A simple and sensitive method for detection of chlormequat chloride residue in wheat was developed using surface-enhanced Raman spectroscopy (SERS) coupled with chemometric methods on a portable Raman spectrometer. Pretreatment of wheat samples was performed using a two-step extraction procedure. Effective and uniform active substrate (gold nanorods) was prepared and mixed with the sample extraction solution for SERS measurement. The limit of detection for chlormequat chloride in wheat extracting solutions and wheat samples was 0.25 mg/L and 0.25
Plant growth regulator can increase crop production, improve quality, and enhance stress resistance [
Chlormequat chloride detection methods as gas [
Spectroscopic methods are promising tools for detection of farm chemical residue because they are simple, rapid, specific, and partially or completely automatic. The commonly used spectroscopic methods include near-infrared spectroscopy (NIR) [
The objective of this study is to develop a simple and sensitive SERS method for quantitative determination of chlormequat chloride in wheat coupled with some chemometric methods on a portable Raman spectrometer, in which pretreatment of wheat samples is performed using a two-step extraction procedure. To the best of our knowledge, this paper is the first to report detection chlormequat chloride in grains using SERS technique.
Wheat samples were purchased from Hefei Zhougudui market. Chlormequat chloride powder (99.6%) was obtained from Beijing Puxi Technology Co., Ltd. Anhydrous methanol was acquired from Sinopharm Chemical Reagent Co., Ltd. Cetyltrimethylammonium bromide (CTAB), hydrogen tetrachloroaurate, trisodium citrate, L-ascorbic acid, sodium borohydride, and silver nitrite were purchased from Aladdin Industrial Corporation.
The pretreatment method for wheat was developed based on the extraction method in gas chromatography (GC). Wheat was first grinded using a pulverizer (Xinrui DFT-150, Changzhou, China) and filtered through 10-mesh sieves. Wheat powder of 5.00g was added with 15 mL of methanol in 50 mL centrifuge tube and then vibrated for 10 min. The mixture was centrifuged at 4000 rpm for 3 min, and the supernatant was moved to the concentrated bottle. Wheat residue was extracted using 10 mL of methanol again, and the supernatant was also moved to the concentrated bottle. The supernatant was evaporated to dry on a Rotavapor (Yarong RE-52A, Shanghai, China) and redissolved in 5 ml of methanol.
Wheat extracting solutions containing different chlormequat chloride were then prepared. The obtained extraction solution was used to dissolve chlormequat chloride powder for getting the solution of 20, 10, 5, 2.5, 1, 0.5, and 0.25 mg/L. Additionally, to simulate actual residue, wheat powder was spiked with chlormequat chloride to yield final residue at 10, 5, 2.5, 1, 0.5, and 0.25
The synthesis of gold nanorods (GNRs) was performed using a seed-mediated growth method previously developed by El-Sayed [
Absorption spectra of the GNRs were recorded on an ultraviolet-visible (UV–Vis) spectrometer (UV-2600, Shimadzu, Japan). Morphologies of the GNRs were surveyed using the scanning electron microscope (SEM) image on a JSM 7500F microscope (JEOL Ltd., Tokyo, Japan). As shown in Figure
Ultraviolet-visible absorption spectrum of the prepared GNRs colloid; the inset is SEM image of GNRs.
The obtained spectra were first baseline-corrected using asymmetric least squares method to eliminate baseline and linear slope effects [
To examine accurate and quantitative determination of analyte further, MLR, PLSR, and SVR were used to develop the regression models. MLR is a regression algorithm that is very efficient in building calibration models when the number of samples is more than that of variables. PLSR is one of the most robust and reliable tools in the development of a multivariate calibration model. Based on the linear algorithm, PLSR is often applied to predict a set of dependent variables from a large set of independent variables. PLSR decomposes the spectral array and concentration array with considering their relationships. Corresponding calculation relationships are strengthened for the better correction model. SVR is a variation of support vector machine with introduction of insensitive loss function. Despite finite sample, SVR still possesses excellent robustness and high sensitivity through balancing complexity and learning ability of model. Meanwhile, with the aid of kernel function, SVR can project data into the high-dimensional space for obtaining higher analysis accuracy. Moreover, RBF was also selected as the kernel function of SVR. Considering the performance of obtained regression models highly depends on penalty coefficient (C) in loss function and width of kernel function (
The characteristic peaks reflect the information of molecular vibration and rotation, and these peaks are the basis for analysis and detection of substance using Raman or SERS technique. To determine the characteristic peaks of chlormequat chloride, pure chlormequat chloride powder was placed on the silicon wafer, and then Raman spectra were obtained through direct laser irradiation on it. The main Raman peaks of chlormequat chloride at 666, 713, 765, 853, and 1447 cm−1 were observed in Figure
Raman spectra of pure chlormequat chloride powder.
However, the characteristic bands of SERS of molecule in complex media may have changes, which is mainly due to influence of Raman active substrate and background signals of complex media. Then, SERS spectra of GNRs, 100 mg/L of chlormequat chloride in methanol, and 20 mg/L of chlormequat chloride in wheat extraction solution were measured and shown in Figure
SERS spectra of gold nanorods (a), 100 mg/L of chlormequat chloride in methanol (b), and 20 mg/L of chlormequat chloride in wheat extraction solution (c).
