done in a text box, a cube with dimensions of 10 cm × 2 cm × 1.5 cm.
The designing paper analytical device (PAD) was folded such that the sampling zone completely covered all the detection spots (Scheme 1b). The PAD was embedded in the square face of the box (Scheme 1c), and a hole was drilled on the opposite side (Scheme 1d). 5.0 mL of pesticide with a certain concentration was poured into a bottle connected to a hole by a plastic tube (Scheme 1e). The analyte was sprayed to the box and the pesticide aerosols released in the box space (Scheme 1f). The aerosols were directed toward the PAD using a stream of N 2 gas (Scheme 1f). The pesticide aerosols were adsorbed by the sampling zone, immediately transferred to the detection zone, and simultaneously interacted with sensing elements (Scheme 1e). This procedure was repeatedly performed for five times. Note that, the flow of N 2 gas was used as a control experiment in this study.
Image processing and data analysis. For each experiment, the photo of the PAD was recorded by a digital camera before and after analysis. The captured photos were given to Image J software to calculate the mean of the color intensities of each sensing element. Three values corresponding to red, green, and blue color elements resulted for each color spot. Then, for each color element, the value of image before analysis was subtracted from that obtained by the image after analysis. Finally, 18 difference values (6 sensing elements × 3 color elements) were collected in a data vector. This process was conducted for both qualitative and quantitative analyses.
In the qualitative detection, the concentration of pesticides was equal to 450.0 ng mL −1 . For each pesticide, five repeated measurements were performed, and the response vectors of the sensor were assembled in a data matrix with a size of 20 × 18. Two popular discrimination methods of principal component analysis (PCA) and hierarchical clustering analysis (HCA) were used to statistically evaluate the obtained matrix and find the unique clusters for each pesticide.
The quantitative analysis was done by employing pesticides at different concentrations in the range of 0-2.0 μg mL −1 . For each concentration, a data vector was obtained. The Euclidean norm of this vector was determined as a total response of E-nose. The analytical features of the sensor were obtained by plotting the amount of Euclidean norm against the pesticide concentrations. The Euclidean norm is the total response of assay which is calculated from the following equation 37,39 : where, n is the number of sensing elements, i indicates the ith sensing elements in the array structure and ∆R, ∆G and ∆B are the difference color values of red, green and blue elements for each sensing elements, respectively.
Finally, the assay was applied to determine simultaneously four thion pesticides in a mixture. The analysis was performed by the principle of partial least square (PLS) method. The PLS model was created by a training set containing 20 standard mixture of four pesticides. The accuracy of the developed model was investigated by five individual mixtures collected in a validation set. The concentration of each pesticide in a certain mixture was set by multilevel partial factorial design reported by Brereton 36 . www.nature.com/scientificreports/ Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/.