杀螟硫磷是一种在农作物上广泛使用的有机磷杀虫剂, 常用于玉米上害虫的防治。 过量或者不合理施用导致的残留积累关系到食品安全和人体健康。 常规检测杀螟硫磷的方法有气相色谱-质谱法、 高效液相色谱法, 其准确性虽好, 但存在需要专业人员介入、 样品前处理复杂、 检测时间长等缺点。 表面增强拉曼光谱(SERS)法具有分析速度快、 检测灵敏度高和特异性好等优点, 被广泛应用于农产品中痕量残留的快速检测。 利用表面增强拉曼光谱结合化学计量学方法实现玉米中杀螟硫磷残留的准确检测。 以两步种子生长法合成的纳米金棒作为拉曼增强基底, 测量600~1 800 cm^-1范围内的拉曼光谱。 对比杀螟硫磷乙醇溶液和金棒的光谱, 确定杀螟硫磷的特征峰在650, 830, 1 082, 1 241, 1 344和1 581 cm^-1处。 采用简单预处理方法快速提取玉米中的杀螟硫磷残留。 将受污染的玉米样品粉碎后, 利用乙醇溶剂对残留进行两次提取, 每次获取的提取液经离心获得上清液, 将上清液合并混匀, 在水浴中蒸发浓缩, 浓缩后的上清液用于采集SERS光谱。 每个浓度制备50个平行样本。 各浓度残留提取液中的残留参考值采用色质联用方法测定。 对比残留提取液的光谱, 1 082, 1 241和1 581 cm^-1处特征峰强度随残留浓度的降低而迅速变弱甚至消失, 650, 830和1 344 cm^-1处的特征峰直至残留浓度为0.48 μg·mL^-1时依然可见。 当浓度低至0.37 μg·mL^-1时, 所测光谱与空白提取液光谱相似。 采用主成分分析(PCA)提取不同浓度杀螟硫磷残留光谱的主体信息, 其中残留为0.37 μg·mL^-1和空白提取液光谱的主成分得分重叠, 进而判断SERS方法对玉米中杀螟硫磷残留的检测限可达到0.48 μg·mL^-1, 低于国家规定的农作物中最大残留限, 体现出SERS检测的高灵敏性。 选取浓度为14.25 μg·mL^-1的50个样本分析其650, 830和1344 cm^-1处的特征峰强度变化可知, 所采集的光谱呈现出较好的重复性, 相对标准偏差(RSD)值仅为3.12%。 对杀螟硫磷残留的定量分析采用支持向量机回归(SVR)实现, Savitzky-Golay卷积平滑和小波变换(WT)用于本次光谱数据的预处理。 校正集和预测集样本的划分采用Kennard-Stone算法实现, 模型的性能采用校正均方根误差(RMSEC)、 校正集决定系数($R^{2}_{c}$)、 预测均方根误差(RMSEP)和预测集决定系数($R^{2}_{p}$)评估。 最优模型为SVR结合WT所构建的, 具有最小的预测误差, 其中校正集的RMSEC=0.103 2 μg·mL^-1, $R^{2}_{c}$ =0.999 74, 预测集的RMSEP=0.134 1 μg·mL^-1, $R^{2}_{p}$=0.999 60。 同时, 最优模型的预测值与色质联用法所测值基本一致, 其预测回收率为95.31%~100.66%。 以上表明, SERS结合化学计量学方法检测玉米中杀螟硫磷残留是准确可行的, 且有望推广到农作物中多种农药残留的检测, 为农产品的安全检测提供一种新思路。

Fenitrothion, an organophosphate insecticide widely appeared in agricultural crop cultivation, is commonly used to prevent and control insect pests in maize. However, excessive or unreasonable application lead to the accumulation of pesticide residues in maize, which concern to agricultural products safety and human health. The routine methods for fenitrothion detection are chromatography-mass spectrometry and high performance liquid chromatography, which are both highly accurate. Nevertheless, the shortcoming of above methods is that they need well-trained personnel, complicated sample preparation, considerable detection time. Surface-enhanced Raman spectroscopy (SERS) has the advantages of rapid speed, high sensitivity, excellent specificity, and extensively applied for rapid detection of trace residues in agricultural products. In this paper, an accurate methodology for detection of fenitrothion residues in maize was developed using surface enhanced Raman spectroscopy and chemometric methods. The gold nanorods solution synthesized by the two-step seed-mediated growth method was used as Raman active substrate. And SERS spectra of 600 to 1 800 cm^-1 were measured. Comparing SERS spectrum of ethanol solution with fenitrothion and gold nanorods, the characteristic peaks of fenitrothion were determined at 650, 830, 1 082, 1 241, 1 344 and 1 581 cm^-1. A simple pretreatment method was developed for extraction of fenitrothion residues in maize. Maize contaminated with fenitrothion was grinded, and then ethanol solution was added to extract fenitrothion residues twice. Next, the two extraction were centrifuged and the supernatant were acquired, followed by mixed, concentrated and evaporated in water bath. The concentrated supernatant was used for SERS measurement. Fifty samples were prepared for each concentration of fenitrothion residues in maize. Reference value of residue in extraction solution was detected by gas chromatography-mass spectrometer. Through observing the spectrum of maize extraction with fenitrothion residues, the characteristic peak intensity of 1 082, 1 241 and 1 581 cm^-1 were rapidly weakened or even disappeared as the fenitrothion residues decreased in different concentration residues extraction whereas the peak at 650, 830 and 1 344 cm^-1 remained visible with fenitrothion of 0.48 μg·mL^-1. Spectra of extraction with 0.37 μg·mL^-1 fenitrothion residues were basically consistent with uncontaminated samples extraction. Principal component analysis (PCA) was adopted to extract the main information of spectra of fenitrothion residues. The principal component scores for spectra of 0.37 μg·mL^-1 fenitrothion residues and uncontaminated samples were overlapped in scatter plot while others were distributed in different positions. It can be further determined from the scatter plot that the detection limit of fenitrothion in maize could reach 0.48 μg·mL^-1, which is lower than the maximum residue limit of China in crops, suggests SERS is of high sensitivity. The intensity variation of characteristic peak of 650, 830 and 1 344 cm^-1 in 50 samples with a concentration of 14.25 μg·mL^-1 fenitrothion residues were analyzed, and the collected spectra showed a good repeatability while the relative standard deviation (RSD) was only 3.12%. Support vector machine regression (SVR) was employed for quantitative analysis of fenitrothion residue. Additionally, Savitzky-Golay convolution smoothing and wavelet transform (WT)were used for the pretreatment of spectral data. The calibration and prediction set of samples were divided by Kennard-Stone algorithm. Quantitative evaluation of model performance was based on root mean square error of correction (RMSEC), coefficient of determination of correction ($R^{2}_{c}$), root mean square error of prediction (RMSEP) and coefficient of determination of prediction ($R^{2}_{p}$). Optimal regression model, which has minimal prediction error, was developed by SVR and WT. The correction set of RMSEC and $R^{2}_{c}$ were 0.103 2 μg·mL^-1 and 0.999 74 while the prediction set of RMSEP and $R^{2}_{p}$ were 0.134 1 μg·mL^-1 and 0.999 60 respectively. Furthermore, the predicted value of optimal model was basically in consonance with GC-MS, and predicted recovery of fenitrothion residues in maize was 95.31%~100.66%. Results demonstrates that SERS combined with chemometric method is feasible to detect fenitrothion residues in maize. This method is expected to be generalized to detect varieties of pesticide residues in other crops, providing a novel approach for the safety detection of agricultural products.