Then, SERS spectra of wheat extraction solution with 20, 10, 5, 2.5, 1, 0.5, or 0.25 mg/L of chlormequat chloride were measured with the uniform GNRs (Figure
Spectra of 20, 10, 5, 2.5, 1, 0.5, or 0.25 mg/L of chlormequat chloride in wheat extraction solution.
Intelligent analysis of spectra using chemometric methods can automatically obtain the information of substances without intervention of professionals, and this process is of significance for simple and rapid detection. SERS spectra are of high dimension and carry useless information for target analyte. The appropriate variable selection and feature extraction can improve the analysis results. Considering the fingerprint properties of SERS, the spectra around characteristic peaks were selected for the intelligent analysis, and the interference can be avoided from the irrelevant information in spectra of other ranges. In particular, for SERS of chlormequat chloride in wheat extraction, the spectra of 653–683, 705–728, and 847–872 cm−1 were selected for the subsequent analysis. Then, KPCA with RBF was adopted to extract the principal feature of processed spectra. Feature extraction was highly dependent on
Predicted results of the model developed using chemometric methods.
Data | MLR | PLSR | KPCA+SVR | |
---|---|---|---|---|
RMSECV (mg/L) | RMSECV (mg/L) | | RMSECV (mg/L) | |
Spectra of 653-683, 705-728, and 847-872 cm−1 | 0.3757 | 0.3758 | 1000 | 4.235 |
5000 | 0.0299 | |||
8000 | 0.0268 | |||
10000 | 0.1131 |
Scatter plot of first two principal component scores obtained by KPCA with
Predicted error of the optimal model built using SVR and KPCA with
In addition, an unbiased estimation for generalization of the model was conducted with an independent testing set. The independent testing set was spectra of wheat extraction with 15, 8, 4, and 2 mg/L obtained through remeasurement, and the representative spectra were shown in Figure
Predicted results of 15, 8, 4, and 2 mg/L of chlormequat chloride in wheat extraction solution using SERS, SVR, and KPCA.
Spiked value ( | Mean predicted value (mg/L) | Standard deviation (mg/L) | Recovery (%) |
---|---|---|---|
15 | 14.87 | 0.066 | 99.1 |
8 | 8.18 | 0.091 | 102.3 |
4 | 3.90 | 0.052 | 97.4 |
2 | 2.21 | 0.102 | 110.3 |
Spectra of 15, 8, 4, and 2 mg/L of chlormequat chloride in wheat extraction solution (a), predicted results by using SVR and KPCA (b).
To simulate actual residue, wheat powder was spiked with chlormequat chloride to yield final residue at 10, 5, 2.5, 1, 0.5, and 0.25
Spectra of wheat samples spiked with chlormequat chloride at 10, 5, 2.5, 1, 0.5, or 0.25
Afterward, all the spectra were processed using KPCA, and the first two principal component scores were used to predict the sample concentration combining with the established model. The experiment results are shown in Table
Predicted results of chlormequat chloride in wheat using SERS, SVR, and KPCA.
Spiked value ( | Mean predicted value (mg/L) | Standard deviation (mg/L) | Recovery (%) |
---|---|---|---|
10 | 9.96 | 0.066 | 99.6 |
5 | 4.74 | 0.042 | 94.7 |
2.5 | 2.614 | 0.064 | 104.6 |
1 | 1.012 | 0.014 | 101.2 |
0.5 | 0.479 | 0.007 | 95.8 |
0.25 | 0.242 | 0.025 | 96.8 |
In this work, a method for detection of chlormequat chloride in wheat was developed using SERS and chemometric methods on a portable Raman spectrometer. The extraction of residue in wheat was performed using a two-step procedure originated from GC detection. As for the spiked wheat samples, the optimal predicted recovery was in the range of 94.7 % to 104.6 %, and standard deviation was from 0.007 mg/L to 0.066 mg/L. These results indicated that the present method is an effective and feasible approach for determination of chlormequat chloride residue in wheat. Meanwhile, with aid of a portable Raman spectrometer, the present method could be executed onsite, which is suitable for rapid residue analysis in grains. However, spectral variation induced by instability of substrate and differences in sample pretreatment should be avoided and resolved prior to application of SERS. In conclusion, SERS with chemometric methods is a potentially powerful approach for detecting chlormequat chloride or other toxic residues in grains which can greatly help improve the safety and quality of agricultural products.
The data used in the article can be downloaded and viewed at the following address:
This article does not contain any studies with human or animal subjects.
The authors declare that they have no conflicts of interest.
This study is supported by Natural Science Foundation of Anhui Province (nos. 1708085QF134 and 1604a0702016), Natural Science Research Project of Anhui Provincial Education Department (no. KJ2017A006), National Natural Science Foundation of China (nos. 31401285 and 61475163), National Key Research and Development Program (no. 4014YFD0800904), Anhui Provincial Science and Technology Project (no. 17030710162), and Open Foundation of Science and Technology on Communication Networks Laboratory (no. XX17641X011-02).